A demand combination prediction method and system based on data center computing power load

Abstract

The present invention relates to the field of computing power load technology, a method and system for demand combination prediction based on data center computing power load, receiving a demand instruction for data center computing power load, starting a computing power load demand device for data center, obtaining a daily identification user data set based on the extracted user site, constructing a daily load curve based on the daily identification user data set, constructing a load fluctuation characteristic matrix of the extracted user site based on the load standard value, load skewness value, load kurtosis value and load trend value, predicting the computing power load of the extracted user site based on the data to be analyzed and the computing power load prediction network, performing classification on the load intensity set, obtaining a high-load site set, a medium-load site set and a low-load site set, and completing the demand combination prediction of the computing power load of the data center. The present invention can use the historical computing power load data of the data center to predict the computing power load demand of the data center, thereby more accurately supplying energy to the data center.

Description

Technical Field

[0001] The present invention relates to the field of computing power load technology, and in particular to a method and system for forecasting demand combination based on data center computing power load. Background Art

[0002] A data center is a facility that houses a large number of servers, storage devices, network equipment, and other computing resources. It provides data storage, processing, and management services for various organizations and enterprises. Computing load refers to the total amount of computing resources required to process data and perform computing tasks in a data center or computing system. In the context of the rapid development of technologies such as big data and artificial intelligence, computing load is the core resource that supports the operation of these technologies, and its performance is directly related to the efficiency and quality of data processing.

[0003] Traditional forecasting models are usually used for short-term and long-term load forecasting, including the Autoregressive Integrated Moving Average (ARIMA) model.

[0004] Although the above content can realize computing load prediction, it may not be able to process large-scale data sets due to insufficient computing power, and cannot fully capture the complexity and nonlinear characteristics of load changes. These problems may lead to a decrease in computing load resource utilization and affect user use. Summary of the invention

[0005] The present invention provides a demand combination prediction method and system based on the computing power load of a data center. Its main purpose is to solve the problem of using the historical computing power load data of the data center to predict the computing power load demand of the data center, so as to more accurately supply energy to the data center.

[0006] To achieve the above object, the present invention provides a demand combination forecasting method based on data center computing load, comprising:

[0007] Receive a demand instruction for a computing power load of a data center, start a computing power load demand device for the data center based on the demand instruction, obtain multiple user sites based on the computing power load demand device for the data center, extract user sites from the multiple user sites in sequence, and perform the following operations on the extracted user sites:

[0008] A daily identification user data set is obtained based on the extracted user site, and a daily load curve is constructed according to the daily identification user data set, wherein the ordinate of the daily load curve is the user computing load, and the abscissa is the daily sequence;

[0009] The daily load curve is normalized to obtain a standard curve, a standard maximum value and a standard minimum value are obtained by using the Min-Max method and the standard curve, a standardized daily load curve is constructed by using the standard maximum value and the standard minimum value, a load standard value of the standardized daily load curve is calculated, and a load kurtosis value and a load skewness value are calculated based on the load standard value of the standardized daily load curve;

[0010] The standardized daily load curve is fitted into a load straight line equation by using a pre-constructed straight line fitting method, the load straight line slope is extracted according to the load straight line equation, and the load trend value of the standardized daily load curve is calculated according to the load straight line slope;

[0011] Constructing a load fluctuation characteristic matrix of the extracted user site based on the load standard value, load skewness value, load kurtosis value and load trend value;

[0012] Calculate the computing load mean, input the computing load mean and the load fluctuation characteristic matrix into a pre-built computing load prediction network to obtain data to be analyzed, and predict the computing load of the extracted user site based on the data to be analyzed and the computing load prediction network;

[0013] Summarizing the computing load to obtain a load intensity set corresponding to multiple user sites, wherein the computing load prediction network is a BP neural network;

[0014] Classify the load intensity set to obtain a high-load site set, a medium-load site set, and a low-load site set, completing the demand combination forecast of the data center computing power load.

[0015] Optionally, the acquiring a daily identification user data set based on the extracted user site includes:

[0016] Extracting a day sequence according to the day sequence;

[0017] Based on the extracted user site, the user computing power load and the total computing power consumption of the user are obtained, and based on the user computing power load and the total computing power consumption of the user, a user PUE evaluation value corresponding to the extracted user site is calculated;

[0018] Based on the extracted user site, the extracted day sequence, the user PUE evaluation value, the user computing power load and the user computing power total consumption, an identification operation is performed on the extracted user site to obtain daily identification user data;

[0019] The daily identified user data is aggregated to obtain the daily identified user data set of the extracted user site, wherein the calculation formula of the user PUE evaluation value is as follows:

[0020]

[0021] Among them, PUE is the user's PUE evaluation value For the user's computing load, The total computing power consumed by the user.

[0022] Optionally, the obtaining of the standard maximum value and the standard minimum value by using the Min-Max method and the standard curve includes:

[0023] According to the daily load curve, the maximum value of the original load data point set and the minimum value of the original load data point set are obtained, the original load data points are extracted from the original load data point set in sequence, and the following operations are performed on the extracted original load data points:

[0024] The Min-Max method is used to normalize the original load data points to obtain the standard value, and the range of the standard value is [0, 1]. The formula is:

[0025] ;

[0026] in, represents the normalized standard value, M represents the original load data point, is the minimum value representing the original load data set, Represents the maximum value of the original load data set.

[0027] The standard values ​​are aggregated to obtain a standard value set, and a standard maximum value and a standard minimum value are extracted from the standard value set, where the standard maximum value is 1 and the standard minimum value is 0.

[0028] Optionally, the calculating of the load standard value of the standardized daily load curve, and the calculating of the load kurtosis value and the load skewness value based on the load standard value of the standardized daily load curve, include:

[0029] The formula for calculating the load standard value is:

[0030]

[0031] Among them, s represents the load standard value, m represents the total number of daily sequences corresponding to the daily sequence in the standardized daily load curve, Represents the average standard user computing load of the standardized daily load curve, represents the standard user computing power load on the i-th day in the standardized daily load curve;

[0032] The load kurtosis value is calculated according to the load standard value of the standardized daily load curve, and the formula is:

[0033]

[0034] Where, k represents the load kurtosis value;

[0035] The calculation formula of the load skewness value of the standardized daily load curve is as follows:

[0036]

[0037] Where p represents the load skewness value of the standardized daily load curve.

