CN118522438B

CN118522438B - Health monitoring method integrating multi-mode biological information

Info

Publication number: CN118522438B
Application number: CN202410584179.3A
Authority: CN
Inventors: 谭丽; 吴树昱; 赵浩东; 曹子健
Original assignee: Shenzhen Ruier Kangyang Technology Co ltd
Current assignee: Shenzhen Ruier Kangyang Technology Co ltd
Priority date: 2024-05-11
Filing date: 2024-05-11
Publication date: 2025-04-08
Anticipated expiration: 2044-05-11
Also published as: CN118522438A

Abstract

The application provides a health monitoring method of integrated multi-mode biological information, which comprises the steps of deploying a multi-mode sensor to collect actual biological signals of a user, including heart rate, blood pressure and body temperature, of the user when the user wears the wearable device for the first time, recording tightness levels and position information of the wearable device, collecting tightness experiences of the user on different position information of the wearable device and corresponding tightness experiences of different tightness levels, correlating the tightness experiences with the tightness levels and the position information, calculating monitoring precision of the wearable device according to the actual biological signals and data of the wearable device, combining the tightness experiences of the current user with the tightness levels and the position information, establishing a regression analysis model for describing association relation between the different tightness levels and the position information and the monitoring precision of the wearable device, combining the minimum monitoring precision corresponding to a target index, determining a recommended wearing position and the recommended tightness level through the regression model, and dynamically adjusting the recommended wearing position and the recommended tightness through real-time monitoring feedback information of the user.

Description

Health monitoring method integrating multi-mode biological information

Technical Field

The invention relates to the technical field of information, in particular to a health monitoring method integrating multi-mode biological information.

Background

In bio-signal monitoring of wearable devices, the wearing position and tightness of the device and the bio-signal accuracy have a crucial influence. However, in practical applications, there is often a conflict between the personal tolerance of the user and the minimum accuracy requirements required for the condition. In order to solve the technical problem, a large amount of personal bearing tightness data of users are firstly required to be collected, and subjective feelings and physiological responses of the users under different wearing positions and tightness are obtained through questionnaires, physiological tests and other modes. And determining the minimum accuracy requirement of biological signal monitoring according to the characteristics of different diseases. In the actual use process, the situation that the personal bearing tightness of the user cannot meet the minimum precision requirement of the symptoms may occur. At this time, on the premise of ensuring the comfort of the user, the acquisition precision of the biological signals is improved by optimizing the wearing position, improving the sensor design and other technical means. Meanwhile, physiological differences and wearing habits of different users are considered, so that the requirements of different users are met. In addition, how to improve the precision of biological signals and simultaneously consider the comfort, portability and sustainability of equipment is also a problem worth deeply discussing.

Disclosure of Invention

The invention provides a health monitoring method integrating multi-mode biological information, which mainly comprises the following steps:

when a user wears the wearing equipment for the first time, deploying a multi-mode sensor to acquire actual biological signals of the user, including heart rate, blood pressure and body temperature, and recording tightness level and position information of the wearing equipment;

collecting tightness feeling of a user corresponding to different tightness grades on the wearable equipment, and associating the tightness feeling with the tightness grade and the position information;

Calculating the monitoring precision of the wearable device according to the actual biological signals and the data of the wearable device, and establishing a regression analysis model by combining the tightness feeling of the current user and the tightness level and the position information to describe the association relation between different tightness levels and position information and the monitoring precision of the wearable device;

Receiving target indexes and personal tightness thresholds corresponding to wearing equipment input by a user, determining a personal tightness level range by combining the personal tightness thresholds and optimal tightness levels corresponding to optimal tightness feeling of the user, and evaluating the matching degree of the wearing equipment to the user based on a preset adjustable range of the wearing equipment;

Determining the minimum monitoring precision according to the target index, determining the highest tightness grade corresponding to the minimum monitoring precision through a regression model, identifying the user as a tolerant user if the highest tightness grade is within the personal tightness grade range, and identifying the user as a sensitive user if the highest tightness grade is outside the personal tightness grade range;

Determining a recommended wearing position and a recommended tightness level through a regression model by combining the minimum monitoring precision corresponding to the target index, and dynamically adjusting the recommended wearing position and the recommended tightness through monitoring feedback information of a user in real time;

analyzing user personalized data by using a long-short-term memory neural network, constructing a model to predict the accuracy of biological signals under different use conditions, comparing a model prediction result with actual acquired data, calculating a mean square error, and dynamically adjusting a wearing strategy according to the error until the error converges to an acceptable range;

The technical scheme provided by the embodiment of the invention can have the following beneficial effects:

The invention discloses a health monitoring method integrating multi-mode biological information, which relates to the optimization of accurate monitoring and personalized wearing experience of a user biological signal through an integrated multi-mode sensor and an intelligent data processing technology. The invention focuses on the influence of the contact tightness and the position of the wearing equipment and the body of the user on the monitoring precision, and how to adjust the equipment according to the personal feeling of the user so as to achieve the optimal wearing effect and data acquisition quality.

In general, the application and monitoring precision of the wearable equipment are obviously improved through the highly personalized data processing and intelligent adjustment mechanism, the use experience of the user is improved, and the method has important practical value and wide market prospect.

Drawings

Fig. 1 is a flow chart of a health monitoring method integrating multi-mode biological information according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.

As shown in fig. 1, a health monitoring method for integrating multi-mode biological information in this embodiment may specifically include:

Step S101, when a user wears the wearable device for the first time, deploying a multi-mode sensor to acquire actual biological signals of the user, including heart rate, blood pressure and body temperature, and recording tightness level and position information of the wearable device.

And acquiring personal basic information of the user, including gender, age, height, weight and the like, and primarily estimating the physiological parameter range of the user according to the information to provide a reference standard for the actual biological signals acquired subsequently. According to the body size of the user and the structural characteristics of the wearing equipment, the optimal tightness level and the wearing position are calculated through simulation, the optimal tightness level and the wearing position are combined to obtain initial sensor wearing parameters, the user is guided to wear correctly, the close fit of the sensor and the skin is ensured, and the accuracy of the acquired signals is improved. In the process of wearing equipment by a user, detecting the motion state of the user in real time, identifying and screening out a relatively static time period, collecting biological signals such as heart rate, blood pressure and body temperature in the time period, extracting characteristic parameters of the signals by using a wavelet analysis and other frequency domain analysis method, analyzing by using a clustering algorithm, obtaining basic physiological indexes under the static state of the user, and establishing a personal health baseline. The method comprises the steps of continuously recording the change condition of biological signals of a user when the user wears equipment at different tightness levels and positions, extracting statistical characteristics and spectrum characteristics of the signals by adopting methods such as time domain analysis and frequency domain analysis, calculating indexes such as mean value, variance, correlation coefficient and the like of characteristic parameters, judging the effectiveness and reliability of acquired data by setting reasonable thresholds, dynamically adjusting working parameters of a sensor according to the effectiveness and reliability, detecting the tightness degree and offset condition of the wearing equipment in real time, and reminding the user to readjust in time if the monitored data exceeds the threshold range so as to ensure the acquisition quality of the biological signals. Aiming at the physiological characteristics of different user groups, such as the elderly, children, chronic patients and the like, a personalized multi-mode sensor combination scheme is adopted, the types, the numbers and the arrangement positions of the sensors are reasonably selected, the comprehensiveness and the comfortableness of monitoring are considered, and the monitoring requirements of different groups are met. Binding each item of biological signal data acquired when a user wears the personal information for the first time with personal basic information, carrying out data desensitization, encryption and other processing, uploading to a cloud platform through a secure channel, ensuring the integrity and non-tamper property of the data by using a blockchain technology, strictly limiting the access rights of the data, and protecting the privacy of the user. And the cloud platform compares and analyzes the received data with health indexes in the medical large database, evaluates the overall health level of the user and forms a personalized health report. And (3) taking the baseline data of the user biological signals obtained in the initial wearing stage as a reference standard for subsequent monitoring, comparing the difference between the current biological signals of the user and the baseline in real time, timely finding out abnormal changes of the physiological state of the user through an abnormal detection algorithm, triggering an early warning mechanism, prompting the user to seek medical attention or adjust the life style in time, and realizing dynamic management and risk prediction of the health state of the user.

