CN118472946B - Smart grid AI joint peak load decision-making method, system, equipment and medium - Google Patents

Abstract

The present invention discloses a smart grid AI joint peak-shaving decision method, system, device and medium, and relates to the field of smart grid optimization technology. According to the geographical location, load characteristics and power supply mode of the power grid, the power grid is divided into multiple independent operating areas, and the historical power demand data of each operating area is obtained to construct a regional demand index vector, and each operating area is divided into three power demand levels: high, medium and low. Further, the external factor index and the regional demand index vector are used to construct a power demand time series prediction model, collect real-time power demand data and real-time external factor index of the operating area, predict the power demand sequence of the current operating area for multiple time steps in the future, and perform graded calibration on it. According to the calibrated power demand prediction sequence, the future power demand change trend is analyzed, and a power grid operation optimization strategy is formulated. This method effectively improves the intelligence level and operation efficiency of the power grid operation, and ensures the smooth operation of the power grid.

Description

Smart grid AI joint peak load decision-making method, system, equipment and medium

Technical Field

The present invention relates to the field of smart grid optimization technology, and more specifically, to a smart grid AI joint peak regulation decision method, system, device and medium.

Background Art

With the continuous expansion of the power grid and the rapid development of smart grid technology, the operation of the power grid is facing more and more challenges. How to achieve efficient, safe and stable operation of the power grid has become a hot issue in power system research. At present, domestic and foreign scholars have carried out a lot of research on power grid load forecasting and optimal scheduling.

The Chinese patent application with publication number CN115689105A discloses a load analysis system and prediction device for power grids. By setting up a big data mining and analysis module for the impact of power outage plans, a power system power outage risk prediction module, and a power grid constraint and intelligent optimization module, it realizes the mining and analysis of factors affecting power outage plans based on big data, the assessment and prevention of power system power outage risks, and the optimization and real-time revision of power outage plans. The device introduces artificial intelligence algorithms in the process of power grid load analysis and prediction, achieving the effect of accurate real-time load prediction of the main grid.

The Chinese patent with the authorization announcement number CN106451438B discloses a load interval prediction method that takes into account smart electricity consumption behavior. This method uses the start time and end time of smart electricity consumption equipment to reflect user behavior, and fully considers the impact of smart electricity consumption behavior on load prediction. Compared with the traditional load prediction that only considers factors such as climate, this method conforms to the development trend of smart grids, increases the influence of user subjective behavior in the original load prediction, and provides a decision-making reference for load prediction after power companies carry out smart electricity consumption projects.

However, most existing technologies perform load forecasting and optimal scheduling for a single region, lack consideration of the overall operating status of the power grid, make insufficient use of external environmental factors and user behavior characteristics, and are unable to adapt to the increasingly complex power grid operating environment.

Summary of the invention

In order to overcome the above-mentioned defects of the prior art, the present invention provides a smart grid AI joint peak-shaving decision method, system, device and medium.

To achieve the above object, the present invention provides the following technical solutions:

Smart grid AI joint peak load decision-making method, including:

Step S1000: Divide the power grid into m independent operation areas according to the geographical location, load characteristics and power supply mode of the power grid; obtain historical power demand data of each operation area, and construct a regional demand index vector according to the historical power demand data; divide each operation area into three power demand levels of high, medium and low according to the regional demand index vector; m is an integer greater than or equal to 1;

Step S2000: Obtain external factor indicators, and construct a power demand time series prediction model based on the external factor indicators and the regional demand indicator vector; collect real-time power demand data and real-time external factor indicators of the operating area, and predict the power demand sequence of the current operating area in the next T time steps based on the real-time power demand data, the real-time external factor indicators and the power demand time series prediction model, and perform hierarchical calibration on the power demand sequence;

Step S3000: Analyze the power demand change trend of the current operating area in the next T time steps according to the calibrated power demand forecast sequence, and formulate a power grid operation optimization strategy according to the power demand change trend.

Furthermore, the step S1000 includes:

Step S1100, dividing the power grid according to its geographical location, load characteristics and power supply mode to obtain m mutually independent operation areas, denoted as A ₁ , A ₂ , ..., A _m ;

According to the geographical location, load characteristics and power supply mode of the power grid, the method of dividing the power grid into m independent operation areas includes:

Determine whether the geographical location has significant topographical differences. If so, give priority to geographical location for division;

Determine whether the load characteristics are obviously different. If so, prioritize the load characteristics for division;

Determine whether there are significant differences in the power supply methods. If so, give priority to the power supply method for classification;

If the geographical location, load characteristics and power supply mode are not significant, the weight allocation method is used to comprehensively consider the geographical location, load characteristics and power supply mode for division;

Step S1200, obtaining historical power demand data of each operating area, and constructing a regional demand index vector according to the historical power demand data;

Step S1300: Based on the regional demand index vector, each operating area is divided into three power demand levels: high, medium, and low.

Furthermore, the step S1200 includes:

Step S1210, using the historical power demand data of the operating area, extracting the average power consumption of different time scales in the operating area as the power consumption index _Qi of the i-th operating area;

Step S1220, based on the historical power demand data of the operating area, a time window is selected, and a load curve corresponding to the time window is extracted; feature extraction is performed on the load curve to obtain a load curve index _Li of the i-th operating area; the time window includes a typical day, a typical month, and a typical year;

Step S1230, based on the historical power demand data of the operation area, analyzing the power consumption behavior of the power users in the operation area, and constructing the power consumption behavior index U _i ;

Step S1240: Aggregate the power consumption index, the load curve index and the power consumption behavior index into a regional demand index vector V _i =(Q _i ,L _i ,U _i ).

Furthermore, the step S1230 includes:

Step S1231: extracting the electricity demand data and attribute information of each electricity user in the operating area from the historical electricity demand data of the operating area, and making a comprehensive profile of the customer based on the attribute information to obtain an attribute profile of the electricity user;

Step S1232, performing association analysis on the attribute profile of the electricity user and its electricity demand data, and mining association rules between the attribute profile and the electricity demand data;

Step S1233, identifying key electricity consumption behavior patterns that have a significant impact on the electricity demand in the operating area according to the association rules between the attribute portrait and the electricity demand data;

Step S1234: construct an index vector U _i reflecting the characteristics of the power consumption behavior of the operation area based on the identified key power consumption behavior pattern, where U _i is the power consumption behavior index of the i-th operation area.