[0038] Optionally, fitting the standardized daily load curve into a load straight line equation by using a pre-constructed straight line fitting method, extracting the load straight line slope according to the load straight line equation, and calculating the load trend value of the standardized daily load curve according to the load straight line slope, includes:

[0039] The data of the standardized daily load curve is obtained, and the data of the standardized daily load curve is plotted into a visual scatter plot. A design matrix is ​​constructed based on the visual scatter plot, and a parameter vector of the load straight line equation is calculated using the least squares method and the design matrix, wherein the formula for calculating the parameter vector of the load straight line equation is:

[0040]

[0041] in, represents the parameter vector of the load line equation, X represents the design matrix, and Y represents the vector of the standard user computing power load. represents the transposed matrix of the design matrix X;

[0042] The load line slope of the load line equation is calculated according to the parameter vector of the load line equation, wherein the calculation formulas of the load line equation and the load line slope are as follows:

[0043] ;

[0044] ;

[0045] Where z represents the linear user computing power load in the load line equation, and x represents the linear time. is represented as the intercept term, It is expressed as the slope of the load line; n represents the total number of scattered points in the visual scatter plot;

[0046] The load trend value of the standardized daily load curve is calculated according to the load straight line slope.

[0047] Optionally, the load fluctuation characteristic matrix is ​​expressed as follows:

[0048]

[0049] in, represents the j-th load fluctuation characteristic matrix, represents the load standard value of the jth load fluctuation characteristic matrix, The load skewness value of the jth load fluctuation characteristic matrix, represents the load kurtosis value of the jth load fluctuation characteristic matrix, Represents the load trend value of the jth load fluctuation characteristic matrix.

[0050] Optionally, calculating the computing load mean includes:

[0051] The total computing load is calculated according to the extracted daily load curve of the user site, the daily sequence number corresponding to the daily load curve is obtained, and the computing load mean is calculated based on the daily sequence number of the total computing load set.

[0052] Optionally, the step of inputting the computing power load mean and the load fluctuation characteristic matrix into a pre-built computing power load prediction network to obtain data to be analyzed includes:

[0053] Obtaining a computing power load original network, wherein the input layer of the computing power load original network is a computing power load mean input node, a load standard value input node, a load skewness value input node, a load kurtosis value input node, and a load trend value input node, and the output layer is a predicted computing power load output node;

[0054] Obtain computing power load training data, train the original computing power load network according to the computing power load training data, and obtain a computing power load prediction network, wherein the computing power load training data includes: a historical computing power load mean value set, a historical load standard value set, a historical load skewness value set, a historical load kurtosis value set, a historical load trend value set, and a historical computing power load set.

[0055] Optionally, the classifying the load intensity set to obtain a high-load site set, a medium-load site set, and a low-load site set includes:

[0056] The preset number of clusters is set to 3, and three load intensities are randomly selected from the load intensity set as the high load centroid, the medium load centroid, and the low load centroid respectively;

[0057] Extracting the load intensities to be clustered in the load intensity set in turn, and calculating the high load clustering distance, the medium load clustering distance and the low load clustering distance between the load intensities to be clustered and the high load centroid, the medium load centroid and the low load centroid;

[0058] Clustering the load intensities to be clustered according to the high load clustering distance, the medium load clustering distance, and the low load clustering distance to obtain a high load clustering cluster, a medium load clustering cluster, and a low load clustering cluster;

[0059] Extracting an iterative high-load centroid, an iterative medium-load centroid and an iterative low-load centroid from the high-load cluster, the medium-load cluster and the low-load cluster respectively;

[0060] The iterative high load centroid, iterative medium load centroid and iterative low load centroid are used to update the high load centroid, medium load centroid and low load centroid respectively, and return to the above step of extracting the load intensities to be clustered in the load intensity set in sequence until the load intensity set completes the extraction of the load intensities to be clustered, and a high load site set, a medium load site set and a low load site set are obtained.

[0061] To achieve the above object, the present invention also provides a demand combination prediction system based on data center computing load, comprising:

[0062] A computing load demand instruction receiving module is used to receive a computing load demand instruction of a data center, start a computing load demand device of the data center based on the demand instruction, obtain multiple user sites based on the computing load demand device of the data center, and extract user sites from the multiple user sites in sequence;

[0063] A load fluctuation characteristic matrix construction module is used to obtain a daily identification user data set based on the extracted user site, and to construct a daily load curve according to the daily identification user data set, wherein the ordinate of the daily load curve is the user computing power load, and the abscissa is the day sequence, and the daily load curve is normalized to obtain a standard curve, and the standard maximum value and the standard minimum value are obtained by using the Min-Max method and the standard curve, and a standardized daily load curve is constructed by using the standard maximum value and the standard minimum value, and the load standard value of the standardized daily load curve is calculated, and the load kurtosis value and the load skewness value are calculated based on the load standard value of the standardized daily load curve, and the standardized daily load curve is fitted into a load straight line equation by using a pre-constructed straight line fitting method, and the load straight line slope is extracted according to the load straight line equation, and the load trend value of the standardized daily load curve is calculated according to the load straight line slope, and the load fluctuation characteristic matrix of the extracted user site is constructed based on the load standard value, load skewness value, load kurtosis value and load trend value;

[0064] A load intensity set composition module is used to calculate a computing power load mean, input the computing power load mean and the load fluctuation characteristic matrix into a pre-built computing power load prediction network to obtain data to be analyzed, predict the computing power load of the extracted user site based on the data to be analyzed and the computing power load prediction network, summarize the computing power load, and obtain a load intensity set corresponding to multiple user sites, wherein the computing power load prediction network is a BP neural network;

[0065] The load site set classification module is used to classify the load intensity set to obtain high-load site sets, medium-load site sets and low-load site sets, and complete the demand combination prediction of the data center computing power load.

[0066] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0067] A memory storing at least one instruction;

[0068] The processor executes the instructions stored in the memory to implement the above-mentioned demand combination prediction method based on data center computing power load.

[0069] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned demand combination prediction method based on data center computing power load.

