Data center energy efficiency anomaly detection method, system and medium based on adaptive prediction interval

Abstract

The present invention discloses a data center energy efficiency anomaly detection method, system and medium based on adaptive prediction interval. The method comprises: selecting energy efficiency indicators, collecting and storing on-site historical data, and calculating energy efficiency values; preprocessing the historical data; establishing a regression prediction model through feature selection and hyperparameter optimization; calculating the hourly energy efficiency prediction error set, calculating its probability density, superimposing the confidence interval, and obtaining an adaptive prediction error interval; preprocessing the real-time operation data and inputting it into the trained regression prediction model to obtain the energy efficiency prediction value at the current moment, superimposing the adaptive prediction error interval, obtaining the energy efficiency adaptive prediction interval at the current moment, comparing whether the actual energy efficiency value is within the interval, and obtaining the energy efficiency anomaly detection result. The present invention can better cover the energy efficiency variation range of the data center, effectively avoid system random errors and modeling errors, realize reliable detection of abnormal energy efficiency, and improve the energy efficiency management level of the data center.

Description

Technical Field

[0001] The present invention belongs to the technical field of data center energy efficiency anomaly detection, and specifically relates to a data center energy efficiency anomaly detection method, system and medium based on adaptive prediction interval. Background Art

[0002] In recent years, with the development and maturity of emerging technologies, the data center market has grown rapidly, and the industry has been included in the national "new infrastructure" category. While data centers are driving the rapid development of the digital economy, their high energy consumption and low energy efficiency are becoming increasingly prominent. During system operation, problems such as equipment anomalies, operational anomalies, and abnormal energy consumption of IT equipment can all lead to abnormal data center energy efficiency. If these anomalies are not promptly detected and countermeasures are not implemented, long-term abnormal system operation will not only result in energy waste, but may also lead to greater anomalies or even failures. Therefore, strengthening the detection of anomalies in data center energy efficiency indicators is of great significance for improving energy utilization efficiency, system operation reliability, and achieving a high-efficiency computer room.

[0003] Currently, machine learning is a common method used to monitor and evaluate data center energy efficiency indicators. Using supervised learning methods to establish predictive models that correlate operating parameters with overall energy efficiency enables energy efficiency monitoring and evaluation. However, energy efficiency prediction models based on operating parameters cannot avoid energy efficiency anomalies caused by abnormal operating parameters, and cannot determine whether energy efficiency is abnormal at a given moment. Furthermore, anomaly detection based on the prediction model's predicted values ​​fails to reflect the range of data center energy efficiency variations, fails to effectively mitigate systemic random errors and modeling errors, and the reliability of anomaly energy efficiency detection results remains questionable. Summary of the Invention

[0004] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the existing technology and provide a data center energy efficiency anomaly detection method, system and medium based on adaptive prediction intervals. According to historical energy efficiency data, outdoor meteorological parameters and time parameters, the reasonable range of data center energy efficiency is predicted, abnormal energy efficiency can be reliably detected, and the energy efficiency management level of the data center can be improved.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] In one aspect of the present invention, a method for detecting anomalies in energy efficiency of a data center based on an adaptive prediction interval is provided, comprising the following steps:

[0007] Select energy efficiency indicators, collect outdoor meteorological parameters, time data and energy consumption data, and calculate energy efficiency data based on energy consumption data;

[0008] Preprocess historical data, including outlier detection and elimination, and missing value interpolation for energy efficiency data; and distorted value elimination and interpolation for outdoor meteorological data;

[0009] Using a feature selection method to perform feature selection on the pre-processed historical data; the historical data includes outdoor meteorological parameters, time data and energy efficiency data;

[0010] Use the data after feature selection to train the regression prediction model, use the model hyperparameter optimization method to determine its optimal hyperparameter combination, and then obtain the trained regression prediction model;

[0011] Use the trained regression prediction model to calculate the predicted value of hourly energy efficiency, subtract the actual value from it, and obtain the prediction error set;

[0012] For the prediction error set, the kernel density estimation algorithm is used to calculate its probability density, and the confidence interval is superimposed to obtain the adaptive prediction error interval;

[0013] The data collected in real time is processed using the same data preprocessing method as that used for historical data, and the features selected in the feature selection method are used to construct an input set. The trained regression prediction model is input to obtain the energy efficiency prediction value at the current moment, and the adaptive prediction error interval is superimposed to obtain the energy efficiency adaptive prediction interval at the current moment. The actual energy efficiency value is compared to see whether it is within the energy efficiency adaptive prediction interval at the current moment to determine whether the energy efficiency is abnormal.