Specifically, when basic information of a user is obtained, the front photo uploaded by the user can be analyzed through an image recognition algorithm, and parameters such as height and weight of the user are estimated. If the convolutional neural network is used for dividing the image, the human body contour features are extracted, and then a pre-trained regression model is combined, and taking height as an example, the estimated error can be controlled within +/-2 cm. According to the sex, age and estimated height and weight of the user, the normal range of physiological parameters of the user can be primarily judged, for example, the normal heart rate of an adult male is 60-100 times per minute, the systolic blood pressure is 90-140mmHg and the like. Meanwhile, finite element simulation software is used, a three-dimensional human body model is built according to the body size of a user, different tightness levels and sensor arrangement schemes are set, and optimal sensor pressure distribution and wearing positions are obtained through repeated iterative computation, for example, a photoelectric volume pulse wave sensor is worn on the radial artery of the wrist, and the pressure is set to be 20-40mmHg. During the process of wearing the equipment by the user, the motion data of the user are collected through the accelerometer and the gyroscope, the motion state of the user is classified by using a Support Vector Machine (SVM) algorithm, and when the user is detected to be in a static state (such as sitting posture and prone posture) for at least 5 minutes, a biological signal collection program is triggered. The acquired signals of electrocardio, pulse wave and the like are preprocessed firstly, including power frequency interference, baseline drift and the like are removed, then time domain features such as heart rate, pulse wave peak intervals and the like are extracted, and frequency domain features such as low frequency/high frequency power ratio (LF/HF) are analyzed through a density clustering algorithm (DBSCAN), so that a normal physiological index range of a user in a static state is obtained. In the continuous monitoring process, the multiscale energy distribution characteristics of biological signals are extracted through a wavelet transformation equal-frequency analysis method, the energy proportions of different frequency bands are calculated, an abnormal threshold (the energy mutation is more than 20%) is set, and the reliability of the signals is judged in real time. For example, 100 heart rate data samples of a user in a static state in one day are analyzed, after the highest and lowest 5% outliers are removed, an average heart rate of 65 times/min and a standard deviation of 3 times/min are calculated and used as a healthy baseline for the heart rate of the user. The method comprises the steps of continuously recording the change condition of biological signals of a user when the user wears equipment at different tightness levels and positions, extracting statistical characteristics and spectrum characteristics of the signals by adopting methods such as time domain analysis and frequency domain analysis, calculating indexes such as mean value, variance, correlation coefficient and the like of characteristic parameters, judging the effectiveness and reliability of acquired data by setting reasonable thresholds, dynamically adjusting working parameters of a sensor according to the effectiveness and reliability, detecting the tightness degree and offset condition of the wearing equipment in real time, and reminding the user to readjust in time if the monitored data exceeds the threshold range so as to ensure the acquisition quality of the biological signals. If the signal-to-noise ratio of the electrocardiosignal of the user is detected to be lower than 20dB or the waveform distortion degree is higher than 10%, the sampling frequency and the gain of the sensor are timely adjusted, and the user is prompted to check the adhesion firmness of the electrode plate. Aiming at the physiological characteristics of different user groups, such as the elderly, children, chronic patients and the like, a personalized multi-mode sensor combination scheme is adopted, the types, the numbers and the arrangement positions of the sensors are reasonably selected, the comprehensiveness and the comfortableness of monitoring are considered, and the monitoring requirements of different groups are met. For the elderly people who need to pay attention to heart health, chest strap type sensors integrating multiple parameters such as electrocardio, blood pressure, blood oxygen and the like can be selected, and measurement accuracy and wearing comfort are considered. Binding each item of biological signal data acquired when a user wears the personal information for the first time with personal basic information, carrying out data desensitization, encryption and other processing, uploading to a cloud platform through a secure channel, ensuring the integrity and non-tamper property of the data by using a blockchain technology, strictly limiting the access rights of the data, and protecting the privacy of the user. If the identity information of the user is hashed, the uploaded physiological data is encrypted by adopting an asymmetric encryption algorithm, and a white list and a multiple identity verification mechanism of data access are set. And the cloud platform compares and analyzes the received data with health indexes in the medical large database, evaluates the overall health level of the user and forms a personalized health report. If the data such as heart rate, blood pressure and blood sugar of the user are compared with the statistical indexes of the people with the same gender and the same age, the percentile of the user on the corresponding indexes is given, and the health risk of the user is evaluated and early warned according to the medical expert knowledge base. And (3) taking the baseline data of the user biological signals obtained in the initial wearing stage as a reference standard for subsequent monitoring, comparing the difference between the current biological signals of the user and the baseline in real time, timely finding out abnormal changes of the physiological state of the user through an abnormal detection algorithm, triggering an early warning mechanism, prompting the user to seek medical attention or adjust the life style in time, and realizing dynamic management and risk prediction of the health state of the user. If the night average heart rate of the user is monitored to be increased by more than 20% from the baseline value in 3 continuous days, early warning information is timely pushed, the user is recommended to pay attention to rest, and medical examination is carried out if necessary.

Step S102, tightness feeling of the user corresponding to different position information and different tightness grades of the wearable device is collected, and the tightness feeling and the tightness grade are associated with the position information.

The method comprises the steps of carrying out subjective evaluation on wearing comfort level of a user by using a Likett meter in an online questionnaire investigation mode, classifying tightness experience into 5 grades, and respectively corresponding to very comfort, general discomfort and very discomfort to obtain subjective tightness experience data of the user under different tightness grades. Meanwhile, pressure distribution data of contact parts of wearing equipment and skin of a user under different tightness grades are collected by adopting a pressure sensor array, subjective tightness feeling scores and objective pressure distribution measurement results are aligned and calibrated by preprocessing methods such as data normalization and feature mapping, a quantitative relation between the subjective tightness feeling scores and the objective pressure distribution measurement results is established, a quantitative relation model of the tightness grade and the pressure distribution is established by using a Gaussian process regression algorithm, and objective measurement of the tightness grade is realized. The human body three-dimensional scanning technology is utilized to acquire body surface contour data of a user, and the three-dimensional structure model of the wearable device is combined, so that the contact area and positive pressure distribution of different wearing positions are calculated through finite element simulation analysis, and the theoretical tightness grade of different positions is obtained. The tightness feeling feedback of users with different ages, sexes and body types is collected, the user attribute with the largest tightness feeling relativity is selected by adopting a characteristic selection algorithm, a decision tree model is constructed, the optimal splitting attribute is selected by indexes such as information gain and the like, a tree structure is generated, and the decision tree is optimized by utilizing a post pruning algorithm, so that the generalization capability is improved. On leaf nodes of the decision tree, adopting an Apriori algorithm to mine association rules of different user attribute combinations and tightness feeling, and establishing personalized tightness-feeling corresponding models for different user groups. Developing tightness feedback APP, after wearing equipment at different tightness levels and positions, a user inputs real-time subjective tightness feeling through the APP, and accumulating a large amount of user feedback data by utilizing crowdsourcing. And designing a data quality evaluation and excitation feedback mechanism, carrying out preprocessing such as outlier detection, noise filtering and the like on feedback data uploaded by a user, eliminating false or invalid feedback, and improving the reliability of the data. And a certain point reward is given to the user with timely and accurate feedback, so that the enthusiasm of the user for participation is improved. And summarizing multi-dimensional data such as subjective tightness feeling, objective pressure distribution, theoretical tightness level, user attributes and the like of a user, and representing and standardizing input features by adopting methods such as one-hot coding, numerical normalization and the like. the LSTM network structure is designed to comprise an input layer, an embedded layer, an LSTM layer and a full connection layer, and super parameters of the network, such as the hidden layer number, the neuron number, the dropout rate and the like, are optimized through a grid search method and the like. Taking cross entropy as a loss function, performing model training by using an Adam optimization algorithm, performing performance evaluation and model tuning on a verification set, and finally forming an end-to-end mapping relation from tightness grade and position information to tightness feeling. Based on the established tightness feeling prediction model, model reasoning is carried out through a forward propagation algorithm according to information such as personal attributes, expected wearing positions and the like input by a new user, and expected tightness feeling distribution under different tightness grades is calculated. By combining a human body three-dimensional model and a wearing equipment virtual assembly technology, wearing effect preview images under different tightness levels are generated by using rendering engines such as Unity 3D and the like, and are intuitively displayed for a user, so that the user is assisted in selecting the optimal tightness level and wearing position, and the user is guided to wear comfortably and effectively.

Specifically, in the on-line questionnaire survey, the tightness feeling is divided into 1 to 5 minutes by adopting a 5-point Likett scale, and the two points correspond to very uncomfortable to very comfortable respectively. Meanwhile, pressure distribution data under different tightness grades are acquired by using a pressure sensor array, pressure values are normalized to be between 0 and 1, and pearson correlation analysis is carried out on the pressure values and subjective scores, so that a correlation coefficient of 0.8 is obtained, and a strong linear relationship between the pressure values and the subjective scores is shown. And a regression model is established by using a Gaussian process regression algorithm and the tightness level as an independent variable and the pressure distribution characteristic as a covariant, and the super-parameters are optimized through maximum likelihood estimation, so that the regression mean square error is 0.1, and the fitting goodness is higher. And (3) carrying out gridding treatment on the three-dimensional body surface contour data of the user to generate a triangular surface patch model containing 10000 nodes, calculating contact pressure distribution at different positions through finite element simulation, and mapping with tightness grades to obtain optimal tightness intervals at different positions. In the construction process of the decision tree model, optimal splitting attributes such as age, sex and the like are selected through the information gain ratio, a tree structure is generated in a recursion mode, the minimum number of samples of leaf nodes is set to be 10, and overfitting is prevented. Model evaluation is carried out by using 10-fold cross validation, and the average accuracy reaches 85%. On leaf nodes, using an Apriori algorithm to mine association rules of user attribute combination and tightness feeling, setting minimum support degree to be 0.05 and minimum confidence degree to be 0.8, and obtaining association rules of Top5, such as 'female and BMI < 18.5= > tightness feeling slightly uncomfortable'. In preprocessing of user feedback data, abnormal value detection is carried out by adopting a Z-score method, and data exceeding 3 times of standard deviation are removed. And (3) carrying out smooth denoising on the data by using a Kalman filtering algorithm, setting a forgetting factor to be 0.9, and dynamically updating noise estimation. In the design of the LSTM network structure, the dimension of an input layer is 10, the dimension of an embedded layer is 32 corresponding to the characteristics of tightness level, position coordinates and the like, the LSTM layer comprises 2 hidden layers, 64 neurons in each layer, the dropout rate is set to be 0.5, and the full-connection layer outputs tightness feeling prediction values. With Adam optimizer, the initial learning rate is 0.001, the momentum factor is 0.9, and L2 regularization is used to prevent overfitting. The average absolute error of the model on the verification set is 0.3, and the accuracy reaches 90%. According to personal attributes and wearing positions of users, tightness feeling distribution under different tightness grades is predicted through model reasoning, virtual wearing effects generated by rendering of a Unity 3D engine are combined, and the tightness grade with the highest expected tightness feeling value and meeting the comfort requirement of the users is selected to generate a personalized wearing scheme.

Step S103, calculating the monitoring precision of the wearable device according to the actual biological signals and the data of the wearable device, and establishing a regression analysis model by combining the tightness feeling and the tightness level and the position information of the current user so as to describe the association relation between different tightness levels and position information and the monitoring precision of the wearable device.