[0070] In order to solve the problems described in the background technology, the present invention receives a demand instruction for the computing power load of a data center, starts a computing power load demand device of a data center based on the demand instruction, obtains multiple user sites based on the computing power load demand device of the data center, and extracts user sites from the multiple user sites in sequence. The present invention obtains data sets of multiple user sites through the computing power demand device of the data center, and can predict the computing power load of multiple user sites and supply energy to multiple user sites. The present invention obtains a daily identified user data set based on the extracted user sites, and constructs a daily load curve based on the daily identified user data set, wherein the ordinate of the daily load curve is the user computing power load, and the abscissa is the daily sequence. The daily load curve is constructed through the daily identified user data set, and the computing power load used by the user in one day can be intuitively viewed through the daily load curve. The present invention obtains a standard curve by normalizing the daily load curve, obtains a standard maximum value and a standard minimum value by using the Min-Max method and the standard curve, constructs a standardized daily load curve by using the standard maximum value and the standard minimum value, calculates the load standard value of the standardized daily load curve, calculates the load kurtosis value and the load skewness value based on the load standard value of the standardized daily load curve, fits the standardized daily load curve into a load straight line equation by using a pre-constructed straight line fitting method, extracts the load straight line slope according to the load straight line equation, calculates the load trend value of the standardized daily load curve according to the load straight line slope, constructs the extracted load fluctuation characteristic matrix of the user site based on the load standard value, the load skewness value, the load kurtosis value and the load trend value, and normalizes different daily load curves through normalization processing for analysis By comparison, the load standard deviation reflects the degree of load fluctuation, the kurtosis value reflects the distribution characteristics of the load, the load skewness value reflects the degree of load skewness, and the load trend value reflects the load fluctuation trend. The above values ​​construct a load fluctuation characteristic matrix to facilitate load prediction; the present invention calculates the mean value of computing power load, inputs the computing power load mean value and the load fluctuation characteristic matrix into the pre-constructed computing power load prediction network, obtains the data to be analyzed, predicts the computing power load of the extracted user site according to the data to be analyzed and the computing power load prediction network, summarizes the computing power load, and obtains the load intensity set corresponding to multiple user sites, wherein the computing power load prediction network is a BP neural network, and by constructing a BP neural network, it can adaptively predict the load characteristics of different user sites and generate a highly adaptable load intensity set. The present invention classifies the load intensity set to obtain a high-load site set, a medium-load site set, and a low-load site set, and completes the demand combination prediction of the computing power load of the data center. Different load intensities can be understood through classification, which helps the data center to allocate computing power resources more effectively, meet the load requirements of each user site, and avoid resource waste. Therefore, the present invention can use the historical computing load data of the data center to predict the computing load demand of the data center, thereby providing energy for the data center more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A schematic diagram of a flow chart of a demand combination prediction and monitoring method based on data center computing load provided by an embodiment of the present invention;

[0072] Figure 2 A functional module diagram of a demand combination prediction and monitoring device based on data center computing load provided by an embodiment of the present invention;

[0073] Figure 3 A schematic diagram of the structure of an electronic device for implementing the demand combination prediction model method based on data center computing power load provided by one embodiment of the present invention.

[0074] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0075] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0076] The embodiment of the present application provides a demand combination prediction method based on the computing power load of a data center. The execution subject of the demand combination prediction method based on the computing power load of a data center includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the demand combination prediction method based on the computing power load of a data center can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0077] Reference Figure 1 FIG. 1 is a flow chart of a method for forecasting a demand combination based on a computing load of a data center provided by an embodiment of the present invention. In this embodiment, the method for forecasting a demand combination based on a computing load of a data center includes:

[0078] S1. Receive a demand instruction for data center computing load, start a data center computing load demand device based on the demand instruction, obtain multiple user sites based on the data center computing load demand device, and extract user sites from the multiple user sites in sequence.

[0079] It needs to be explained that demand instructions are instructions or commands issued by the data center to meet specific needs. User sites are the places where users access the Internet. Computing load refers to the workload of the data center at a certain moment.

[0080] For example, there is a data center. During the period from 9 am to 5 pm from Monday to Friday, the data center load demand surges due to a large number of users working online and processing data. In order to cope with the energy consumption during this peak period, the data center issues a demand instruction to increase energy supply to meet the energy demand. When a small number of users are working online and processing data, the load demand decreases. In order to meet the energy consumption during this period, the data center issues a demand instruction to reduce energy supply to save energy.

[0081] S2. Obtain a daily identified user data set based on the extracted user site, and construct a daily load curve according to the daily identified user data set, wherein the ordinate of the daily load curve is the user computing power load, and the abscissa is the daily sequence.

[0082] It should be explained that the daily load curve is a curve used to describe the changes in user computing load corresponding to the daily sequence within a daily sequence.

[0083] Furthermore, the step of acquiring a daily identification user data set based on the extracted user site includes:

[0084] Extracting a day sequence according to the day sequence;

[0085] Based on the extracted user site, the user computing power load and the total computing power consumption of the user are obtained, and the user PUE evaluation value corresponding to the extracted user site is calculated based on the user computing power load and the total computing power consumption of the user;

[0086] Based on the extracted user site, the extracted day sequence, the user PUE evaluation value, the user computing power load, and the user computing power total consumption, an identification operation is performed on the extracted user site to obtain daily identification user data;

[0087] The daily identified user data is aggregated to obtain the daily identified user data set of the extracted user site, wherein the calculation formula of the user PUE evaluation value is as follows:

[0088] ;

[0089] Among them, PUE is the user's PUE evaluation value, For the user's computing load, The total computing power consumed by the user.

[0090] It should be explained that the total user computing power consumption is the total energy consumed by all computing devices in the data center facility when performing tasks. The PUE evaluation value is a key indicator for measuring the energy efficiency of a data center. The PUE evaluation value is defined as the ratio of the total equipment energy consumption of the data center to the energy consumption of IT equipment. The extracted date sequence is the extracted data points recorded by date, which is used to analyze trends over time. The identification operation is an operation that marks or identifies data.

[0091] It is understandable that the PUE value represents the energy utilization efficiency of the user site. When supplying energy, more energy is needed to be supplied to user sites with low energy utilization efficiency to meet the computing load demand, and relatively less energy is provided to user sites with high energy utilization efficiency to meet the computing load demand. Therefore, in order to more accurately provide energy to meet the computing load demands of different user sites, it is necessary to calculate the PUE value.

[0092] For example, suppose we want to calculate the PUE evaluation value of user X, and it is known that the total computing power consumption of user X is = 100kWh, the computing load of user X is =250kWh, then the PUE is 2.5, so the total computing power consumption used by user X is 2.5 times the computing power load.

[0093] S3. Normalize the daily load curve to obtain a standard curve, use the Min-Max method and the standard curve to obtain the standard maximum value and the standard minimum value, use the standard maximum value and the standard minimum value to construct a standardized daily load curve, and calculate the load standard value of the standardized daily load curve.