Furthermore, the step S2000 includes:

Step S2100, obtaining external factor indicators, constructing an external factor indicator vector, and forming a regional power total indicator vector according to the external factor indicator vector and the regional demand indicator vector;

Step S2200, constructing a power demand time series forecasting model based on the regional power total index vector;

Step S2300, collecting real-time power demand data and real-time external factor indicators of the operating area, constructing a real-time regional demand indicator vector and a real-time external factor indicator vector of the operating area, and forming a real-time regional power total indicator vector of the current operating area according to the real-time regional demand indicator vector and the real-time external factor indicator vector;

Step S2400, input the real-time regional power total index vector into the power demand time series forecasting model, obtain the power demand forecast sequence of the current operating area in the next T time steps as the initial power demand forecast sequence; according to the power demand level of the operating area, the initial power demand forecast sequence is graded and calibrated to obtain the calibrated power demand forecast sequence.

Furthermore, the method of performing hierarchical calibration on the initial power demand forecast sequence according to the power demand level of the operating area to obtain the calibrated power demand forecast sequence includes:

;

in:

: Calibrated electricity demand forecast series;

: Initial power demand forecast sequence;

: Dynamic weighting coefficient matrix;

: Power demand level correlation matrix;

: Cross-level demand mean matrix;

The dynamic weight coefficient matrix is obtained by solving the following optimization problem :

Minimize the objective; ;

Constraints: , ;

in:

: Indicates the importance of the predicted value of the g-th time step when calibrating the predicted value of the t-th time step in the future;

: represents the Frobenius norm of the matrix;

T: represents the total number of time steps.

Furthermore, the step S2200 includes:

Step S2210, taking the regional power total index vector as a sample data set, and dividing the sample data set into a training set and a validation set; dividing the sample data into subsequences of fixed length according to the time series, and taking the first T-1 time steps contained in each subsequence as input, and the last time step as the true label;

Step S2220, constructing a power demand time series forecasting model using a neural network model, designing a neural network model structure, wherein the neural network model structure includes an input layer, a plurality of hidden layers, and an output layer; initializing neural network model parameters, including an input weight matrix and a bias vector;

Step S2230, the input subsequence of the training set is sent to the neural network model, and the output of each layer is calculated by forward propagation; the input layer receives the subsequence, outputs the first characteristic information of the power demand, and passes the first characteristic information of the power demand to the hidden layer; the hidden layer continues to extract the second characteristic information of the power demand from the first characteristic information of the power demand by weighted summation and calculation of the activation function, and passes the second characteristic information of the power demand to the next layer; the output layer of the neural network model outputs the predicted value of the power demand for the next T time steps; the mean square error loss function is used to calculate the deviation between the predicted value and the true label;

Step S2240, through the back propagation algorithm, update the parameters of the neural network based on gradient descent to minimize the loss function, use the optimizer to control the step size and direction of the parameter update, repeat steps S2230-S2240 until the performance indicators on the verification set converge to obtain the final power demand timing forecasting model.

Furthermore, the step S3000 includes:

Step S3100, analyzing the power demand change trend of the current operating area in the next T time steps according to the calibrated power demand forecast sequence; when the power demand shows an upward trend, executing step S3200; when the power demand shows a downward trend, executing step S3300; when the power demand shows a stable trend, executing step S3400;

Step S3200, when the power demand shows an upward trend, determine whether the power demand increase rate exceeds the increase rate threshold; if it exceeds the increase rate threshold, start the emergency plan to ensure the smooth operation of the power grid; if it does not exceed the increase rate threshold, optimize the power grid operation mode according to the demand forecast result to improve the power grid operation efficiency; the emergency plan includes increasing power supply and peak load shaving;

Step S3300: when the power demand shows a downward trend, analyze the reasons for the decline in power demand; if it is caused by weather factors, adjust the grid operation mode; if it is caused by major events and holidays, evaluate the impact scope and duration of major events and holidays, formulate response plans, and reduce the impact on grid operation;

Step S3400: when the power demand shows a stable trend, evaluate the sustainability of the stable power demand; if it is expected that the demand will remain stable for a period of time in the future, maintain the current grid operation mode; if it is expected that there will be large fluctuations in the future, formulate a fluctuation plan in advance and make preparations for it;

Step S3500, monitor the real-time operation status of the power grid, evaluate the operation effect of the power grid, and provide real-time feedback.

Furthermore, the calculation method of the rising rate threshold includes:

;

in:

: Rising rate threshold;

: Standard deviation of the rate of change of electricity demand;

: adjustment coefficient of average electricity demand change rate;

: Adjustment coefficient of standard deviation of electricity demand change rate;

: The rate of change of power demand during the jth period;

: The number of time periods used to calculate the rate of change of power demand;

: Average electricity demand change rate.

A smart grid AI joint peak-shaving decision system, which is used to implement the above-mentioned smart grid AI joint peak-shaving decision method, includes:

Regional division module: used to divide the power grid into m independent operating areas according to the geographical location, load characteristics and power supply mode of the power grid; obtain the historical power demand data of each operating area, and construct the regional demand index vector according to the historical power demand data; divide each operating area into three power demand levels of high, medium and low according to the regional demand index vector; m is an integer greater than or equal to 1;

External factor index acquisition module: used to acquire external factor index, construct external factor index vector, and form regional power total index vector according to the external factor index vector and regional demand index vector;

Real-time data collection module: used to collect real-time power demand data and real-time external factor indicators of the operating area, construct the real-time regional demand indicator vector and the real-time external factor indicator vector of the operating area, and form the real-time regional power total indicator vector of the current operating area according to the real-time regional demand indicator vector and the real-time external factor indicator vector;

Electricity demand time series forecasting model construction module: used to construct an electricity demand time series forecasting model based on the regional electricity total index vector;

Power demand time series forecasting module: used to input the real-time regional power total index vector into the power demand time series forecasting model to obtain the power demand forecast sequence of the current operating area in the next T time steps as the initial power demand forecast sequence;

Power demand forecast sequence calibration module: used to perform graded calibration on the initial power demand forecast sequence according to the power demand level of the operating area to obtain a calibrated power demand forecast sequence;

Power grid operation optimization decision module: It is used to analyze the power demand change trend in the current operating area in the next T time steps according to the calibrated power demand forecast sequence, and formulate power grid operation optimization strategy according to the power demand change trend.

