Load forecasting method, device, equipment and storage medium

Abstract

The present application discloses a load forecasting method, device, equipment and storage medium, the method comprising: obtaining a load data of n electrical equipment in a historical time period; obtaining b environmental data of n electrical equipment in a historical time period through m environmental sensors; constructing a basic measurement matrix according to a load data; constructing a first augmented measurement matrix according to b environmental data; determining a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix; determining T linear eigenvalue statistical difference values ​​according to the coupling analysis matrix; determining n first load data sets according to the T linear eigenvalue statistical difference values; inputting the n first load data sets into a load forecasting model, obtaining a target first load data set of each electrical equipment in the n electrical equipment in a future time period, and obtaining n target first load data sets. The present application is conducive to improving the accuracy of load forecasting.

Description

Technical Field

[0001] This application relates to the technical field of power prediction, and particularly to a load prediction method, device, equipment and storage medium. Background Art

[0002] With the continuous development of modern power systems, load prediction has become a key link in power system planning, operation and management. Accurate load prediction directly affects the reliability and economy of power systems, helps optimize the allocation of power resources, and avoids power outages or waste caused by supply-demand imbalances. In the context of the rapid development of renewable energy, load prediction can effectively cope with the volatility of energy and improve the stability and operation efficiency of power systems.

[0003] In existing load prediction methods, load data of electrical equipment of power users in a historical time period is collected, and methods such as statistical analysis, regression analysis, and neural network prediction are used to predict the load data of the electrical equipment in a future time period. However, during the operation of power systems, load data is affected by various factors. Only predicting the load data of power users in a future time period based on the load data of power users in a historical time period results in poor accuracy of load prediction. Summary of the Invention

[0004] To solve the above problems existing in the prior art, embodiments of this application provide a load prediction method, device, equipment and storage medium. A basic measurement matrix is constructed according to the load data of n electrical equipment in the same physical space in a historical time period, and a first augmented measurement matrix is constructed according to the environmental data of this physical space, so as to determine a coupling analysis matrix. The linear eigenvalue statistic difference determined by the coupling analysis matrix is used to determine n first load data sets, and the n first load data sets are input into a load prediction model to obtain a target first load data set for each of the n electrical equipment in a future time period. Considering the influence of environmental data on the load data of electrical equipment and the historical load data of electrical equipment, the load is predicted, improving the accuracy of load prediction.

[0005] In a first aspect, an embodiment of this application provides a load prediction method, including:

[0006] Obtain T load data of each of the n electrical equipment in a historical time period to obtain a load data; obtain T environmental data of the n electrical equipment in the historical time period through each of the m environmental sensors to obtain b environmental data; the n electrical equipment are in the same physical space; n, m, and T are all integers greater than 1, a = n T, b = m T;

[0007] Construct a basic measurement matrix according to the a load data;

[0008] Construct a first augmented measurement matrix according to the b environmental data;

[0009] Determine a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix;

[0010] Determine T linear eigenvalue statistic differences according to the coupling analysis matrix;

[0011] Determine n first load data sets according to the T linear eigenvalue statistic differences; the n first load data sets correspond to the n electrical devices one by one;

[0012] Input the n first load data sets into a load prediction model to obtain a target first load data set for each of the n electrical devices in a future time period, and obtain n target first load data sets.

[0013] In a second aspect, an embodiment of the present application provides a load prediction device, including:

[0014] An acquisition unit, configured to acquire T load data of each of the n electrical devices in a historical time period to obtain a load data; acquire T environmental data of the n electrical devices in the historical time period through each of the m environmental sensors to obtain b environmental data; the n electrical devices are in the same physical space; n, m, and T are all integers greater than 1, a = n T, b = m T;

[0015] A processing unit, configured to construct a basic measurement matrix according to the a load data;

[0016] Construct a first augmented measurement matrix according to the b environmental data;

[0017] Determine a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix;

[0018] Determine T linear eigenvalue statistic differences according to the coupling analysis matrix;

[0019] Determine n first load data sets according to the T linear eigenvalue statistic differences; the n first load data sets correspond to the n electrical devices one by one;

[0020] Input the n first load data sets into a load prediction model to obtain a target first load data set for each of the n electrical devices in a future time period, and obtain n target first load data sets.

[0021] In a third aspect, an embodiment of the present application provides an electronic device, including: a processor and a memory. The processor is connected to the memory. The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory so that the electronic device executes the method described in the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, and the computer program is executed by a processor to implement the method described in the first aspect.

[0023] In a fifth aspect, an embodiment of the present application provides a computer program product including a computer program, and the computer program implements the method described in the first aspect when executed by a processor.

[0024] Implementing the embodiments of the present application has the following beneficial effects:

[0025] In the embodiment of the present application, first, T load data of each of the n electrical devices in the same physical space within a historical time period are obtained, resulting in a load data. T environmental data of the n electrical devices within the historical time period are obtained by each of the m environmental sensors, resulting in b environmental data. Then, based on the a load data, a basic measurement matrix is constructed, and based on the b environmental data, a first augmented measurement matrix is constructed. According to the basic measurement matrix and the first augmented measurement matrix, a coupling analysis matrix can be determined. Further, according to the coupling analysis matrix, T linear eigenvalue statistic differences are determined, and based on the T linear eigenvalue statistic differences, n first load data sets are determined. Finally, the n first load data sets are input into a load prediction model to obtain a target first load data set for each of the n electrical devices within a future time period, resulting in n target first load data sets. Thus, through the basic measurement matrix constructed from the load data of the electrical devices in the same physical space within the historical time period, and the first augmented measurement matrix constructed from the environmental data in this physical space, a coupling analysis matrix can be determined. The linear eigenvalue statistic differences determined by the coupling analysis matrix can represent the influence index of the environmental data on the load data of the electrical devices, thereby determining the first load data set considering the influence of the environmental data. By inputting the first load data set into the load prediction model to predict the target first load data set of the electrical devices within the future time period, the accuracy of load prediction can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0027] Figure 1 Schematic diagram of an application scenario of a load prediction method provided by an embodiment of the present application;

[0028] Figure 2 Schematic diagram of the process of a load prediction method provided by an embodiment of the present application;

[0029] Figure 3 Schematic diagram of the structure of a preset load prediction model provided by an embodiment of the present application;

[0030] Figure 4 Schematic diagram of the structure of a gated recurrent unit provided by an embodiment of the present application;

[0031] Figure 5 Schematic diagram of a load prediction device provided by an embodiment of the present application;

[0032] Figure 6 Schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0034] The terms "first", "second", "third", and "fourth" in the specification, claims, and accompanying drawings of the present application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and "having", and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products, or devices.

[0035] References herein to "embodiments" mean that the particular features, results, or characteristics described in connection with the embodiments can be included in at least one embodiment of the present application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0036] First, refer to Figure 1 , Figure 1 which is a schematic diagram of an application scenario of a load forecasting method provided for an embodiment of the present application. As shown in Figure 1 the application scenario includes: a distribution network, electrical equipment, and a load forecasting device.

