CN118966888A

CN118966888A - Method and system for predicting PUE value of data center

Info

Publication number: CN118966888A
Application number: CN202411028007.4A
Authority: CN
Inventors: 顾乡; 宋大伟; 姚佳
Original assignee: Beijing Longkun Software Technology Co ltd
Current assignee: Beijing Longkun Software Technology Co ltd
Priority date: 2024-07-30
Filing date: 2024-07-30
Publication date: 2024-11-15

Abstract

The invention relates to the technical field of energy efficiency management, in particular to a method and a system for predicting a PUE value of a data center. The method comprises the following steps: performing real-time environmental enthalpy difference processing on a target data center to generate environmental enthalpy difference matrix data; performing air conditioner power consumption load prediction according to the environmental enthalpy difference matrix data to generate air conditioner power consumption prediction curve data; performing PUE value calculation according to the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data, and performing confidence interval correction processing to generate corrected PUE value prediction curve data; and performing intelligent consumption reduction regulation strategy processing according to the modified PUE value prediction curve data to generate intelligent consumption reduction regulation strategy data. According to the invention, the PUE value of the data center is predicted based on the air conditioner power consumption load analysis, and the accuracy and reliability of the PUE prediction are effectively improved through the confidence interval correction.

Description

Method and system for predicting PUE value of data center

Technical Field

The invention relates to the technical field of energy efficiency management, in particular to a method and a system for predicting a PUE value of a data center.

Background

With the rapid development of technologies such as cloud computing, big data and artificial intelligence, a data center is used as a core infrastructure of information technology, and the scale and energy consumption of the data center are continuously increased. The energy consumption of the data center mainly comprises aspects of IT equipment, a refrigerating system, a power supply and distribution system and the like, wherein the refrigerating system occupies a relatively large area. In order to evaluate energy efficiency of a data center, power use efficiency is commonly used as an evaluation index in the industry. The PUE value is an important index for measuring the energy efficiency of the data center, and is defined as the ratio of the total energy consumption of the data center to the energy consumption of IT equipment. The closer the PUE value is to 1, the higher the energy utilization efficiency of the data center. However, the conventional method for predicting the PUE value of the data center generally performs static PUE prediction based on historical data, and it is difficult to accurately reflect the changing environment and load condition of the data center, and meanwhile, there is a problem that influence of environmental factors on the PUE is not considered, resulting in limited prediction accuracy.

Disclosure of Invention

Based on the above, the present invention provides a method and a system for predicting PUE values of a data center, so as to solve at least one of the above technical problems.

To achieve the above object, a method for predicting PUE values for a data center, comprising the steps of:

Step S1: collecting weather forecast data of a target data center to obtain original weather forecast data; performing environmental air enthalpy value processing according to the original weather forecast data to generate environmental enthalpy value prediction time sequence data; performing real-time environmental enthalpy difference processing on the target data center through the environmental enthalpy value prediction time sequence data to generate environmental enthalpy difference matrix data;

step S2: acquiring historical operation data; establishing a mapping relation between the power consumption of the air conditioning system and the environmental enthalpy difference according to the historical operation data, and constructing an air conditioning power consumption prediction model; transmitting the environmental enthalpy difference matrix data to an air conditioner power consumption prediction model to perform air conditioner power consumption load prediction, and generating air conditioner power consumption prediction curve data;

step S3: carrying out IT equipment operation load distribution on preset business load data, carrying out equipment time sequence power consumption analysis based on historical operation data, and generating IT equipment power consumption prediction time sequence data; constructing a machine room thermodynamic propagation network according to the IT equipment power consumption prediction time sequence data, and performing potential energy-saving space positioning according to the air conditioner power consumption prediction curve data to obtain potential energy-saving space positioning data;

step S4: calculating a PUE value of a prediction time point according to the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data, and performing confidence interval correction processing to generate corrected PUE value prediction curve data; performing out-of-standard node identification according to the modified PUE value prediction curve data to obtain out-of-standard PUE value node data;

Step S5: and performing intelligent consumption reduction regulation and control strategy processing on the node data of the exceeding PUE value through the potential energy-saving space positioning data to generate intelligent consumption reduction regulation and control strategy data.

According to the method, the accuracy of response to environmental changes is improved by collecting and processing weather forecast data of a target data center place, the environmental air enthalpy value in a future time period is calculated by utilizing the obtained original weather forecast data and combining with a calculation formula of the air enthalpy value, enthalpy difference change between the external environment and the internal environment of the data center can be accurately reflected, environmental enthalpy value forecast data is compared with environmental parameters monitored in the data center in real time, and heat dissipation demand differences of different areas and time periods of the data center can be more intuitively reflected. And constructing a mapping model between the power consumption of the air conditioning system and the environmental enthalpy difference by utilizing the historical operation data, wherein the model can predict the expected energy consumption of the air conditioning system under different enthalpy difference conditions, so that the energy use is planned and optimized in advance, and unnecessary energy waste is reduced. By applying the environmental enthalpy difference matrix data to this model, a future power consumption prediction curve of the air conditioning system can be obtained. Based on the preset service load and the historical operation data analysis, the data center manager is helped to accurately grasp the energy consumption trend of future IT equipment. The constructed machine room thermodynamic propagation network model is combined with air conditioner power consumption prediction data, so that the temperature distribution in the data center can be predicted, the machine room thermodynamic propagation network can be constructed according to the IT equipment power consumption prediction data, and the heat transfer process in the data center can be simulated. By combining the air conditioner power consumption prediction curve data, the area with supercooling or uneven heat dissipation in the data center can be identified, namely the potential energy-saving space. The PUE value (energy use efficiency) of the predicted time point is calculated by comprehensively considering the predicted energy consumption of the air conditioning system and the IT equipment, confidence interval correction is carried out, the accuracy and reliability of a predicted result are improved, and the identified out-of-standard PUE value node data directly indicate a problem area in the energy consumption management of the data center. By utilizing potential energy-saving space positioning data, a personalized energy-saving control scheme is designed and implemented aiming at the identified out-of-standard PUE value node, so that the whole energy consumption of the data center is effectively reduced, the energy utilization efficiency is improved, the operation sustainability of the data center is further promoted, the operation cost is reduced, and the competitiveness is enhanced. Therefore, the method for predicting the PUE value of the data center carries out environmental air enthalpy difference value processing through future weather forecast data of the data center, establishes a mapping relation between the power consumption of the air conditioning system and the environmental enthalpy difference by considering historical operation data, and realizes the prediction of the power consumption load of the future air conditioner; and carrying out business load power consumption on the IT equipment, and combining the air conditioner predicted power consumption load data to realize the calculation of the PUE predicted value of the data center, and formulating a corresponding intelligent consumption reduction regulation strategy according to the predicted PUE value, so that the PUE value of the data center is effectively reduced, and the aims of energy conservation and emission reduction are realized.

Preferably, step S1 comprises the steps of:

Step S11: performing geographic position positioning processing on a target data center to generate target area positioning data;

Step S12: acquiring future weather forecast data of a week per hour granularity based on the target area positioning data, thereby obtaining original weather forecast data;

step S13: carrying out wet air enthalpy value calculation element identification according to the original weather forecast data to generate key weather forecast element data;

Step S14: carrying out hour-level environmental air enthalpy value calculation according to the key weather forecast element data to obtain environmental enthalpy value prediction time sequence data;

step S15: extracting environmental parameters of an air inlet and an air outlet in real time from a target data center through preset temperature monitoring point data, calculating the enthalpy value of the air in real time, and generating real-time monitoring point environmental enthalpy value data;

Step S16: environmental enthalpy difference value calculation is carried out on environmental enthalpy value data of the real-time monitoring points by utilizing environmental enthalpy value prediction time sequence data, and enthalpy difference time sequence matrix data is generated; and performing matrix transposition on the enthalpy difference time sequence matrix data to obtain environment enthalpy difference matrix data.

According to the invention, the target data center is accurately positioned in the geographic position, so that the pertinence and the accuracy of the follow-up meteorological data acquisition are ensured. Based on the accurate positioning of the target area, weather forecast data of the granularity of each hour of the future week is obtained, the short-term weather change trend can be reflected more finely, and more accurate external condition information is provided for a data center. The identified key weather forecast elements (such as temperature, humidity, etc.) are the core parameters for calculating the enthalpy value of the humid air. The hour-level environmental air enthalpy value prediction time sequence data obtained through calculation of the key meteorological elements can reflect the change trend of the environmental air enthalpy value in a period of time in the future. The environmental parameters of the air inlet and the air outlet are monitored in real time, and the enthalpy value is calculated, so that the real-time performance of the environmental monitoring of the data center is enhanced, the actual heat load condition in the data center is obtained, and the actual change condition of the environmental enthalpy value in the data center can be reflected. By utilizing the predicted environmental enthalpy value data and the real-time monitoring point enthalpy value to calculate the enthalpy difference, the data center can instantly know the deviation degree of future prediction and the current state, for example, if the predicted enthalpy difference is obviously increased, the cooling is required to be increased in the future to avoid overheating, otherwise, the cooling is reduced to save energy.