Physiological signals such as electrocardio, pulse and blood oxygen of a user are acquired in real time by adopting a multi-mode sensor, motion data such as acceleration, angular speed and the like of a wearable device are synchronously recorded, time alignment and sampling rate conversion are carried out on heterogeneous data from different sources through data preprocessing, key characteristics such as time domain, frequency domain and the like are extracted, and fusion and characteristic representation of the multi-mode data are realized. And by utilizing a deep learning model such as a convolutional neural network and the like, quality evaluation and anomaly detection are carried out on the physiological signals, noise, distortion and invalid fragments in the signals are automatically identified, and the reliability and the effectiveness of the data are improved. And dynamically screening out high-quality physiological signal fragments according to the signal quality evaluation result, comparing the high-quality physiological signal fragments with the monitoring result of the reference medical equipment, and calculating measurement errors and deviations of the wearable equipment on different physiological indexes to obtain quantitative evaluation of monitoring accuracy. Subjective comfort scores of the user at different tightness levels and wearing positions are collected by adopting the modes of questionnaire investigation, man-machine interaction and the like, and the corresponding relation between tightness feeling and physiological parameters of the user is established. Through feature selection algorithms, such as a filtering method based on mutual information, an embedding method based on L1 regularization and the like, the correlation and importance among factors such as tightness level, position information, user tightness feeling and the like and monitoring precision are evaluated, key influence factors are screened out, and the dimension of a feature space is reduced. For the selected continuous characteristics, the data scaling and distribution conversion are carried out by adopting methods such as minimum-maximum normalization and Z-score normalization, and for the discrete characteristics, the quantitative representation of qualitative indexes is realized by adopting methods such as single-heat coding and serial number coding. And constructing a multiple regression analysis model by using the monitoring precision as a dependent variable and key factors such as tightness level, position information, user tightness feeling and the like as independent variables and using a machine learning algorithm. Before modeling, multiple collinearity problems are diagnosed by calculating indexes such as a pearson correlation coefficient matrix, a variance expansion factor and the like among the features, and redundant features with high correlation are combined or eliminated. Regression coefficients are estimated by a least square method, and weights and contribution degrees of independent variables are estimated through hypothesis testing and significance analysis. And describing interaction effect and nonlinear relation between independent variables by taking the introduction of second-order, third-order and other higher-order items into consideration. For complex nonlinear modes, kernel function methods such as polynomial kernels, gaussian kernels and the like are adopted to map original features to a high-dimensional space, so that the expression capacity of the model is enhanced. Through cross verification and grid search, super parameters of the model, such as regularization coefficients, kernel function parameters and the like, are optimized, and generalization performance and robustness of the model are improved. Based on the regression model obtained through training, the monitoring precision change trend and sensitivity under different tightness grades, position information and user tightness feeling are analyzed, the optimal wearing parameter combination is predicted, and the monitoring quality of physiological signals is maximized while the user comfort level is ensured. And a resampling method such as Bootstrap is adopted to evaluate the confidence interval and uncertainty of model prediction, so that a statistical basis is provided for reliability estimation of monitoring precision. Through dynamic updating and online learning of the model, parameters and structures of the regression model are continuously optimized, the method is suitable for changes of user states and wearing environments, and self-adaptive adjustment and dynamic prediction of monitoring precision are realized.

Specifically, when the data of the multi-mode sensor are fused, a Dynamic Time Warping (DTW) algorithm is adopted to time align physiological signals with different sampling rates such as electrocardio, pulse and the like, and a mapping function of the minimum accumulated distance is found through optimizing path searching, so that the synchronization and registration of the signals are realized. Extracting time domain features such as average value, standard deviation, peak-to-peak value, etc., and frequency domain features such as power spectrum density, band energy ratio, etc., and constructing 20-dimensional feature vector. The convolutional neural network is adopted for signal quality evaluation, 1000 labeling samples are used for training, the signals comprise 500 high-quality signals and 500 noisy signals, and the average accuracy of the model reaches 95% through 5-fold cross verification. During monitoring accuracy calculation, selecting high-quality signal segments for 5 minutes continuously, and carrying out Root Mean Square Error (RMSE) analysis on the high-quality signal segments and the measurement results of medical reference equipment to obtain the RMSE of indexes such as heart rate, blood oxygen saturation and the like of 2 times/minute and 1 percent respectively. For quantitative representation of tightness feeling, a fuzzy comprehensive evaluation method is adopted to convert 5 language grades from loose to tight into membership values of a [0,1] interval, and comprehensive scores of tightness feeling are obtained through weighted average. During feature selection, a maximum correlation minimum redundancy (mRMR) algorithm is adopted, and 10 optimal feature subsets are selected through correlation between features and monitoring accuracy and redundancy among the features, so that feature dimensions are reduced by 90%. When the regression model is constructed, a second-order polynomial kernel function is adopted, the regression coefficients of the tightness level, the position information and the tightness feeling are respectively 0.5, 0.3 and-0.2 through maximum likelihood estimation, and the P values are smaller than 0.05 through hypothesis test, so that the method has significance. By adopting 10-fold cross validation, the determination coefficient R ² of the model is 0.85, and the mean square error is 0.1, which shows that the model has high fitting goodness and strong generalization capability.

Step S104, receiving target indexes and personal tightness thresholds corresponding to the wearing equipment input by the user, determining a personal tightness level range by combining the personal tightness thresholds and the optimal tightness level corresponding to the optimal tightness feeling of the user, and evaluating the matching degree of the wearing equipment to the user based on a preset adjustable range of the wearing equipment.

Through a human-computer interaction interface, functional requirements of a user on the wearable equipment and expected physiological monitoring indexes such as heart rate, blood oxygen, respiratory rate and the like are obtained, meanwhile, the user is allowed to customize a comfort level range, and acceptable minimum and maximum tightness values are input to form a personal tightness threshold. A fuzzy comprehensive evaluation method is adopted to construct a multi-level fuzzy evaluation matrix with tightness feeling as a target layer, physiological parameters, physical pressure, wearing parts and the like as standard layers and different tightness grades as scheme layers. And determining the weight coefficient and membership function of each evaluation index through an expert knowledge base and user investigation data, and carrying out multilevel fuzzy synthesis operation by adopting a weighted average method to obtain the quantized mapping relation between the subjective tightness feeling and the objective tightness grade of the user. And analyzing tightness feeling feedback data of the existing users by using a k-means clustering algorithm, determining an optimal clustering number k by using an elbow rule and contour coefficient evaluation, and iteratively optimizing a clustering center by using Euclidean distance to measure similarity between tightness feeling feature vectors to obtain optimal tightness feeling modes of different crowds. An Apriori association rule mining algorithm is adopted, crowd clustering labels are consequent, individual attributes such as age, gender and body type are antecedent, minimum support and confidence threshold values are set, significant association rules are extracted, and recommendation mapping of user portraits and optimal tightness levels is formed. and in the process of determining the personalized tightness level, adopting a decision fusion strategy of weighted voting. The method comprises the steps of giving a higher weight coefficient to a user self-defined tightness threshold value, reflecting the priority of subjective demands of the user, giving a next highest weight coefficient to the optimal tightness level based on crowd image recommendation, and optimizing objective monitoring performance. And (3) carrying out weighted fusion on the two tightness grades by setting a weighted weight vector and a threshold value to obtain a personalized tightness grade interval of the user. By referring to patent documents, consultative material suppliers and the like, mechanical performance parameters of TPU, silicone rubber and other materials used by the wearable equipment, such as Young modulus, poisson ratio and the like, are obtained. And (3) establishing a tightness level prediction model based on material mechanics by combining structural parameters in a product design drawing, such as bandwidth, thickness and the like. And adopting a finite element simulation method to carry out numerical simulation on stress deformation of the wearable equipment under different tightness grades, analyzing stress strain distribution characteristics, and evaluating whether the mechanical properties of the material meet the tightness grade requirements of users. Designing a matching degree evaluation function, comprehensively considering a personal tightness threshold value, an optimal tightness level of a user, and the material performance and the adjustable range of wearable equipment, and constructing a multi-criterion attribute decision matrix. And calculating Euclidean distances between each candidate wearable device and the ideal optimal solution and between each candidate wearable device and the negative ideal solution by adopting a weighted normalization TOPSIS (Technique for Order Preference by Similarity to an Ideal Solution) method, and obtaining a closeness coefficient as a matching degree quantization index. the higher the matching degree is, the higher the matching degree between the wearable equipment and the user needs is. Meanwhile, a user feedback mechanism is introduced, satisfaction degree scores of users on recommendation results are collected, and a collaborative filtering algorithm is adopted to update and optimize the matching degree evaluation model on line.

Specifically, in the man-machine interaction interface, through controls such as a sliding bar and a digital input box, a user is enabled to input target indexes such as expected heart rate monitoring precision (such as + -5 times/min), blood oxygen saturation error (such as + -2%), and acceptable tightness threshold (such as 0.1-0.5 newton). In the fuzzy comprehensive evaluation process, a fuzzy evaluation matrix comprising 10 criterion layer indexes and 5 tightness levels is constructed through expert scoring and user questionnaire investigation, a triangular membership function is adopted, the comprehensive evaluation value of tightness feeling is calculated through a weighted average operator, and pearson correlation analysis is carried out on the comprehensive evaluation value and the actually measured physiological parameter, so that a correlation coefficient r=0.8 is obtained, and the subjective and objective evaluation result has good consistency. In the clustering analysis, min-max normalization is carried out on tightness feeling data of 1000 users, an optimal clustering number k=4 is determined through an average contour coefficient and Calinski-Harabaz indexes, an initial clustering center selection is carried out by adopting a k-means++ algorithm, 4 user groups are obtained through iteration 20 times of convergence, association rule mining is carried out by adopting an Apriori algorithm, the minimum support degree is set to be 0.1, the minimum confidence degree is set to be 0.8, and 10 obvious association rules are obtained, such as 'age >60 years old, BMI < 18.5- > optimal tightness grade is 2'. And when the personalized tightness level is determined, setting the user-defined threshold weight to be 0.6, and the crowd recommended tightness weight to be 0.4, and obtaining tightness level intervals [2,4] of the user through weighted average. And when the material mechanics is modeled, the strain distribution cloud picture of the TPU material under the pressure of 0.1-0.5 newton is obtained through finite element simulation analysis, the strain peak value is less than 5%, and the requirement of the user tightness grade is met. During matching degree evaluation, 5 alternative sport bracelets are selected, a weighted normalization matrix is constructed from 6 dimensions such as performance, comfort level and appearance, euclidean distance between each alternative bracelet and an ideal solution is calculated through a TOPSIS algorithm, a matching degree coefficient is obtained, the highest matching degree of the bracelet A is 0.85, and the bracelet A is recommended to a user for try-on and purchase. Meanwhile, 5-star satisfaction degree scores of the user on the recommendation result are collected and used as explicit feedback data of a collaborative filtering algorithm, the matching degree assessment model is updated on line, and accuracy and individuation degree of follow-up recommendation are improved.