[0037] Among them, the distribution network is a power grid in the power system that connects transmission lines and electrical equipment of power users, and is used to distribute the electric energy in the transmission lines or power plants to the electrical equipment of power users. Electrical equipment is equipment in the power system that consumes electric energy. For example, industrial manufacturing equipment, commercial electrical equipment, residential electrical equipment, charging piles for electric vehicles, and so on. The electrical equipment in the embodiments of the present application is all in the same physical space. For example, multiple charging piles in the same physical space. Since the electrical equipment is in the same physical space, the electrical equipment all operates in the same environmental data. Among them, an intelligent electricity meter is provided on each electrical equipment, and the intelligent electricity meter can collect the power data of the electrical equipment, including: current, voltage, load data, and so on. The power data of the electrical equipment collected by the intelligent electricity meter can be fed back to the distribution network so that the distribution network can adjust the power supply to the electrical equipment according to the power data of the electrical equipment. A variety of environmental sensors are also provided in the physical space where the electrical equipment is located, such as temperature sensors, humidity sensors, wind speed detectors, and so on, which are respectively used to detect the temperature, humidity, wind speed, and so on in the physical space. The intelligent electricity meter of each electrical equipment and a variety of environmental sensors in the physical space are all communicatively connected to the load forecasting device.

[0038] Among them, the load prediction device can be a server for data transceiver, data processing, and data storage functions. The server can provide high-performance computing services for other devices in the network. Exemplarily, the server can include: application server, virtual private server (VPS), complex instruction set computer (CISC), and so on. This application does not make specific limitations. The load prediction device can also be a processor with data transceiver and data processing capabilities. The processor can be connected to the memory and run the corresponding computer program by reading and executing the computer program instructions in the memory. Exemplarily, the processor can include: digital signal processor (DSP), central processing unit (CPU), microcontroller unit (MCU), and so on. This application does not make specific limitations.

[0039] In the embodiments of this application, the load prediction device can send a data acquisition instruction to the smart meter of the electrical equipment to obtain the load data of the electrical equipment returned by the smart meter. The load prediction device can also send a data acquisition instruction to the environmental sensor in the physical space to obtain the environmental data returned by the environmental sensor. Optionally, the load prediction device can also be communicatively connected to the power distribution equipment in the power distribution network to obtain the power data corresponding to the power distribution equipment.

[0040] It should be noted that in the existing load prediction methods for electrical equipment, usually based on the load data of the electrical equipment in the historical time period, methods such as statistical analysis method, regression analysis method, neural network prediction, etc. are used to predict the load data of the electrical equipment in the future time period. However, since the load data between the electrical equipment in the same physical space will have an impact, and the environmental data will also have an impact on the load data of the electrical equipment, only predicting the load data of the electrical equipment in the future time period based on the load data of the electrical equipment in the historical time period, the accuracy of load prediction is relatively poor.

[0041] Therefore, applied to the above scenario, in the load prediction method provided by this application, the load prediction device obtains T load data of each of the n electrical equipment in the historical time period, resulting in a load data; the load prediction device obtains T environmental data of the n electrical equipment in the historical time period through each of the m environmental sensors, resulting in b environmental data;

[0042] The load prediction device constructs a basic measurement matrix based on the a load data;

[0043] The load forecasting device constructs a first augmented measurement matrix based on b environmental data;

[0044] The load forecasting device determines a coupling analysis matrix based on the basic measurement matrix and the first augmented measurement matrix;

[0045] The load forecasting device determines T linear eigenvalue statistic differences based on the coupling analysis matrix;

[0046] The load forecasting device determines n first load data sets based on the T linear eigenvalue statistic differences;

[0047] The load forecasting device inputs the n first load data sets into the load forecasting model to obtain a target first load data set for each of the n electrical devices in the future time period, and obtains n target first load data sets.

[0048] It can be seen that when applied to the above scenario, the load forecasting device can construct a basic measurement matrix based on the load data of n electrical devices in the same physical space, construct a first augmented measurement matrix based on the environmental data in this physical space, and determine a coupling analysis matrix based on the basic measurement matrix and the first augmented measurement matrix. The linear eigenvalue statistic differences determined by the coupling analysis matrix can represent the influence of environmental data on the load data of electrical devices. Therefore, based on the linear eigenvalue statistic differences, n first load data sets input to the load forecasting model are determined to predict the load data of n electrical devices in the future time period. The predicted load data combines the influence of environmental data on the load data, improving the accuracy of load forecasting. Moreover, by simultaneously predicting n electrical devices in the same physical space, the influence of the load data between electrical devices is combined, further improving the accuracy of load forecasting.

[0049] Refer to Figure 2 , Figure 2 which is a schematic flowchart of a load forecasting method provided by an embodiment of the present application. This method is applied to the load forecasting device in the above scenario. As Figure 2 shown, this method includes but is not limited to the following steps:

[0050] 201: Obtain T load data of each of the n electrical devices in the historical time period to obtain a load data.

[0051] In the embodiment of the present application, the n electrical devices are in the same physical space. The historical time period represents a preset time period between the current moment. The load data is the electric power consumed by the electrical device corresponding to the load data. Both n and T are integers greater than 1, and a = n T.

[0052] Specifically, within a historical time period, the smart meter of each of the n electrical devices will collect the load data of the electrical device at T sampling moments within the historical time period according to a preset collection period of the load data. That is, the smart meter of each electrical device collects T load data, resulting in a load data. The load prediction device can obtain the above a load data from the smart meter of each of the n electrical devices.

[0053] 202: Obtain T environmental data of the n electrical devices within a historical time period through each of the m environmental sensors, resulting in b environmental data.

[0054] In the embodiment of the present application, the environmental sensors may include: temperature sensors, humidity sensors, wind speed detectors, and so on. The environmental data may include: temperature, humidity, wind speed, and so on. m is an integer greater than 1, and b = m T.

[0055] Specifically, taking the example of obtaining T temperatures through a temperature sensor, within a historical time period, the temperature sensor can collect the temperatures of the physical spaces where the n electrical devices are located at T collection moments according to a preset temperature collection period. The load prediction device can obtain the T temperatures collected by the temperature sensor from the temperature sensor. Similarly, the load prediction device can obtain T environmental data of the physical spaces where the n electrical devices are located within a historical time period from each of the m environmental sensors, resulting in b environmental data.

[0056] 203: Construct a basic measurement matrix based on the a load data.

[0057] In the embodiment of the present application, the basic measurement matrix is a matrix with a dimension of n×T. The T data in each row of the basic measurement matrix are the T load data of the electrical device corresponding to that row within the historical time period. The n data in each column of the basic measurement matrix are the n load data of the n electrical devices at the sampling moment corresponding to that column, and each electrical device corresponds to one load data in that column. The load prediction device can construct the above basic measurement matrix based on the a load data of the n electrical devices within the historical time period.

[0058] 204: Construct a first augmented measurement matrix based on the b environmental data.

[0059] In an embodiment of the present application, the first augmented measurement matrix is a matrix with a dimension of m×T. The T data in each row of the first augmented measurement matrix are T environmental data collected by the environmental sensor corresponding to that row within the historical time period. Each column of the first augmented measurement matrix is composed of m environmental data collected by m environmental sensors at the acquisition moment corresponding to that column, and each environmental sensor collects one environmental data for that column. The load prediction device can construct the first augmented measurement matrix according to the b environmental data obtained above.

[0060] 205: Determine a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix.

[0061] In an embodiment of the present application, the coupling analysis matrix is a matrix with a dimension of (n + k m)×T. Among them, the coupling analysis matrix is formed by splicing matrices after dimension expansion of the basic measurement matrix and the first augmented measurement matrix. The data in the coupling analysis matrix include: a load data, b environmental data, and multiple white noises. The load prediction device can perform dimension expansion on the first augmented measurement matrix and splice the expanded first augmented measurement matrix with the basic measurement matrix to obtain the coupling analysis matrix.