[0094] It should be explained that the standard curve adjusts the value of the daily load curve to a unified standard, and its value range is [0, 1]. The Min-Max method is a common data minimum and maximum standardization method, which is used to scale the data to a specific value range. The load standard value represents the value obtained by the degree of deviation between the user's computing power load and the daily sequence within a certain time range.

[0095] Furthermore, the method of using the Min-Max method and the standard curve to obtain the standard maximum value and the standard minimum value includes:

[0096] According to the daily load curve, the maximum value of the original load data point set and the minimum value of the original load data point set are obtained, the original load data points are extracted from the original load data point set in sequence, and the following operations are performed on the extracted original load data points:

[0097] The Min-Max method is used to normalize the original load data points to obtain the standard value, and the range of the standard value is [0, 1]. The formula is:

[0098] ;

[0099] in, represents the normalized standard value, M represents the original load data point, is the minimum value representing the original load data set, Represents the maximum value of the original load data set.

[0100] The standard values ​​are aggregated to obtain a standard value set, and a standard maximum value and a standard minimum value are extracted from the standard value set, where the standard maximum value is 1 and the standard minimum value is 0.

[0101] It can be understood that the original load data point is the load data point on the daily load curve, the standard value is the value of the load data point after standardization of the original load data point, and the standard value set is the set obtained by summarizing the values ​​of the load data point after standardization of the original load data point.

[0102] S4. Calculate the load kurtosis value and the load skewness value based on the load standard value of the standardized daily load curve, fit the standardized daily load curve into a load straight line equation using a pre-constructed straight line fitting method, extract the load straight line slope according to the load straight line equation, calculate the load trend value of the daily load curve according to the load straight line slope, and construct the extracted load fluctuation characteristic matrix of the user site based on the load standard value, load skewness value, load kurtosis value and load trend value.

[0103] It should be explained that the load kurtosis value is a statistical value that describes the peak characteristics of the load distribution in the data center. The load skewness value is the value of the degree of skewness of the load distribution. Linear fitting is a way of fitting a curve into a straight line. The load line slope refers to the slope of the linear equation of the load and time relationship after linear fitting. The load trend value is a numerical value that represents the development trend of the computing load of the extracted user site. The load trend value is equal to the load line slope. The load fluctuation characteristic matrix is ​​a matrix composed of the load standard value, load skewness value, load kurtosis value and load trend value.

[0104] Further, the calculating of the load standard value of the standardized daily load curve, and the calculating of the load kurtosis value and the load skewness value based on the load standard value of the standardized daily load curve, include:

[0105] The formula for calculating the load standard value is:

[0106]

[0107] Among them, s represents the load standard value, m represents the total number of daily sequences corresponding to the daily sequence in the standardized daily load curve, Represents the average standard user computing load of the standardized daily load curve, represents the standard user computing power load on the i-th day in the standardized daily load curve;

[0108] The load kurtosis value is calculated according to the load standard value of the standardized daily load curve, and the formula is:

[0109]

[0110] Where, k represents the load kurtosis value;

[0111] The calculation formula of the load skewness value of the standardized daily load curve is as follows:

[0112] ;

[0113] Where p represents the load skewness value of the standardized daily load curve.

[0114] Furthermore, the method of fitting the standardized daily load curve into a load straight line equation by using a pre-constructed straight line fitting method, extracting the load straight line slope according to the load straight line equation, and calculating the load trend value of the standardized daily load curve according to the load straight line slope includes:

[0115] The data of the standardized daily load curve is obtained, and the data of the standardized daily load curve is plotted into a visual scatter plot. A design matrix is ​​constructed based on the visual scatter plot, and a parameter vector of the load straight line equation is calculated using the least squares method and the design matrix, wherein the formula for calculating the parameter vector of the load straight line equation is:

[0116]

[0117] in, represents the parameter vector of the load line equation, X represents the design matrix, and Y represents the vector of the standard user computing power load. represents the transposed matrix of the design matrix X;

[0118] The load line slope of the load line equation is calculated according to the parameter vector of the load line equation, wherein the calculation formulas of the load line equation and the load line slope are as follows:

[0119] ;

[0120] ;

[0121] Where z represents the linear user computing power load in the load line equation, and x represents the linear time. is represented as the intercept term, It is expressed as the slope of the load line; n represents the total number of scattered points in the visual scatter plot;

[0122] The load trend value of the standardized daily load curve is calculated according to the load straight line slope.

[0123] It can be understood that the visualized scatter plot is an image that represents the data points of the standardized daily load curve with scattered points on a plane rectangular coordinate system. The visualized scatter plot is used to show the distribution between the daily sequence and the user computing power load, wherein the horizontal coordinate of the scatter points in the visualized scatter plot is the daily sequence in the standardized daily load curve, and the vertical coordinate of the scatter points is the standard user computing power load corresponding to the daily sequence of the standardized daily load curve. The data of the standardized daily load curve is the daily sequence of the standardized daily load curve and the user computing power load. The standard user computing power load is the value in the standard curve that represents the trend of the user computing power load, and the linear user computing power load is the vertical coordinate in the load straight line equation, which is used to reflect the user computing power load in the load straight line equation.

[0124] It can be understood that the load straight line equation is a straight line obtained by fitting the standardized daily load curve into a linear function, and the linear time is the horizontal coordinate in the load straight line equation, and the linear time is a continuous time value obtained by fitting the daily sequence of the standardized daily load curve corresponding to the daily sequence. Since the load straight line equation is a continuous equation, the discrete daily sequence is also converted into a continuous time value after being input into the load straight line equation.

[0125] Furthermore, the load fluctuation characteristic matrix is ​​expressed as follows:

[0126] ;

[0127] in, represents the j-th load fluctuation characteristic matrix, represents the load standard value of the jth load fluctuation characteristic matrix, The load skewness value of the jth load fluctuation characteristic matrix, represents the load kurtosis value of the jth load fluctuation characteristic matrix, Represents the load trend value of the jth load fluctuation characteristic matrix.

[0128] S5. Calculate the mean value of the computing load, input the mean value of the computing load and the load fluctuation characteristic matrix into a pre-built computing load prediction network to obtain the data to be analyzed, predict the computing load of the extracted user site based on the data to be analyzed and the computing load prediction network, summarize the computing load, and obtain a load intensity set corresponding to multiple user sites, wherein the computing load prediction network is a BP neural network.

[0129] It should be explained that the data to be analyzed is the data obtained by inputting the mean value of computing power load and the load fluctuation characteristic matrix into the pre-built computing power load prediction network. The load intensity set is a set obtained by summarizing multiple computing power loads.