[0014] As a preferred technical solution, the energy efficiency indicators include: power utilization rate, cooling load factor, power supply load factor and renewable energy utilization rate;

[0015] The energy consumption data includes parameter data directly collected by sensors for calculating energy efficiency data;

[0016] The energy efficiency data includes: the energy efficiency value at the current moment, the energy efficiency value at the previous moment, and the energy efficiency value at the previous 24 moments;

[0017] The outdoor meteorological parameters include: outdoor temperature and outdoor humidity;

[0018] The time data includes: year, month, date, time, season type and working day type; wherein the season type includes: cooling season, transition season, non-cooling season; working day type includes: working day, non-working day.

[0019] As a preferred technical solution, the method for detecting and eliminating outliers in energy efficiency data is as follows: s Guidelines:

[0020] ;

[0021] ;

[0022] L 1= m -3 s ;

[0023] L 2= m +3 s ;

[0024] in, X i For a certain column i Energy efficiency data, N is the total number of samples, m is the mean of the data in this column, s is the standard deviation of the data in this column, L 1 is the lower limit of the abnormal threshold, L 2 is the upper limit of the abnormal threshold, and the abnormal value judgment standard is that the sample value is less than the lower limit L 1 or greater than the upper limit of judgment L 2;

[0025] Remove abnormal values ​​from the energy efficiency value at the current moment, the energy efficiency value at the previous moment, and the energy efficiency value at the previous 24 moments respectively;

[0026] The method for interpolating missing values ​​of energy efficiency data is the interpolation method:

[0027] X i = X i+24k ;

[0028] in, X i For a certain column i Energy efficiency data, that is, the energy efficiency value at a certain moment, X i+24k Before k The energy efficiency value at the same time of the day, when k =1, if X i+24 If it is a normal value, take X i+24 Assign the value of X i ,like X i+24 If it is an abnormal value, then look at k =2, X i+24*2 Is it a normal value? If so, assign it to X i If not, see k =3, until Xi+24k Until the normal value is reached, the normal value X i+24k Assign to X i ;

[0029] Interpolate missing values ​​for the energy efficiency values ​​of the previous moment and the previous 24 moments;

[0030] The method for removing distortion values ​​from outdoor meteorological data is a discriminant method:

[0031] ;

[0032] in, Y i For a certain column i The outlier criterion for outdoor meteorological data is that the difference between the sample value and the mean of the data before and after is greater than 5;

[0033] The method for interpolating missing values ​​of outdoor meteorological data is the mean interpolation method:

[0034] ;

[0035] in, Y i For a certain column i outdoor meteorological data, take the mean of the data before and after and assign it to Y i .

[0036] As a preferred technical solution, the feature selection method adopts one of the Pearson correlation coefficient method, the embedding method, the mutual information method and the recursive feature elimination method.

[0037] As a preferred technical solution, the regression prediction model is specifically:

[0038] The regression prediction model is fitted using an artificial intelligence regression algorithm, using one of vector machines, decision trees, neural networks, ensemble learning, deep learning, and generative adversarial networks;

[0039] The hyperparameter optimization method of the regression prediction model adopts one of the random search, grid search, Bayesian optimization and evolutionary algorithm.

[0040] As a preferred technical solution, the specific process of fitting the regression prediction model is as follows:

[0041] The data samples are randomly divided into training sets and test sets according to a certain ratio for model training and validation testing;

[0042] Build an artificial intelligence algorithm regression prediction model and use default parameters to train and validate the data set;

[0043] Taking the regression prediction model evaluation index as the optimization target, the hyperparameter optimization method is used to optimize the hyperparameters of the regression prediction model. The verification method is k fold cross validation;

[0044] Save the regression prediction model optimized through the above steps;

[0045] After the update time period T Finally, use the latest accumulated data set to repeat the above steps and update the regression prediction model.

[0046] As a preferred technical solution, the kernel density estimation algorithm is used to calculate the probability density of the prediction error set, and the confidence interval is superimposed to obtain the adaptive prediction error interval. The specific process is as follows:

[0047] Import the data sample prediction error set and use the hyperparameter optimization method to optimize the bandwidth of the kernel density estimation model. The verification method is k fold cross validation;

[0048] Use optimal bandwidth h A kernel density estimation model is established to calculate the probability density of each prediction error value e. Its probability density function is:

[0049] ;

[0050] in, K ( e ) is a kernel function, and one of the Gaussian kernel function, gamma kernel function, uniform kernel function and triangular kernel function is used;

[0051] The confidence level for calculating the probability density is 1- α The confidence interval of f α / 2 , f 1-α / 2 ], based on the functional relationship between probability density and prediction error value, an adaptive prediction error interval is constructed [ e 1, e 2];

[0052] After the update time period T Finally, the above steps are repeated using the latest accumulated data set to update the adaptive prediction error interval.