Step S105, determining the lowest monitoring precision according to the target index, determining the highest tightness grade corresponding to the lowest monitoring precision through a regression model, identifying the user as a tolerant user if the highest tightness grade is within the personal tightness grade range, and identifying the user as a sensitive user if the highest tightness grade is outside the personal tightness grade range.

According to application scenes and function positioning of the wearable equipment, such as sports health monitoring, sleep quality assessment and the like, the medical significance of physiological parameters, technical characteristics of the wearable sensor and other factors are comprehensively considered, and target indexes of monitoring of various physiological parameters are determined. Physiological parameter monitoring data of different subjects under different tightness levels, materials, wearing positions and other conditions are acquired through experiments, preprocessing operations such as cleaning, outlier processing and normalization are performed on the data, and a multiple linear regression model taking the tightness levels, the material characteristics, the wearing positions and the like of wearing equipment as independent variables and monitoring errors of the physiological parameters as dependent variables is constructed. And (3) adopting a stepwise regression method to select features, optimizing model parameters through cross verification and regularization technology, and adopting F test and t test to perform significance analysis on regression coefficients to obtain quantitative relation among tightness level, material characteristics, wearing position and monitoring precision. Substituting the monitoring target index into a regression equation, calculating the minimum tightness level required by realizing the target precision, comprehensively considering factors such as the adjustment precision of the tightness level, the individual difference of physiological parameters and the like, determining reasonable tightness level margin through sensitivity analysis and Monte Carlo simulation experiments, and finally obtaining the highest tightness level threshold for guaranteeing the monitoring precision. Personal attribute information such as age, sex, height weight and the like filled in by a user during registration and subjective tightness experience scores fed back during initial wearing are obtained, and feature engineering and standardization processing are carried out on user data. And (3) clustering the users by adopting a k-means clustering algorithm, and determining the optimal clustering number through an elbow rule and a contour coefficient to obtain user subgroups with different tightness feeling characteristics. And establishing a classification rule between the user attribute and the clustering result by using a C4.5 decision tree algorithm, and mining the association modes of attribute combinations such as different ages, sexes, BMI and the like and the optimal tightness level interval to form a personalized user tightness feeling portrait. Comparing the highest tightness level threshold with the personal tightness level interval of the user, marking the user as tolerant if the highest threshold falls in the personal interval, which means that the user can adapt to higher wearing tightness to achieve better monitoring precision, and identifying the user as sensitive if the highest threshold exceeds the upper limit of the personal interval, which means that the user is sensitive to wearing pressure and needs to ensure wearing comfort preferentially. And according to the field expert knowledge and the historical user feedback data, performing experience setting and iterative optimization on the tightness grade deviation threshold value, and dynamically adjusting the boundary conditions of user classification. For a tolerant user, according to the personal optimal tightness level interval and actual wearing habit data, a fuzzy reasoning and reinforcement learning algorithm is adopted to give out targeted tightness level adjustment suggestions, and the user is guided to properly improve tightness on the premise of ensuring wearing comfort so as to meet the requirement of monitoring precision. For a sensitive user, recommending a lower tightness wearing scheme corresponding to the lower limit of the personal tightness level interval, providing suggestions for wearing position selection, wearing duration control and the like for relieving uncomfortable feeling of pressure, prompting the user that the monitoring effect is possibly reduced under the condition of excessively low tightness, and helping the user to balance and choose between comfortableness and functionality.

Specifically, physiological parameter monitoring data of different subjects under different tightness levels, materials, wearing positions and other conditions are collected through experiments, pretreatment operations such as cleaning, outlier processing and normalization are carried out on the data, and a multiple linear regression model with the tightness levels, the material characteristics, the wearing positions and the like of wearing equipment as independent variables and monitoring errors of the physiological parameters as dependent variables is constructed. For example, 30 subjects were selected, each with heart rate and blood oxygen saturation monitoring at 3 different levels of tightness (relaxed, moderate, tight), 2 materials (nylon, silicone), 3 wearing positions (wrist, upper arm, ankle), and repeated 3 times under each combination condition, obtaining 30×3×2×3×3=1620 sets of data. Feature selection is carried out by adopting a stepwise regression method, model parameters are optimized through 5-fold cross validation and L1 regularization technology, F test and t test are used for carrying out significance analysis (significance level alpha=05) on regression coefficients, quantitative relations among tightness grades, material characteristics, wearing positions and monitoring precision are obtained, for example, heart rate monitoring errors=5×tightness grades-2×material hardness +8×wearing positions +6 blood oxygen saturation monitoring errors=1×tightness grades-9×material air permeability +1×wearing positions +5 are substituted into regression equations, minimum tightness grades required for achieving target precision are calculated, and then adjusting precision (for example, each grade interval 5) of tightness grades is comprehensively considered, And determining that the reasonable tightness level margin is 5 by sensitivity analysis and 1000 Monte Carlo simulation experiments according to factors such as individual differences (such as standard deviation 2 times/min) of physiological parameters and the like, and finally obtaining the highest tightness level threshold for guaranteeing the monitoring precision, wherein the highest tightness level threshold is not more than 3 levels for heart rate monitoring and not more than 4 levels for blood oxygen monitoring. Personal attribute information such as age, sex, height weight and the like of 1000 users filled in during registration and subjective tightness experience scores (1-5 points) fed back during initial wearing are obtained, and characteristic engineering and z-score standardization processing are carried out on user data. And (3) clustering the users by adopting a k-means clustering algorithm, and determining that the optimal clustering number is 4 through an elbow rule and a contour coefficient to obtain four user subgroups with different tightness feeling characteristics, namely comfort type (45%), general type (30%), sensitivity type (20%), and extremely sensitivity type (5%). And establishing a classification rule between the user attribute and the clustering result by using a C5 decision tree algorithm, for example, if age <30andBMI <24th comfort (confidence coefficient 86%) If age is more than or equal to 60and gender=female th sensitivity (confidence coefficient 78%), mining association modes of attribute combinations such as different ages, sexes, BMI and the like and the optimal tightness level interval, and forming a personalized user tightness feeling image, for example, the optimal interval of a comfortable user is 2-4 levels, and the optimal interval of a sensitive user is 1-2 levels. Comparing the highest tightness level threshold with the personal tightness level interval of the user, marking the user as tolerant if the highest threshold falls in the personal interval, which means that the user can adapt to higher wearing tightness to achieve better monitoring precision, and identifying the user as sensitive if the highest threshold exceeds the upper limit of the personal interval, which means that the user is sensitive to wearing pressure and needs to ensure wearing comfort preferentially. And according to field expert knowledge and historical user feedback data, performing experience setting and iterative optimization on a tightness grade deviation threshold, for example, dynamically adjusting the deviation from an initial grade 5 to a grade 8, so that the boundary condition of user classification is more accurate. For a tolerant user, according to the optimal tightness level interval (such as 2-4 levels) of the individual and actual wearing habit data (such as average 5 levels), a fuzzy reasoning and Q-learning reinforcement learning algorithm is adopted, and a targeted tightness level adjustment suggestion is given, for example, a user is recommended to properly improve the tightness level from 5 levels to 3 levels so as to achieve a better heart rate monitoring effect, and meanwhile, the wearing feeling is still in a comfortable interval, so that the user is guided to properly improve the tightness on the premise of ensuring wearing comfort so as to meet the monitoring precision requirement. For sensitive users, a lower tightness wearing scheme corresponding to the lower limit (such as level 1) of the personal tightness level interval is recommended, and suggestions such as wearing position selection (such as the inner side of a wrist), wearing duration control (such as resting for 10 minutes every 2 hours) and the like for relieving uncomfortable feeling of pressure are given, and meanwhile, the users are prompted to' under the lower tightness level 1, heart rate monitoring errors can exceed 8 times/minute, blood oxygen monitoring errors can exceed 5%, and you are asked to trade off.

For the tolerant user, generating an adjustment instruction of the wearable device, wherein the adjustment instruction is used for adjusting the tightness level of the wearable device to be higher than the highest tightness level and closest to the optimal tightness level, if the lowest monitoring precision is lower than a threshold value, the adjustment instruction comprises reducing the data acquisition frequency of the wearable device so as to prolong the service life, and if the lowest monitoring precision is higher than the threshold value, the adjustment instruction comprises increasing the data acquisition frequency of the wearable device.