Preferably, step S2 comprises the steps of:

step S21: acquiring historical operation data through a target data center, wherein the historical operation data comprises historical air conditioner operation data, historical PUE data and historical IT equipment data;

Step S22: performing operation state layering processing according to the historical air conditioner operation data to generate historical air conditioner layering state data;

Step S23: establishing a mapping relation between the power consumption of the air conditioning system and the environmental enthalpy difference based on the historical air conditioning layering state data, and generating enthalpy difference-power consumption mapping relation data;

Step S24: constructing an air conditioner load prediction model for enthalpy difference-power consumption mapping relation data through a preset multiple nonlinear regression model to obtain an initial air conditioner power consumption prediction model;

Step S25: performing model fitting optimization on the initial air conditioner power consumption prediction model by using the historical air conditioner layering state data to generate an air conditioner power consumption prediction model;

Step S26: and transmitting the environmental enthalpy difference matrix data to an air conditioner power consumption prediction model to perform air conditioner system power consumption load prediction, and generating air conditioner power consumption prediction curve data.

The historical operation data of the target data center collected by the invention covers information of various aspects such as an air conditioning system, a PUE value, IT equipment and the like. By layering historical air conditioner operation data, energy consumption modes in different operation states can be clearly distinguished, and the refinement is helpful for identifying efficiency characteristics of the air conditioning system under different working conditions. Enthalpy difference-power consumption mapping relation data constructed based on historical air conditioner layering state data are directly related to the change of the power consumption of an air conditioning system and the enthalpy difference of an external environment, and complex environmental influences are converted into key conversion of operational model parameters. The complex nonlinear relation between the energy consumption of the air conditioning system and a plurality of variables can be captured and expressed by modeling the enthalpy difference-power consumption mapping relation by adopting a multi-element nonlinear regression model, and compared with a simple linear model, the complex nonlinear relation can provide prediction capability which is closer to reality, and the prediction precision of the model is enhanced. The initial model is fitted and optimized through the historical air conditioner layering state data, so that the prediction model can be better adapted to the actual running condition of the data center air conditioning system, model errors are reduced, and the generalization capability and the prediction stability of the model are improved. The environmental enthalpy difference matrix data is applied to the optimized air conditioner power consumption prediction model, the predicted energy consumption trend of the air conditioner system in a future period is provided, the energy use strategy is planned and adjusted in advance, and the energy efficiency is maximized.

Preferably, step S22 comprises the steps of:

Step S221: performing data cleaning on the historical air conditioner operation data to obtain cleaning historical air conditioner operation data;

Step S222: performing time sequence characteristic association processing according to the cleaning historical air conditioner operation data to respectively obtain historical air conditioner system power consumption characteristic data and contemporaneous historical meteorological environment characteristic data;

step S223: labeling running state of the historical air conditioning system power consumption characteristic data to generate labeled running state data;

Step S224: performing two-dimensional space temperature and humidity mapping processing based on contemporaneous historical meteorological environment characteristic data to generate historical temperature and humidity interval data;

Step S225: carrying out temperature and humidity interval screening on the labeled operation state data through historical temperature and humidity interval data to obtain preliminary layered operation data;

Step S226: and carrying out interval data merging processing on the power consumption characteristic data of the historical air conditioning system through a preset data quantity threshold based on the preliminary layered operation data to generate historical layered state data.

According to the invention, the historical air conditioner operation data is subjected to data cleaning, so that errors, repeated or inconsistent records in the historical air conditioner operation data are removed, and the accuracy and the effectiveness of subsequent analysis are ensured. Through carrying out time sequence characteristic association processing on the cleaned data, the internal connection between the operation of the air conditioning system and the external environment condition is revealed. The running state label marking is carried out on the power consumption characteristic data of the air conditioning system, the abstract running data are converted into the information with the clear state identification, the energy consumption modes of the air conditioning system under different running states can be analyzed conveniently, and the data interpretation and the practicability are enhanced. Through the two-dimensional space temperature and humidity mapping processing of the historical meteorological environment characteristic data, the formed temperature and humidity interval data intuitively shows the distribution situation of meteorological conditions, and is beneficial to identifying temperature and humidity combinations with the most obvious influence on the energy consumption of an air conditioning system. The air conditioner operation state is subdivided based on the environmental conditions, so that the air conditioner operation characteristics in different environmental intervals can be deeply understood. And the data quantity threshold value is set to combine the preliminary layered data, so that the data sparseness problem caused by too thin layering is avoided while the data distinction degree is maintained.

Preferably, step S3 comprises the steps of:

step S31: extracting IT device basic data from the historical operation data to generate historical IT basic device data, wherein the historical IT basic device data comprises IT device layout data and historical IT device power consumption characteristic data;

step S32: performing digital processing according to the IT equipment layout data to obtain digital twin machine room data;

step S33: carrying out IT equipment operation load distribution on preset business load data, carrying out cycle time sequence power consumption analysis according to historical IT equipment power consumption characteristic data, and generating IT equipment power consumption prediction time sequence data;

Step S34: carrying out time sequence azimuth power consumption mapping processing on the IT equipment power consumption time sequence data through the digital twin machine room data to generate heat source space distribution matrix data;

Step S35: carrying out heat conduction dynamic simulation on the digital twin machine room data based on the heat source space distribution matrix data, thereby constructing a machine room thermodynamic propagation network;

Step S36: and comprehensively grading the refrigerating energy efficiency of the machine room thermodynamic propagation network through the air conditioner power consumption prediction curve data, and performing potential energy-saving space positioning to obtain potential energy-saving space positioning data.

The invention extracts IT equipment basic information from the historical operation data and reflects the configuration and operation state of the IT equipment in the data center. Through the digital processing of the IT equipment layout data, digital twin machine room data is created, and the virtual model can highly simulate the structure and equipment distribution of the entity data center and provide a virtual environment for subsequent heat distribution simulation and energy saving analysis. According to preset business load data, work load distribution is carried out on the IT equipment, the running state of each equipment is determined, the running state of the IT equipment can directly influence the power consumption of the IT equipment, and accurate load distribution can improve the accuracy of the power consumption prediction of the IT equipment. And carrying out cycle time sequence power consumption analysis based on the historical IT equipment power consumption characteristic data to reflect the power consumption change trend of the IT equipment in a future period of time. And the digital twin machine room data is utilized to carry out space-time mapping on the IT equipment power consumption time sequence data, so that the heat energy distribution condition in the data center is intuitively displayed. The thermal transmission network constructed in the digital twin machine room is dynamically simulated through heat conduction, the actual transmission process of heat in the data center is simulated, and the dynamic model can accurately predict the temperature distribution in the machine room. Comprehensive refrigeration energy efficiency scoring is carried out on a machine room thermodynamic propagation network through air conditioner power consumption prediction curve data, the operation efficiency of a refrigeration system of a data center is reflected, potential energy-saving space is identified according to the comprehensive refrigeration energy efficiency scoring result, and a target area can be provided for an intelligent consumption reduction regulation strategy.

Preferably, step S36 comprises the steps of:

step S361: setting a heat source boundary condition according to a machine room thermodynamic propagation network, and generating heat source boundary condition data;

Step S362: carrying out time sequence step load characteristic decomposition on the air conditioner power consumption prediction curve data to generate time sequence air conditioner load characteristic data;

Step S363: setting air-conditioning supply air flow conditions according to the time sequence air-conditioning load characteristic data to obtain air flow cooling condition data;

Step S364: carrying out time sequence airflow numerical simulation by utilizing digital twin machine room data through heat source boundary condition data and airflow cooling condition data, and carrying out finite element diffusion analysis to obtain machine room time sequence airflow field data;

step S365: calculating the temperature difference of the cold and hot channels according to the time sequence airflow field data of the machine room, analyzing the temperature distribution of the machine room, and generating cooling efficiency data of the machine room;

step S366: performing flow field topology analysis according to the machine room time sequence airflow field data, performing equipment winding flow degree evaluation, and generating machine room refrigeration efficiency evaluation data;

step S367: and comprehensively grading the digital twin machine room data based on the machine room cooling efficiency data and the machine room refrigerating efficiency evaluation data, and performing potential energy-saving space positioning to obtain potential energy-saving space positioning data.