[0130] Furthermore, the computing power load mean value includes:

[0131] The total computing load is calculated according to the extracted daily load curve of the user site, the daily sequence number corresponding to the daily load curve is obtained, and the computing load mean is calculated based on the daily sequence number of the total computing load set.

[0132] It should be explained that the total computing load is the sum of the user computing loads corresponding to the day number in the daily load curve. The day number is the number of days that have passed in the day sequence. The average computing load is the representative value of the computing load of the extracted user site in one day. The average computing load is equal to the total computing load divided by the day number.

[0133] Furthermore, the computing load mean and the load fluctuation characteristic matrix are input into a pre-built computing load prediction network to obtain data to be analyzed, including:

[0134] Obtaining a computing power load original network, wherein the input layer of the computing power load original network is a computing power load mean input node, a load standard value input node, a load skewness value input node, a load kurtosis value input node, and a load trend value input node, and the output layer is a predicted computing power load output node;

[0135] Obtain computing power load training data, train the original computing power load network according to the computing power load training data, and obtain a computing power load prediction network, wherein the computing power load training data includes: a historical computing power load mean value set, a historical load standard value set, a historical load skewness value set, a historical load kurtosis value set, a historical load trend value set, and a historical computing power load set.

[0136] It can be understood that the original computing load network is the BP neural network before data input and iteration, the computing load mean value input node is the neural node for inputting the computing load mean value, the load standard value input node is the neural node for inputting the load standard value, the load skewness value input node is the neural node for inputting the load skewness value, the load kurtosis value input node is the neural node for inputting the load kurtosis value, the load trend value input node is the neural node for inputting the load trend value, and the predicted computing load output node is the neural node for outputting the computing load in the original computing load network. The computing load prediction network is the BP neural network after data input and iteration. The historical computing power load mean value set is a set consisting of the computing power load mean values ​​of the input samples of the computing power load prediction network, the historical load standard value set is a set consisting of the load standard values ​​of the input samples of the computing power load prediction network, the historical load skewness value set is a set consisting of the load skewness values ​​of the input samples of the computing power load prediction network, the historical load kurtosis value set is a set consisting of the load kurtosis values ​​of the input samples of the computing power load prediction network, the historical load trend value set is a set consisting of the load trend values ​​of the input samples of the computing power load prediction network, and the historical computing power load set is a set consisting of the computing power load of the input samples of the computing power load prediction network.

[0137] It can be understood that by constructing a computing load prediction network model based on the BP neural network, after inputting the expected prediction time and the expected prediction user site, the computing load of the user site can be predicted. Furthermore, the computing load prediction network model can analyze and predict the computing load of multiple sites, and function according to the computing load of each user site in the multiple user sites, thereby realizing energy supply distribution according to the load demand of the user site.

[0138] S6. Classify the load intensity set to obtain a high-load site set, a medium-load site set, and a low-load site set, and complete the demand combination forecast of the data center computing power load.

[0139] It should be explained that the high-load site set is a set of sites with high load consumption requirements, the medium-load site set is a set of sites with medium load consumption requirements, and the low-load site set is a set of sites with low load consumption requirements.

[0140] Furthermore, the classification of the load intensity set to obtain a high-load site set, a medium-load site set and a low-load site set includes:

[0141] The preset number of clusters is set to 3, and three load intensities are randomly selected from the load intensity set as the high load centroid, the medium load centroid, and the low load centroid respectively;

[0142] Extracting the load intensities to be clustered in the load intensity set in turn, and calculating the high load clustering distance, the medium load clustering distance and the low load clustering distance between the load intensities to be clustered and the high load centroid, the medium load centroid and the low load centroid;

[0143] Clustering the load intensities to be clustered according to the high load clustering distance, the medium load clustering distance, and the low load clustering distance to obtain a high load clustering cluster, a medium load clustering cluster, and a low load clustering cluster;

[0144] Extracting an iterative high-load centroid, an iterative medium-load centroid and an iterative low-load centroid from the high-load cluster, the medium-load cluster and the low-load cluster respectively;

[0145] The iterative high load centroid, iterative medium load centroid and iterative low load centroid are used to update the high load centroid, medium load centroid and low load centroid respectively, and return to the above step of extracting the load intensities to be clustered in the load intensity set in sequence until the load intensity set completes the extraction of the load intensities to be clustered, and a high load site set, a medium load site set and a low load site set are obtained.

[0146] It should be explained that the number of clusters is the number of divided sets. The high-load centroid is the load intensity corresponding to the cluster center of the computing load corresponding to the high-load site set that is finally divided, the medium-load centroid is the load intensity corresponding to the cluster center of the computing load corresponding to the medium-load site set that is finally divided, and the low-load centroid is the load intensity corresponding to the cluster center of the computing load corresponding to the low-load site set that is finally divided. The load intensity to be clustered is the load intensity that needs to be clustered and extracted from the load intensity set. The high-load clustering distance is the distance between the load intensity to be clustered and the high-load centroid. Optionally, the distance in the high-load clustering distance is calculated using the absolute difference, the square of the absolute difference or other distances. The embodiment of the present invention is not limited here, and the medium-load clustering distance and the low-load clustering distance use the same calculation method as the high-load clustering distance. Therefore, by the same token, the medium-load clustering distance is the distance between the load intensity to be clustered and the medium-load centroid, and the low-load clustering distance is the distance between the load intensity to be clustered and the low-load centroid. The high-load clustering cluster is the cluster that is clustered into the high-load site set, the medium-load clustering cluster is the cluster that is clustered into the medium-load site set, and the low-load clustering cluster is the cluster that is clustered into the low-load site set. The iteration high-load centroid is the high-load centroid after iteration, the iteration medium-load centroid is the medium-load centroid after iteration, and the iteration low-load centroid is the low-load centroid after iteration.

[0147] It is understandable that through classification, we can understand the load intensity of different sites. By setting the number of clusters, we can obtain high-load clustering clusters, medium-load clustering clusters and low-load clustering clusters. The iterative high-load centroid, iterative medium-load centroid and iterative low-load centroid extracted by clustering are updated. This helps the data center to allocate computing resources more effectively, meet the needs of each load site, and avoid waste of resources.