The comfort degree scoring and experience feedback of the tolerant user to different tightness degrees are obtained through user investigation and physiological experiments, and the optimal tightness degree interval and monitoring precision requirements of the user are determined by combining the physiological signal monitoring precision of the wearable device under different tightness degrees and stored in a user information database, so that a basis is provided for subsequent tightness degree adjustment. The method comprises the steps of reading an optimal tightness level interval and monitoring precision requirements of a user from a database, comparing an interval upper limit with a preset highest tightness level threshold, taking the threshold as a recommended tightness level if the upper limit is larger than or equal to the threshold, and taking the upper limit as the recommended tightness level if the upper limit is smaller than the threshold. And designing a fuzzy control algorithm to generate a tightness adjustment instruction of the wearing equipment according to the recommended tightness level. The tightness grade is taken as an input variable, control parameters (such as voltage, current and the like) of an executing mechanism are taken as output variables, membership functions such as Gaussian, triangular and the like are defined, and a fuzzy rule base is constructed, such as 'IF tightness grade is high THEN voltage is big'. The tightness level is mapped to a specific control parameter value by fuzzy reasoning and defuzzification processes. And sending the adjustment instruction to an embedded control unit of the wearing equipment, and driving an executing mechanism such as a piezoelectric material, a memory alloy and the like by the control unit according to the instruction to change the tightness degree of the wearing equipment in real time. And a PID control algorithm based on pressure sensor feedback is introduced, the contact pressure between the wearing equipment and the skin is monitored in real time, the pressure error is calculated, the output of the executing mechanism is regulated until the target tightness level is reached, and the control precision is within a preset range. According to the minimum monitoring precision threshold value, the data acquisition frequency of the wearable equipment is adaptively adjusted, and on the premise of meeting the monitoring requirement, the energy consumption is minimized, and the service life of the battery is prolonged. And if the current monitoring precision is lower than the threshold value, triggering a frequency-reducing control logic, reducing the data sampling rate by 1-5 gears according to the ratio of the precision difference value to the threshold value, wherein each gear corresponds to the sampling frequency change of 10Hz, and if the current monitoring precision is higher than the threshold value, triggering an frequency-increasing control logic, and increasing the data sampling rate by 1-3 gears until the highest allowable sampling frequency is reached. The wearable device generally comprises various types of sensors such as acceleration, gyroscope, PPG, ECG and the like, the sensors are distributed on different body positions such as wrists, breasts, ankles and the like, the acquisition frequency is different from 1Hz to 1kHz, and the data characteristics are different. And a multi-sensor data fusion algorithm such as Kalman filtering, bayesian estimation and the like is adopted to comprehensively analyze and noise inhibit physiological signal data acquired by different sensors, so that the data quality and the monitoring accuracy are improved. The data contribution degree of each sensor is dynamically adjusted through a self-adaptive weight distribution strategy, when the monitoring precision is reduced, the fusion weight of the high confidence sensor is increased, for example, the weight of the PPG is increased by 20%, the weight of the ECG is increased by 10%, and when the monitoring precision is increased, the fusion weight of each sensor is balanced, and the data redundancy and the energy consumption are optimized.

Specifically, 100 tolerant users are subjected to tracking investigation for 1 month, wearing equipment with different tightness grades (1-5 grades) every day, recording comfort degree scoring (1-10 grades) of the users on each tightness grade, and testing heart rate monitoring errors of the equipment under each tightness grade. Comprehensive analysis shows that the comfort degree score is highest (average 9.2 points) under the condition of 3-4 levels of tightness for 80% of users, and the heart rate monitoring error is lowest (< 5%), so that the optimal tightness level interval of the user group is determined to be 3-4 levels, and the minimum monitoring precision requirement is 95%. For a certain target user, the user information database is queried to obtain the 4-level upper limit of the optimal tightness level, and the optimal tightness level is compared with the 5-level of the preset highest tightness threshold, and 4-level is selected as the recommended tightness level of the user due to 4<5. According to the recommended tightness level, a control algorithm containing 25 fuzzy rules is designed, wherein the input variable is the tightness level (1-5 level), and the output variable is the driving voltage (0-5V) and the current (0-100 mA) of the actuating mechanism, such as the rule "IF tightness level is4THEN voltage is4Vand current is80mA". Defuzzification is carried out on the fuzzy reasoning result through a gravity center method, so that a control instruction with the driving voltage of 3.8V and the current of 76mA is obtained, and the control instruction is issued to a control unit of the wearable device. After the control unit receives the instruction, the driving voltage is gradually increased from 0V to 3.8V, meanwhile, the output of the pressure sensor is monitored, the pressure error is calculated in real time through a PID algorithm, the voltage is regulated, the voltage is finally stabilized at 3.85V, the tightness level of the wearable device is4, the contact pressure between the wearable device and the skin is (30+/-1) mmHg, and the requirement of the recommended tightness level is met. In the monitoring process, if the current monitoring precision is lower than 95%, triggering the frequency-reducing control logic, reducing the data sampling rate from 100Hz to 60Hz according to the precision difference ratio (such as 2%), and if the monitoring precision is higher than 98%, triggering the frequency-increasing control logic, and increasing the sampling rate from 60Hz to 1 gear. Meanwhile, multisource heterogeneous data of an acceleration sensor (50 Hz), a gyroscope (100 Hz), a PPG (500 Hz) and an ECG (1 kHz) are fused through a Kalman filtering algorithm, fusion weights of the sensors are adjusted in a self-adaptive mode, when the monitoring precision is lower than 95%, the weight of the PPG is improved to 0.5 from 0.2, the weight of the ECG is improved to 0.3, when the monitoring precision is higher than 98%, the weight of the PPG and the weight of the ECG are respectively reduced to 0.4 and 0.25, the weight of the acceleration and the weight of the gyroscope are set to 0.175, and the energy consumption is reduced while the monitoring precision is ensured. Finally, the user wears the wearing equipment with optimal adjustment and continuously monitors for 1 week, the average monitoring precision is stabilized at (97+/-1)%, the wearing time is improved by 5% compared with that before wearing, the battery endurance time is prolonged from 2 days to 3 days, and the comfort and the monitoring requirement of the user are met.

For sensitive users, generating regulation and control instructions of the wearable equipment, wherein the regulation and control instructions are used for reducing the data acquisition frequency of the wearable equipment, adjusting the tightness level of the wearable equipment to the highest tightness level when the wearable equipment performs data acquisition, and adjusting the tightness level of the wearable equipment to be closest to the optimal tightness level when the wearable equipment does not perform data acquisition.

And reading the personal optimal tightness level and the data acquisition frequency requirement of the identified sensitive user from the user information database, and establishing a quantized relation model of the sensitivity, the optimal tightness level and the data acquisition frequency according to the sensitivity characteristics and the physiological parameters of the user. And (3) fitting a nonlinear mapping relation among the sensitivity, the tightness level and the acquisition frequency by constructing a ternary quadratic polynomial regression equation, and estimating model parameters by adopting a least square method to obtain a quantitative prediction function of the sensitivity, the optimal tightness level and the data acquisition frequency. From this function, the highest level of tightness that the user can accept for a long time is calculated, as well as the amplitude of the data acquisition frequency that needs to be reduced. Generating a data acquisition instruction according to the down-conversion amplitude requirement, wherein the instruction content comprises parameters such as a sampling period of a sensor, the size of a data buffer area, the time interval of data packaging and the like. The nonlinear dynamic relation between the frequency-reducing amplitude and the acquisition parameters is described by adopting a fractional calculus model, the mathematical expression of the model is D ^a _xf(t)＝t^n-a/Γ(n-α)·(dⁿf(t)/dtⁿ, wherein alpha is fractional calculus, the speed and smoothness of the frequency-reducing process are reflected, n is an integer order, and Gamma is a Gamma function. By solving a fractional differential equation, a nonlinear mapping relation between the frequency-reducing amplitude and the acquisition parameters is established, an optimal model parameter combination is searched by adopting a parameter self-adaptive optimization algorithm such as a self-adaptive weight PSO algorithm, and each parameter value in the acquisition instruction is dynamically adjusted, so that the frequency-reducing requirement is met, and meanwhile, the integrity and the instantaneity of data are ensured. And generating a tightness adjustment instruction, and adjusting the tightness level to be the highest when the wearable device is detected to start data acquisition. The highest tightness level is converted into control variables of an actuating mechanism, such as motor rotation speed, deformation of the shape memory alloy and the like by adopting a fuzzy control algorithm. A fuzzy rule base of tightness-control variables is established, a Mamdani reasoning method is used for mapping the language variable of tightness level into fuzzy membership, fuzzy synthesis is performed, fuzzy output is performed by using a gravity center method, and specific actuator control variable values are obtained. The form of the fuzzy rule base is IF tightness grade ISHIGHTHEN motor rotating speed ISFASTAND alloy deformation isLarge, and smooth mapping from tightness grade to execution control quantity is realized by defining proper membership function and reasoning rule. And immediately generating a tightness callback instruction after detecting that the data acquisition task of the wearable equipment is finished, and reducing the tightness level from the highest value to the value closest to the optimal tightness level. By building a three-layer feedforward neural network, taking the tightness grade as input and the user comfort grade as output, a nonlinear regression model of the tightness grade and the user comfort grade is built. The network structure adopts the arrangement of 5-10-1, namely 5 neurons in the input layer, 10 neurons in the hidden layer and 1 neuron in the output layer, and the activation function is selected from hyperbolic tangent function tanh. The mean square error is used as a loss function, the gradient descent method is used for carrying out iterative optimization on the network weight, the convergence speed and stability of the training process are controlled by setting super parameters such as the learning rate, the momentum factor and the like, and the uncomfortable feeling of the user when the tightness level deviates from the optimal value is minimized. And taking the callback tightness level as the input of the neural network, and predicting the optimal callback tightness level through forward propagation calculation so as to enable the callback tightness level to be closest to the optimal tightness interval of the user. And integrating the data acquisition instruction and the tightness adjustment instruction, and generating a unified regulation instruction sequence according to a time sequence, wherein the instruction sequence comprises the attributes of instruction type, execution time, parameter value and the like. By constructing a directed acyclic graph model of an instruction sequence, the timeliness and harmony of instruction execution are evaluated by using a discrete event simulation algorithm, such as an event scheduling method. For each node in the graph, a corresponding event object is generated, including attributes such as event type, timestamp, priority and the like, and the events are inserted into an event queue according to the sequence of the timestamp. The event is triggered and processed step by step through an event circulation mechanism, the state and the clock are updated, and the discovered command conflict and error are detected and corrected in time, such as the triggering time, the priority and the like of the event are adjusted, so that the feasibility and the consistency of the regulation command sequence are ensured. And issuing the regulation instruction sequence to an embedded control module of the wearable device, wherein the control module packages operations such as instruction analysis, parameter setting, execution control and the like into independent task threads based on the real-time operating system FreeRTOS. Synchronous communication among threads is realized through mechanisms such as semaphores, message queues and the like, and exception handling and fault tolerance control are realized through mechanisms such as interrupt service routines, watchdog timers and the like. And analyzing the instruction sequence into a parameter setting task and an execution control task by adopting a producer-consumer model, and respectively storing the parameter setting task and the execution control task into a task queue. The task scheduler dynamically allocates CPU time and resources according to the priority and time constraint of the tasks, coordinates concurrent execution of the tasks, and ensures real-time performance and reliability. In the task execution process, the monitoring and exception handling of the task state are realized through a hook function and a callback mechanism, and smooth completion of the instruction is ensured. And through an embedded wireless communication module, the information such as the execution state, user feedback, abnormal log and the like of the wearable equipment is uploaded to the cloud server in real time by adopting an MQTT protocol. Preprocessing and feature extraction are carried out on the collected user data and the collected equipment working data, and a multi-dimensional time sequence data set is constructed. And adopting an increment learning algorithm, such as an online sequence extreme learning machine (OS-ELM), and carrying out real-time modeling and increment updating on the data to dynamically optimize the regulation strategy. The OS-ELM randomly generates hidden layer weights, and online updates the output layer weights by using a sequential learning strategy, so that the whole network is not required to be retrained, and the method is suitable for processing streaming data and concept drift problems. By setting forgetting factors and regularization parameters, the plasticity and stability of the model are balanced, and the variation trend of the user behavior pattern and the device performance characteristic is tracked. And generating a regulation strategy improvement instruction, and transmitting the regulation strategy improvement instruction to the wearable equipment to form closed-loop feedback control, so that the intelligent level and the user experience are continuously improved. Meanwhile, the optimization effect is evaluated by adopting methods such as cross validation, A/B test and the like, and the self-adaptive adjustment and decision optimization of the online learning process are supported.