An electronic device comprises a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the above-mentioned smart grid AI joint peak regulation decision method is implemented.

A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the above-mentioned smart grid AI joint peak regulation decision method.

Compared with the prior art, the present invention has the following beneficial effects:

The present invention uses artificial intelligence technology to construct a power demand time series forecasting model, which can accurately predict the power demand change trend in multiple time steps in the future. This method not only improves the accuracy of power demand forecasting, but also can dynamically adapt to the actual conditions of different operating areas.

The present invention uses a hierarchical calibration mechanism to correct the initial forecast sequence according to the power demand level to ensure that the forecast results are more in line with actual needs. This calibration mechanism optimizes the dynamic weighting coefficient matrix to make the power demand forecast more refined and personalized, effectively reducing the forecast error.

The present invention considers the impact of external factors (such as weather, major events, etc.) on power demand, constructs an external factor index vector, and combines it with the regional demand index vector to form a comprehensive index for power demand forecasting. This multi-dimensional comprehensive analysis method greatly improves the comprehensiveness and reliability of the forecast.

The present invention formulates a smart grid operation optimization strategy based on the calibrated power demand forecast sequence. According to the rising, falling or stable trend of power demand, corresponding measures are taken, such as launching emergency plans, optimizing operation modes, adjusting power grid configuration, etc., to ensure the stable and efficient operation of the power grid.

The present invention monitors the operation status of the power grid in real time, evaluates the operation effect of the power grid, and makes real-time feedback adjustments. This closed-loop control mechanism not only improves the reliability of power grid operation, but also can respond to emergencies in a timely manner to ensure the safety and stability of the power grid.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG1 is a flow chart of a smart grid AI joint peak load decision method in the present invention;

FIG2 is a flow chart of a method for constructing a regional demand index vector in the smart grid AI joint peak load decision method of the present invention;

FIG3 is a functional module diagram of the smart grid AI joint peak-shaving decision system in the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Example 1

As shown in FIG1 , this embodiment provides a smart grid AI joint peak load decision method, including:

Step S1000: Divide the power grid into m independent operation areas according to the geographical location, load characteristics and power supply mode of the power grid; obtain historical power demand data of each operation area, and construct a regional demand index vector according to the historical power demand data; divide each operation area into three power demand levels of high, medium and low according to the regional demand index vector; m is an integer greater than or equal to 1;

Further, step S1000 includes:

Step S1100, dividing the power grid according to its geographical location, load characteristics and power supply mode to obtain m mutually independent operation areas, denoted as A ₁ , A ₂ , ..., A _m ;

According to the geographical location, load characteristics and power supply mode of the power grid, the method of dividing the power grid into m independent operation areas includes:

Determine whether the geographical location has significant topographical differences. If so, give priority to geographical location for division;

Determine whether the load characteristics are obviously different. If so, prioritize the load characteristics for division;

Determine whether there are significant differences in the power supply methods. If so, give priority to the power supply method for classification;

If the geographical location, load characteristics and power supply mode are not significant, the weight allocation method is used to comprehensively consider the geographical location, load characteristics and power supply mode for division;

The process of dividing the power grid satisfies the following mathematical model:

;

in:

m: indicates the total number of operating areas;

A _i : represents the i-th operating area;

x: represents the sample point in the area;

: represents the center point of the i-th operating area;

: represents the sample point x and the center of the region Distance measurement, you can choose common measurement functions such as Euclidean distance and Mahalanobis distance;

Sample points in a region refer to some representative points selected in a certain operating area of the power grid, which can be used to describe the power demand or other related characteristics of the region. Some key geographical locations can be selected as sample points based on the geographical distribution of the power grid; some typical load points can also be selected as sample points based on the load characteristics in the region, such as large industrial users, commercial areas, residential areas, etc. The center point of the i-th operating area refers to the geometric or characteristic center point of each area after the power grid is divided into multiple independent operating areas.

The geographical location needs to consider the geographical distribution and terrain characteristics of the region, such as mountainous areas, plains, etc. The load characteristics need to be based on the type and change characteristics of the load in the region, such as industrial areas, commercial areas and residential areas. The power supply mode is to analyze the power supply mode of the region, such as centralized power supply or decentralized power supply. Through a comprehensive analysis of these factors, the power grid can be reasonably divided into several independent operating areas, each with independent power demand characteristics. By minimizing the sum of the distances between samples and the center in each area, high similarity of samples within the region and low coupling between regions can be achieved, achieving the optimal effect of regional division. Reasonable regional division can fully consider the differences and independence of different regions, avoid the estimation bias caused by "one size fits all", and provide a more accurate decision-making basis for subsequent prediction and optimization.

Step S1200, obtaining historical power demand data of each operating area, and constructing a regional demand index vector according to the historical power demand data;

Further, referring to FIG. 2 , step S1200 includes:

Step S1210, using the historical power demand data of the operating area, extracting the average power consumption of different time scales in the operating area as the power consumption index _Qi of the i-th operating area;

Step S1220, based on the historical power demand data of the operating area, a time window is selected, and a load curve corresponding to the time window is extracted; feature extraction is performed on the load curve to obtain a load curve index _Li of the i-th operating area; the time window includes a typical day, a typical month, and a typical year;

Step S1230, based on the historical power demand data of the operation area, analyzing the power consumption behavior of the power users in the operation area, and constructing the power consumption behavior index U _i ;

Further, step S1230 includes:

Step S1231: extracting the electricity demand data and attribute information of each electricity user in the operating area from the historical electricity demand data of the operating area, and making a comprehensive profile of the customer based on the attribute information to obtain an attribute profile of the electricity user;

Step S1232, performing association analysis on the attribute profile of the electricity user and its electricity demand data, and mining association rules between the attribute profile and the electricity demand data;

Step S1233, identifying key electricity consumption behavior patterns that have a significant impact on the electricity demand in the operating area according to the association rules between the attribute portrait and the electricity demand data;

Step S1234, based on the identified key electricity consumption behavior pattern, construct an index vector U _i reflecting the electricity consumption behavior characteristics of the operation area, where U _i is the electricity consumption behavior index of the i-th operation area;