[0148] In order to solve the problems described in the background technology, the present invention receives a demand instruction for the computing power load of a data center, starts a computing power load demand device of a data center based on the demand instruction, obtains multiple user sites based on the computing power load demand device of the data center, and extracts user sites from the multiple user sites in sequence. The present invention obtains data sets of multiple user sites through the computing power demand device of the data center, and can predict the computing power load of multiple user sites and supply energy to multiple user sites. The present invention obtains a daily identified user data set based on the extracted user sites, and constructs a daily load curve based on the daily identified user data set, wherein the ordinate of the daily load curve is the user computing power load, and the abscissa is the daily sequence. The daily load curve is constructed through the daily identified user data set, and the computing power load used by the user in one day can be intuitively viewed through the daily load curve. The present invention obtains a standard curve by normalizing the daily load curve, obtains a standard maximum value and a standard minimum value by using the Min-Max method and the standard curve, constructs a standardized daily load curve by using the standard maximum value and the standard minimum value, calculates the load standard value of the standardized daily load curve, calculates the load kurtosis value and the load skewness value based on the load standard value of the standardized daily load curve, fits the standardized daily load curve into a load straight line equation by using a pre-constructed straight line fitting method, extracts the load straight line slope according to the load straight line equation, calculates the load trend value of the standardized daily load curve according to the load straight line slope, constructs the extracted load fluctuation characteristic matrix of the user site based on the load standard value, the load skewness value, the load kurtosis value and the load trend value, and normalizes different daily load curves through normalization processing for analysis By comparison, the load standard deviation reflects the degree of load fluctuation, the kurtosis value reflects the distribution characteristics of the load, the load skewness value reflects the degree of load skewness, and the load trend value reflects the load fluctuation trend. The above values ​​construct a load fluctuation characteristic matrix to facilitate load prediction; the present invention calculates the mean value of computing power load, inputs the computing power load mean value and the load fluctuation characteristic matrix into the pre-constructed computing power load prediction network, obtains the data to be analyzed, predicts the computing power load of the extracted user site according to the data to be analyzed and the computing power load prediction network, summarizes the computing power load, and obtains the load intensity set corresponding to multiple user sites, wherein the computing power load prediction network is a BP neural network, and by constructing a BP neural network, it can adaptively predict the load characteristics of different user sites and generate a highly adaptable load intensity set. The present invention classifies the load intensity set to obtain a high-load site set, a medium-load site set, and a low-load site set, and completes the demand combination prediction of the computing power load of the data center. Different load intensities can be understood through classification, which helps the data center to allocate computing power resources more effectively, meet the load requirements of each user site, and avoid resource waste. Therefore, the present invention can use the historical computing load data of the data center to predict the computing load demand of the data center, thereby providing energy for the data center more accurately.

[0149] like Figure 21 is a functional module diagram of a demand combination prediction system based on data center computing load provided by an embodiment of the present invention.

[0150] The demand combination forecasting system 100 based on data center computing power load of the present invention can be installed in an electronic device. According to the functions to be implemented, the demand combination forecasting system 100 based on data center computing power load can include a computing power load demand instruction receiving module 101, a load fluctuation characteristic matrix building module 102, a load intensity set composition module 103 and a load site set classification module 104. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0151] A computing load demand instruction receiving module 101 is used to receive a demand instruction of a computing load of a data center, start a computing load demand device of the data center based on the demand instruction, obtain multiple user sites based on the computing load demand device of the data center, and extract user sites from the multiple user sites in sequence;

[0152] A load fluctuation characteristic matrix construction module 102 is used to obtain a daily identification user data set based on the extracted user site, and to construct a daily load curve according to the daily identification user data set, wherein the ordinate of the daily load curve is the user computing power load, and the abscissa is the daily sequence, and the daily load curve is normalized to obtain a standard curve, and the standard maximum value and the standard minimum value are obtained by using the Min-Max method and the standard curve, and a standardized daily load curve is constructed by using the standard maximum value and the standard minimum value, and the load standard value of the standardized daily load curve is calculated, and the load kurtosis value and the load skewness value are calculated based on the load standard value of the standardized daily load curve, and the standardized daily load curve is fitted into a load straight line equation by using a pre-constructed straight line fitting method, and the load straight line slope is extracted according to the load straight line equation, and the load trend value of the standardized daily load curve is calculated according to the load straight line slope, and the load fluctuation characteristic matrix of the extracted user site is constructed based on the load standard value, the load skewness value, the load kurtosis value and the load trend value;

[0153] The load intensity set composition module 103 is used to calculate the computing load mean, input the computing load mean and the load fluctuation characteristic matrix into the pre-built computing load prediction network, obtain the data to be analyzed, predict the computing load of the extracted user site according to the data to be analyzed and the computing load prediction network, summarize the computing load, and obtain the load intensity set corresponding to multiple user sites, wherein the computing load prediction network is a BP neural network;

[0154] The load site set classification module 104 is used to classify the load intensity set to obtain a high-load site set, a medium-load site set and a low-load site set, thereby completing the demand combination prediction of the computing power load of the data center.

[0155] In detail, the modules in the demand combination forecasting system 100 based on data center computing load in the embodiment of the present invention are used in the same manner as above. Figure 1 The same technical means are used as the demand combination prediction method based on data center computing power load described in, and can produce the same technical effects, so they will not be repeated here.

[0156] like Figure 3 1 is a schematic diagram of the structure of an electronic device for implementing a demand combination prediction method based on data center computing power load provided by an embodiment of the present invention.

[0157] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a demand combination prediction method program based on data center computing power load.

[0158] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 also includes an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the demand combination prediction method program based on the computing power load of the data center, but also can be used to temporarily store data that has been output or is to be output.

[0159] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect the various components of the entire electronic device, and executes or executes the programs or modules stored in the memory 11 (such as the demand combination prediction method program based on the computing power load of the data center, etc.), and calls the data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0160] The bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0161] Figure 3 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0162] For example, although not shown, the electronic device 1 may also include a processor 10 that is logically connected to power the various components, so that functions such as charging management, discharging management, and power consumption management are implemented through a power management device. The power supply may also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0163] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0164] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0165] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0166] The demand combination prediction method program based on data center computing load stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0167] Receive a demand instruction for a computing power load of a data center, start a computing power load demand device for the data center based on the demand instruction, obtain multiple user sites based on the computing power load demand device for the data center, extract user sites from the multiple user sites in sequence, and perform the following operations on the extracted user sites:

[0168] A daily identification user data set is obtained based on the extracted user site, and a daily load curve is constructed according to the daily identification user data set, wherein the ordinate of the daily load curve is the user computing load, and the abscissa is the daily sequence;