The heat source boundary condition data set by the invention provides an accurate starting point for air flow and heat conduction simulation, and ensures that the simulation process can accurately reflect the actual distribution condition of the heat source in the data center. The time sequence air conditioner load characteristic data can reflect the running state and load change condition of the air conditioner system in different time periods. The air flow cooling condition data is obtained according to the air flow condition setting of the air conditioner, and specific parameter setting is provided for simulating how the air flow effectively takes heat and reducing the temperature of a machine room. Time sequence airflow numerical simulation and finite element diffusion analysis are carried out by using heat source boundary condition data and airflow cooling condition data, and dynamic changes of airflow in a data center are depicted in detail. The cold and hot channel temperature difference calculation and the machine room temperature distribution analysis performed according to the machine room time sequence airflow field data directly reflect the execution effect of the current refrigeration strategy, and provide quantitative indexes for evaluating the refrigeration efficiency and guiding the temperature control strategy adjustment. Specific conditions of airflow flowing around the equipment are disclosed, airflow obstruction and hot spot areas are facilitated to be identified, and the method has important effects of optimizing equipment layout and reducing ineffective refrigeration. According to the comprehensive refrigeration energy efficiency scoring result, a potential energy-saving space is identified, the current refrigeration efficiency of the data center is quantitatively evaluated, and the space capable of realizing energy saving through measures such as layout improvement, air flow management optimization and the like is pointed out.

Preferably, step S4 comprises the steps of:

Step S41: performing time sequence alignment processing on the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data, and performing time point total power consumption calculation to obtain total power consumption time sequence data of the data center;

step S42: calculating a PUE value of a prediction time point of the IT equipment power consumption prediction time sequence data through the total power consumption time sequence data of the data center, and generating predicted PUE instantaneous sequence data;

step S43: performing time sequence value smoothing on the predicted PUE instantaneous sequence data to generate circumferential trend PUE value prediction curve data;

Step S44: performing confidence interval correction processing on the circumferential trend PUE value prediction curve data through historical operation data to generate corrected PUE value prediction curve data;

step S45: and carrying out node-by-node scanning comparison on the corrected PUE value prediction curve data through a preset PUE target threshold value, and marking curve nodes which are larger than or equal to the preset PUE target threshold value as out-of-standard PUE value node data.

According to the invention, the time sequence alignment processing is carried out on the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data, so that the calculated total power consumption time sequence data of the data center can accurately reflect the energy consumption overall view of each time point. The predicted PUE instantaneous sequence data obtained through calculation is directly reflected by comprehensively analyzing the total power consumption time sequence data of the data center and the IT equipment power consumption prediction time sequence data, so that the energy use efficiency of the data center at different time points is directly reflected. And (3) carrying out time sequence value smoothing processing on the predicted PUE instantaneous sequence data, eliminating random fluctuation in the predicted PUE instantaneous sequence data, and reflecting the actual periodical periodic variation trend. Confidence interval correction is carried out by utilizing historical operation data, so that the reliability of the corrected PUE value prediction curve data is improved, the prediction result is ensured to be more stable in statistical sense, and the accuracy of the PUE value is improved. And a comprehensive physical examination of the energy efficiency performance of the data center can quickly identify the period of low energy use efficiency in the future.

Preferably, step S44 includes the steps of:

step S441: extracting historical PUE value characteristics from the historical operation data to obtain historical PUE value characteristic data;

Step S442: carrying out distribution statistical analysis according to the historical PUE value characteristic data to obtain PUE characteristic distribution data;

step S443: performing time sequence node confidence interval calculation on the circumferential trend PUE value prediction curve data through the PUE feature distribution data based on a preset PUE reference value to obtain dynamic PUE value confidence interval data;

step S444: performing PUE value anomaly detection on the circumferential trend PUE value prediction curve data through dynamic PUE value confidence interval data to generate abnormal PUE value node data;

Step S445: and carrying out time window PUE value correlation analysis on the abnormal PUE value node data through the circumferential trend PUE value prediction curve data, and carrying out abnormal PUE value correction processing to obtain corrected PUE value prediction curve data.

The invention discloses the internal law and the characteristic of the change of the PUE value of the data center along with time by deep mining of the historical operation data. Distribution statistical analysis based on historical PUE value characteristic data reveals the distribution range and the concentration trend of the PUE values, and is helpful for understanding the normalcy and the abnormal limit of the change of the PUE values. And utilizing the PUE characteristic distribution data and the dynamic PUE value confidence interval data calculated by the preset PUE reference value to endow each time sequence node on the prediction curve with a reliability index. And (3) carrying out PUE value anomaly detection on the circumferential trend PUE value prediction curve data, wherein marked abnormal PUE value node data directly point to an energy efficiency inefficiency period caused by prediction model deviation, data anomaly or actual operation problem. The time window PUE value correlation analysis can judge whether the abnormal PUE value node is caused by data fluctuation or real running state change, and the abnormal PUE value correction processing not only corrects the error of the prediction model, but also improves the fit degree of the prediction result and the actual situation, and ensures the accuracy and the practicability of the PUE value prediction.

Preferably, step S5 comprises the steps of:

step S51: performing actual running risk assessment on the data center through the over-standard PUE value node data to generate PUE over-standard risk assessment data;

Step S52: performing out-of-standard risk judgment according to the PUE out-of-standard risk assessment data, and continuously executing a current operation regulation strategy by the data center when the PUE out-of-standard risk assessment data is lower than a preset risk threshold value;

Step S53: when the PUE exceeding risk assessment data is higher than or equal to a preset risk threshold, carrying out consumption reduction target formulation on the node data of the exceeding PUE value, and generating predicted PUE value consumption reduction target data;

step S54: carrying out key consumption reduction parameter identification on potential energy-saving space positioning data by using predicted PUE value consumption reduction target data, and carrying out consumption reduction parameter sensitivity analysis to obtain key consumption reduction adjustment parameter data;

Step S55: and carrying out parameter combination adjustment according to the key consumption reduction adjustment parameter data, and carrying out multi-objective consumption reduction optimization treatment, thereby obtaining the PUE intelligent regulation strategy.

The invention provides a quantitative evaluation standard for the operation safety of a data center through risk evaluation on the node data with the exceeding PUE value. According to the comparison of the PUE exceeding risk assessment data and the preset risk threshold, the automatic judgment and maintenance of the operation regulation strategy of the data center are realized, the existing strategy is continued when the energy efficiency is good, and the resource waste caused by unnecessary adjustment is avoided. When the PUE value is identified to have the exceeding risk, a consumption reduction target is formulated through analysis of the exceeding nodes, a clear direction and a quantification target are provided for a subsequent energy saving strategy, and more effective improvement measures are promoted. Through key consumption reduction parameter identification and sensitivity analysis, not only are main factors influencing the PUE value clear, but also the sensitivity of the factors to energy efficiency improvement is disclosed. The parameter combination adjustment and optimization are carried out through the multi-objective optimization algorithm, so that the optimization of a single parameter is considered, the mutual influence among a plurality of consumption reduction targets is balanced, the comprehensive optimality of a regulation strategy is ensured, and the data center is facilitated to realize the remarkable reduction of energy consumption and the continuous improvement of energy efficiency ratio while the service performance is ensured.

The present invention also provides a prediction system for a PUE value of a data center, performing the prediction method for a PUE value of a data center as described above, the prediction system for a PUE value of a data center comprising:

the environment enthalpy value preprocessing module is used for acquiring weather forecast data of the target data center to obtain original weather forecast data; performing environmental air enthalpy value processing according to the original weather forecast data to generate environmental enthalpy value prediction time sequence data; performing real-time environmental enthalpy difference processing on the target data center through the environmental enthalpy value prediction time sequence data to generate environmental enthalpy difference matrix data;

The air conditioner power consumption prediction module is used for acquiring historical operation data; establishing a mapping relation between the power consumption of the air conditioning system and the environmental enthalpy difference according to the historical operation data, and constructing an air conditioning power consumption prediction model; transmitting the environmental enthalpy difference matrix data to an air conditioner power consumption prediction model to perform air conditioner power consumption load prediction, and generating air conditioner power consumption prediction curve data;

the energy-saving space positioning module is used for carrying out IT equipment operation load distribution on preset business load data, carrying out equipment time sequence power consumption analysis based on historical operation data and generating IT equipment power consumption prediction time sequence data; constructing a machine room thermodynamic propagation network according to the IT equipment power consumption prediction time sequence data, and performing potential energy-saving space positioning according to the air conditioner power consumption prediction curve data to obtain potential energy-saving space positioning data;

the PUE value prediction module is used for calculating a PUE value at a prediction time point according to the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data, and performing confidence interval correction processing to generate corrected PUE value prediction curve data; performing out-of-standard node identification according to the modified PUE value prediction curve data to obtain out-of-standard PUE value node data;

the intelligent consumption reduction strategy generation module is used for performing intelligent consumption reduction regulation strategy processing on the node data of the exceeding PUE value through the potential energy-saving space positioning data to generate intelligent consumption reduction regulation strategy data.