[0053] As a preferred technical solution, the energy efficiency prediction value at the current moment is obtained, and the adaptive prediction error interval is superimposed to obtain the energy efficiency adaptive prediction interval at the current moment, specifically:

[0054] The energy efficiency prediction value at the current moment b, and the adaptive prediction error interval [ e 1, e 2] to obtain the current energy efficiency adaptive prediction interval [ b+e 1, b+e 2];

[0055] The comparison of whether the actual energy efficiency value is within the current energy efficiency adaptive prediction range to determine whether the energy efficiency is abnormal is specifically as follows:

[0056] Determine the actual energy efficiency value at the current moment c Whether it is in the current energy efficiency adaptive prediction range [ b+e 1, b+e 2], if not, it is an energy efficiency abnormality.

[0057] Another aspect of the present invention further provides a data center energy efficiency anomaly detection system based on adaptive prediction intervals, which is applied to the above-mentioned data center energy efficiency anomaly detection method based on adaptive prediction intervals, and includes a data acquisition module, a data preprocessing module, a feature selection module, a regression prediction model training module, an adaptive prediction error interval construction module, and an anomaly detection module;

[0058] The data acquisition module is used to select energy efficiency indicators, collect outdoor meteorological parameters, time data and energy consumption data through sensors, and calculate energy efficiency data based on the energy consumption data;

[0059] The data preprocessing module is used to preprocess historical data, including detecting and eliminating outliers and interpolating missing values ​​for energy efficiency data; and eliminating and interpolating distorted values ​​for outdoor meteorological data;

[0060] The feature selection module is used to perform feature selection on the pre-processed historical data using a feature selection method; the historical data includes outdoor meteorological parameters, time data and energy efficiency data;

[0061] The model training module is used to train the regression prediction model using the data after feature selection, and use the model hyperparameter optimization method to determine its optimal hyperparameter combination, thereby obtaining a trained regression prediction model;

[0062] After the update time period T Finally, use the latest accumulated data set to repeat the above steps and update the regression prediction model;

[0063] The adaptive prediction error interval construction module is used to generate a prediction error set, calculate its probability density using a kernel density estimation algorithm, and superimpose confidence intervals to obtain an adaptive prediction error interval;

[0064] After the update time period TFinally, the above steps are repeated using the latest accumulated data set to update the adaptive prediction error interval;

[0065] The anomaly detection module is used to input the data to be tested into the trained regression prediction model to obtain the energy efficiency prediction value at the current moment, superimpose the adaptive prediction error interval to obtain the energy efficiency adaptive prediction interval at the current moment, compare whether the actual energy efficiency value is within the energy efficiency adaptive prediction interval at the current moment to determine whether the energy efficiency is abnormal, and obtain the anomaly detection result.

[0066] Another aspect of the present invention further provides a storage medium storing a program, which, when executed by a processor, implements the above-mentioned data center energy efficiency anomaly detection method based on adaptive prediction interval.

[0067] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0068] (1) The data center energy efficiency anomaly detection method based on a regression prediction model and an adaptive prediction interval proposed in the present invention can quickly and accurately evaluate the real-time operating energy efficiency status of a complex nonlinear data center energy consumption system, which is beneficial for data center operation and maintenance personnel to quickly make corresponding decisions based on the evaluation results of the real-time operating energy efficiency status, prevent the system operation from further deteriorating, and improve the system's energy efficiency;

[0069] (2) The anomaly detection method of the regression prediction model superimposed on the adaptive prediction interval proposed in this invention can better cover the range of changes in data center energy efficiency, more comprehensively reflect the operating status of the data center, effectively avoid system random errors and modeling errors, and improve the credibility of the abnormal energy efficiency detection results of the data center;

[0070] (3) The adaptive prediction error interval constructed based on the kernel density estimation algorithm proposed in this invention can adaptively determine the upper and lower limits of the interval according to the distribution of the sample data itself, thereby reducing the false alarm rate of anomaly detection;

[0071] (4) The present invention regularly uses real-time data sets to dynamically update the existing anomaly detection model to improve the accuracy of anomaly detection;

[0072] (5) The method and system proposed in the present invention are generally applicable to data centers with stable operation and can be applied to the energy efficiency management of data centers, and have good scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 This is a flow chart of a data center energy efficiency anomaly detection method based on adaptive prediction intervals according to an embodiment of the present invention;

[0074] Figure 2 This is the energy efficiency data interpolation graph according to the embodiment of the present invention;

[0075] Figure 3 1 is a schematic structural diagram of a data center energy efficiency anomaly detection system based on adaptive prediction intervals according to an embodiment of the present invention;

[0076] Figure 4 It is a structural diagram of the storage medium described in an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0078] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. For example, the selection of data center energy efficiency indicators is not limited to the PUE in the embodiment, the selection of regression prediction models is not limited to the LightGBM in the embodiment, and the selection of hyperparameter optimization methods is not limited to the Bayesian optimization and grid search in the embodiment. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0079] Example 1:

[0080] This example uses a data center in southern China as an example to illustrate the implementation process of the present invention. This data center has established an energy consumption monitoring platform, which can collect data on energy efficiency-related parameters required for this example through sensors. This example selects historical energy efficiency-related data from the data center from December 2020 to July 2022.