Extract attribute information of each electricity user in the operating area from historical power demand data, such as industry category (steel, electrolytic aluminum, chemical industry, etc.), production process (such as continuous production, periodic production), energy efficiency level (such as energy efficiency grade), etc., to make a comprehensive portrait of the customer. This attribute information can usually be obtained from business systems such as power marketing and energy efficiency management. Perform correlation analysis on the attribute portrait of the electricity user and its power demand data to mine the association rules between customer attributes and power demand. For example, it can be found that "the power demand of customers in the steel industry usually presents a continuous and stable characteristic" and "the power consumption of customers in the electrolytic aluminum industry increases significantly during the low electricity price period". Classic algorithms such as Apriori and FP-Growth can be used for association rule mining. According to the mined association rules, identify the key power consumption behavior patterns that have a significant impact on regional power demand. For example, "continuous production in the steel industry" and "cross-power consumption in the electrolytic aluminum industry". These key behavior patterns are usually closely related to customer attributes and can explain the inherent driving mechanism of regional power demand. Based on the identified key power consumption behavior patterns, construct an indicator vector U _i that reflects the characteristics of regional power consumption behavior. For example, indicators such as "number of customers in the steel industry" and "proportion of electricity consumption in the steel industry" can be used to characterize the key behavioral pattern of "continuous production in the steel industry".

Step S1230 can comprehensively and accurately characterize the characteristics of electricity consumption behavior in each region through detailed analysis of electricity users in the operating area. Specifically, by extracting the demand data and attribute information of electricity users, a comprehensive portrait of customers is established, laying the foundation for subsequent association analysis. By correlating customer attribute portraits with electricity demand data, the association rules between the two are mined, revealing the inherent laws of regional power demand. Based on these association rules, key electricity behavior patterns that have a significant impact on power demand are identified, and these patterns can explain the driving factors of regional power demand. Finally, through the identified key behavior patterns, an indicator vector U _i reflecting the characteristics of regional power consumption behavior is constructed, which provides an important basis for subsequent power demand prediction and optimization. Overall, step S1230 makes power demand assessment more accurate and comprehensive through multi-dimensional and multi-level analysis, thereby providing reliable data support and scientific decision-making basis for the joint peak regulation decision of smart grids.

Step S1240, summarizing the power consumption index, load curve index and power consumption behavior index into a regional demand index vector V _i =(Q _i ,L _i ,U _i );

Step S1200 uses the historical power demand data in the operating area to extract the average power consumption at different time scales (year, month, day, hour) to reflect the overall level of demand. The longer the historical data, the more reliable the statistical results. Based on the historical power demand data of the operating area, select time windows such as typical days, typical months, and typical years to extract the corresponding load curve. Use signal processing methods such as Fourier transform and wavelet transform to perform feature analysis on the load curve to obtain a series of feature vectors reflecting the shape, periodicity, and mutation of the curve, such as fundamental component, high-order harmonic component, peak factor, etc. _Li is used to represent the set of load feature vectors of the i-th operating area, and _Li is used as the load curve index of the i-th operating area. "Typical day", "typical month" and "typical year" are concepts used to describe representative power consumption characteristics in a specific time period. They extract typical data that can represent the overall power consumption law in the time period by statistically analyzing power consumption data on different time scales. The load curve index reveals the regular characteristics of regional demand from the perspective of dynamic changes. The richer the historical data, the more comprehensive and accurate the feature extraction. By constructing a multi-element indicator system, the characteristics of the power demand level in the operating area can be characterized from multiple angles such as power consumption, load curve, and power consumption behavior, thereby improving the comprehensiveness and accuracy of the evaluation and providing more abundant and reliable data support for subsequent predictions and optimization.

Step S1300, based on the regional demand index vector, each operating area is divided into three power demand levels: high, medium and low;

Based on the regional demand index vector, the method of dividing each operating area into three power demand levels of high, medium and low includes:

Using the fuzzy C-means algorithm, the objective function is as follows:

;

in:

m: indicates the total number of operating areas;

e: fuzziness index, usually greater than 1;

: The membership degree of the operating area A _i to the power demand level k;

V _i : regional demand index vector of operating area A _i ;

C _k : is the center vector of power demand level k;

: represents Euclidean distance;

k: power demand level, 1≤k≤3;

J _e : objective function value of fuzzy C-means algorithm;

Through the fuzzy C-means algorithm, each operating area is divided into three power demand levels of high, medium and low, and the power demand level _Di of area _Ai is obtained, where _Di values 1, 2, and 3 correspond to low, medium and high power demand levels respectively. The fuzzy C-means algorithm belongs to the common knowledge in the technical field, and the specific process of the fuzzy C-means algorithm is not repeated in this embodiment.

Step S1300 uses the fuzzy C-means algorithm to accurately divide each operating area into three power demand levels: high, medium, and low. The fuzzy C-means algorithm calculates the degree of membership and processes the fuzziness and uncertainty in the data, making the classification results more accurate and flexible. This grading method can accurately reflect the differences in power demand in each region and avoid the deviation caused by the "one-size-fits-all" traditional method. By dividing the power demand levels, the optimal allocation of power grid resources can be achieved and the operating efficiency and stability of the power system can be improved. In addition, this classification also provides more scientific and detailed data support for subsequent power demand forecasting and optimization decisions, making the joint peak-shaving decisions of smart grids more accurate and reliable.

Step S2000: Obtain external factor indicators, and construct a power demand time series prediction model based on the external factor indicators and the regional demand indicator vector; collect real-time power demand data and real-time external factor indicators of the operating area, and predict the power demand sequence of the current operating area in the next T time steps based on the real-time power demand data, the real-time external factor indicators and the power demand time series prediction model, and perform hierarchical calibration on the power demand sequence;

Furthermore, step S2000 includes:

Step S2100, obtaining external factor indicators, constructing an external factor indicator vector, and forming a regional power total indicator vector according to the external factor indicator vector and the regional demand indicator vector;

Furthermore, step S2100 includes:

Step S2110, normalizing the regional demand index vector to obtain a dimensionless standardized regional demand index vector;

Step S2120, obtaining external factor indicators, constructing an external factor indicator vector, and splicing the external factor indicator vector with the standardized regional demand indicator vector to form a regional power total indicator vector; the external factor indicators include weather indicators, holiday indicators, and major event indicators;

Specifically, the Min-Max standardization method is used to standardize the regional demand index vector to obtain a dimensionless standardized regional demand index vector, eliminating the impact of the dimension differences of different indicators. Indicators reflecting external factors such as weather, holidays, and major events are collected from multiple data sources, including but not limited to weather indicators, holiday indicators, and major event indicators; weather indicators, such as temperature, humidity, wind speed, precipitation, etc., which can significantly affect electricity demand, especially under extreme weather conditions. Holiday indicators, such as statutory holidays, weekends, etc., holidays usually lead to significant changes in electricity consumption patterns. Major event indicators, such as major engineering construction, large-scale exhibitions, emergencies, etc., these events may cause sudden changes in electricity demand.