[0062] Exemplarily, determining the coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix may include:

[0063] Obtain an initial white noise intensity regulation matrix when performing dimension expansion on the first augmented measurement matrix;

[0064] Determine the first signal-to-noise ratio corresponding to the first augmented measurement matrix;

[0065] Perform dimension expansion on the first augmented measurement matrix according to the initial white noise intensity regulation matrix to obtain a second augmented measurement matrix;

[0066] Determine the second signal-to-noise ratio of the second augmented measurement matrix;

[0067] If the second signal-to-noise ratio is consistent with the first signal-to-noise ratio, splice the basic measurement matrix and the second augmented measurement matrix to obtain the coupling analysis matrix;

[0068] If the second signal-to-noise ratio is inconsistent with the first signal-to-noise ratio, adjust the initial white noise intensity regulation matrix according to the second signal-to-noise ratio and the first signal-to-noise ratio to obtain a target white noise intensity regulation matrix;

[0069] Perform dimension expansion on the first augmented measurement matrix according to the target white noise intensity regulation matrix to obtain a third augmented measurement matrix;

[0070] Splice the basic measurement matrix and the third augmented measurement matrix to obtain the coupling analysis matrix.

[0071] In the embodiment of the present application, the initial white noise intensity regulation matrix is a white noise matrix with a dimension of (k m - m)×T. The initial white noise intensity regulation matrix can be preset according to the actual test results. The second augmented measurement matrix is a matrix with a dimension of (k m)×T. The second augmented measurement matrix includes all the data in the first augmented measurement matrix and all the data in the initial white noise intensity regulation matrix.

[0072] Specifically, the load prediction device first obtains the initial white noise intensity regulation matrix when expanding the dimension of the first augmented measurement matrix. Then, the load prediction device determines the transpose matrix of the first augmented measurement matrix, and takes the product of the transpose matrix of the first augmented measurement matrix and the first augmented measurement matrix as the first square matrix. The load prediction device determines the transpose matrix of the initial white noise intensity regulation matrix, and takes the product of the transpose matrix of the initial white noise intensity regulation matrix and the initial white noise intensity regulation matrix as the second square matrix. Then, the trace of the first square matrix and the trace of the second square matrix are determined. The ratio of the trace of the first square matrix to the trace of the second square matrix is taken as the first ratio. The ratio of the first ratio to the square of the signal-to-noise ratio adjustment parameter is taken as the first signal-to-noise ratio corresponding to the first augmented measurement matrix. Among them, the signal-to-noise ratio adjustment parameter can be preset according to the actual test results.

[0073] Then, the load prediction device expands the dimension of the first augmented measurement matrix according to the initial white noise intensity regulation matrix to expand the dimension of the first augmented measurement matrix to (k m)×T to obtain the second augmented measurement matrix. Among them, the second augmented measurement matrix includes m rows of environmental data and (k - m) rows of white noise. The introduction of white noise is to prevent the repeated data in the first augmented measurement matrix from interfering with the subsequent processing process. Then, the load prediction device determines the second signal-to-noise ratio of the second augmented measurement matrix. Exemplarily, determining the second signal-to-noise ratio of the second augmented measurement matrix includes:

[0074] The second signal-to-noise ratio is calculated through the following formula (1), specifically as follows:

[0075] Formula (1)

[0076] In formula (1), SNR represents the second signal-to-noise ratio, Tr() represents calculating the trace of the matrix, Y represents the second augmented measurement matrix, represents the transpose matrix of the second augmented measurement matrix, represents the signal-to-noise ratio adjustment parameter, represents the initial white noise intensity regulation matrix, represents the transpose matrix of the initial white noise intensity regulation matrix.

[0077] As can be seen from Equation (1), the load forecasting device first determines the transpose matrix of the second augmented measurement matrix, and takes the product of the transpose matrix of the second augmented measurement matrix and the second augmented measurement matrix as a third square matrix. Then, it determines the trace of the third square matrix. The trace of the third square matrix and the trace of the second square matrix are used as a second ratio. The ratio of the second ratio to the square of the signal-to-noise ratio adjustment parameter is used as the second signal-to-noise ratio of the second augmented measurement matrix. It should be noted that during the process of dimension expansion of the first augmented measurement matrix, the signal-to-noise ratio of the matrix needs to be kept consistent to determine a more accurate coupling analysis matrix and improve the accuracy of load forecasting.

[0078] If the second signal-to-noise ratio is consistent with the first signal-to-noise ratio, the load forecasting device splices the basic measurement matrix and the second augmented measurement matrix to obtain a coupling analysis matrix. Among them, since the basic measurement matrix is a matrix with dimensions of n×T, and the second augmented measurement matrix is a matrix with dimensions of (k m)×T, the basic measurement matrix and the second augmented measurement matrix need to be spliced by columns to obtain a coupling analysis matrix.

[0079] If the second signal-to-noise ratio is inconsistent with the first signal-to-noise ratio, the load forecasting device will determine the difference between the first signal-to-noise ratio and the second signal-to-noise ratio. According to the difference between the first signal-to-noise ratio and the second signal-to-noise ratio and the above Equation (1), it determines the trace of the target white noise intensity regulation matrix. The data in the initial white noise intensity regulation matrix is adjusted according to the trace of the target white noise intensity regulation matrix to obtain the target white noise intensity regulation matrix.

[0080] Furthermore, the load forecasting device inserts the data in the target white noise intensity regulation matrix into the matrix after the first augmented measurement matrix is expanded according to the target white noise intensity regulation matrix to obtain a third augmented measurement matrix. Finally, the basic measurement matrix and the third augmented measurement matrix are spliced by columns to obtain a coupling analysis matrix.

[0081] It can be seen that in the embodiments of the present application, by obtaining the initial white noise intensity regulation matrix during the dimension expansion of the first augmented measurement matrix and determining the first signal-to-noise ratio of the first augmented measurement matrix. Then, according to the initial white noise intensity regulation matrix, the first augmented measurement matrix is dimensionally expanded to obtain a second augmented measurement matrix, and the second signal-to-noise ratio of the second augmented measurement matrix is determined. By determining whether the second signal-to-noise ratio is consistent with the first signal-to-noise ratio during the dimension expansion process, the first augmented measurement matrix can be dimensionally expanded to avoid interference caused by duplicate data in the first augmented measurement matrix to the subsequent processing process, and the expanded matrix and the basic measurement matrix are spliced to obtain a coupling analysis matrix, thereby combining load data with environmental data and improving the accuracy of load forecasting.

[0082] 206: Determine T differences of linear eigenvalue statistics according to the coupling analysis matrix.

[0083] In an embodiment of the present application, the difference of linear eigenvalue statistics is the difference between the linear eigenvalue statistic (Linear Eigenvalue Statistic, LES) of the coupling analysis matrix and the linear eigenvalue statistic of the reference matrix. The reference matrix is a matrix obtained by dimension expansion of the basic measurement matrix and having the same dimension as the coupling analysis matrix. The load prediction device can determine T differences of linear eigenvalue statistics by determining T linear eigenvalue statistics of the coupling analysis matrix and T linear eigenvalue statistics of the reference matrix.

[0084] Exemplarily, determining T differences of linear eigenvalue statistics according to the coupling analysis matrix may include:

[0085] Perform dimension expansion on the basic measurement matrix according to the dimension of the coupling analysis matrix to obtain a reference matrix;

[0086] Obtain the size of the rolling window;

[0087] According to the size of the rolling window, obtain T first window data from the coupling analysis matrix;

[0088] Determine a first eigenvalue set corresponding to each of the T first window data in the T first window data to obtain T first eigenvalue sets;

[0089] Determine T first linear eigenvalue statistics according to the T first eigenvalue sets;

[0090] According to the T first window data, obtain T second window data from the reference matrix;

[0091] Determine a second eigenvalue set corresponding to each of the T second window data in the T second window data to obtain T second eigenvalue sets;

[0092] Determine T second linear eigenvalue statistics according to the T second eigenvalue sets;

[0093] Determine T differences of linear eigenvalue statistics according to the T first linear eigenvalue statistics and the T second linear eigenvalue statistics.