[0081] like Figure 1 As shown, this embodiment provides a data center energy efficiency anomaly detection method based on adaptive prediction interval, including the following steps:

[0082] S1. Data acquisition: Select specific energy efficiency indicators, collect and store outdoor meteorological parameters, time data and energy consumption data, and calculate energy efficiency data;

[0083] S11. The energy efficiency indicators include: power utilization efficiency (PUE), cooling load factor (CLF), power load factor (PLF), and renewable energy utilization rate (RER). In this embodiment, PUE is selected as the energy efficiency indicator of this embodiment;

[0084] S12. Collect and store historical energy efficiency-related data through sensors, including parameter data directly collected by sensors for calculating energy efficiency data, specifically: total power consumption, total power consumption of IT equipment, outdoor temperature, outdoor humidity, year, month, date, time, season type, and weekday type;

[0085] S13. Calculate the energy efficiency index PUE hourly:

[0086] ;

[0087] This can provide the PUE data for the previous moment, the PUE for the previous moment, and the PUE for the previous 24 moments.

[0088] S2. Data preprocessing: Preprocess historical data, including outlier detection and removal, and missing value interpolation for energy efficiency data; and distorted value removal and interpolation for outdoor meteorological data. Specifically, it includes:

[0089] S21. Perform outlier detection on the PUE at the previous moment and the PUE at the previous 24 moments, respectively. The outlier judgment standard is the 3σ criterion. Calculate the mean μ and standard deviation σ of each characteristic parameter in the original data:

[0090] ;

[0091] ;

[0092] L 1= m -3 s ;

[0093] L 2= m +3 s ;

[0094] in, X i For a certain column i Energy efficiency data, N is the total number of samples, m is the mean of the data in this column, s is the standard deviation of the data in this column, L 1 is the lower limit of the abnormal threshold, L 2 is the upper limit of the abnormal threshold, and the abnormal value judgment standard is that the sample value is less than the lower limit L 1 or greater than the upper limit of judgment L 2;

[0095] If a feature or a data does not fall within [ m -3 s , m +3 s] interval, it is judged as an outlier, and the data point where the outlier is located is considered an outlier, and the data is removed;

[0096] S22, such as Figure 2 The figure shows the energy efficiency data interpolation diagram. The abnormal points eliminated by S21 need to be interpolated. The interpolation method is:

[0097] X i = X i+24k ;

[0098] in, X i For a certain column i Energy efficiency data, that is, the energy efficiency value at a certain moment, X i+24k Before k The energy efficiency value at the same time of the day, when k =1, if X i+24 If it is a normal value, take X i+24 Assign the value of X i ,like X i+24 If it is an abnormal value, then look at k =2, X i+24*2 Is it a normal value? If so, assign it to X i If not, see k =3, until X i+24k Until the normal value is reached, the normal value X i+24k Assign to X i ;

[0099] The PUE at the previous moment and the PUE at the previous 24 moments after data cleaning and interpolation are used as candidate feature variables;

[0100] S23. Detect the distortion values ​​of the outdoor temperature and outdoor humidity using a discrimination method:

[0101] ;

[0102] in, Y i For a certain column i The outlier criterion for outdoor meteorological data is that the difference between the sample value and the mean of the data before and after is greater than 5;

[0103] If the difference between a certain feature data and the mean of the data before and after it is greater than 5, it is judged as an outlier, and the data point where the outlier is located is considered an outlier and is removed;

[0104] S24: The abnormal points eliminated by S23 need to be interpolated. The interpolation method is the mean interpolation method:

[0105] ;

[0106] in, Y i For a certain column i outdoor meteorological data, take the mean of the data before and after and assign it to Y i ;

[0107] The outdoor temperature and outdoor humidity after data cleaning and interpolation are used as candidate characteristic variables;

[0108] S25. Arrange all characteristic variables and the current PUE into a table;

[0109] S26: Detect abnormal values ​​of the current PUE. The abnormal value judgment standard is 3. s Criteria, calculate the mean in the original data m and standard deviation s ;

[0110] If a data does not fall within [ m -3 s , m +3 s ] interval, it is judged as an outlier, and the data point where the outlier is located is considered an outlier, and the row of data is removed. Finally, after preprocessing, a total of 14513 valid data are obtained.