[0169] The daily load curve is normalized to obtain a standard curve, a standard maximum value and a standard minimum value are obtained by using the Min-Max method and the standard curve, a standardized daily load curve is constructed by using the standard maximum value and the standard minimum value, a load standard value of the standardized daily load curve is calculated, and a load kurtosis value and a load skewness value are calculated based on the load standard value of the standardized daily load curve;

[0170] The standardized daily load curve is fitted into a load straight line equation by using a pre-constructed straight line fitting method, the load straight line slope is extracted according to the load straight line equation, and the load trend value of the standardized daily load curve is calculated according to the load straight line slope;

[0171] Constructing a load fluctuation characteristic matrix of the extracted user site based on the load standard value, load skewness value, load kurtosis value and load trend value;

[0172] Calculate the computing load mean, input the computing load mean and the load fluctuation characteristic matrix into a pre-built computing load prediction network to obtain data to be analyzed, and predict the computing load of the extracted user site based on the data to be analyzed and the computing load prediction network;

[0173] Summarizing the computing load to obtain a load intensity set corresponding to multiple user sites, wherein the computing load prediction network is a BP neural network;

[0174] Classify the load intensity set to obtain a high-load site set, a medium-load site set, and a low-load site set, completing the demand combination forecast of the data center computing power load.

[0175] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0176] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0177] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0178] Receive a demand instruction for a computing power load of a data center, start a computing power load demand device for the data center based on the demand instruction, obtain multiple user sites based on the computing power load demand device for the data center, extract user sites from the multiple user sites in sequence, and perform the following operations on the extracted user sites:

[0179] A daily identification user data set is obtained based on the extracted user site, and a daily load curve is constructed according to the daily identification user data set, wherein the ordinate of the daily load curve is the user computing load, and the abscissa is the daily sequence;

[0180] The daily load curve is normalized to obtain a standard curve, a standard maximum value and a standard minimum value are obtained by using the Min-Max method and the standard curve, a standardized daily load curve is constructed by using the standard maximum value and the standard minimum value, a load standard value of the standardized daily load curve is calculated, and a load kurtosis value and a load skewness value are calculated based on the load standard value of the standardized daily load curve;

[0181] The standardized daily load curve is fitted into a load straight line equation by using a pre-constructed straight line fitting method, the load straight line slope is extracted according to the load straight line equation, and the load trend value of the standardized daily load curve is calculated according to the load straight line slope;

[0182] Constructing a load fluctuation characteristic matrix of the extracted user site based on the load standard value, load skewness value, load kurtosis value and load trend value;

[0183] Calculate the computing load mean, input the computing load mean and the load fluctuation characteristic matrix into a pre-built computing load prediction network to obtain data to be analyzed, and predict the computing load of the extracted user site based on the data to be analyzed and the computing load prediction network;

[0184] Summarizing the computing load to obtain a load intensity set corresponding to multiple user sites, wherein the computing load prediction network is a BP neural network;

[0185] Classify the load intensity set to obtain a high-load site set, a medium-load site set, and a low-load site set, completing the demand combination forecast of the data center computing power load.

[0186] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, and actual implementation may have other division methods.

[0187] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0188] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0189] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0190] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A demand combination forecasting method based on data center computing load, characterized in that: The method comprises: Receive a demand instruction for a computing power load of a data center, start a computing power load demand device for the data center based on the demand instruction, obtain multiple user sites based on the computing power load demand device for the data center, extract user sites from the multiple user sites in sequence, and perform the following operations on the extracted user sites: A daily identification user data set is obtained based on the extracted user site, and a daily load curve is constructed according to the daily identification user data set, wherein the ordinate of the daily load curve is the user computing load, and the abscissa is the daily sequence; The daily load curve is normalized to obtain a standard curve, a standard maximum value and a standard minimum value are obtained by using the Min-Max method and the standard curve, a standardized daily load curve is constructed by using the standard maximum value and the standard minimum value, a load standard value of the standardized daily load curve is calculated, and a load kurtosis value and a load skewness value are calculated based on the load standard value of the standardized daily load curve; The standardized daily load curve is fitted into a load straight line equation by using a pre-constructed straight line fitting method, the load straight line slope is extracted according to the load straight line equation, and the load trend value of the standardized daily load curve is calculated according to the load straight line slope; Constructing a load fluctuation characteristic matrix of the extracted user site based on the load standard value, load skewness value, load kurtosis value and load trend value; Calculate the computing load mean, input the computing load mean and the load fluctuation characteristic matrix into a pre-built computing load prediction network to obtain data to be analyzed, and predict the computing load of the extracted user site based on the data to be analyzed and the computing load prediction network; Summarizing the computing load to obtain a load intensity set corresponding to multiple user sites, wherein the computing load prediction network is a BP neural network; Classify the load intensity set to obtain high-load site sets, medium-load site sets, and low-load site sets, and complete the demand combination forecast of the data center computing power load; The step of calculating the load standard value of the standardized daily load curve and calculating the load kurtosis value and the load skewness value based on the load standard value of the standardized daily load curve includes: The formula for calculating the load standard value is: ； Among them, s represents the load standard value, m represents the total number of daily sequences corresponding to the daily sequence in the standardized daily load curve, Represents the average standard user computing load of the standardized daily load curve, represents the standard user computing power load on the i-th day in the standardized daily load curve; The load kurtosis value is calculated according to the load standard value of the standardized daily load curve, and the formula is: ； Where, k represents the load kurtosis value; The calculation formula of the load skewness value of the standardized daily load curve is as follows: ； Where p represents the load skewness value of the standardized daily load curve.

2. The demand combination forecasting method based on data center computing load as claimed in claim 1, characterized in that: The step of acquiring a daily identification user data set based on the extracted user site includes: Extracting a day sequence according to the day sequence; Based on the extracted user site, the user computing power load and the total computing power consumption of the user are obtained, and the user PUE evaluation value corresponding to the extracted user site is calculated based on the user computing power load and the total computing power consumption of the user; Based on the extracted user site, the extracted day sequence, the user PUE evaluation value, the user computing power load, and the user computing power total consumption, an identification operation is performed on the extracted user site to obtain daily identification user data; The daily identified user data is aggregated to obtain the daily identified user data set of the extracted user site, wherein the calculation formula of the user PUE evaluation value is as follows: ； Among them, PUE is the user's PUE evaluation value, For the user's computing load, The total computing power consumed by the user.