[0094] In an embodiment of the present application, the T first linear eigenvalue statistics correspond one-to-one to the T first window data. The T first window data correspond one-to-one to the T second window data. The T second linear eigenvalue statistics correspond one-to-one to the T second window data.

[0095] Specifically, the load prediction device first expands the dimension of the basic measurement matrix to (n + k according to the dimension of the coupling analysis matrix An (m)×T matrix to obtain a reference matrix. Among them, the data in the first n rows of the reference matrix are the data of the basic measurement matrix, and the data in the last (k m) rows of the data are the added preset white noise.

[0096] Then, obtain the size of the rolling window. Among them, the rolling window is used to obtain window data from the coupling analysis matrix or the reference matrix, and the size of the rolling window can be preset in advance. The load prediction device will obtain T first window data from the coupling analysis matrix according to the size of the rolling window. It can be understood that the first window data is a matrix, and the dimension of this matrix is determined by the size of the rolling window. Any two of the T first window data may include duplicate data.

[0097] Furthermore, the load prediction device determines the first eigenvalue set corresponding to each of the T first window data, obtaining T first eigenvalue sets. The first eigenvalue set includes multiple first eigenvalues of the first window data corresponding to this first eigenvalue set, and the number of first eigenvalues is determined by the size of the rolling window. The load prediction device determines the sum of the multiple first eigenvalues in each first eigenvalue set as the first linear eigenvalue statistic of the first window data corresponding to this first eigenvalue set. In this way, T first linear eigenvalue statistics can be determined.

[0098] Furthermore, the load prediction device obtains, according to the T first window data, the second window data corresponding to each of the T first window data from the reference matrix, obtaining T second window data. Then, determine the second eigenvalue set corresponding to each of the T second window data, obtaining T second eigenvalue sets. The second eigenvalue set includes multiple second eigenvalues of the second window data corresponding to this second eigenvalue set. The load prediction device determines the sum of the multiple second eigenvalues in each second eigenvalue set as the second linear eigenvalue statistic of the second window data corresponding to this second eigenvalue set. In this way, T second linear eigenvalue statistics can be determined.

[0099] Finally, the load prediction device subtracts the T first linear eigenvalue statistics from the T second linear eigenvalue statistics correspondingly according to the corresponding relationship between the T first window data and the T second window data, obtaining T linear eigenvalue statistic differences. It should be noted that the linear eigenvalue statistic difference can represent the coupling between the environmental data and the load data, and also represents the influence index of the environmental data on the load data. The larger the linear eigenvalue statistic difference, the better the coupling between the environmental data and the load data, and the smaller the influence index of the environmental data on the load data.

[0100] It can be seen that a reference matrix is obtained by expanding the dimension of the basic measurement matrix. Then, according to the size of the rolling window, T first window data are obtained from the coupling analysis matrix, and a first eigenvalue set of each first window data in the T first window data is determined, so as to determine a first linear eigenvalue statistic of the first window data according to the first eigenvalue set of each first window data. Next, according to the T first window data, T second window data are obtained from the reference matrix, and a second eigenvalue set corresponding to each second window data is determined, obtaining T second eigenvalue sets. According to the T second eigenvalue sets, T second linear eigenvalue statistics are determined. Finally, according to the T first linear eigenvalue statistics and the T second linear eigenvalue statistics, T linear eigenvalue statistic differences can be determined. Thus, through the linear eigenvalue statistic differences between the coupling analysis matrix and the reference matrix, the influence of environmental data on load data can be determined to improve the accuracy of load forecasting.

[0101] 207: Determine n first load data sets according to the T linear eigenvalue statistic differences.

[0102] In the embodiment of the present application, the n first load data sets correspond to n electrical devices one by one. Each first load data set corresponding to an electrical device includes T load data adjusted according to the environmental data for the electrical device. The load forecasting device can determine an influence index of the environmental data on the load data according to the T linear eigenvalue statistic differences, and thus adjust the load data corresponding to the n electrical devices according to the influence index corresponding to each load data, and further obtain the n first load data sets.

[0103] Exemplarily, determining the n first load data sets according to the T linear eigenvalue statistic differences may include:

[0104] Determine n second load data sets according to a load data;

[0105] Determine the linear eigenvalue statistic differences less than a preset difference in the T linear eigenvalue statistic differences, obtaining c linear eigenvalue statistic differences;

[0106] Obtain c first window data from the T first window data according to the c linear eigenvalue statistic differences;

[0107] Determine the influence index of the environmental data on the load data in each of the c first window data according to the c linear eigenvalue statistic differences;

[0108] According to the influence index of the environmental data in each of the c first window data on the load data, determine the sum of the influence indexes corresponding to each load data among the d load data of the c first window data, and obtain d total influence indexes;

[0109] According to the d total influence indexes and the d load data, determine d adjusted load data;

[0110] According to the d adjusted load data, update the n second load data sets to obtain n first load data sets.

[0111] In the embodiments of the present application, each second load data set corresponds to an electrical device. c is an integer greater than 1 and less than or equal to T. The d load data are all the load data in the c first window data.

[0112] Specifically, the load prediction device first extracts the above-mentioned a load data from the coupling analysis matrix, and uses the T load data corresponding to each of the n electrical devices as a load data set corresponding to the electrical device, and obtains n second load data sets.

[0113] Then, the load prediction device determines the linear eigenvalue statistic differences less than the preset difference among the T linear eigenvalue statistic differences, and obtains c linear eigenvalue statistic differences. Among them, the linear eigenvalue statistic difference being less than the preset difference means that the coupling between the environmental data and the load data in the first window data corresponding to the linear eigenvalue statistic difference is less than the preset coupling, and the influence index of the environmental data in the first window data corresponding to the linear eigenvalue statistic difference on the load data is greater than the preset influence index. The load prediction device will obtain c first window data corresponding to the c linear eigenvalue statistic differences from the T first window data according to the c linear eigenvalue statistic differences.

[0114] Furthermore, the load prediction device determines the influence index of the environmental data in each of the c first window data on the load data according to the c linear eigenvalue statistic differences. Among them, the mapping relationship between the linear eigenvalue statistic difference and the influence index can be preset according to the actual test results. It can be understood that there may be one or more environmental data and one or more load data in each first window data, and the influence index of each environmental data corresponding to the same first window data on each load data is the same. Among them, there are a total of d load data in the c first window data, and the d load data can be any d load data among the T load data.

[0115] Further, the load prediction device will calculate the sum of the influence indices corresponding to each load data among the d load data, obtaining d total influence indices. Based on the d total influence indices and the d load data, d adjusted load data are determined, where the adjusted load data is equal to the product of the total influence index and the load data.

[0116] Finally, the load prediction device updates the d load data in the n second load data sets corresponding to the d adjusted load data according to the d adjusted load data. For example, each load data in the d load data is subtracted by the adjusted load data corresponding to this load data to update the n second load data sets, obtaining n first load data sets.