Step S2100 can significantly improve the accuracy of power demand forecasting by standardizing and integrating external factors. The standardized regional demand index vector eliminates the impact of differences in the dimensions of different indicators, allowing each indicator to be compared and calculated on a unified scale. The introduction of external factor index vectors enables the model to consider the impact of factors such as weather changes, holidays and major events on power demand, enhancing the adaptability and predictive ability of the model. This comprehensive processing method not only improves the accuracy of power demand forecasting, but also enables the power system to respond more flexibly to changes in the external environment, thereby improving the stability and operating efficiency of the power grid.

Step S2200, constructing a power demand time series forecasting model based on the regional power total index vector;

Further, step S2200 includes:

Step S2210, taking the regional power total index vector as a sample data set, and dividing the sample data set into a training set and a validation set; dividing the sample data into subsequences of fixed length according to the time series, and taking the first T-1 time steps contained in each subsequence as input, and the last time step as the true label;

Step S2220, constructing a power demand time series forecasting model using a neural network model, designing a neural network model structure, wherein the neural network model structure includes an input layer, a plurality of hidden layers, and an output layer; initializing neural network model parameters, including an input weight matrix and a bias vector;

Step S2230, the input subsequence of the training set is sent to the neural network model, and the output of each layer is calculated by forward propagation; the input layer receives the subsequence, outputs the first characteristic information of the power demand, and passes the first characteristic information of the power demand to the hidden layer; the hidden layer continues to extract the second characteristic information of the power demand from the first characteristic information of the power demand by weighted summation and calculation of the activation function, and passes the second characteristic information of the power demand to the next layer; the output layer of the neural network model outputs the predicted value of the power demand for the next T time steps; the mean square error loss function is used to calculate the deviation between the predicted value and the true label;

Step S2240, updating the parameters of the neural network based on gradient descent through the back propagation algorithm to minimize the loss function, using the optimizer to control the step size and direction of the parameter update, repeating steps S2230-S2240 until the performance indicators on the validation set converge, and obtaining the final power demand time series forecasting model;

Step S2300, collecting real-time power demand data and real-time external factor indicators of the operating area, constructing a real-time regional demand indicator vector and a real-time external factor indicator vector of the operating area, and forming a real-time regional power total indicator vector of the current operating area according to the real-time regional demand indicator vector and the real-time external factor indicator vector;

Step S2400, inputting the real-time regional power total index vector into the power demand time series forecasting model, obtaining the power demand forecast sequence of the current operating area in the next T time steps as the initial power demand forecast sequence; performing graded calibration on the initial power demand forecast sequence according to the power demand level of the operating area, and obtaining the calibrated power demand forecast sequence;

According to the power demand level of the operating area, the initial power demand forecast sequence is graded and calibrated to obtain the calibrated power demand forecast sequence, which includes:

;

in:

: Calibrated electricity demand forecast series;

: Initial power demand forecast sequence;

: Dynamic weighting coefficient matrix, , T represents the total number of time steps, that is is a T×T matrix;

: Power demand level correlation matrix, , is a 3×3 symmetric matrix that describes the correlation between different power demand levels, and R represents a real number set, that is, The elements in the matrix can be any real numbers:

: Cross-level demand mean matrix, , is a 3×T matrix, representing the cross-level conditional means at different power demand levels and time steps, and R represents a set of real numbers;

Dynamic weighting coefficient matrix Elements Indicates the importance of the predicted value at the g-th time step when calibrating the predicted value at the t-th time step in the future. The matrix can be obtained by solving the following optimization problem:

Minimize the objective; ;

Constraints: , ;

in:

: Indicates the importance of the predicted value of the g-th time step when calibrating the predicted value of the t-th time step in the future;

: represents the Frobenius norm of the matrix;

The goal of this optimization problem is to minimize the difference between the calibrated power demand forecast sequence and the initial power demand forecast sequence, while ensuring The sum of each row of the matrix is 1 to maintain the scale of the predicted values.

;

in, represents the correlation coefficient between the ith power demand level and the jth power demand level, .

;

in, It represents the cross-level conditional mean of electricity demand level i at the tth time step in the future, that is, the expected electricity demand of the i-th level at the tth time step when the electricity demand of other levels is known.

Step S2400 introduces a dynamic weighting coefficient matrix λ to assign different levels of importance to the prediction values of different time steps, so as to more flexibly integrate time series information. By solving the λ matrix through the optimization problem, the weights of the prediction values of each time step can be adaptively adjusted to suppress the influence of outliers and improve the robustness of the prediction. The power demand level correlation matrix Ω is introduced to consider the correlation effect between different power demand levels. The Ω matrix can be obtained by statistically analyzing historical data. In real scenarios, there is often a certain correlation between different levels of power demand. For example, the peak power consumption in a high-level demand area may cause an increase in demand in adjacent level areas. By characterizing this correlation effect through the Ω matrix, when calibrating the prediction value, not only the demand characteristics of this level are considered, but also the influence of other levels, which enhances the systematic nature of the prediction. The cross-level demand mean matrix is introduced to correct the prediction value of this level given the power demand of other levels. Established through multivariate regression analysis The matrix can explore the implicit constraints between different power demand levels and dynamically calibrate the predicted value of this level. For example, when the demand in the high-level area is very high, the actual demand in the medium-level area may also be higher than the expected value. The matrix can describe this cross-level conditional mean and make adaptive adjustments to the forecast sequence. The dynamic weighting coefficient matrix λ ensures the time-varying and adaptability of the weighting, and flexibly integrates the two types of information at different time scales, making full use of the intelligence of the prediction model and taking into account the differences in demand levels, thus achieving dynamic optimization of the prediction effect.

The power demand time series forecasting model based on the regional power total index vector can effectively use historical data and external factors to accurately predict future power demand. The standardized regional power total index vector combined with the powerful learning ability of the neural network model can capture complex time series relationships and feature interactions, thereby improving the accuracy and reliability of the forecast. In addition, the forward propagation and back propagation algorithms are used to optimize the model parameters, so that the model can converge quickly, further enhancing the model's forecasting performance. This method can not only provide high-precision power demand forecasts, but also provide important decision support for the operation and dispatch of power systems.