Drawings

FIG. 1 is a flow chart of the steps of the method for predicting PUE values for a data center of the present invention;

FIG. 2 is a detailed flowchart illustrating the implementation of step S1 in FIG. 1;

FIG. 3 is a flowchart illustrating the detailed implementation of step S3 in FIG. 1;

FIG. 4 is a schematic diagram of the PUE value prediction and consumption reduction flow of the data center;

The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

The following is a clear and complete description of the technical method of the present invention, taken in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

To achieve the above object, referring to fig. 1 to 4, the present invention provides a method for predicting PUE values of a data center, comprising the steps of:

In the embodiment of the present invention, as described with reference to fig. 1, the step flow diagram of the method for predicting PUE values of a data center of the present invention is shown, and in the embodiment, the method for predicting PUE values of a data center includes the following steps:

In the embodiment of the invention, the longitude and latitude information of the position of the data center is utilized to call a public weather data interface (such as a Chinese weather network API) to acquire weather forecast data of a period of time (for example, 72 hours) in the future. The data content includes, but is not limited to, temperature, humidity, air pressure, etc. And according to the acquired weather forecast data, calculating the environmental air enthalpy value at a corresponding time point by using a standard enthalpy value calculation formula or calling a professional enthalpy value calculation library (such as PsychrometricChart library). And arranging the calculated environmental air enthalpy values at all time points according to a time sequence to form environmental enthalpy value prediction time sequence data. And acquiring an air enthalpy value in the data center, and calculating a difference value between the air enthalpy value and a value of a corresponding time point in the environmental enthalpy value prediction time sequence data to obtain an environmental enthalpy difference value of each time point. And arranging the calculated environmental enthalpy difference values at each time point according to a time sequence, and dividing according to specific areas of the data center to form environmental enthalpy difference matrix data.

In the embodiment of the invention, the power consumption data of the air conditioning system in a historical period (for example, the past month) and the environmental enthalpy difference data of the corresponding period are obtained from the monitoring system of the data center. The data content includes, but is not limited to, total power consumption of the air conditioning system, power consumption of each air conditioning unit, environmental enthalpy difference of each area of the data center, and the like. Air conditioning system power consumption data for a historical period of time (for example, the past month) and environmental enthalpy difference data for the corresponding period of time are obtained from a monitoring system of a data center. The data content includes, but is not limited to, total power consumption of the air conditioning system, power consumption of each air conditioning unit, environmental enthalpy difference of each area of the data center, and the like. And inputting the environmental enthalpy difference matrix data into a constructed air conditioner power consumption prediction model to predict the power consumption load of each time point data center air conditioner system in a future period of time.

In the embodiment of the invention, the environmental enthalpy difference matrix data generated in the step S1 are input into a constructed air conditioner power consumption prediction model to predict the power consumption load of each time point data center air conditioner system in a future period of time. Based on historical operation data, performing time sequence analysis on the power consumption of each IT device by using a statistical analysis method or a machine learning method, establishing a relation model between the power consumption of the IT device and a business load, and predicting the power consumption of each IT device at each time point in a future period. And arranging the predicted power consumption data of each IT device at each time point according to a time sequence to form IT device power consumption prediction time sequence data. And inputting the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data into a machine room thermodynamic propagation network model for simulation calculation, analyzing the temperature distribution conditions of different areas, identifying the area with too low temperature or larger temperature fluctuation, and positioning the area as a potential energy-saving space.

According to the embodiment of the invention, the PUE value of each time point in the predicted time period is calculated according to the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data. And taking uncertainty of model prediction into consideration, carrying out confidence interval correction on the PUE value obtained by prediction by using a statistical method (such as a Bootstrap method) to obtain corrected PUE value prediction curve data. And identifying nodes exceeding the target value in the modified PUE value prediction curve data according to the PUE target value set by the data center, and marking the nodes as out-of-standard PUE value node data.

In the embodiment of the invention, the identified out-of-standard PUE value node data is analyzed to determine the out-of-standard reasons, such as low efficiency of an air conditioning system, excessive load of IT equipment, unreasonable airflow organization and the like. And formulating a targeted intelligent consumption reduction regulation strategy according to the exceeding reason and the potential energy-saving space positioning data. Comprising the following steps: the method aims at solving the problems of low efficiency of an air conditioning system, excessive load of IT equipment and unreasonable air flow organization. And converting the formulated regulation strategy into executable instructions or parameter configuration to generate intelligent consumption reduction regulation strategy data.

Preferably, step S1 comprises the steps of:

As an example of the present invention, referring to fig. 2, a detailed implementation step flow diagram of step S1 in fig. 1 is shown, where step S1 includes:

In the embodiment of the invention, the geographic position information of the target data center is obtained from the data center management system or related documents, and the geographic position information comprises longitude and latitude coordinates, for example, the longitude and latitude data can be queried from a basic information table of the data center. And converting the acquired longitude and latitude coordinate information into a standard geographic information format, such as GeoJSON format, and storing the standard geographic information format as a target area positioning data file.

In the embodiment of the invention, the weather forecast data of the granularity per hour of the future week is acquired by calling a public weather forecast API (such as a Chinese weather network API) by utilizing the target area positioning data. When the API is called, corresponding request parameters including longitude and latitude coordinates of the target area, a predicted time range (a week in the future), a predicted time granularity (per hour), and the like need to be set. The weather forecast API usually returns data in JSON or XML format, and needs to parse the data using a corresponding parsing library to extract the required weather forecast elements, such as temperature, humidity, air pressure, etc. And arranging the weather forecast elements of each time point obtained through analysis according to a time sequence to form original weather forecast data.

In the embodiment of the invention, key weather forecast elements required for calculation are identified according to a calculation formula of the enthalpy value of the humid air, and the key weather forecast elements generally comprise dry bulb temperature, relative humidity and air pressure. Extracting dry bulb temperature, relative humidity and air pressure data at corresponding time points from the original weather forecast data, for example, extracting dry bulb temperature values of each hour from the original data, wherein the units are degrees celsius (DEG C); extracting relative humidity values per hour in percent (%) from the raw data; and inquiring a standard barometer according to the altitude of the target data center, or calling a weather service API to obtain an actual barometric pressure value of corresponding time, wherein the unit is hundred Pa (hPa). The extracted key weather forecast elements are arranged according to the time sequence to form a new data table or file as key weather forecast element data.

In the embodiment of the present invention, a suitable calculation formula of the enthalpy value of the humid air is selected or a specialized calculation library of the enthalpy value (such as PsychrometricChart library) is called, for example, the formula provided in ASHRAEHandbook may be selected for calculation. Substituting the key weather forecast element data into a selected enthalpy value calculation formula or library function, and calculating to obtain the ambient air enthalpy value of each hour. And arranging the calculated environmental air enthalpy values at each time point according to a time sequence to form environmental enthalpy value prediction time sequence data, and storing the environmental air enthalpy value prediction time sequence data into a time sequence data format, wherein the time sequence data comprises two columns of data of a time stamp and the environmental enthalpy value corresponding to the time.

In the embodiment of the invention, real-time data of preset temperature monitoring points are acquired from a data center monitoring system, wherein the real-time data comprise dry bulb temperature, relative humidity, air pressure and the like of each monitoring point. These monitoring points are typically located at key locations in the data center, such as air intakes, outlets, and the like. Substituting the dry bulb temperature, the relative humidity and the air pressure data of each monitoring point acquired in real time into a calculation formula or a library function, and calculating to obtain the ambient air enthalpy value of the current time of each monitoring point. And correlating the calculated real-time environmental air enthalpy value of each monitoring point with the corresponding position information thereof to form real-time monitoring point environmental enthalpy value data.