[0111] S3. Data feature selection: performing feature selection on the pre-processed historical data using a feature selection method; the historical data includes outdoor meteorological parameters, time data, and energy efficiency data;

[0112] Feature selection methods include: Pearson correlation coefficient method, embedding method, mutual information method, and recursive feature elimination method;

[0113] This example uses LightGBM for feature selection, an embedded feature selection method. First, the algorithm is trained on the initial feature set, and the importance of each feature is obtained using LightGBM's feature importance property. Next, features are pruned to obtain the final feature set: PUE for the previous moment, outdoor temperature, PUE for the previous 24 moments, time of day, outdoor humidity, and month.

[0114] S4, model training: Use the data after feature selection in S3 to train the regression prediction model, use the model hyperparameter optimization method to determine its optimal hyperparameter combination, and then obtain the trained regression prediction model;

[0115] S41. Randomly divide the data sample into training set and test set at a ratio of 4:1, thus obtaining an input training set of 11610 rows and 6 columns. D train_x , 11610 rows and 1 column output training set D train_y , 2903 rows and 6 columns of input test set D test_x , 2903 rows and 1 column output test set D test_y , which facilitates model training and validation testing in subsequent steps;

[0116] S42, build an artificial intelligence algorithm regression model, the model fitting uses artificial intelligence regression algorithm, including: vector machine, decision tree, neural network, ensemble learning, deep learning, generative adversarial network, this embodiment uses the LightGBM algorithm based on gradient boosting decision tree to build a regression prediction model, and uses the default parameters to train the data set. D train_x 、 D train_y Train and use the dataset D test_x 、 D test_y Perform validation tests to obtain model evaluation parameters R 2 =0.92;

[0117] S43. Taking the regression model evaluation index as the optimization target, the hyperparameter optimization method is used to optimize the hyperparameters of the regression model. The verification method is k fold cross validation;

[0118] Selecting regression model evaluation metrics R 2 To optimize the goal;

[0119] Model hyperparameter optimization methods include random search, grid search, Bayesian optimization, and evolutionary algorithm. This embodiment uses the Bayesian optimization algorithm to optimize the hyperparameters of the LightGBM algorithm.

[0120] The verification method is k fold cross validation, where k =10.

[0121] The model fitting coefficients obtained after Bayesian optimization in this embodiment R 2=0.93, the optimized LightGBM parameters are learning_rate=0.08, max_depth=12, n_estimators=111, num_leaves=41;

[0122] Save the model after optimization in the above steps.

[0123] S5. Generate a prediction error set: Use the regression prediction model trained in S4 to calculate the predicted value of the hourly energy efficiency, subtract the actual value from the predicted value, and obtain the prediction error set errs;

[0124] The input set of 14513 rows and 6 columns D x Import the trained Bayes-LightGBM model to get the predicted value D y_pred , calculate the prediction error value e :

[0125] e = D y - D y_pred ;

[0126] in, D y The obtained prediction error set errs is a data set with 14513 rows and 1 column.

[0127] S6. Constructing an adaptive prediction error interval: For the prediction error set in S5, the kernel density estimation algorithm is used to calculate its probability density, and the confidence interval is superimposed to obtain the adaptive prediction error interval;

[0128] S61, import the data sample prediction error set errs, use the hyperparameter optimization method grid search method to optimize the bandwidth of the kernel density estimation model, and the verification method is k fold cross validation, take k =5, after optimization, the optimal bandwidth h =0.1;

[0129] S62, use the optimal bandwidth h Establish a kernel density estimation model and calculate the prediction error values e The probability density of , its probability density function is:

[0130] ;

[0131] in, K ( e) is a kernel function, and the available kernel functions include Gaussian kernel function, gamma kernel function, uniform kernel function, and triangular kernel function. In this embodiment, Gaussian kernel function is selected;

[0132] S63, according to the confidence level of probability density is 1- α , construct the adaptive prediction error interval, and take α =0.1, and the final adaptive prediction error interval is (-0.0221, 0.0221).

[0133] S7, anomaly detection: Use the same data preprocessing method as step S2 to process the data collected in real time, and use the features selected in step S3 to construct an input set, input the trained regression prediction model, obtain the energy efficiency prediction value at the current moment, superimpose the adaptive prediction error interval, obtain the energy efficiency adaptive prediction interval at the current moment, and judge whether the actual energy efficiency value at the current moment is within the energy efficiency adaptive prediction interval at the current moment. If not, it is an energy efficiency anomaly, and output the detection result.

[0134] S8, model update, after the update time period T =60×24=1440h later, the above steps are repeated using the latest accumulated data set to update the regression prediction model and the adaptive prediction error interval.

[0135] Through the technical solution of this embodiment, the data center energy efficiency anomaly detection results are highly reliable, which helps to realize online real-time anomaly detection of data center energy efficiency, promptly discover anomalies and issue alarms, and remind management personnel to take corresponding countermeasures in time, which helps to manage data center energy efficiency and establish an efficient computer room.