Step S3000: Analyze the power demand change trend of the current operating area in the next T time steps according to the calibrated power demand forecast sequence, and formulate a power grid operation optimization strategy according to the power demand change trend;

Furthermore, step S3000 includes:

Step S3100, analyzing the power demand change trend of the current operating area in the next T time steps according to the calibrated power demand forecast sequence; when the power demand shows an upward trend, executing step S3200; when the power demand shows a downward trend, executing step S3300; when the power demand shows a stable trend, executing step S3400;

The demand change trend can be obtained by performing slope analysis on the forecast sequence. When the slope is positive, it means that the demand is increasing; when the slope is negative, it means that the demand is decreasing; when the slope is close to 0, it means that the demand is stable.

Step S3200, when the power demand shows an upward trend, determine whether the power demand increase rate exceeds the increase rate threshold; if it exceeds the increase rate threshold, start the emergency plan to ensure the smooth operation of the power grid; if it does not exceed the increase rate threshold, optimize the power grid operation mode according to the demand forecast result to improve the power grid operation efficiency; the emergency plan includes increasing power supply and peak load shaving;

When demand rises too fast and exceeds the grid's regulation capacity, it may cause grid instability, so emergency measures need to be taken in a timely manner. The formulation of emergency plans needs to comprehensively consider multiple factors such as power dispatch, load control, and equipment maintenance to ensure the smooth operation of the grid. When optimizing the operation mode of the grid, the demand forecast results can be used to reasonably arrange power supply start and stop, load switching, etc., to improve the economy and reliability of grid operation.

The calculation method of the rise rate threshold includes:

;

in:

: Rising rate threshold;

: The standard deviation of the rate of change of electricity demand reflects the volatility of demand changes. The larger the standard deviation, the more unstable the demand changes;

: adjustment coefficient of average electricity demand change rate;

: Adjustment coefficient of standard deviation of electricity demand change rate;

: The rate of change of power demand during the jth period;

: The number of time periods used to calculate the rate of change of power demand;

: Average electricity demand change rate, which indicates the average value of the electricity demand change rate over a period of time, reflecting the overall trend of demand changes;

The formula provides a dynamically adjusted threshold for the rate of increase in electricity demand by comprehensively considering the average value and volatility of electricity demand changes, thereby improving the accuracy of predictions and the sensitivity of emergency response. The formula can not only more accurately reflect the actual changes in electricity demand, but also help to promptly initiate emergency plans when demand increases and ensure the smooth operation of the power grid. It can also optimize the operation of the power grid when demand changes do not exceed the threshold for the rate of increase, improve operating efficiency and reduce energy waste. In addition, the formula adjusts the coefficient ( and ) can be flexibly adjusted and has strong adaptability. It can be applied to different power grid environments and demand patterns, while maintaining a certain degree of calculation simplicity, which is convenient for actual system implementation.

Step S3300: when the power demand shows a downward trend, analyze the reasons for the decline in power demand; if it is caused by weather factors, adjust the grid operation mode; if it is caused by major events and holidays, evaluate the impact scope and duration of major events and holidays, formulate response plans, and reduce the impact on grid operation;

Step S3400: when the power demand shows a stable trend, evaluate the sustainability of the stable power demand; if it is expected that the demand will remain stable for a period of time in the future, maintain the current grid operation mode; if it is expected that there will be large fluctuations in the future, formulate a fluctuation plan in advance and make preparations for it;

Step S3500, monitor the real-time operation status of the power grid, evaluate the operation effect of the power grid, and provide real-time feedback.

When the demand decline is caused by weather factors, such as moderate climate in spring and autumn, the demand for electricity decreases, the power generation plan can be adjusted accordingly, unnecessary power generation can be reduced, and operating costs can be reduced. When the demand decline is caused by major events and holidays, it is necessary to assess the scope and duration of the event and formulate a response plan. For example, a certain area is about to usher in the national holiday Spring Festival. According to historical data, the demand for electricity during the Spring Festival dropped significantly, because many enterprises and factories stopped work and took holidays, and residents' household electricity consumption also decreased. The grid operator needs to assess the scope and duration of this impact and finds that the impact is mainly concentrated in industrial and commercial areas and is expected to last for a week. In response to this situation, the operator can adjust the power generation plan, reduce unnecessary power generation, reduce operating costs, and formulate a flexible power dispatching strategy to ensure a quick response when demand recovers and ensure the stable operation of the power grid. When it is expected that demand will remain stable for a period of time in the future, the current grid operation mode can be maintained, such as maintaining the current power generation plan, maintenance arrangements, etc. However, considering the uncertainty of power demand, emergency preparations need to be made even in stable periods. When large fluctuations are expected in the future, such as major holidays, extreme weather and other factors that may cause drastic changes in electricity demand, it is necessary to formulate plans for fluctuations in advance, such as preparing sufficient peak-shaving power sources, formulating orderly electricity consumption plans, etc., to improve the power grid's ability to respond to emergencies.

Step S3000 formulates a grid operation optimization strategy based on the calibrated power demand forecast sequence, which can achieve efficient allocation of power resources, avoid waste of power resources and overload of the power grid, and improve the overall operation efficiency of the power grid. By optimizing power supply and scheduling, it can effectively respond to fluctuations in power demand, reduce peak loads, and reduce grid operation costs. At the same time, the optimization strategy can improve the reliability and stability of the power grid, reduce the risk of power outages and the occurrence rate of failures. Through reasonable power demand management and scheduling, it can promote the use of renewable energy and improve the sustainable development capacity of the power grid.