[0117] It can be seen that the load prediction device can obtain c first window data from the T first window data by determining c linear eigenvalue statistic differences less than the preset difference among the T linear eigenvalue statistic differences. Then, according to the c linear eigenvalue statistic differences, the influence indices of the environmental data on the load data in the c first window data are determined. According to the total influence indices corresponding to each load data among the d load data, d adjusted load data can be determined to update the n second load data sets according to the d adjusted load data, obtaining n first load data sets. Thus, it is determined that the n first load data sets are all load data sets considering the influence of environmental data on load data. Performing load prediction based on the n first load data sets can improve the accuracy of load prediction.

[0118] 208: Input the n first load data sets into the load prediction model to obtain the target first load data sets of each of the n electrical devices in the future time period, obtaining n target first load data sets.

[0119] In the embodiment of the present application, the load prediction model is obtained by training a preset load prediction model. The preset load prediction model can be a neural network model based on the Gated Recurrent Unit (GRU) and the multi-task learning (MTL) framework. The input of the load prediction model is the n first load data sets corresponding to the n electrical devices, and the output of the load prediction model is the n target first load data sets of the n electrical devices in the future time period.

[0120] Exemplarily, before inputting the n target first load data sets into the load prediction model, it further includes:

[0121] Obtain multiple load association matrices from the coupling analysis matrix;

[0122] Determine the set of eigenvalues corresponding to each load correlation matrix among multiple load correlation matrices to obtain multiple third eigenvalue sets;

[0123] Take any one electrical device corresponding to each load correlation matrix as the associated electrical device corresponding to the load correlation matrix, and take the load data corresponding to the associated electrical device of each load correlation matrix as the associated load data of the load correlation matrix;

[0124] Fill the associated load data of each load correlation matrix among multiple load correlation matrices into preset white noise to obtain multiple associated reference matrices;

[0125] Determine the set of eigenvalues corresponding to each associated reference matrix among multiple associated reference matrices to obtain multiple fourth eigenvalue sets;

[0126] Determine the correlation degree between any two electrical devices among n electrical devices according to multiple third eigenvalue sets and multiple fourth eigenvalue sets;

[0127] Adjust the n first load data sets according to the correlation degree between any two electrical devices among n electrical devices to obtain the adjusted n first load data sets.

[0128] In the embodiments of the present application, the load correlation matrix includes the load data corresponding to any two electrical devices among n electrical devices. It should be noted that the load data between n electrical devices in the same physical space will be correlated. Therefore, it is necessary to adjust the n first load data sets input to the load prediction model according to the correlation degree between any two load data.

[0129] Specifically, the load prediction device will obtain the load data corresponding to any two electrical devices from the coupling analysis matrix to construct a load correlation matrix and obtain multiple load correlation matrices. Then, determine the set of eigenvalues corresponding to each load correlation matrix among multiple load correlation matrices to obtain multiple third eigenvalue sets. The third eigenvalue set includes multiple third eigenvalues of the load correlation matrix corresponding to the third eigenvalue set. Next, the load prediction device takes any one electrical device corresponding to each load correlation matrix as the associated electrical device corresponding to the load correlation matrix, and takes the load data corresponding to the associated electrical device of each load correlation matrix as the associated load data of the load correlation matrix.

[0130] Furthermore, the load prediction device fills the associated load data of each load correlation matrix among multiple load correlation matrices into preset white noise to obtain multiple associated reference matrices. Determine the set of eigenvalues corresponding to each associated reference matrix among multiple associated reference matrices to obtain multiple fourth eigenvalue sets. The fourth eigenvalue set includes multiple fourth eigenvalues of the associated reference matrix corresponding to the fourth eigenvalue set.

[0131] Further, the load prediction device determines a plurality of correlation differences according to a plurality of third eigenvalue sets and a plurality of fourth eigenvalue sets, where the correlation difference is equal to the difference between the sum of a plurality of third eigenvalues in the third eigenvalue set and the sum of a plurality of fourth eigenvalues in the fourth eigenvalue set. One third eigenvalue set and one fourth eigenvalue set determine one correlation difference, and the plurality of correlation differences correspond one-to-one to the plurality of load correlation matrices. According to each correlation difference among the plurality of correlation differences, the correlation degree between two electrical devices corresponding to the load correlation matrix corresponding to the correlation difference can be determined. Among them, the mapping relationship between the correlation difference and the correlation degree can be preset. In this way, the correlation degree between any two of the n electrical devices can be determined.

[0132] Finally, according to the correlation degree between any two of the n electrical devices, the sum of the correlation degrees corresponding to each electrical device among the n electrical devices can be determined, and the total correlation degree corresponding to the electrical device can be obtained. According to the total correlation degree of each electrical device among the n electrical devices, the load correlation quantity of the electrical device in the future time period can be determined. Among them, the mapping relationship between the total correlation degree and the load correlation quantity can be preset. According to the load correlation quantity corresponding to each electrical device among the n electrical devices, the n first load data sets can be adjusted to obtain the adjusted n first load data sets. For example, the load data in the first load data set is added to the load correlation quantity of the electrical device corresponding to the first load data set to obtain the adjusted first load data set corresponding to the electrical device, so as to obtain the adjusted n first load data sets.

[0133] Thus, the adjusted n first load data sets are load data sets that combine the correlation degrees between all electrical devices in the same physical space. By predicting the load data sets of the n electrical devices in the future time period through the adjusted n first load data sets, the accuracy of load prediction can be improved.

[0134] Based on this, the adjusted n first load data sets can be input into the load prediction model. Exemplarily, the load prediction model is obtained by training a preset load prediction model, and the method further includes:

[0135] Obtain a preset load prediction model;

[0136] Obtain a plurality of training sets, test sets, and validation sets;

[0137] Use the n third load data sets of each training set in the plurality of training sets as the input of the preset load prediction model, and use the n target third load data sets of the training set as the output of the preset load prediction model to train the preset load prediction model to obtain a candidate load prediction model;

[0138] Input n fourth load datasets into the candidate load prediction model to obtain n expected load datasets;

[0139] Determine n mean squared errors based on the n expected load datasets and the n target fourth load datasets;

[0140] Determine the sum of the n mean squared errors as the model loss;

[0141] Adjust the model parameters of the candidate load prediction model according to the model loss to obtain the load prediction model.

[0142] In the embodiments of the present application, the training set includes n third load datasets of n electrical devices in the first time period and n target third load datasets in the second time period. Each electrical device corresponds to a third load dataset and a target third load dataset. The test set includes n fourth load datasets of n electrical devices in the third time period. The validation set includes n target fourth load datasets of n electrical devices in the fourth time period. Each electrical device corresponds to a fourth load dataset and a target fourth load dataset. The time length of the first time period, the time length of the third time period are equal to the time length of the historical time period, and the time length of the second time period, the time length of the fourth time period are equal to the time length of the future time period.

[0143] It should be noted that the interval time length between the historical time period and the future time period is equal to the interval time length between the first time period and the second time period, and is also equal to the interval time length between the third time period and the fourth time period. In this way, the trained load prediction model can determine the n target first load datasets in the future time period based on the n first load datasets in the historical time period.

[0144] Specifically, the load prediction device first obtains a preset load prediction model, and obtains multiple training sets, test sets and validation sets. As Figure 3 shown, the preset load prediction model may include: an input layer, a shared layer, multiple hidden layers and multiple output layers. The preset load prediction model includes multiple tasks, such as Figure 3 shown Task 1, Task 2 and Task 3. It can be understood that Figure 3 the preset load prediction model shown is only exemplary, and the number of tasks is determined by the number of electrical devices. Each task corresponds to an electrical device, and each task is used to predict the load dataset of the electrical device corresponding to the task.