[0136] It should be noted that, for the sake of convenience, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously.

[0137] Example 2:

[0138] Based on the same concept as the data center energy efficiency anomaly detection method based on an adaptive prediction interval in the above-mentioned embodiment, the present invention also provides a data center energy efficiency anomaly detection system based on an adaptive prediction interval, which can be used to execute the above-mentioned data center energy efficiency anomaly detection method based on a regression prediction model and an adaptive prediction interval. For ease of explanation, the structural diagram of an embodiment of a data center energy efficiency anomaly detection system based on a regression prediction model and an adaptive prediction interval only shows the parts related to the embodiment of the present invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation of the device, and may include more or fewer components than shown in the diagram, or combine certain components, or arrange the components differently.

[0139] like Figure 3 As shown, in another embodiment of the present application, a data center energy efficiency anomaly detection system 100 based on an adaptive prediction interval is provided, which includes a data acquisition module 101, a data preprocessing module 102, a feature selection module 103, a regression prediction model training module 104, an adaptive prediction error interval construction module 105, and an anomaly detection module 106;

[0140] The data acquisition module 101 is used to select specific energy efficiency indicators, collect and store outdoor meteorological parameters, time data and energy consumption data through sensors, and calculate energy efficiency data based on the energy consumption data;

[0141] The data preprocessing module 102 is used to preprocess historical data, including outlier detection and elimination, and missing value interpolation for multiple energy efficiency data; and distortion value elimination and interpolation for acquired outdoor meteorological data;

[0142] A feature selection module 103 is configured to perform feature selection on the pre-processed historical data using a feature selection method; the historical data includes outdoor meteorological parameters, time data, and energy efficiency data;

[0143] The model training module 104 is used to train the regression prediction model using the feature-selected data, determine the optimal hyperparameter combination using a model hyperparameter optimization method, and thereby obtain a trained regression prediction model;

[0144] After the update time period T Finally, use the latest accumulated data set to repeat the above steps and update the regression prediction model;

[0145] The adaptive prediction error interval construction module 105 is used to generate a prediction error set, calculate its probability density using a kernel density estimation algorithm, and superimpose the confidence interval to obtain an adaptive prediction error interval;

[0146] After the update time period TFinally, the above steps are repeated using the latest accumulated data set to update the adaptive prediction error interval;

[0147] The anomaly detection module 106 is used to input the data to be tested into the trained regression prediction model to obtain the energy efficiency prediction value at the current moment, superimpose the adaptive prediction error interval to obtain the energy efficiency adaptive prediction interval at the current moment, compare whether the actual energy efficiency value is within the energy efficiency adaptive prediction interval at the current moment to determine whether the energy efficiency is abnormal, and obtain the anomaly detection result.

[0148] It should be noted that the data center energy efficiency anomaly detection system based on adaptive prediction intervals of the present invention corresponds one-to-one to the data center energy efficiency anomaly detection method based on adaptive prediction intervals of the present invention. The technical features and beneficial effects described in the embodiment of the above-mentioned data center energy efficiency anomaly detection method based on adaptive prediction intervals are applicable to the embodiment of a data center energy efficiency anomaly detection system based on adaptive prediction intervals. For specific contents, please refer to the description in the embodiment of the method of the present invention. No further details will be given here. This is hereby declared.

[0149] It should be noted here that the system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above. The system is a data center energy efficiency anomaly detection method based on adaptive prediction intervals applied to the above embodiment.

[0150] Example 3:

[0151] like Figure 4 As shown, in another embodiment of the present application, a storage medium is further provided, storing a program, which, when executed by a processor, implements a data center energy efficiency anomaly detection method based on an adaptive prediction interval, specifically:

[0152] Select energy efficiency indicators, collect outdoor meteorological parameters, time data and energy consumption data, and calculate energy efficiency data based on energy consumption data;

[0153] Preprocess historical data, including outlier detection and elimination, and missing value interpolation for energy efficiency data; and distorted value elimination and interpolation for outdoor meteorological data;

[0154] Using a feature selection method to perform feature selection on the pre-processed historical data; the historical data includes outdoor meteorological parameters, time data and energy efficiency data;

[0155] Use the data after feature selection to train the regression prediction model, use the model hyperparameter optimization method to determine its optimal hyperparameter combination, and then obtain the trained regression prediction model;

[0156] Use the trained regression prediction model to calculate the predicted value of hourly energy efficiency, subtract the actual value from it, and obtain the prediction error set;

[0157] For the prediction error set, the kernel density estimation algorithm is used to calculate its probability density, and the confidence interval is superimposed to obtain the adaptive prediction error interval;

[0158] The data collected in real time is processed using the same data preprocessing method as that used for historical data, and the features selected in the feature selection method are used to construct an input set. The trained regression prediction model is input to obtain the energy efficiency prediction value at the current moment, and the adaptive prediction error interval is superimposed to obtain the energy efficiency adaptive prediction interval at the current moment. The actual energy efficiency value is compared to see whether it is within the energy efficiency adaptive prediction interval at the current moment to determine whether the energy efficiency is abnormal.