Example 2

This embodiment provides a smart grid AI joint peak load decision system based on the first embodiment, as shown in FIG3 , including:

Regional division module: used to divide the power grid into m independent operating areas according to the geographical location, load characteristics and power supply mode of the power grid; obtain the historical power demand data of each operating area, and construct the regional demand index vector according to the historical power demand data; divide each operating area into three power demand levels of high, medium and low according to the regional demand index vector; m is an integer greater than or equal to 1;

External factor index acquisition module: used to acquire external factor index, construct external factor index vector, and form regional power total index vector according to the external factor index vector and regional demand index vector;

Real-time data collection module: used to collect real-time power demand data and real-time external factor indicators of the operating area, construct the real-time regional demand indicator vector and the real-time external factor indicator vector of the operating area, and form the real-time regional power total indicator vector of the current operating area according to the real-time regional demand indicator vector and the real-time external factor indicator vector;

Electricity demand time series forecasting model construction module: used to construct an electricity demand time series forecasting model based on the regional electricity total index vector;

Power demand time series forecasting module: used to input the real-time regional power total index vector into the power demand time series forecasting model to obtain the power demand forecast sequence of the current operating area in the next T time steps as the initial power demand forecast sequence;

Power demand forecast sequence calibration module: used to perform graded calibration on the initial power demand forecast sequence according to the power demand level of the operating area to obtain a calibrated power demand forecast sequence;

Power grid operation optimization decision module: It is used to analyze the power demand change trend in the current operating area in the next T time steps according to the calibrated power demand forecast sequence, and formulate power grid operation optimization strategy according to the power demand change trend.

Example 3

This embodiment discloses an electronic device, including a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the smart grid AI joint peak-shaving decision method provided above is implemented.

Since the electronic device introduced in this embodiment is an electronic device used to implement the smart grid AI joint peak-shaving decision method in the embodiment of this application, based on the smart grid AI joint peak-shaving decision method introduced in the embodiment of this application, the technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel in this field implement the electronic device used in the smart grid AI joint peak-shaving decision method in the embodiment of this application, it belongs to the scope of protection of this application.

Example 4

This embodiment discloses a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the above-mentioned smart grid AI joint peak-shaving decision method is implemented.

The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters, weights and thresholds in the formula are set by technicians in this field according to actual conditions.

The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired network or a wireless network. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD) or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. The intelligent power grid AI joint peak regulation decision method is characterized by comprising the following steps:

Step S1000: dividing the power grid into m mutually independent operation areas according to the geographic position, load characteristics and power supply modes of the power grid; acquiring historical power demand data of each operation area, and constructing an area demand index vector according to the historical power demand data; dividing each operation area into three power demand levels of high, medium and low according to the area demand index vector; m is an integer greater than or equal to 1;

Step S2000: acquiring external factor indexes, and constructing a power demand time sequence prediction model according to the external factor indexes and the regional demand index vector; collecting real-time power demand data and real-time external factor indexes of an operation area, predicting power demand sequences of T time steps in the future of the current operation area according to the real-time power demand data, the real-time external factor indexes and a power demand time sequence prediction model, and carrying out hierarchical calibration on the power demand sequences;

Step S3000: analyzing the power demand change trend of T time steps in the current operation area according to the calibrated power demand prediction sequence, and formulating a power grid operation optimization strategy according to the power demand change trend;

the step S2000 includes:

Step S2100, obtaining an external factor index, constructing an external factor index vector, and forming a regional power total index vector according to the external factor index vector and the regional demand index vector;

Step S2200, constructing a power demand time sequence prediction model based on the regional power total index vector;

step S2300, collecting real-time power demand data and real-time external factor indexes of an operation area, constructing a real-time area demand index vector and a real-time external factor index vector of the operation area, and forming a real-time area power total index vector of the current operation area according to the real-time area demand index vector and the real-time external factor index vector;

step S2400, inputting a real-time regional power total index vector into a power demand time sequence prediction model to obtain a power demand prediction sequence of T time steps in the future of the current operation region, and using the power demand prediction sequence as an initial power demand prediction sequence; according to the power demand level of the operation area, carrying out hierarchical calibration on the initial power demand prediction sequence to obtain a calibrated power demand prediction sequence;

The method for carrying out hierarchical calibration on the initial power demand prediction sequence according to the power demand level of the operation area to obtain the calibrated power demand prediction sequence comprises the following steps:

；

wherein:

: a calibrated power demand prediction sequence;

: an initial power demand prediction sequence;

: a dynamic weighting coefficient matrix;

: a power demand level correlation matrix;

: a cross-level demand average matrix;

Obtaining a dynamic weighting coefficient matrix by solving the following optimization problem ：

Minimizing the target;；

constraint conditions: ，；

wherein:

: indicating the importance of the predicted value of the nth time step in the future when the predicted value of the nth time step is calibrated;

: a Fr Luo Beini Usness norm representing a matrix;

t: representing the total number of time steps.

2. The smart grid AI joint peak shaver deciding method according to claim 1, wherein the step S1000 includes:

step S1100, dividing the power grid according to the geographic position, the load characteristic and the power supply mode of the power grid to obtain m mutually independent operation areas, and recording the m mutually independent operation areas as A ₁,A₂,…,A_m;

The method for dividing the power grid into m mutually independent operation areas according to the geographic position, the load characteristic and the power supply mode of the power grid comprises the following steps:

Judging whether the geographic positions have obvious terrain differences or not, and if so, preferentially considering the geographic positions for division;

judging whether the load characteristics are obviously different, if so, preferentially considering the load characteristics for division;

judging whether the power supply modes have obvious differences or not, if so, preferentially considering the power supply modes for division;

if the geographic position, the load characteristic and the power supply mode are not obvious, dividing by comprehensively considering the geographic position, the load characteristic and the power supply mode by adopting a method for distributing weights;

Step S1200, acquiring historical power demand data of each operation area, and constructing an area demand index vector according to the historical power demand data;

in step S1300, each operation area is divided into three power demand levels, high, medium and low, based on the area demand index vector.

3. The smart grid AI joint peak shaver deciding method according to claim 2, wherein the step S1200 includes:

step S1210, extracting the average value of the electricity consumption of different time scales in the operation area by using the historical electricity demand data of the operation area as the electricity consumption index Q _i of the ith operation area;

Step S1220, selecting a time window based on the historical power demand data of the operation area, and extracting a load curve corresponding to the time window; extracting characteristics of the load curve to obtain a load curve index L _i of an ith operation area; the time window includes typical days, typical months, and typical years;

Step S1230, based on the historical power demand data of the operation area, analyzing the power consumption behavior of the power consumption clients in the operation area, and constructing a power consumption behavior index U _i;

step S1240, the electricity consumption index, the load curve index and the electricity consumption behavior index are summarized as the region demand index vector V _i=(Q_i,L_i,U_i).