[0145] When training a preset load prediction model with multiple training sets, n third load data sets of each training set in the multiple training sets are used as the input of the preset load prediction model and input into the input layer of the preset load prediction model. The n target third load data sets of the training set are used as the outputs of multiple output layers of the preset load prediction model to train the preset load prediction model. Among them, each output layer in the preset load prediction model is respectively used to output the target third load data set of each of the n electrical devices. Among them, the shared layer mainly includes a GRU layer, which is used to extract common features from the input n third load data sets and transmit the n third load data sets and the common features to the hidden layers of n tasks respectively to predict the target third load data sets of the n electrical devices. The hidden layer of each task is used to transmit the load data set of this task, and finally the target third load data set of this task is output through the output layer of this task. Among them, the hidden layer includes a GRU layer and a fully connected layer. It can be understood that the number of hidden layers of each task can be multiple, Figure 3 Only for exemplification, this application does not limit this.

[0146] Exemplarily, the structure of the GRU can be as Figure 4 shown. Among them, the GRU includes an update unit and a reset unit. The update unit is used to output the update data set corresponding to this update unit according to the output data set of the previous hidden layer. The reset unit is used to output the reset data set corresponding to this reset unit according to the common feature. According to the update data set and the reset data set of the GRU, the output data set of this GRU can be determined. Exemplarily, Figure 4 The processing process of the GRU shown can be expressed by the following formula (2):

[0147] Formula (2)

[0148] Among them, z1 represents the update data set output by the update unit, and r1 represents the reset data set output by the reset unit. represents the activation function of the GRU, such as Figure 4 shown by " ", which represents this activation function. For example, this activation function can be the Sigmoid function. represents the weight matrix connected by the update unit, represents the weight matrix connected by the reset unit. Among them, the weight matrix connected by the update unit and the weight matrix connected by the reset unit can be preset. h0 represents the output data set of the previous hidden layer, and x1 represents the common feature. represents the output data set of the next candidate hidden layer. represents the weight matrix connected by the next candidate hidden layer. tanh() represents calculating the hyperbolic tangent value. h1 represents the output data set of the GRU.

[0149] Based on this, by training the above-mentioned preset load prediction model with multiple training sets, the GRU layers in the above-mentioned shared layer and multiple hidden layers can be trained, and finally a candidate load prediction model can be obtained.

[0150] Furthermore, the load prediction device inputs the n fourth load data sets in the test set into the above-mentioned candidate load prediction model, and n expected load data sets can be obtained. According to the n expected load data sets and the n target fourth load data sets in the validation set, n mean square errors can be determined, where one expected load data set and one target fourth load data set determine one mean square error. The load prediction device can use the sum of the n mean square errors as the model loss. According to the model loss, the model parameters can be adjusted to obtain the load prediction model. For example, the model parameters can include: learning rate, the number of GRU layers, and so on.

[0151] Thus, by training the preset load prediction model with multiple training sets to obtain a candidate load prediction model, calculating the model loss of the candidate load prediction model through the test set and the validation set, and adjusting the model parameters of the candidate load prediction model according to the model loss, a load prediction model can be obtained to improve the accuracy of load prediction.

[0152] Based on this, by inputting the n first load data sets into the load prediction model, the target first load data set of each of the n electrical devices in the future time period can be obtained, and n target first load data sets are obtained. Among them, the processing method of the n first load data sets in the load prediction model is similar to the processing method of the n third load data sets in the preset load prediction model, which will not be elaborated here.

[0153] In summary, in the embodiments of the present application, first, T load data of each of the n electrical devices in the same physical space within a historical time period are obtained, resulting in a load data. T environmental data of the n electrical devices within the historical time period are obtained by each of the m environmental sensors, resulting in b environmental data. Then, based on the a load data, a basic measurement matrix is constructed, and based on the b environmental data, a first augmented measurement matrix is constructed. According to the basic measurement matrix and the first augmented measurement matrix, a coupling analysis matrix can be determined. Further, according to the coupling analysis matrix, T linear eigenvalue statistic differences are determined, and according to the T linear eigenvalue statistic differences, n first load data sets are determined. Finally, the n first load data sets are input into a load prediction model to obtain a target first load data set for each of the n electrical devices within a future time period, resulting in n target first load data sets. Thus, through the basic measurement matrix constructed by the load data of the electrical devices in the same physical space within the historical time period, and the first augmented measurement matrix constructed by the environmental data in this physical space, a coupling analysis matrix can be determined. The linear eigenvalue statistic differences determined by the coupling analysis matrix can represent the influence index of the environmental data on the load data of the electrical devices, thereby determining the first load data sets considering the influence of the environmental data. By inputting the first load data sets into the load prediction model to predict the target first load data sets of the electrical devices within the future time period, the accuracy of load prediction can be improved.

[0154] Refer to Figure 5 , Figure 5 which is a schematic diagram of a load prediction device provided by an embodiment of the present application. The load prediction device 500 can be the load prediction device in any of the above embodiments. The load prediction device 500 includes an acquisition unit 501 and a processing unit 502.

[0155] The acquisition unit 501 is configured to obtain T load data of each of the n electrical devices within a historical time period, resulting in a load data; obtain T environmental data of the n electrical devices within the historical time period by each of the m environmental sensors, resulting in b environmental data; the n electrical devices are in the same physical space; n, m, and T are all integers greater than 1, a = n T, b = m T;

[0156] The processing unit 502 is configured to construct a basic measurement matrix according to the a load data;

[0157] construct a first augmented measurement matrix according to the b environmental data;

[0158] determine a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix;

[0159] Determine T differences in linear eigenvalue statistics according to the coupling analysis matrix;

[0160] Determine n first load data sets according to the T differences in linear eigenvalue statistics; the n first load data sets correspond one-to-one with n electrical devices;

[0161] Input the n first load data sets into the load prediction model to obtain the target first load data sets of each of the n electrical devices in the future time period, and obtain n target first load data sets.

[0162] In a possible embodiment, in terms of determining the coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix, the processing unit 502 is specifically configured to:

[0163] Obtain the initial white noise intensity regulation matrix when performing dimensionality expansion on the first augmented measurement matrix;

[0164] Determine the first signal-to-noise ratio corresponding to the first augmented measurement matrix;

[0165] Perform dimensionality expansion on the first augmented measurement matrix according to the initial white noise intensity regulation matrix to obtain a second augmented measurement matrix;

[0166] Determine the second signal-to-noise ratio of the second augmented measurement matrix;

[0167] If the second signal-to-noise ratio is consistent with the first signal-to-noise ratio, splice the basic measurement matrix and the second augmented measurement matrix to obtain the coupling analysis matrix;

[0168] If the second signal-to-noise ratio is inconsistent with the first signal-to-noise ratio, adjust the initial white noise intensity regulation matrix according to the second signal-to-noise ratio and the first signal-to-noise ratio to obtain the target white noise intensity regulation matrix;

[0169] Perform dimensionality expansion on the first augmented measurement matrix according to the target white noise intensity regulation matrix to obtain a third augmented measurement matrix;

[0170] Splice the basic measurement matrix and the third augmented measurement matrix to obtain the coupling analysis matrix.

[0171] In a possible embodiment, in terms of determining the second signal-to-noise ratio of the second augmented measurement matrix, the processing unit 502 is specifically configured to:

[0172] Calculate the second signal-to-noise ratio through the following formula (1), specifically as follows:

[0173] Formula (1)

[0174] Wherein, SNR represents the second signal-to-noise ratio, Tr() represents calculating the trace of a matrix, Y represents the second augmented measurement matrix, represents the transpose matrix of the second augmented measurement matrix, represents the signal-to-noise ratio adjustment parameter, represents the initial white noise intensity regulation matrix, represents the transpose matrix of the initial white noise intensity regulation matrix.