Specifically, through carrying out statistical analysis on historical wearing data of a sensitive user, the optimal tightness level is 1.5, the highest acceptable tightness level is 2.5, and the optimal data acquisition frequency is 10Hz. The relationship between sensitivity, tightness level and acquisition frequency is fitted based on a ternary quadratic polynomial regression model y＝a₀+a₁x₁+a₂x₂+a₃x₁ ²+a₄x₂ ²+a₅x₁x₂, where y is the acquisition frequency, x ₁ is the sensitivity, and x ₂ is the tightness level. Regression coefficients a ₀＝15.2,a₁＝-3.1,a₂＝-4.6,a₃＝0.8,a₄＝1.2,a₅ = -0.5 were estimated using least squares, model decision coefficients R ² =0.92. Substituting the sensitivity value of 0.8 of the user into the model predicts that the optimal acquisition frequency is 12Hz, and the default acquisition frequency needs to be reduced from 20Hz to 8Hz. The fractional calculus model D ^a _xf(t)＝t^n-a/Γ(n-α)·(dⁿf(t)/dtⁿ) is adopted to describe the down-conversion process, α=0.5 and n=2 are taken, the sampling period is increased from 0.05s to 0.08s, the buffer area size is reduced from 1KB to 0.6KB, and the data packing time interval is prolonged from 1s to 1.5 s. When the data acquisition is detected to start, according to a fuzzy rule of IF tightness grade ISHIGHTHEN motor rotating speed ISFASTAND alloy deformation isLarge, setting membership of linguistic _variable tightness grade as mu (High) =0.8, obtaining fuzzy output mu (Fast) =0.7 and mu (Large) =0.6 through Mamdani reasoning, performing inverse gelatinization by using a gravity center method, obtaining a motor rotating speed control value of 800rpm, an alloy deformation control value of 3mm, and driving an executing mechanism to adjust the tightness grade to 2.3. After data acquisition is finished, the trained three-layer neural network is used for predicting the optimal callback tightness level, the network structure is 5-10-1, 5 neurons of the input layer correspond to the age, sex, BMI, heart rate, blood pressure and other characteristics of a user, 10 neurons of the hidden layer are used, 1 neuron of the output layer gives out the predicted tightness level, the learning rate is set to be 0.01, the momentum factor is 0.9, and after 1000 epochs neurons are trained, the average prediction error on a test set is 0.15. And inputting the characteristics of the user into a neural network, and performing forward propagation calculation to obtain the optimal callback degree level of the user as 1.7. The generated acquisition instructions and tightness instructions are integrated in time sequence, an instruction sequence directed acyclic graph is constructed, discrete event simulation is conducted by using an event scheduling method, each instruction event comprises types, time stamps and priority attributes, such as { type: "acquisition instructions": "timestamp:" 00:05", priority:2}, { type:" tightness instructions ":" timestamp: "00:03", priority:1} and the like, the events are scheduled according to the minimum time stamp and highest priority principle, the total simulation time is 15min, the average instruction execution delay is 0.5s, and the priority conflict rate is 2%. The embedded control module analyzes the received instruction sequence into an acquisition task and a tightness task based on FreeRTOS operation systems, and respectively enters task queues with priorities of 2 and 1. The task scheduler adopts a preemptive priority scheduling algorithm to allocate independent kernels for tightness tasks, the acquisition tasks are executed in a time slice rotation mode, the time slice length is set to be 50ms, and the context switching time is less than 1ms. Synchronization among tasks is realized through binary semaphores, asynchronous communication is realized through a ring buffer zone, task-level and system-level exception handling is realized through a hardware timer and a watchdog mechanism, and instantaneity and reliability are ensured. The wearable device sends the collected user data and the collected device state data to the cloud server in a JSON format according to the reliability level of QoS1 through an MQTT protocol, the transmission delay is less than 500ms, and the packet loss rate is less than 0.1%. The server side uses an OS-ELM algorithm to model the received multidimensional time series data online, the dimension of an input layer is 10, the characteristics of heart rate, blood oxygen, step number and the like are included, 100 hidden layer neurons are initialized, input weights in a random generation interval [ -1,1] are updated once every 1min according to a sequential block learning strategy, a regularization factor C=10, a forgetting factor lambda=0.9, and after continuous learning for 6h, the root mean square error of the model is reduced to below 0.2. Based on the prediction result of the OS-ELM model, generating an optimization instruction for adjusting parameters such as sampling rate, tightness level, data fusion strategy and the like, and issuing the optimization instruction to the wearable device for execution, so that end-to-end closed-loop control is realized. Through the cross verification and A/B test of 10 users, the user satisfaction is improved by 15%, the battery endurance time is prolonged by 20%, and the average monitoring precision is improved.

And S106, determining the recommended wearing position and the recommended tightness level through a regression model by combining the minimum monitoring precision corresponding to the target index. And dynamically adjusting the recommended wearing position and the recommended tightness by monitoring feedback information of the user in real time.

According to different application scenes and monitoring targets, such as exercise heart rate monitoring, sleep blood oxygen monitoring and the like, the minimum monitoring precision requirement of corresponding physiological parameters is determined and used as a reference for evaluating wearing positions and tightness grades. The minimum monitoring precision threshold value is quantitatively set by consulting the IEEEP1708 heart rate monitoring wearable equipment standard, the ISO80601-2-61 pulse oximeter standard and other industry specifications. Physiological parameter data under different wearing positions and tightness grades are acquired through the sensor array, such as wearing equipment on the positions of a wrist, an upper arm, an ankle and the like in a tightness-adjustable mode, signals such as an electrocardiogram, a photoelectric volume pulse wave and acceleration are synchronously recorded, the acquired data are cleaned, denoised and feature extracted, and a high-quality monitoring precision sample set is constructed. For different physiological parameters, a multiple nonlinear regression model is established, such as support vector regression of a polynomial kernel function is adopted, the wearing position, the tightness level, the physiological characteristics of a user and the like are taken as input characteristics, the monitoring precision of the physiological parameters is taken as an output target, characteristic selection is carried out based on an L1 regularized minimum angle regression method, the model super-parameters are optimized through grid search, the generalization performance of the model is evaluated by using cross verification, and a nonlinear regression equation of the monitoring precision about the wearing position and the tightness level is obtained. On the basis of the regression model, the multi-objective optimization problem under the constraint of monitoring precision is built. Taking wearing positions and tightness grades as decision variables, taking monitoring precision, wearing comfort, equipment stability and the like as optimization targets, establishing a weighted objective function, and carrying out normalization processing and weight assignment on a plurality of targets by adopting a TOPSIS method. And setting the wearing position and the value range of the tightness level as constraint conditions, and converting the constraint conditions into a multi-element nonlinear programming problem with inequality constraint. And solving by adopting a genetic algorithm, encoding the wearing position and the tightness level into binary gene strings, carrying out iterative search on an optimal solution through genetic operations such as selection, intersection, variation and the like, comprehensively considering the monitoring precision and the weighting performance of other targets by an fitness function, and outputting an optimal wearing position and tightness level combination meeting the minimum monitoring precision requirement after iteration for 1000 generations. Developing a user feedback acquisition module, pushing a questionnaire through a mobile phone APP, and acquiring subjective scores of a user on the current wearing position and the tightness level, wherein the score dimension comprises comfort, compression feeling, limited activity and the like, and the higher the score is, the better the experience is indicated. Meanwhile, physiological feedback of a user is detected in real time through a pressure sensor, a heart rate sensor and the like which are arranged in the wearable equipment, characteristic indexes such as skin surface pressure distribution and heart rate variability are extracted, states such as fatigue and pressure level of the user are classified and quantified by using a machine learning algorithm, and a multidimensional user feedback vector is constructed. And calculating the similarity of different users on wearing positions and tightness level preferences by adopting a collaborative filtering algorithm based on the articles, generating a user-wearing preference matrix, and predicting the acceptance degree of the user-wearing preference matrix for a new wearing scheme according to the user history preference. And evaluating the comprehensive performance of the current recommended scheme by combining the physiological parameter monitoring precision and the weighted score fed back by the user. A dynamic adjustment strategy based on deep reinforcement learning is constructed, a user feedback vector and monitoring precision are used as a state space, adjustment values of wearing positions and tightness grades are used as an action space, and a comprehensive performance improvement value is used as a reward function. And (3) establishing a state-action value function approximate model by adopting a DQN algorithm, constructing a Q network by taking a long short-term memory network (LSTM) as a backbone, inputting state characteristics, and outputting the Q value of each action. And (3) performing offline training on the network by generating experience playback data through balanced exploration and utilization of an epsilon-greedy strategy, and using a target network to alleviate overfitting. And the error test and the learning are continuously carried out, the mapping strategy from the state to the action is optimized, and the self-adaptive dynamic adjustment of the wearing position and the tightness level in a real-time environment is realized. After each adjustment, the optimization performance is evaluated and recorded, the user experience feedback of the new scheme is collected, key improvement points are extracted, the state representation, action design and rewarding function of the reinforcement learning model are iteratively updated, meanwhile, the performance difference before and after the optimization is objectively evaluated through methods such as an A/B test and the like, the optimization knowledge and data are continuously accumulated, a self-perfected closed-loop optimization mechanism is formed, the user comfort is maximized on the premise of guaranteeing the monitoring precision, and the self-adaptive wearing optimization of man-machine cooperation is realized.