3. The demand combination forecasting method based on data center computing load as claimed in claim 2 is characterized in that: The method of using the Min-Max method and the standard curve to obtain the standard maximum value and the standard minimum value includes: According to the daily load curve, the maximum value of the original load data point set and the minimum value of the original load data point set are obtained, the original load data points are extracted from the original load data point set in sequence, and the following operations are performed on the extracted original load data points: The Min-Max method is used to normalize the original load data points to obtain the standard value, and the range of the standard value is [0, 1]. The formula is: ； in, represents the normalized standard value, M represents the original load data point, is the minimum value representing the original load data set, Represents the maximum value of the original load data set; The standard values ​​are aggregated to obtain a standard value set, and a standard maximum value and a standard minimum value are extracted from the standard value set, where the standard maximum value is 1 and the standard minimum value is 0.

4. The demand combination forecasting method based on data center computing load as claimed in claim 1, characterized in that: The method of fitting the standardized daily load curve into a load straight line equation by using a pre-constructed straight line fitting method, extracting the load straight line slope according to the load straight line equation, and calculating the load trend value of the standardized daily load curve according to the load straight line slope includes: The data of the standardized daily load curve is obtained, and the data of the standardized daily load curve is plotted into a visual scatter plot. A design matrix is ​​constructed based on the visual scatter plot, and a parameter vector of the load straight line equation is calculated using the least squares method and the design matrix, wherein the formula for calculating the parameter vector of the load straight line equation is: ； in, represents the parameter vector of the load line equation, X represents the design matrix, and Y represents the vector of the standard user computing power load. represents the transposed matrix of the design matrix X; The load line slope of the load line equation is calculated according to the parameter vector of the load line equation, wherein the calculation formulas of the load line equation and the load line slope are as follows: ； ； Where z represents the linear user computing power load in the load line equation, and x represents the linear time. is represented as the intercept term, It is expressed as the slope of the load line; n represents the total number of scattered points in the visual scatter plot; The load trend value of the standardized daily load curve is calculated according to the load straight line slope.

5. The demand combination forecasting method based on data center computing load as claimed in claim 4 is characterized in that: The load fluctuation characteristic matrix is ​​expressed as follows: ； in, represents the j-th load fluctuation characteristic matrix, represents the load standard value of the jth load fluctuation characteristic matrix, The load skewness value of the jth load fluctuation characteristic matrix, represents the load kurtosis value of the jth load fluctuation characteristic matrix, Represents the load trend value of the jth load fluctuation characteristic matrix.

6. The demand combination forecasting method based on data center computing load as claimed in claim 2, characterized in that: The calculation of the computing load mean value includes: The total computing load is calculated according to the extracted daily load curve of the user site, the daily sequence number corresponding to the daily load curve is obtained, and the computing load mean is calculated based on the daily sequence number of the total computing load set.

7. The demand combination forecasting method based on data center computing load as claimed in claim 6, characterized in that: The step of inputting the computing load mean value and the load fluctuation characteristic matrix into a pre-built computing load prediction network to obtain data to be analyzed includes: Obtaining a computing power load original network, wherein the input layer of the computing power load original network is a computing power load mean input node, a load standard value input node, a load skewness value input node, a load kurtosis value input node, and a load trend value input node, and the output layer is a predicted computing power load output node; Obtain computing power load training data, train the original computing power load network according to the computing power load training data, and obtain a computing power load prediction network, wherein the computing power load training data includes: a historical computing power load mean value set, a historical load standard value set, a historical load skewness value set, a historical load kurtosis value set, a historical load trend value set, and a historical computing power load set.

8. The demand combination forecasting method based on data center computing load as claimed in claim 7, characterized in that: The classifying of the load intensity set to obtain a high-load site set, a medium-load site set and a low-load site set includes: The preset number of clusters is set to 3, and three load intensities are randomly selected from the load intensity set as the high load centroid, the medium load centroid, and the low load centroid respectively; Extracting the load intensities to be clustered in the load intensity set in turn, and calculating the high load clustering distance, the medium load clustering distance and the low load clustering distance between the load intensities to be clustered and the high load centroid, the medium load centroid and the low load centroid; Clustering the load intensities to be clustered according to the high load clustering distance, the medium load clustering distance, and the low load clustering distance to obtain a high load clustering cluster, a medium load clustering cluster, and a low load clustering cluster; Extracting an iterative high-load centroid, an iterative medium-load centroid and an iterative low-load centroid from the high-load cluster, the medium-load cluster and the low-load cluster respectively; The iterative high load centroid, iterative medium load centroid and iterative low load centroid are used to update the high load centroid, medium load centroid and low load centroid respectively, and return to the above step of extracting the load intensities to be clustered in the load intensity set in sequence until the load intensity set completes the extraction of the load intensities to be clustered, and a high load site set, a medium load site set and a low load site set are obtained.

9. A demand combination forecasting system based on data center computing load, characterized in that: A method for forecasting demand combination based on data center computing load according to any one of claims 1 to 8, the system comprising: A computing load demand instruction receiving module is used to receive a computing load demand instruction of a data center, start a computing load demand device of the data center based on the demand instruction, obtain multiple user sites based on the computing load demand device of the data center, and extract user sites from the multiple user sites in sequence; A load fluctuation characteristic matrix construction module is used to obtain a daily identification user data set based on the extracted user site, and to construct a daily load curve according to the daily identification user data set, wherein the ordinate of the daily load curve is the user computing power load, and the abscissa is the day sequence, and the daily load curve is normalized to obtain a standard curve, and the standard maximum value and the standard minimum value are obtained by using the Min-Max method and the standard curve, and a standardized daily load curve is constructed by using the standard maximum value and the standard minimum value, and the load standard value of the standardized daily load curve is calculated, and the load kurtosis value and the load skewness value are calculated based on the load standard value of the standardized daily load curve, and the standardized daily load curve is fitted into a load straight line equation by using a pre-constructed straight line fitting method, and the load straight line slope is extracted according to the load straight line equation, and the load trend value of the standardized daily load curve is calculated according to the load straight line slope, and the load fluctuation characteristic matrix of the extracted user site is constructed based on the load standard value, load skewness value, load kurtosis value and load trend value; A load intensity set composition module is used to calculate a computing power load mean, input the computing power load mean and the load fluctuation characteristic matrix into a pre-built computing power load prediction network to obtain data to be analyzed, predict the computing power load of the extracted user site based on the data to be analyzed and the computing power load prediction network, summarize the computing power load, and obtain a load intensity set corresponding to multiple user sites, wherein the computing power load prediction network is a BP neural network; The load site set classification module is used to classify the load intensity set to obtain high-load site sets, medium-load site sets and low-load site sets, and complete the demand combination prediction of the data center computing power load.

Images