[0175] In a possible embodiment, in determining T linear eigenvalue statistic differences according to the coupling analysis matrix, the processing unit 502 is specifically configured to:

[0176] According to the dimension of the coupling analysis matrix, perform dimension expansion on the basic measurement matrix to obtain a reference matrix;

[0177] Obtain the size of the rolling window;

[0178] According to the size of the rolling window, obtain T first window data from the coupling analysis matrix;

[0179] Determine the first eigenvalue set corresponding to each of the T first window data in the T first window data to obtain T first eigenvalue sets;

[0180] According to the T first eigenvalue sets, determine T first linear eigenvalue statistics; the T first linear eigenvalue statistics correspond one-to-one with the T first window data;

[0181] According to the T first window data, obtain T second window data from the reference matrix; the T first window data correspond one-to-one with the T second window data;

[0182] Determine the second eigenvalue set corresponding to each of the T second window data in the T second window data to obtain T second eigenvalue sets;

[0183] According to the T second eigenvalue sets, determine T second linear eigenvalue statistics; the T second linear eigenvalue statistics correspond one-to-one with the T second window data;

[0184] According to the T first linear eigenvalue statistics and the T second linear eigenvalue statistics, determine the T linear eigenvalue statistic differences.

[0185] In a possible embodiment, in determining n first load data sets according to the T linear eigenvalue statistic differences, the processing unit 502 is specifically configured to:

[0186] Determine n second load data sets according to a load data; each second load data set corresponds to an electrical device;

[0187] Determine the linear eigenvalue statistic differences among the T linear eigenvalue statistics that are less than a preset difference, and obtain c linear eigenvalue statistic differences; c is an integer greater than 1 and less than or equal to T;

[0188] According to the c linear eigenvalue statistic differences, obtain c first window data from the T first window data;

[0189] According to the c linear eigenvalue statistic differences, determine the influence index of the environmental data on the load data in each of the c first window data;

[0190] According to the influence index of the environmental data on the load data in each of the c first window data, determine the sum of the influence indexes corresponding to each of the d load data in the c first window data, and obtain d total influence indexes; the d load data are all the load data in the c first window data;

[0191] According to the d total influence indexes and the d load data, determine d adjusted load data;

[0192] According to the d adjusted load data, update the n second load data sets to obtain n first load data sets.

[0193] In a possible embodiment, before inputting the n first load data sets into the load prediction model, the processing unit 502 is further configured to:

[0194] Obtain a plurality of load correlation matrices from the coupling analysis matrix; the load correlation matrix includes the load data corresponding to any two of the n electrical devices;

[0195] Determine the eigenvalue set corresponding to each load correlation matrix in the plurality of load correlation matrices to obtain a plurality of third eigenvalue sets;

[0196] Take any one of the electrical devices corresponding to each load correlation matrix as the associated electrical device corresponding to the load correlation matrix, and take the load data corresponding to the associated electrical device of each load correlation matrix as the associated load data of the load correlation matrix;

[0197] Fill the associated load data of each load correlation matrix in the plurality of load correlation matrices with preset white noise to obtain a plurality of associated reference matrices;

[0198] Determine the eigenvalue set corresponding to each associated reference matrix in the plurality of associated reference matrices to obtain a plurality of fourth eigenvalue sets;

[0199] According to the plurality of third eigenvalue sets and the plurality of fourth eigenvalue sets, determine the correlation degree between any two of the n electrical devices;

[0200] Adjust the n first load data sets according to the correlation degree between any two of the n electrical devices, so as to obtain the adjusted n first load data sets.

[0201] In a possible embodiment, the processing unit 502 is further configured to:

[0202] Obtain a preset load prediction model;

[0203] Obtain multiple training sets, test sets and validation sets; the training set includes the n third load data sets of the n electrical devices in the first time period and the n target third load data sets in the second time period, and each electrical device corresponds to a third load data set and a target third load data set; the test set includes the n fourth load data sets of the n electrical devices in the third time period; the validation set includes the n target fourth load data sets of the n electrical devices in the fourth time period, and each electrical device corresponds to a fourth load data set and a target fourth load data set; the time length of the first time period and the time length of the third time period are equal to the time length of the historical time period, and the time length of the second time period and the time length of the fourth time period are equal to the time length of the future time period;

[0204] Use the n third load data sets of each training set in the multiple training sets as the input of the preset load prediction model, and use the n target third load data sets of the training set as the output of the preset load prediction model to train the preset load prediction model to obtain a candidate load prediction model;

[0205] Input the n fourth load data sets into the candidate load prediction model to obtain n expected load data sets;

[0206] Determine n mean square errors according to the n expected load data sets and the n target fourth load data sets;

[0207] Determine the sum of the n mean square errors as the model loss;

[0208] Adjust the model parameters of the candidate load prediction model according to the model loss to obtain a load prediction model.

[0209] Refer to Figure 6 , Figure 6 is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 6 shown, the electronic device 600 includes a transceiver 601, a processor 602 and a memory 603. They are connected through a bus 604. The memory 603 is used to store computer programs and data, and can transmit the data stored in the memory 603 to the processor 602. The electronic device 600 may be the load prediction device 500.

[0210] The processor 602 is used to read the computer program in the memory 603 and perform the following operations:

[0211] Obtain T load data of each of the n electrical devices in the historical time period to obtain a load data; obtain T environmental data of the n electrical devices in the historical time period through each of the m environmental sensors to obtain b environmental data; the n electrical devices are in the same physical space; n, m, and T are all integers greater than 1, and a = n T, and b = m T;

[0212] Construct a basic measurement matrix according to the a load data;

[0213] Construct a first augmented measurement matrix according to the b environmental data;

[0214] Determine a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix;

[0215] Determine the difference of T linear eigenvalue statistics according to the coupling analysis matrix;

[0216] Determine n first load data sets according to the difference of T linear eigenvalue statistics; the n first load data sets correspond one-to-one with the n electrical devices;

[0217] Input the n first load data sets into the load prediction model to obtain the target first load data sets of each of the n electrical devices in the future time period, and obtain n target first load data sets.

[0218] The above mainly introduces the solution of the embodiment of the present application from the perspective of the execution process on the method side. It can be understood that in order for the electronic device to implement the above functions, it includes the corresponding hardware structure and / or software module for executing each function. Those skilled in the art should easily realize that, combined with the units and algorithm steps of each example described in the embodiments provided in this article, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed in the way of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0219] The embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement part or all of the steps of any one of the methods described in the above method embodiments.

[0220] An embodiment of the present application further provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. The computer program is operable to cause a computer to execute some or all of the steps of any one of the methods described in the above method embodiments.

[0221] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0222] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0223] In several embodiments provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0224] The unit described as a separated component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0225] In addition, in each embodiment of the present application, the functional units can be integrated into one processing unit, or each unit exists physically alone, or two or more units are integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software program module.

[0226] When the integrated unit is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned memory includes: various media such as USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0227] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable memory, and the memory can include: flash drives, read-only memories (abbreviation: ROM, English: Read-Only Memory), random access memories (abbreviation: RAM, English: Random Access Memory), magnetic disks, or optical discs, etc.