In the embodiment of the invention, environmental enthalpy value prediction time sequence data and environmental enthalpy value data of real-time monitoring points are matched, and enthalpy difference values between each monitoring point and the environment at the same time point are calculated. For example, 14: environmental enthalpy value of 00 and 14: and 00, calculating the difference value of the environmental enthalpy value obtained by each monitoring point in real time. The enthalpy difference values of each monitoring point at each time point are arranged according to time sequence to form a matrix, each row represents one monitoring point, and each column represents the enthalpy difference value of one time point, namely enthalpy difference time sequence matrix data. And performing matrix transposition operation on the generated enthalpy difference time sequence matrix data to obtain environmental enthalpy difference matrix data, namely, each row represents a time point, and each column represents an enthalpy difference value of a monitoring point.

Preferably, step S2 comprises the steps of:

In embodiments of the present invention, the time frame of historical operating data is obtained, such as data from the past year or from the past quarter. A monitoring system, database or other data storage system connected to the target data center, and stored in a unified data format. Wherein, historical air conditioner operation data: the system comprises the running state (such as a switching-on and switching-off state, a refrigerating mode, a fan rotating speed and the like) of each air conditioning unit, chilled water supply and return water temperature, cooling water supply and return water temperature, total power consumption of an air conditioning system, power consumption of each air conditioning unit and the like; historical PUE data: the method comprises the steps of including PUE values of all time points, and calculating total energy consumption and IT equipment energy consumption data required by the PUE values; Historical IT device data: including operational status (e.g., CPU utilization, memory occupancy, etc.) and power consumption data of IT devices such as servers, network devices, storage devices, etc. According to the characteristics of the operation characteristics and the historical operation data of the air conditioning system, proper indexes are selected to layer the operation state of the air conditioner, for example, indexes such as the water supply and return temperature, the outdoor environment temperature, the air conditioner load rate and the like of the air conditioning system can be layered. Classifying the historical air conditioner operation data, marking a corresponding air conditioner layering state label for each time point, and generating the historical air conditioner layering state data. For example, the layering status data may be stored as a new data table, and the historical air conditioning layering status data may be correlated with the environmental enthalpy difference data for the corresponding time period to form a multi-dimensional dataset comprising air conditioning layering status, environmental enthalpy difference, and air conditioning system power consumption. Grouping the data according to the air conditioner layering states, and counting the corresponding relation between the environmental enthalpy difference and the power consumption of the air conditioning system in each air conditioner layering state. For example, the average air conditioning system power consumption corresponding to different environmental enthalpy differences in each air conditioning layered state can be calculated. And (3) arranging the corresponding relation between the environmental enthalpy difference and the power consumption of the air conditioning system in the different air conditioning layering states obtained through statistics into a data table form to form enthalpy difference-power consumption mapping relation data. And selecting a proper multiple nonlinear regression model as an air conditioner load prediction model, such as Support Vector Regression (SVR), random forest Regression (RF) and the like, wherein the models can better treat nonlinear relation and multivariable prediction problems. The enthalpy difference-power consumption mapping relation data is divided into a training set and a test set, for example, according to 7:3, wherein 70% of the data is used for model training and 30% of the data is used for model evaluation. And performing model training by using the selected multiple nonlinear regression model and training set data to obtain an initial air conditioner power consumption prediction model. For example, the corresponding model training function may be called for model training using the Scikit-learn library in Python. And evaluating the initial air conditioner power consumption prediction model by using test set data, calculating indexes such as prediction errors, decision coefficients (R-squared) and the like of the model, and evaluating the fitting degree and the prediction accuracy of the model. And according to the model evaluation result, adjusting parameters of the model, such as kernel functions, penalty coefficients and the like of the SVR model, or adjusting the number, depth and the like of decision trees of the RF model, so as to improve the prediction accuracy of the model. And inputting the environmental enthalpy difference matrix data into an air conditioner power consumption prediction model obtained by optimization, and predicting the power consumption load of each time point data center air conditioner system in a future period according to the input environmental enthalpy difference data and the air conditioner power consumption prediction model. And arranging the predicted air conditioner power consumption load data at each time point according to a time sequence to form air conditioner power consumption prediction curve data.

Preferably, step S22 comprises the steps of:

In the embodiment of the invention, whether missing values exist in historical air conditioner operation data or not is checked, for example, air conditioner power consumption data at certain time points are missing. For the missing values, interpolation, mean, regression, etc. may be used to fill in, for example, the missing values between adjacent time points using linear interpolation. An outlier in the historical air conditioner operation data, such as an air conditioner power consumption data anomaly due to a sensor failure, is identified. For the outliers, deletion, replacement, correction, and the like methods may be employed, for example, the outliers are identified and deleted using the box graph method. It is ensured that the historical air conditioning operating data and the historical weather environmental data are aligned in time, for example, all data statistics are performed in units of hours. If the temporal granularity of the data is different, a data resampling or interpolation process is required, such as aggregating minute data into hour data. And extracting the power consumption characteristic data of the air conditioning system from the historical air conditioning operation data, such as the total power consumption of the air conditioning system, the power consumption of each air conditioning unit, the chilled water supply and return water temperature and the like. Contemporaneous historical meteorological environment characteristic data, such as dry bulb temperature, relative humidity, dew point temperature, etc., are extracted from the historical meteorological environment data. And setting reasonable label rules according to understanding and analyzing the running state of the air conditioning system, and labeling the running state label of the power consumption characteristic data of the historical air conditioning system. For example, the operation states may be classified into three levels of low load, medium load, and high load according to the magnitude of the total power consumption of the air conditioning system, and are respectively denoted as "L", "M", and "H". And constructing a two-dimensional space temperature and humidity coordinate system by taking the dry bulb temperature as an abscissa and the relative humidity as an ordinate. And mapping the dry bulb temperature and relative humidity data in the contemporaneous historical meteorological environment characteristic data into a two-dimensional space temperature and humidity coordinate system to form a scatter diagram. According to the distribution condition of the scatter diagram, the two-dimensional space temperature and humidity coordinate system is divided into sections, for example, the temperature is divided into a plurality of sections, the humidity is divided into a plurality of sections, and a plurality of temperature and humidity sections are formed. And matching the tagged running state data with the historical temperature and humidity interval data, and finding out a temperature and humidity interval corresponding to the air conditioner running state tag at each time point. And screening the temperature and humidity intervals according to a preset rule, for example, selecting the temperature and humidity intervals with more data points or selecting the temperature and humidity intervals with stronger relevance with the running state of the air conditioning system. And setting a data quantity threshold according to actual conditions, and judging whether the primary hierarchical operation data needs to be subjected to section data merging processing or not. And combining the temperature and humidity intervals with the data volume smaller than the threshold value with the adjacent temperature and humidity intervals until the combined interval data volume reaches the threshold value or the combination cannot be continued. And updating the preliminary hierarchical operation data according to the result of the interval merging to generate final historical hierarchical state data.

Preferably, step S3 comprises the steps of:

As an example of the present invention, referring to fig. 3, a detailed implementation step flow diagram of step S3 in fig. 1 is shown, where step S3 includes:

In the embodiment of the invention, IT equipment layout data is obtained from a data center infrastructure management system (DCIM) or related documents, wherein the data center infrastructure management system (DCIM) comprises information such as cabinet positions, positions (U-positions) of servers in cabinets, equipment types, equipment sizes and the like. And extracting power consumption data of the IT equipment from the historical operation data, wherein the power consumption data comprises power consumption values of various IT equipment such as a server, network equipment, storage equipment and the like, and corresponding information such as acquisition time and the like.

In the embodiment of the invention, CAD drawing or other structural data of a data center machine room are imported into a modeling tool to construct a 3D model of the machine room, wherein the 3D model comprises structural elements such as walls, floors, doors and windows, air-conditioning air-supply and return openings and the like. And adding the 3D model of the IT equipment into a machine room model according to the IT equipment layout data, and setting the size, the position, the orientation and other attributes of the equipment. The custom model may be created using a library of models provided by modeling software, or based on actual device dimensions. Corresponding material properties, such as thermal conductivity of walls, reflectivity of floors, etc., are set for each element in the machine room model.

In the embodiment of the invention, the service load is distributed to specific IT equipment according to preset service load data and in combination with a data center resource scheduling strategy, for example, a virtual machine is deployed on a designated physical server or a database query request is distributed to a designated database server. The historical IT device power consumption characteristic data is analyzed, and a periodic rule of the IT device power consumption is identified, such as power consumption differences of weekdays and weekends, power consumption fluctuation of different time periods every day and the like. Historical data may be analyzed using time series analysis methods such as autoregressive moving average (ARMA), seasonal decomposition (STL), and the like. And predicting the power consumption of each IT device at each time point in a future period according to the IT device work load distribution result and Zhou Shixu power consumption analysis result, and generating IT device power consumption prediction time sequence data. A power consumption prediction model may be constructed using a machine learning method, such as a Support Vector Machine (SVM), random Forest (RF), etc., and trained and evaluated using historical data.