4. The smart grid AI joint peak shaver deciding method according to claim 3, wherein the step S1230 comprises:

step S1231: extracting power consumption demand data and attribute information of each power consumption client in the operation area from historical power demand data of the operation area, and carrying out omnibearing portrait on the client based on the attribute information to obtain attribute portrait of the power consumption client;

s1232, carrying out association analysis on the attribute portrait of the electricity consumer and the electricity demand data thereof, and mining association rules between the attribute portrait and the electricity demand data;

Step S1233, a key electricity utilization behavior mode which has significant influence on the electricity demand of the operation area is identified according to the association rule between the attribute portrait and the electricity demand data;

Step S1234, based on the identified key electricity behavior mode, an index vector U _i,U_i reflecting the electricity behavior characteristics of the operation region is constructed as the electricity behavior index of the ith operation region.

5. The smart grid AI joint peak shaver deciding method according to claim 1, wherein the step S2200 includes:

Step S2210, taking the regional power total index vector as a sample data set, and dividing the sample data set into a training set and a verification set; dividing sample data into subsequences with fixed length according to time sequences, wherein the first T-1 time steps contained in each subsequence are used as input, and the last time step is used as a real tag;

Step S2220, constructing a power demand time sequence prediction model by using a neural network model, and designing a neural network model structure, wherein the neural network model structure comprises an input layer, a plurality of hidden layers and an output layer; initializing neural network model parameters, including an input weight matrix and a bias vector;

Step S2230, the input subsequence of the training set is sent into a neural network model, and the output of each layer is calculated through forward propagation; the input layer receives the subsequence, outputs first characteristic information of the power demand, and transmits the first characteristic information of the power demand to the hidden layer; the hidden layer continuously extracts second characteristic information of the power demand from the first characteristic information of the power demand through weighted summation and calculation of an activation function, and transmits the second characteristic information of the power demand to the next layer; the output layer of the neural network model outputs the predicted value of the power demand of T time steps in the future; calculating the deviation between the predicted value and the real label by using a mean square error loss function;

step S2240, updating the parameters of the neural network based on gradient descent through a back propagation algorithm, minimizing the loss function, controlling the step length and the direction of parameter updating through an optimizer, and repeating the steps S2230-S2240 until the performance index on the verification set converges, thus obtaining the final power demand time sequence prediction model.

6. The smart grid AI joint peak shaver deciding method according to claim 1, wherein the step S3000 comprises:

Step S3100, analyzing the power demand change trend of the current operation area in the future T time steps according to the calibrated power demand prediction sequence; when the power demand shows an ascending trend, step S3200 is performed; when the power demand shows a decreasing trend, step S3300 is performed; when the power demand shows a smooth trend, step S3400 is performed;

Step S3200, when the power demand presents an ascending trend, judging whether the ascending rate of the power demand exceeds an ascending rate threshold; if the rising speed threshold value is exceeded, starting an emergency plan, and guaranteeing the stable operation of the power grid; if the rising speed threshold value is not exceeded, optimizing the power grid operation mode according to the demand prediction result, and improving the power grid operation efficiency; the emergency plan comprises increasing power supply and peak clipping and valley filling;

Step S3300, when the power demand shows a decreasing trend, analyzing the reason of the decreasing power demand; if the weather factor causes, the running mode of the power grid is adjusted; if the event is caused by a major event and a holiday, evaluating the influence range and duration of the major event and the holiday, and making a corresponding scheme to reduce the impact on the operation of the power grid;

Step S3400, when the power demand presents a steady trend, evaluating the sustainability of the power demand steady; if the demand is expected to remain stable for a period of time in the future, maintaining the current power grid operation mode; if larger fluctuation is expected to occur in the future, a fluctuation situation plan is formulated in advance, and preparation for coping is made;

step S3500, monitoring the real-time running state of the power grid, evaluating the running effect of the power grid and feeding back in real time.

7. The smart grid AI joint peak shaver deciding method according to claim 6, wherein the calculating method of the rising rate threshold value comprises:

；

wherein:

: a rise rate threshold;

: standard deviation of the rate of change of power demand;

: an adjustment factor for the rate of change of average power demand;

: an adjustment coefficient of the standard deviation of the power demand change rate;

: the rate of change of power demand during the jth period;

: a number of time periods for calculating a rate of change of the power demand;

: average rate of change of power demand.

8. A smart grid AI joint peak shaver decision system for implementing the smart grid AI joint peak shaver decision method according to any one of claims 1 to 7, characterized by comprising:

Region dividing module: the method is used for dividing the power grid into m mutually independent operation areas according to the geographic position, the load characteristics and the power supply mode of the power grid; acquiring historical power demand data of each operation area, and constructing an area demand index vector according to the historical power demand data; dividing each operation area into three power demand levels of high, medium and low according to the area demand index vector; m is an integer greater than or equal to 1;

The external factor index acquisition module is used for: the method comprises the steps of obtaining external factor indexes, constructing an external factor index vector, and forming a regional power total index vector according to the external factor index vector and a regional demand index vector;

and the real-time data acquisition module is used for: the method comprises the steps of collecting real-time power demand data and real-time external factor indexes of an operation area, constructing a real-time area demand index vector and a real-time external factor index vector of the operation area, and forming a real-time area power total index vector of the current operation area according to the real-time area demand index vector and the real-time external factor index vector;

The power demand time sequence prediction model building module comprises: the power demand time sequence prediction model is used for constructing a power demand time sequence prediction model based on the regional power total index vector;

A power demand timing prediction module: the method comprises the steps of inputting a real-time regional power total index vector into a power demand time sequence prediction model to obtain a power demand prediction sequence of T time steps in the future of a current operation region, and using the power demand prediction sequence as an initial power demand prediction sequence;

the power demand prediction sequence calibration module: for performing a hierarchical calibration of the initial power demand prediction sequence according to the power demand level of the operating region, obtaining a calibrated power demand prediction sequence;

The power grid operation optimization decision module: for analyzing the trend of power demand change of the current operating region by T time steps in the future according to the calibrated power demand prediction sequence, and formulating a power grid operation optimization strategy according to the power demand change trend.

9. An electronic device comprising a memory, a central processor and a computer program stored on the memory and executable on the central processor, characterized in that the central processor implements the smart grid AI joint peak shaver decision method according to any of claims 1-7 when executing the computer program.

10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program, which when executed implements the smart grid AI joint peak shaver decision method according to any one of claims 1-7.