Specifically, according to the ieee p1708 standard, the heart rate monitoring accuracy is defined as the root mean square error between the measured value and the reference value, requiring that the measurement error of the static heart rate does not exceed ±3 times/min. The photoelectric volume pulse wave sensors with adjustable pressure are worn at 6 different positions of wrists, upper arms, ankles and the like of 100 subjects, 24-hour dynamic heart rate data are collected in three tight grades of loose, medium and tight grades, a medical grade electrocardiograph monitor is synchronously used for collecting reference heart rate data, preprocessing such as trend removal, band-pass filtering and the like is carried out on 500 ten thousand heart rate sample points collected, and time domain and frequency domain characteristics are extracted. A support vector regression model of polynomial kernel function K (x, x ') = (γ x t x' +r) d was used, where the kernel function parameter γ=0.1, r=1, d=3, the least angle regression method was used to select the 8 most relevant feature subsets from the 50 features, and the model parameters c=10, epsilon=0.1 were optimized by 5-fold cross-validation with an average absolute error of 1.5 times/min over the test set. Setting the monitoring precision target to be +/-2 times/min, combining the comfort level and the stability target, and constructing a three-target weighting function f=0.5×acc+0.3× comf +0.2×stab, wherein acc represents the precision score, comf represents the comfort level score, stab represents the stability score, and the weight coefficient is determined by a hierarchical analysis method. Searching an optimal solution by adopting a genetic algorithm, encoding 6 wearing positions and 3 tightness grades into 9-bit binary strings, setting the population scale as 50, setting the crossover probability as 0.8, setting the variation probability as 0.1, and iterating for 1000 generations to obtain the optimal wearing scheme which is the middle tightness of the upper arm, wherein the comprehensive objective function value is 0.93. The subjective scores of 1000 users are carried out through the questionnaire, the comfort level is evaluated in three dimensions of pressure distribution uniformity, compression feeling and activity obstruction feeling, and the stability is evaluated in the success rate of heart rate and blood oxygen measurement, so that 5 ten thousand user feedback vectors are formed. And calculating the similarity between users by using the pearson correlation coefficient, predicting the preference of the users to different wearing schemes by using a collaborative filtering algorithm based on the articles, and for a new user, matching the similar users according to the personal attribute characteristics of the new user, and generating a personalized recommendation scheme. The LSTM-DQN model is built to dynamically adjust the wearing position and tightness, the state quantity comprises 12 dimensions of user feedback scoring, monitoring precision, physiological pressure characteristics and the like, the action space comprises upward/downward adjustment of 1-2 gears or tightness, the reward function is the comprehensive performance improvement percentage, and the fault tolerance penalty item is set. After 10000 training iterations, the average rewarding value is increased from 0.2 to 0.85, and the converged strategy network can recommend the optimal fine tuning scheme in real time according to the user state. By applying the strategy to continuously optimize the wearing schemes of 100 users, the average monitoring precision is improved by 2%, the comfort level score is improved by 15%, and the user satisfaction reaches 95%. The tracking comparison of the optimized group and the control group is carried out for 1 month through the A/B test, the wearing compliance of equipment of the optimized group is higher than 30%, the continuous monitoring time is prolonged by 6 hours/day, and the effectiveness of the self-adaptive optimization mechanism is verified.

Step S107, analyzing the personalized data of the user by applying the long-term and short-term memory neural network, constructing a model to predict the accuracy of the biological signals under different using conditions, comparing the model prediction result with the actual acquired data, calculating the mean square error, and dynamically adjusting the wearing strategy according to the error size until the error converges to an acceptable range.

And collecting multi-dimensional personalized data such as historical wearing records, physiological parameters, activity types and the like of the user, and expanding sample diversity by adopting data enhancement technologies such as random rotation, mirror image overturning and the like. The data are subjected to cleaning, alignment and standardization processing, category distribution of the data in different scenes is balanced through methods such as up-sampling and down-sampling, key features reflecting wearing habits of users are extracted, such as wearing duration, adjusting frequency, comfort feedback and the like, a time sequence feature matrix is constructed, and the time sequence feature matrix is randomly divided into a training set, a verification set and a test set. And designing a framework of the long-short-term memory neural network model, wherein the number of nodes of an input layer is consistent with the characteristic dimension, the number of nodes of an output layer is consistent with the type of biological signals, and the framework comprises an LSTM hidden layer. And optimizing the super parameters by using a random search algorithm, traversing parameter combinations such as learning rate, hidden unit number, batch size and the like in a specified range, and selecting the model configuration with the minimum verification set error. User behavior feature representations are extracted from the encoder by training the LSTM as pre-training initialization parameters for the subsequent model. And training a model by using a back propagation algorithm, and introducing L2 regularization term to control complexity of the model by taking the mean square error as a loss function. The Adam optimizer is adopted to adaptively adjust the learning rate, and the gradient clipping threshold value is set to be 1.0, so that gradient explosion is avoided. Dropout regularization is applied to the input layer and LSTM layer with probabilities of 0.2 and 0.5, respectively, alleviating overfitting. And after each epoch is finished, evaluating the performance of the model on a verification set, triggering an early-stopping mechanism if the model is not improved for 5 times, and storing model parameters with optimal performance. The data set is divided by using 10-fold cross validation, training is repeated 10 times, the prediction results of all models are averaged, and the model robustness is improved. And (3) applying the trained LSTM model, and predicting the acquisition accuracy of the biological signals under different use scenes according to the conditions of real-time wearing data, environmental conditions, physiological states and the like of the user. And introducing a sequential attention mechanism, dynamically adjusting the attention weights of different time steps through a mask matrix, and highlighting the influence of key moments. And a causal convolution is applied in the time dimension, so that the long-term dependency relationship of the user behavior is effectively captured. And carrying out confidence calibration on the output result of the model, setting an uncertainty interval according to the historical prediction deviation, and improving the interpretation of the prediction. And acquiring biological signal data acquired by actually wearing equipment by a user, performing feature engineering processing on the data, extracting indexes corresponding to the prediction output of the model, calculating a plurality of evaluation indexes such as mean square error and average absolute error between a predicted value and a true value, evaluating the prediction performance of the model at multiple angles, and identifying a time period and a use condition with larger prediction error. And triggering an adaptive model adjustment mechanism aiming at the condition that the model prediction error exceeds the threshold value continuously for a plurality of times. And searching optimal configuration in a super-parameter space such as LSTM unit number, learning rate, regularization parameter and the like by adopting a grid search algorithm, and avoiding overfitting by an early-stop method and model integration. Pruning and quantization optimization are carried out on the LSTM model, redundant connection and low importance parameters are removed, weight is weighted into 8-bit fixed point number, and calculation cost is reduced while performance is ensured. And fine-tuning the parameters of the updated model on the training set, and reducing the difference between the predicted value and the true value. And according to the feedback information of the prediction error, combining the domain knowledge and the user feedback, formulating a rule and a reward function for dynamic optimization of the wearing strategy, and modeling the optimization problem as a Markov decision process. The state space comprises user attributes, equipment parameters, environmental factors and the like, the action space comprises adjustment of wearing positions, tightness, sampling frequency and the like, and the reward function comprehensively considers factors such as signal accuracy, wearing comfort and the like. And applying DeepQ-Network and other reinforcement learning algorithms, and learning the optimal wearing parameter adjustment strategy by exploring and utilizing balance to generate targeted optimization suggestions. And an online strategy updating mechanism based on incremental learning is designed, and the strategy network is subjected to fine adjustment regularly according to newly acquired user data and feedback, so that the adaptability and the effectiveness of the strategy are improved. Building a human-computer interaction feedback closed loop, pushing wearing optimization suggestions through a smart phone APP, collecting subjective scores of users on optimization effects, recording behavior logs of the users actively adjusting wearing modes, screening samples with the most obvious model lifting effect by using an active learning algorithm, and adding the samples into a training set to participate in model iteration. Based on user portrayal and behavior analysis, personalized wearing optimization strategies are customized for different user groups, and the pertinence and the acceptance of optimization are improved. Continuously monitoring the optimized prediction error change, drawing a learning curve, and judging to be converged to an acceptable level when the continuous repeated iteration error descending amplitude is lower than a certain threshold value, so as to complete the self-adaptive wearing strategy optimization task.