In the embodiment of the invention, the space of the digital twin machine room model is divided into uniform grids, for example, the machine room space is divided into grids with the size of 1 cubic meter. According to the predicted IT device power consumption time sequence data and the position information of the IT device in the digital twin machine room model, mapping the IT device power consumption of each time point to a corresponding space grid, calculating the total power consumption of each grid, and for the IT devices crossing a plurality of grids, carrying out power consumption distribution among the grids according to the device volume. And arranging the power consumption values of all grids at each time point into a matrix form, and generating heat source space distribution matrix data, wherein each element of the matrix represents the power consumption value of the grid corresponding to the corresponding time point.

In the embodiment of the invention, according to the actual situation of the machine room, boundary conditions for heat conduction simulation, such as the temperature of the wall of the machine room, the temperature and wind speed of an air-conditioning air supply outlet, the heat dissipation of personnel and equipment in the machine room and the like, are set. And (3) using Computational Fluid Dynamics (CFD) software or a library to call a corresponding solver to perform heat conduction simulation on the digital twin machine room model, and performing temperature distribution and airflow flowing conditions at various positions in the computer room. And (3) iteratively calculating the temperature field and the flow field of each time step according to the set time step until the simulation is finished.

According to the embodiment of the invention, the refrigeration energy efficiency score of each space grid at each time point is calculated according to the air conditioner power consumption prediction curve data and the machine room thermodynamic propagation network simulation result. The calculation method of the refrigeration energy efficiency score can be defined according to actual conditions, and factors such as deviation of grid temperature and target temperature, heat density of IT equipment in the grid, air-conditioning air supply quantity and the like are considered. And analyzing the refrigeration energy efficiency scoring result of each time point, and identifying the area with lower refrigeration efficiency, namely the potential energy-saving space. For example, a grid with a refrigeration energy efficiency score below a preset threshold may be marked as a potential energy savings space. For example, the results show that grid refrigeration energy efficiency scores near the window area are low, indicating that the air conditioning system is low in refrigeration efficiency in the area and that energy is wasted. Thus, the area near the window can be marked as potentially energy saving space.

Preferably, step S36 comprises the steps of:

In the embodiment of the invention, main heat sources in a digital twin machine room model, such as a high-power-consumption server, network equipment and the like, are identified according to the heat source spatial distribution matrix data. Respective thermal boundary conditions, such as heat flux density, surface temperature, etc., are set for each heat source. The settings may be based on actual power consumption and heat dissipation of the IT equipment or with reference to data provided by the equipment manufacturer. And determining a time sequence step, for example, 1 hour, according to the time granularity of the air conditioner power consumption prediction curve data, and dividing the air conditioner power consumption prediction curve data according to the set time step to obtain an air conditioner power consumption value of each time step. The air conditioner power consumption value for each time step may be considered as an average air conditioner load characteristic over that time period. And (3) associating the time sequence air conditioner load characteristic data with air conditioner air supply outlets in the digital twin machine room model, for example, determining which air conditioner air supply outlets are in an open state and the corresponding air supply amount in each time step according to control logic of an air conditioning system. According to the air supply quantity and air supply temperature of the air conditioner air supply opening, setting the air flow parameters of each air conditioner air supply opening in each time step, such as wind speed, wind direction, temperature and the like. Heat source boundary condition data and air flow cooling condition data are input into CFD software. And loading boundary condition data of each time step in sequence according to the set time step, performing sequential airflow numerical simulation, and calculating temperature field and flow field distribution in a machine room of each time step. In the simulation process of each time step, a finite element method can be adopted to carry out numerical solution on heat conduction and air flow diffusion, so that the simulation precision is improved. And identifying a cold channel and a hot channel area in the digital twin machine room model according to the machine room layout and the airflow organization mode. According to the time sequence air flow field data of the machine room, calculating the average temperature difference of the cold channel and the hot channel of each time step, and taking the average temperature difference as one of indexes for measuring the cooling efficiency of the machine room. Analyzing the temperature distribution condition of each area in the machine room, and identifying the area with overhigh or overlow temperature and the area with uneven temperature distribution. Analyzing the temperature distribution condition of each area in the machine room, and identifying the area with overhigh or overlow temperature and the area with uneven temperature distribution. And analyzing the bypass condition of the air flow passing through the IT equipment, and evaluating the heat dissipation efficiency of the equipment. For example, an average wind speed, turbulence intensity, etc. around the device may be calculated to assess the ability of the airflow to remove heat. And (3) comprehensively grading the refrigerating energy efficiency of each region in the digital twin machine room model according to the machine room cooling efficiency data and the machine room refrigerating efficiency evaluation data. The scoring rule can be defined according to actual demands, for example, indexes such as temperature difference of cold and hot channels, uniformity of temperature distribution, degree of equipment bypass and the like are weighted and averaged. And identifying a region with lower score, namely a potential energy-saving space, according to the comprehensive refrigeration energy efficiency scoring result. For example, areas with scores below a preset threshold may be marked as potential energy savings space, generating potential energy savings space location data.

Preferably, step S4 comprises the steps of:

In the embodiment of the invention, whether the time stamps of the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time sequence data are consistent or not is checked, and if not, time sequence alignment processing is needed. Interpolation, resampling, etc. methods can be used to adjust both data sets to the same temporal granularity and temporal range. And accumulating the air conditioner power consumption prediction data and the IT equipment power consumption prediction data which are aligned in time sequence at each time point to obtain the total power consumption time sequence data of the data center. PUE = total data center energy consumption/IT equipment energy consumption, according to the definition of PUE. Substituting the total power consumption data and the IT equipment power consumption prediction data of each time point into a PUE calculation formula to obtain the PUE value of each time point, and forming the prediction PUE instantaneous sequence data. The predicted PUE transient sequence data is processed by selecting a suitable time series data smoothing method, such as a moving average method, an exponential smoothing method, a wavelet transformation method, and the like. Smoothing parameters such as moving average window size, exponential smoothing factor, etc. are set according to the selected method and characteristics of the data. The predicted PUE transient sequence data is smoothed using selected methods and parameters to reduce random fluctuations in the data, highlighting long-term trends in the data. And connecting the smoothed PUE values according to a time sequence to form circumferential trend PUE value prediction curve data. From the historical operating data, confidence intervals, e.g., 95% confidence intervals, for PUE values are calculated. The confidence interval calculation method can adopt a statistical method, such as a t distribution method, a Bootstrap method and the like. And comparing the circumferential trend PUE value prediction curve data with the calculated confidence interval, and if the value of the prediction curve falls outside the confidence interval, correcting the prediction curve so as to fall in the confidence interval. The correction method can adopt linear scaling, nonlinear mapping and the like. The PUE target threshold is set, for example, 1.1 or 1.2, according to the energy saving target and the actual situation of the data center. And carrying out node-by-node scanning on the modified PUE value prediction curve data, and comparing the predicted PUE value of each time point with a set target threshold value. If the predicted PUE value at a certain point in time is greater than or equal to the set target threshold, the node is marked as out-of-standard PUE value node data.

Preferably, step S44 includes the steps of:

In the embodiment of the invention, the historical operation data is cleaned to remove abnormal values and missing values, for example, the abnormal values are identified and removed by using a box diagram method, and the missing values are supplemented by using a linear interpolation method. Selecting a feature associated with the PUE value, for example: time characteristics: including hours, days of the week, months, etc., for reflecting the periodic variation law of PUE values. Environmental characteristics: including temperature, humidity, air pressure, etc., for reflecting the influence of environmental factors on the PUE value. IT load characteristics: including server utilization, network traffic, etc., to reflect the impact of IT device loads on PUE values. The historical PUE value feature data is grouped according to the extracted features, for example, by time features such as hours, days of the week, months, etc., or by different environmental features and IT load feature combinations. Statistical indicators of the PUE values of each group, such as average values, standard deviations, quantiles, etc., are calculated for describing the distribution of the PUE values under different characteristic conditions. According to historical operation data or industry experience, a reasonable PUE reference value, for example, 1.1 is preset, and the characteristic value of each time point in the circumferential trend PUE value prediction curve data is matched with the PUE characteristic distribution data to find out a corresponding characteristic group. And calculating a dynamic PUE value confidence interval of each time point according to the PUE value statistical index of the matched feature group and a preset PUE reference value, for example, calculating the confidence interval by using a normal distribution hypothesis or a t distribution hypothesis. And (3) organizing the predicted PUE value, the corresponding upper limit of the confidence interval and the corresponding lower limit of the confidence interval of each time point into a data structure, and generating dynamic PUE value confidence interval data. If the predicted PUE value at a certain time point in the circumferential trend PUE value prediction curve data falls outside the dynamic confidence interval at the time point, the PUE value at the time point is considered to have abnormality. The time point determined to be abnormal is marked as an abnormal PUE value node, and the type of abnormality is recorded, for example, above the upper confidence interval limit or below the lower confidence interval limit. A reasonable time window is set, for example, 3 hours or 6 hours, according to the change law of the PUE value. For each abnormal PUE value node, PUE values in the preceding and following time windows thereof are extracted, and their correlation with adjacent time points is analyzed, for example, autocorrelation coefficients, partial autocorrelation coefficients, and the like are calculated. If there is a strong correlation between the abnormal PUE value and the PUE value at the adjacent time point, the abnormal value may be corrected by using a moving average method, an exponential smoothing method, or the like. If the abnormal PUE value has a weak correlation with the PUE value at an adjacent point in time, it is possible to consider predicting and correcting the abnormal value using a more complex model such as a state space model, a neural network, or the like.