[0159] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0160] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A data center energy efficiency anomaly detection method based on adaptive prediction interval, characterized by: The steps include: Select energy efficiency indicators, collect outdoor meteorological parameters, time data and energy consumption data, and calculate energy efficiency data based on energy consumption data; Preprocess historical data, including outlier detection and elimination, and missing value interpolation for energy efficiency data; and distorted value elimination and interpolation for outdoor meteorological data; The method for detecting and eliminating outliers in energy efficiency data is as follows: σ Guidelines: ； ； L 1= μ -3 σ ； L 2= μ +3 σ ； in, X i For a certain column i Energy efficiency data, N is the total number of samples, μ is the mean of the data in this column, σ is the standard deviation of the data in this column, L 1 is the lower limit of the abnormal threshold, L 2 is the upper limit of the abnormal threshold, and the abnormal value judgment standard is that the sample value is less than the lower limit L 1 or greater than the upper limit of judgment L 2; Remove abnormal values ​​from the energy efficiency value at the current moment, the energy efficiency value at the previous moment, and the energy efficiency value at the previous 24 moments respectively; The method for interpolating missing values ​​of energy efficiency data is the interpolation method: X i = X i+24k ； in, X i For a certain column i Energy efficiency data, that is, the energy efficiency value at a certain moment, X i+24k Before k The energy efficiency value at the same time of the day, when k =1, if X i+24 If it is a normal value, take X i+24 Assign the value of X i ,like X i+24 If it is an abnormal value, then look at k =2, X i+24*2 Is it a normal value? If so, assign it to X i If not, see k =3, until X i+24k Until the normal value is reached, the normal value X i+24k Assign to X i ; Interpolate missing values ​​for the energy efficiency values ​​of the previous moment and the previous 24 moments; The method for removing distortion values ​​from outdoor meteorological data is a discriminant method: ； in, Y i For a certain column i The outlier criterion for outdoor meteorological data is that the difference between the sample value and the mean of the data before and after is greater than 5; The method for interpolating missing values ​​of outdoor meteorological data is the mean interpolation method: ； in, Y i For a certain column i outdoor meteorological data, take the mean of the data before and after and assign it to Y i ; Using a feature selection method to perform feature selection on the pre-processed historical data; the historical data includes outdoor meteorological parameters, time data and energy efficiency data; Use the data after feature selection to train the regression prediction model, use the model hyperparameter optimization method to determine its optimal hyperparameter combination, and then obtain the trained regression prediction model; Use the trained regression prediction model to calculate the predicted value of hourly energy efficiency, subtract the actual value from it, and obtain the prediction error set; For the prediction error set, the kernel density estimation algorithm is used to calculate its probability density, and the confidence interval is superimposed to obtain the adaptive prediction error interval; The data collected in real time is processed using the same data preprocessing method as that used for historical data, and the features selected in the feature selection method are used to construct an input set. The trained regression prediction model is input to obtain the energy efficiency prediction value at the current moment, and the adaptive prediction error interval is superimposed to obtain the energy efficiency adaptive prediction interval at the current moment. The actual energy efficiency value is compared to see whether it is within the energy efficiency adaptive prediction interval at the current moment to determine whether the energy efficiency is abnormal.

2. The data center energy efficiency anomaly detection method based on adaptive prediction interval according to claim 1 is characterized in that: The energy efficiency indicators include: electricity utilization rate, cooling load factor, power supply load factor and renewable energy utilization rate; The energy consumption data includes parameter data directly collected by sensors for calculating energy efficiency data; The energy efficiency data includes: the energy efficiency value at the current moment, the energy efficiency value at the previous moment, and the energy efficiency value at the previous 24 moments; The outdoor meteorological parameters include: outdoor temperature and outdoor humidity; The time data includes: year, month, date, time, season type and working day type; wherein the season type includes: cooling season, transition season, non-cooling season; working day type includes: working day, non-working day.

3. The data center energy efficiency anomaly detection method based on adaptive prediction interval according to claim 1 is characterized in that: The feature selection method adopts one of the Pearson correlation coefficient method, the embedding method, the mutual information method and the recursive feature elimination method.