Specifically, wearing records of 1000 users for 3 months continuously are extracted from a wearable device database, each record contains 10 dimension characteristics such as user ID, timestamp, acceleration, heart rate, step number and the like, nearest neighbor interpolation is carried out on missing values, and box line diagram-based filtering is carried out on the missing values. And (3) up-sampling a few types of samples by adopting an SMOTE algorithm, so that the number of the samples of each type is balanced. And 20 statistical features reflecting the habit of the user, such as wearing duration, heart rate variability and the like, are extracted, and are combined with attribute features, such as age, gender, BMI and the like, of the user to form a 40-dimensional feature vector, and Min-Max normalization processing is adopted. The training set, the verification set and the test set are divided into 8:1:1, and the data of each user are randomly distributed in a ratio of 7:2:1. An LSTM model is built by using Keras frames, the input layer contains 40 nodes, the LSTM layer contains 128 units, the output layer contains 5 nodes by using a tanh activation function, and predicted values of 5 biological signals such as blood pressure, heart rate and the like are respectively represented. And carrying out 50 rounds of random searches in a parameter space with the hidden unit number of [64,128,256], the learning rate of [0.001,0.005,0.01] and the batch size of [32,64,128], and selecting the optimal super-parameter combination according to the average absolute error of the verification set. An Adam optimizer training model is used, the initial learning rate is 0.005, the learning rate is attenuated to be 0.5 times of the original learning rate every 10 epochs, and the L2 regularization coefficient is set to be 0.01. Training 50 epochs, triggering early stop when the error of the verification set is not reduced for 5 times continuously, and taking the model parameter with the lowest error of the verification set as a final model. The 5 biological signals of the test set were predicted with average absolute errors of 2 times/min, 5 mg, 1%, 5 steps/min, 0.5 degrees celsius, respectively. And introducing a sequence attention mechanism, calculating the similarity of the Query vector and the Key vector through a dot product to obtain attention weight, and carrying out weighted summation on the hidden state of the LSTM output to generate a feature representation focused on the Key moment. Extracting long-term dependence of user behaviors in a time dimension by adopting causal convolution, wherein the convolution kernel is 24 hours, the step length is 1 hour, and the receptive field range is enlarged by using hole convolution with the expansion rate of 2. And performing Platt calibration on the model predicted value, estimating a confidence interval according to the prediction error distribution, and considering that the confidence is high when the proportion of the actual value falling in the interval exceeds 95%. And triggering the self-adaptive adjustment of the model under the condition that the continuous 3-day prediction error exceeds 3 times of the average error level, optimizing the super-parameters such as LSTM unit number, learning rate and the like through grid search, pruning the weights of the hidden layer and the output layer, keeping the parameters of sensitivity Top20% based on Hessian matrix analysis, keeping the rest cut-off as 0, and compressing the model scale. The updated model is trimmed for 10 epochs, so that the prediction error is reduced to be less than 50% of the original prediction error. And optimizing a wearing strategy by applying a DQN algorithm, wherein a state space comprises 20 characteristics such as a user type, a history error, current parameters and the like, an action space comprises wearing positions and tightness of 1-5 gears which are adjusted up and down, and a reward function is set as a weighted sum of a prediction accuracy lifting quantity and user feedback comfort level. After 10000 times of training, the average rewarding value is converged from 0.2 to 0.85, the optimal strategy is applied to a new user, the new user is prompted to wear the bracelet at the position 2cm above the wrist bone, the tightness is adjusted to 3 grades, the sampling frequency is increased by 20%, the prediction error is reduced by 15 percentage points, and the user satisfaction is improved.

The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should make equivalent substitutions or modifications according to the technical scheme of the present invention and the inventive concept thereof, and should be covered by the scope of the present invention.

Claims

1. A health monitoring method integrating multimodal bioinformation, characterized in that the method comprises:

When a user wears a wearable device for the first time, a multimodal sensor is deployed to collect the user's actual biosignals and record the tightness level and location information of the wearable device;

Collect users' tightness perception of wearable devices at different positions and corresponding to different tightness levels, and associate tightness perception with tightness level and position information; including dividing tightness perception into 5 levels, collecting users' subjective evaluation of wearing comfort, and obtaining users' subjective tightness perception data at different tightness levels; collecting users' pressure distribution data at the contact point between the wearable device and the skin at different tightness levels, aligning and calibrating the subjective tightness perception score with the objective pressure distribution measurement results, establishing a quantitative relationship between the two, and establishing a quantitative relationship model between tightness level and pressure distribution; obtaining users' body surface contour data, combining with the three-dimensional structure model of the wearable device, and obtaining theoretical tightness levels at different positions through the contact area and positive pressure distribution at different wearing positions; collecting tightness perception feedback from different users, selecting the user attributes with the greatest correlation with tightness perception, building a decision tree model, selecting the optimal splitting attribute through the information gain index, and generating a tree structure;

The monitoring accuracy of the wearable device is calculated based on the actual biological signals and wearable device data. A regression analysis model is established based on the tightness perception, tightness level, and location information of the current user to describe the correlation between different tightness levels and location information and the monitoring accuracy of the wearable device.

Receive the target index and personal tightness threshold corresponding to the wearable device input by the user, combine the personal tightness threshold and the best tightness level corresponding to the user's best tightness feeling, determine the personal tightness level range, and evaluate the matching degree of the wearable device to the user based on the preset adjustable range of the wearable device; including obtaining the user's functional requirements and expected physiological monitoring indicators for the wearable device, while allowing the user to customize the comfort range to form a personal tightness threshold; adopt a fuzzy comprehensive evaluation method to construct a multi-level fuzzy evaluation matrix with tightness feeling as the target layer, physiological parameters, physical pressure and wearing parts as the criterion layer, and different tightness levels as the solution layer; determine the weight coefficient and membership function of each evaluation indicator, use the weighted average method to perform multi-level fuzzy synthesis operation, and obtain the quantitative mapping relationship between the user's subjective tightness feeling and the objective tightness level; analyze the user's tightness feedback data, determine the optimal clustering number k through the elbow rule and silhouette coefficient evaluation, use the Euclidean distance to measure the similarity between the tightness feeling feature vectors, iteratively optimize the clustering center, and obtain the best tightness feeling mode for different groups of people;

Determine the minimum monitoring accuracy according to the target indicator, and determine the highest tightness level corresponding to the minimum monitoring accuracy through the regression model. If the highest tightness level is within the personal tightness level range, the user is identified as a tolerant user; if the highest tightness level is outside the personal tightness level range, the user is identified as a sensitive user;

Combined with the minimum monitoring accuracy corresponding to the target indicator, the recommended wearing position and recommended tightness level are determined through the regression model, and the recommended wearing position and recommended tightness are dynamically adjusted through real-time monitoring of user feedback information;

Long short-term memory neural networks are used to analyze user personalized data, and a model is built to predict the accuracy of biological signals under different usage conditions. The model prediction results are compared with the actual collected data, the mean square error is calculated, and the wearing strategy is dynamically adjusted according to the error size until the error converges to an acceptable range.

2. The method according to claim 1, characterized in that when the user wears the wearable device for the first time, deploying a multimodal sensor to collect the user's actual biological signals and record the tightness level and location information of the wearable device, comprising:

Obtain the user's basic personal information, preliminarily estimate the user's physiological parameter range, and provide a reference benchmark for the actual biological signals collected;

Calculate the optimal tightness level and wearing position, obtain the initial wearing parameters, guide users to wear correctly, and improve the accuracy of collected signals;

Detect the user's motion status, identify and filter out relatively static time periods, collect biological signals within the time period, extract characteristic parameters of the signals, obtain basic physiological indicators of the user in a static state, and establish a personal health baseline;

Continuously record the changes in biological signals when the user wears the device at different tightness levels and positions, extract the statistical characteristics and spectral characteristics of the signal, calculate the mean, variance and correlation coefficient indicators of the characteristic parameters, set thresholds, judge the validity and reliability of the collected data, and adjust the working parameters of the sensor accordingly, and detect the tightness and offset of the wearable device in real time;

The biosignal data collected when the user wears the device for the first time is bound to the user’s basic personal information and uploaded to the cloud platform.

3. The method according to claim 1 is characterized in that the monitoring accuracy of the wearable device is calculated based on the actual biological signal and the wearable device data, and a regression analysis model is established in combination with the tightness perception and tightness level and location information of the current user to describe the correlation between different tightness levels and location information and the monitoring accuracy of the wearable device, including:

Collect the user's physiological signals and synchronously record the motion data of the user's wearable device;

Use convolutional neural network deep learning models to perform quality assessment and anomaly detection on physiological signals;

According to the signal quality evaluation results, high-quality physiological signal segments are screened out, and the screened high-quality physiological signal segments are compared with the monitoring results of the reference medical equipment, and the measurement errors and deviations of the wearable device on different physiological indicators are calculated to obtain a quantitative evaluation of the monitoring accuracy;

By means of questionnaire survey and human-computer interaction, the subjective comfort scores of users at different tightness levels and wearing positions were collected, and the corresponding relationship between tightness perception and user physiological parameters was established.

4. The method according to claim 1, characterized in that the minimum monitoring accuracy is determined according to the target indicator, and the maximum tightness level corresponding to the minimum monitoring accuracy is determined through a regression model, and if the maximum tightness level is within the personal tightness level range, the user is identified as a tolerant user, and if the maximum tightness level is outside the personal tightness level range, the user is identified as a sensitive user; comprising:

Determine the target indicators for monitoring various physiological parameters based on the application scenarios and functional positioning of wearable devices;

The physiological parameter monitoring data of different subjects under different tightness levels, materials and wearing positions were collected, and the data were cleaned, outliers processed and normalized. A multivariate linear regression model was constructed with the tightness level, material characteristics and wearing position of the wearable device as independent variables and the monitoring error of the physiological parameters as the dependent variable.

The stepwise regression method was used for feature selection, and the regression coefficient was analyzed for significance, and the quantitative relationship between tightness level, material characteristics, wearing position and monitoring accuracy was obtained.

Substitute the monitoring target indicators into the regression equation, calculate the minimum tightness level required to achieve the target accuracy, determine the reasonable tightness level margin, and finally obtain the maximum tightness level threshold to ensure monitoring accuracy;

The highest tightness level threshold is compared with the user's personal tightness level range. If the highest threshold falls within the personal range, the user is marked as tolerant, indicating that the user can adapt to a higher wearing tightness to achieve better monitoring accuracy;

If the highest threshold exceeds the upper limit of the personal range, the user will be identified as sensitive, indicating that the user is more sensitive to wearing pressure and needs to prioritize wearing comfort.

5. The method according to claim 4 is characterized in that, for tolerant users, an adjustment instruction of the wearable device is generated to adjust the tightness level of the wearable device to be higher than the maximum tightness level and closest to the optimal tightness level. If the minimum monitoring accuracy is lower than a threshold, the adjustment instruction includes reducing the data collection frequency of the wearable device to extend its service life; if the minimum monitoring accuracy is higher than the threshold, the adjustment instruction includes increasing the data collection frequency of the wearable device.

6. The method according to claim 4 is characterized in that, for the sensitive user, a control instruction of the wearable device is generated to reduce the data collection frequency of the wearable device, and the tightness level of the wearable device is adjusted to the highest tightness level when the wearable device is collecting data, and the tightness level of the wearable device is adjusted to be closest to the optimal tightness level when the wearable device is not collecting data.