Preferably, step S5 comprises the steps of:

In the embodiment of the invention, according to the actual running condition of the data center, a risk assessment index related to the exceeding of the PUE is selected, for example: power supply system load factor: exceeding PUEs generally means increased data center energy consumption, resulting in excessive power system load rates and increased trip risk. Refrigeration system load factor: the high PUE reflects low efficiency of the refrigeration system or excessive heat dissipation density of IT equipment, resulting in excessive load factor of the refrigeration system, affecting refrigeration effect and equipment life. Risk of temperature violations: exceeding the PUE leads to an increase in the temperature of the machine room, increasing the risk of exceeding the operating temperature range of the equipment. Using a suitable risk assessment model, such as a Bayesian network or decision tree, the risk of overstandard for each node is calculated from the overstandard PUE value node data and the risk assessment index. According to the bearing capacity of the data center to risks, a PUE exceeding risk threshold is preset, for example, when the risk level reaches 'high' or the risk probability exceeds 20%, intervention measures are needed. And comparing the PUE exceeding risk assessment value with a preset risk threshold. If the PUE exceeding risk assessment value is lower than a preset risk threshold value, the data center continues to execute the current operation regulation strategy; if the PUE exceeding risk assessment value is higher than or equal to a preset risk threshold, a subsequent step needs to be performed. And analyzing specific reasons of the exceeding of the PUE, such as low efficiency of an air conditioning system, excessive load of IT equipment, unreasonable airflow organization and the like by combining the information of the node data of the exceeding PUE, historical operation data, potential energy-saving space positioning data and the like. And setting a reasonable PUE consumption reduction target according to the exceeding degree of the PUE and the exceeding reason obtained by analysis, for example, reducing the PUE value below a target threshold or reducing a certain proportion. And associating the predicted PUE value consumption reduction target data with the potential energy saving space positioning data, for example, matching the PUE consumption reduction target at each time point with the corresponding potential energy saving space. And identifying key parameters related to the PUE consumption reduction target, such as an air conditioner temperature set value, an air conditioner air supply quantity, a server load rate and the like, according to information, such as space positions, refrigeration energy efficiency scores, existing energy saving measures and the like, contained in the potential energy saving space positioning data. Sensitivity analysis is performed on the identified key parameters, for example, using orthogonal test methods, sobol sequences, etc., and the degree of influence of each parameter on the PUE value, and interactions between the parameters are analyzed. According to the sensitivity analysis result, selecting parameters which have larger influence on the PUE value and are easy to adjust as key consumption reduction adjustment parameters, and generating key consumption reduction adjustment parameter data, such as a data table containing information such as parameter names, parameter value ranges and the like. And generating a plurality of parameter combinations according to the key consumption reduction adjustment parameter data, for example, searching a parameter space by using an optimization algorithm such as a genetic algorithm, a particle swarm algorithm and the like, and generating the parameter combinations meeting constraint conditions. each parameter combination is simulated using a digital twin model or other simulation tool, its effect on the PUE value is evaluated, and simulation results, such as PUE value changes, energy consumption changes, temperature changes, etc., are recorded. The optimal parameter combination is converted into an executable control strategy, such as adjusting an air conditioner temperature set value to a specified value, transferring a server load to other physical equipment, and the like, and generating a PUE intelligent regulation strategy.

The dynamic PUE value prediction method is constructed by combining weather forecast data, historical operation data and a digital twin machine room model, and the mapping relation between the environment enthalpy value and the air conditioner power consumption is established by utilizing the weather forecast data and the historical operation data, and the influence of the environment factors on the PUE value is considered. And then, simulating the power consumption and heat distribution of the IT equipment by constructing a digital twin machine room model, analyzing the thermodynamic propagation condition of the machine room, and further identifying a potential energy-saving space. And finally, carrying out confidence interval correction and abnormal value processing on the PUE value prediction data, and generating an intelligent regulation strategy according to the risk assessment result, so that dynamic prediction and optimization of the PUE value of the data center are realized, and the accuracy and precision of the PUE value prediction are improved. On the premise of not influencing the normal operation of the data center, the predicted PUE value is calculated, and meanwhile, a plurality of cold source parameters are adjusted and modified, so that the energy consumption of the cold source is reduced, and data support is provided for adjusting the energy consumption index.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the PUE value of a data center, comprising the following steps:

Step S1: collecting weather forecast data for the target data center to obtain original weather forecast data; processing ambient air enthalpy values according to the original weather forecast data to generate ambient enthalpy value prediction time series data; performing real-time ambient enthalpy difference processing on the target data center through the ambient enthalpy value prediction time series data to generate ambient enthalpy difference matrix data;

Step S2: Acquire historical operation data; establish a mapping relationship between the power consumption of the air-conditioning system and the environmental enthalpy difference according to the historical operation data, and construct an air-conditioning power consumption prediction model; transmit the environmental enthalpy difference matrix data to the air-conditioning power consumption prediction model to perform air-conditioning power consumption load prediction, and generate air-conditioning power consumption prediction curve data;

Step S3: Allocate IT equipment workloads for preset business load data, and perform equipment time series power consumption analysis based on historical operation data to generate IT equipment power consumption prediction time series data; construct a computer room thermal propagation network based on IT equipment power consumption prediction time series data, and locate potential energy-saving spaces through air conditioning power consumption prediction curve data to obtain potential energy-saving space location data;

Step S4: Calculate the PUE value at the predicted time point according to the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time series data, and perform confidence interval correction processing to generate corrected PUE value prediction curve data; identify the nodes exceeding the standard according to the corrected PUE value prediction curve data to obtain the node data of the PUE value exceeding the standard;

Step S5: Perform intelligent power consumption reduction control strategy processing on the node data with excessive PUE value through the potential energy-saving space positioning data to generate intelligent power consumption reduction control strategy data.

2. The method for predicting the PUE value of a data center according to claim 1, wherein step S1 comprises the following steps:

Step S11: Perform geographic location positioning processing on the target data center to generate target area positioning data;

Step S12: acquiring hourly granular weather forecast data for the next week based on the target area positioning data, thereby obtaining original weather forecast data;

Step S13: identifying the calculation elements of the moist air enthalpy value according to the original weather forecast data, and generating key weather forecast element data;

Step S14: Calculate the hourly ambient air enthalpy value according to the key meteorological forecast element data to obtain the ambient enthalpy value prediction time series data;

Step S15: extracting the real-time air inlet and outlet environmental parameters of the target data center through the preset temperature monitoring point data, and calculating the real-time ambient air enthalpy value to generate real-time monitoring point ambient enthalpy value data;

Step S16: Calculate the environmental enthalpy difference of the environmental enthalpy data of the real-time monitoring point using the environmental enthalpy prediction time series data to generate enthalpy difference time series matrix data; perform matrix transposition processing on the enthalpy difference time series matrix data to obtain environmental enthalpy difference matrix data.

3. The method for predicting the PUE value of a data center according to claim 1, wherein step S2 comprises the following steps:

Step S21: acquiring historical operation data through the target data center, wherein the historical operation data includes historical air conditioning operation data, historical PUE data, and historical IT equipment data;

Step S22: performing operation status stratification processing according to historical air-conditioning operation data to generate historical air-conditioning stratified status data;

Step S23: establishing a mapping relationship between the power consumption of the air conditioning system and the environmental enthalpy difference based on the historical air conditioning stratification state data, and generating enthalpy difference-power consumption mapping relationship data;

Step S24: constructing an air conditioning load prediction model for the enthalpy difference-power consumption mapping relationship data through a preset multivariate nonlinear regression model to obtain an initial air conditioning power consumption prediction model;

Step S25: Optimizing the model fitting of the initial air conditioning power consumption prediction model by using the historical air conditioning stratification state data to generate an air conditioning power consumption prediction model;

Step S26: transmitting the environmental enthalpy difference matrix data to the air conditioning power consumption prediction model to perform air conditioning system power consumption load prediction and generate air conditioning power consumption prediction curve data.