4. The data center energy efficiency anomaly detection method based on adaptive prediction interval according to claim 1 is characterized in that: The regression prediction model is specifically: The regression prediction model is fitted using an artificial intelligence regression algorithm, using one of vector machines, decision trees, neural networks, ensemble learning, deep learning, and generative adversarial networks; The hyperparameter optimization method of the regression prediction model adopts one of the random search, grid search, Bayesian optimization and evolutionary algorithm.

5. The data center energy efficiency anomaly detection method based on adaptive prediction interval according to claim 4 is characterized in that: The specific process of fitting the regression prediction model is as follows: The data samples are randomly divided into training sets and test sets according to a certain ratio for model training and validation testing; Build an artificial intelligence algorithm regression prediction model and use default parameters to train and validate the data set; Taking the regression prediction model evaluation index as the optimization target, the hyperparameter optimization method is used to optimize the hyperparameters of the regression prediction model. The verification method is k fold cross validation; Save the regression prediction model optimized through the above steps; After the update time period T Finally, use the latest accumulated data set to repeat the above steps and update the regression prediction model.

6. The data center energy efficiency anomaly detection method based on adaptive prediction interval according to claim 1 is characterized in that: The kernel density estimation algorithm is used to calculate the probability density of the prediction error set, and the confidence interval is superimposed to obtain the adaptive prediction error interval. The specific process is as follows: Import the data sample prediction error set and use the hyperparameter optimization method to optimize the bandwidth of the kernel density estimation model. The verification method is k fold cross validation; Use optimal bandwidth h A kernel density estimation model is established to calculate the probability density of each prediction error value e. Its probability density function is: ； in, K ( e ) is a kernel function, and one of the Gaussian kernel function, gamma kernel function, uniform kernel function and triangular kernel function is used; The confidence level for calculating the probability density is 1- α The confidence interval of f α / 2 , f 1-α / 2 ], based on the functional relationship between probability density and prediction error value, an adaptive prediction error interval is constructed [ e 1, e 2]; After the update time period T Finally, the above steps are repeated using the latest accumulated data set to update the adaptive prediction error interval.

7. The data center energy efficiency anomaly detection method based on adaptive prediction interval according to claim 1, characterized in that: The energy efficiency prediction value at the current moment is obtained, and the adaptive prediction error interval is superimposed to obtain the energy efficiency adaptive prediction interval at the current moment, specifically: The energy efficiency prediction value at the current moment b , and the adaptive prediction error interval [ e 1, e 2] to obtain the current energy efficiency adaptive prediction interval [ b+e 1, b+e 2]; The comparison of whether the actual energy efficiency value is within the current energy efficiency adaptive prediction range to determine whether the energy efficiency is abnormal is specifically as follows: Determine the actual energy efficiency value at the current moment c Whether it is in the current energy efficiency adaptive prediction range [ b+e 1, b+e 2], if not, it is an energy efficiency abnormality.

8. Data center energy efficiency anomaly detection system based on adaptive prediction interval, characterized by: A data center energy efficiency anomaly detection method based on adaptive prediction intervals applied to any one of claims 1-7, comprising a data acquisition module, a data preprocessing module, a feature selection module, a regression prediction model training module, an adaptive prediction error interval construction module, and an anomaly detection module; The data acquisition module is used to select energy efficiency indicators, collect outdoor meteorological parameters, time data and energy consumption data through sensors, and calculate energy efficiency data based on the energy consumption data; The data preprocessing module is used to preprocess historical data, including detecting and eliminating outliers and interpolating missing values ​​for energy efficiency data; and eliminating and interpolating distorted values ​​for outdoor meteorological data; The feature selection module is used to perform feature selection on the pre-processed historical data using a feature selection method; the historical data includes outdoor meteorological parameters, time data and energy efficiency data; The model training module is used to train the regression prediction model using the data after feature selection, and use the model hyperparameter optimization method to determine its optimal hyperparameter combination, thereby obtaining a trained regression prediction model; After the update time period T Finally, use the latest accumulated data set to repeat the above steps and update the regression prediction model; The adaptive prediction error interval construction module is used to generate a prediction error set, calculate its probability density using a kernel density estimation algorithm, and superimpose confidence intervals to obtain an adaptive prediction error interval; After the update time period T Finally, the above steps are repeated using the latest accumulated data set to update the adaptive prediction error interval; The anomaly detection module is used to input the data to be tested into the trained regression prediction model to obtain the energy efficiency prediction value at the current moment, superimpose the adaptive prediction error interval to obtain the energy efficiency adaptive prediction interval at the current moment, compare whether the actual energy efficiency value is within the energy efficiency adaptive prediction interval at the current moment to determine whether the energy efficiency is abnormal, and obtain the anomaly detection result.

9. A storage medium storing a program, characterized in that: When the program is executed by a processor, the data center energy efficiency anomaly detection method based on adaptive prediction interval according to any one of claims 1 to 7 is implemented.

Images