4. The method for predicting the PUE value of a data center according to claim 3, wherein step S22 comprises the following steps:

Step S221: cleaning the historical air-conditioning operation data to obtain cleaned historical air-conditioning operation data;

Step S222: performing time series feature correlation processing according to the historical air conditioning operation data of the cleaning, and obtaining the historical air conditioning system power consumption feature data and the historical meteorological environment feature data of the same period;

Step S223: labeling the historical air-conditioning system power consumption characteristic data with an operation status label to generate labeled operation status data;

Step S224: performing two-dimensional spatial temperature and humidity mapping processing based on historical meteorological environment characteristic data of the same period to generate historical temperature and humidity interval data;

Step S225: screening the labeled operation status data by temperature and humidity intervals using the historical temperature and humidity interval data to obtain preliminary stratified operation data;

Step S226: Based on the preliminary stratified operation data, the historical air-conditioning system power consumption characteristic data is subjected to interval data merging processing through a preset data volume threshold to generate historical stratified status data.

5. The method for predicting the PUE value of a data center according to claim 1, wherein step S3 comprises the following steps:

Step S31: extracting IT equipment basic data from historical operation data to generate historical IT basic equipment data, wherein the historical IT basic equipment data includes IT equipment layout data and historical IT equipment power consumption characteristic data;

Step S32: Perform digital processing according to the IT equipment layout data to obtain digital twin computer room data;

Step S33: performing IT equipment workload distribution on the preset business load data, and performing weekly time series power consumption analysis based on historical IT equipment power consumption characteristic data to generate IT equipment power consumption prediction time series data;

Step S34: Perform time-series azimuth power consumption mapping processing on the IT equipment power consumption time series data through the digital twin computer room data to generate heat source spatial distribution matrix data;

Step S35: Performing a dynamic simulation of heat conduction on the digital twin computer room data based on the heat source spatial distribution matrix data, thereby constructing a computer room thermal propagation network;

Step S36: Perform a comprehensive cooling energy efficiency score on the heat dissipation network of the computer room through the air conditioning power consumption prediction curve data, and locate the potential energy-saving space to obtain the potential energy-saving space location data.

6. The method for predicting the PUE value of a data center according to claim 5, wherein step S36 comprises the following steps:

Step S361: setting heat source boundary conditions according to the heat propagation network of the computer room to generate heat source boundary condition data;

Step S362: performing time-series step load feature decomposition on the air-conditioning power consumption prediction curve data to generate time-series air-conditioning load feature data;

Step S363: setting the air supply air flow condition of the air conditioner according to the time-series air conditioning load characteristic data to obtain the air flow cooling condition data;

Step S364: numerically simulate the time-series airflow using the digital twin computer room data through the heat source boundary condition data and the airflow cooling condition data, and perform finite element diffusion analysis to obtain the time-series airflow field data of the computer room;

Step S365: Calculate the temperature difference between the hot and cold channels according to the time-series airflow field data of the computer room, and analyze the temperature distribution of the computer room to generate the cooling efficiency data of the computer room;

Step S366: Performing flow field topology analysis based on the time-series airflow field data of the computer room, and evaluating the degree of equipment bypass flow, to generate computer room cooling efficiency evaluation data;

Step S367: Based on the computer room cooling efficiency data and the computer room refrigeration efficiency evaluation data, the digital twin computer room data is scored for comprehensive refrigeration energy efficiency, and potential energy-saving space positioning is performed to obtain potential energy-saving space positioning data.

7. The method for predicting the PUE value of a data center according to claim 1, wherein step S4 comprises the following steps:

Step S41: performing time series alignment processing on the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time series data, and calculating the total power consumption at a time point to obtain the total power consumption time series data of the data center;

Step S42: Calculate the PUE value at the predicted time point for the IT equipment power consumption prediction time series data using the total power consumption time series data of the data center to generate predicted PUE instantaneous series data;

Step S43: performing time series value smoothing processing on the predicted PUE instantaneous series data to generate weekly trend PUE value prediction curve data;

Step S44: performing confidence interval correction processing on the weekly trend PUE value prediction curve data using historical operation data to generate corrected PUE value prediction curve data;

Step S45: Scan and compare the modified PUE value prediction curve data node by node using the preset PUE target threshold, and mark the curve nodes that are greater than or equal to the preset PUE target threshold as excessive PUE value node data.

8. The method for predicting the PUE value of a data center according to claim 7, wherein step S44 comprises the following steps:

Step S441: extracting historical PUE value features from historical operation data to obtain historical PUE value feature data;

Step S442: Perform distribution statistics analysis based on historical PUE value characteristic data to obtain PUE characteristic distribution data;

Step S443: Based on the preset PUE benchmark value, the weekly trend PUE value prediction curve data is subjected to time series node confidence interval calculation through the PUE characteristic distribution data to obtain dynamic PUE value confidence interval data;

Step S444: performing PUE value anomaly detection on the weekly trend PUE value prediction curve data through the dynamic PUE value confidence interval data to generate abnormal PUE value node data;

Step S445: Perform time window PUE value correlation analysis on abnormal PUE value node data through weekly trend PUE value prediction curve data, and perform abnormal PUE value correction processing to obtain corrected PUE value prediction curve data.

9. The method for predicting the PUE value of a data center according to claim 1, wherein step S5 comprises the following steps:

Step S51: Performing actual operation risk assessment on the data center through the node data of the exceeded PUE value to generate PUE exceeded risk assessment data;

Step S52: judging the risk of exceeding the standard according to the PUE exceeding the standard risk assessment data. When the PUE exceeding the standard risk assessment data is lower than the preset risk threshold, the data center continues to execute the current operation control strategy.

Step S53: When the PUE exceeding standard risk assessment data is higher than or equal to the preset risk threshold, a consumption reduction target is set for the exceeding standard PUE value node data to generate predicted PUE value consumption reduction target data;

Step S54: using the predicted PUE value consumption reduction target data to identify key consumption reduction parameters for the potential energy-saving space positioning data, and performing a consumption reduction parameter sensitivity analysis to obtain key consumption reduction adjustment parameter data;

Step S55: adjusting the parameter combination according to the key power-saving adjustment parameter data, and performing multi-objective power-saving optimization processing, thereby obtaining a PUE intelligent control strategy.

10. A system for predicting the PUE value of a data center, characterized in that it is used to execute the method for predicting the PUE value of a data center as claimed in claim 1, and the system for predicting the PUE value of a data center comprises:

The environmental enthalpy value preprocessing module is used to collect weather forecast data for the target data center to obtain original weather forecast data; process the ambient air enthalpy value according to the original weather forecast data to generate environmental enthalpy value prediction time series data; perform real-time environmental enthalpy difference processing on the target data center through the environmental enthalpy value prediction time series data to generate environmental enthalpy difference matrix data;

The air conditioning power consumption prediction module is used to obtain historical operation data; establish a mapping relationship between the air conditioning system power consumption and the environmental enthalpy difference based on the historical operation data, and build an air conditioning power consumption prediction model; transmit the environmental enthalpy difference matrix data to the air conditioning power consumption prediction model to predict the air conditioning power consumption load and generate air conditioning power consumption prediction curve data;

The energy-saving space positioning module is used to distribute IT equipment workloads based on preset business load data, and to analyze equipment power consumption in time series based on historical operation data to generate IT equipment power consumption prediction time series data; to build a computer room heat propagation network based on IT equipment power consumption prediction time series data, and to locate potential energy-saving spaces through air conditioning power consumption prediction curve data to obtain potential energy-saving space positioning data;

The PUE value prediction module is used to calculate the PUE value at the prediction time point according to the air conditioner power consumption prediction curve data and the IT equipment power consumption prediction time series data, and perform confidence interval correction processing to generate the corrected PUE value prediction curve data; identify the nodes exceeding the standard according to the corrected PUE value prediction curve data to obtain the node data of the PUE value exceeding the standard;

The intelligent consumption reduction strategy generation module is used to perform intelligent consumption reduction control strategy processing on the node data with excessive PUE value through the potential energy-saving space positioning data, and generate intelligent consumption reduction control strategy data.