[0228] The embodiments of the present application have been introduced in detail above. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A load forecasting method, characterized in that: include: Obtain T load data of each of n electrical devices in a historical time period to obtain a load data; obtain T environmental data of the n electrical devices in the historical time period through each of m environmental sensors to obtain b environmental data; the n electrical devices are in the same physical space; n, m, T are all integers greater than 1, a=n T,b=m T; According to the a load data, construct a basic measurement matrix; Constructing a first augmented measurement matrix according to the b environmental data; Determining a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix; Determining T linear eigenvalue statistical difference values ​​according to the coupling analysis matrix; Determining n first load data sets according to the T linear eigenvalue statistical difference values; The n first load data sets correspond one-to-one to the n electrical devices; Inputting the n first load data sets into a load prediction model to obtain a target first load data set for each of the n electrical devices in a future time period, thereby obtaining n target first load data sets; Before inputting the n first load data sets into the load forecasting model, the method further includes: Acquire multiple load association matrices from the coupling analysis matrix; the load association matrix includes load data corresponding to any two electrical devices among the n electrical devices; Determine an eigenvalue set corresponding to each load association matrix in the plurality of load association matrices to obtain a plurality of third eigenvalue sets; Any electric device corresponding to each load association matrix is ​​used as the associated electric device corresponding to the load association matrix, and the load data corresponding to the associated electric device of each load association matrix is ​​used as the associated load data of the load association matrix; Filling the associated load data of each load association matrix in the multiple load association matrices with preset white noise to obtain multiple associated reference matrices; Determine an eigenvalue set corresponding to each correlation reference matrix in the plurality of correlation reference matrices to obtain a plurality of fourth eigenvalue sets; Determining, according to the plurality of third eigenvalue sets and the plurality of fourth eigenvalue sets, a correlation degree between any two of the n electrical devices; Based on the correlation between any two of the n electrical devices, determine the sum of the correlations corresponding to each of the n electrical devices to obtain the total correlation corresponding to each of the n electrical devices; based on the total correlation of each of the n electrical devices, determine the load correlation amount of each of the n electrical devices in the future time period; based on the load correlation amount corresponding to each of the n electrical devices, adjust the n first load data sets to obtain adjusted n first load data sets.

2. The method according to claim 1, characterized in that The step of determining a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix comprises: Obtaining an initial white noise intensity control matrix when performing dimension expansion on the first augmented measurement matrix; Determining a first signal-to-noise ratio corresponding to the first augmented measurement matrix; According to the initial white noise intensity control matrix, dimensionally expand the first augmented measurement matrix to obtain a second augmented measurement matrix; determining a second signal-to-noise ratio of the second augmented measurement matrix; If the second signal-to-noise ratio is consistent with the first signal-to-noise ratio, concatenating the basic measurement matrix with the second augmented measurement matrix to obtain the coupling analysis matrix; If the second signal-to-noise ratio is inconsistent with the first signal-to-noise ratio, adjusting the initial white noise intensity control matrix according to the second signal-to-noise ratio and the first signal-to-noise ratio to obtain a target white noise intensity control matrix; According to the target white noise intensity control matrix, dimensionally expand the first augmented measurement matrix to obtain a third augmented measurement matrix; The basic measurement matrix and the third augmented measurement matrix are concatenated to obtain the coupling analysis matrix.

3. The method according to claim 2, characterized in that The determining a second signal-to-noise ratio of the second augmented measurement matrix includes: The second signal-to-noise ratio is calculated by the following formula, which is as follows: Wherein, SNR represents the second signal-to-noise ratio, Tr() represents the trace of the calculation matrix, Y represents the second augmented measurement matrix, represents the transposed matrix of the second augmented measurement matrix, represents the signal-to-noise ratio adjustment parameter, represents the initial white noise intensity control matrix, represents the transposed matrix of the initial white noise intensity control matrix.

4. The method according to any one of claims 1 to 3, characterized in that Determining T linear eigenvalue statistical difference values ​​according to the coupling analysis matrix includes: According to the dimension of the coupling analysis matrix, the dimension of the basic measurement matrix is ​​expanded to obtain a reference matrix; Get the size of the scroll window; According to the size of the rolling window, obtaining T first window data from the coupling analysis matrix; Determine a first eigenvalue set corresponding to each first window data in the T first window data to obtain T first eigenvalue sets; Determine T first linear eigenvalue statistics according to the T first eigenvalue sets; the T first linear eigenvalue statistics correspond one-to-one to the T first window data; According to the T first window data, T second window data are acquired from the reference matrix; the T first window data correspond to the T second window data one by one; Determine a second eigenvalue set corresponding to each second window data in the T second window data to obtain T second eigenvalue sets; Determine T second linear eigenvalue statistics according to the T second eigenvalue sets; the T second linear eigenvalue statistics correspond one-to-one to the T second window data; The T linear eigenvalue statistical differences are determined according to the T first linear eigenvalue statistics and the T second linear eigenvalue statistics.

5. The method according to claim 4, characterized in that The step of determining n first load data sets according to the T linear eigenvalue statistical difference values ​​comprises: Determine n second load data sets according to the a load data; each second load data set corresponds to an electrical device; Determine the linear eigenvalue statistical difference values ​​that are less than the preset difference value among the T linear eigenvalue statistical difference values, and obtain c linear eigenvalue statistical difference values; c is an integer greater than 1 and less than or equal to T; Obtaining c first window data from the T first window data according to the c linear eigenvalue statistical difference values; Determine, according to the c linear eigenvalue statistical difference values, an influence index of the environmental data in each of the c first window data on the load data; According to the influence index of the environmental data in each of the c first window data on the load data, determine the sum of the influence index corresponding to each load data in the d load data of the c first window data, and obtain d total influence indexes; the d load data are all the load data in the c first window data; Determining d adjusted load data according to the d total impact indexes and the d load data; The n second load data sets are updated according to the d adjusted load data to obtain the n first load data sets.

6. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: Obtaining a preset load forecasting model; Acquire multiple training sets, test sets and validation sets; the training set includes n third load data sets of the n electrical devices in a first time period and n target third load data sets in a second time period, and each electrical device corresponds to a third load data set and a target third load data set; the test set includes n fourth load data sets of the n electrical devices in a third time period; the validation set includes n target fourth load data sets of the n electrical devices in a fourth time period, and each electrical device corresponds to a fourth load data set and a target fourth load data set; the time length of the first time period and the time length of the third time period are equal to the time length of the historical time period, and the time length of the second time period and the time length of the fourth time period are equal to the time length of the future time period; Taking n third load data sets of each training set in the multiple training sets as input of the preset load forecasting model, taking n target third load data sets of the training set as output of the preset load forecasting model, training the preset load forecasting model, and obtaining a candidate load forecasting model; Inputting the n fourth load data sets into the candidate load forecasting model to obtain n expected load data sets; Determining n mean square errors according to the n expected load data sets and the n target fourth load data sets; Determine the sum of the n mean square errors as the model loss; According to the model loss, the model parameters of the candidate load forecasting model are adjusted to obtain the load forecasting model.

7. A load forecasting device, characterized in that: include: an acquisition unit, configured to acquire T load data of each of n electrical devices in a historical time period, and obtain a load data; acquire T environmental data of the n electrical devices in the historical time period through each of m environmental sensors, and obtain b environmental data; the n electrical devices are in the same physical space; n, m, T are all integers greater than 1, a=n T,b=m T; A processing unit, used for constructing a basic measurement matrix according to the a load data; Constructing a first augmented measurement matrix according to the b environmental data; Determining a coupling analysis matrix according to the basic measurement matrix and the first augmented measurement matrix; Determining T linear eigenvalue statistical difference values ​​according to the coupling analysis matrix; Determining n first load data sets according to the T linear eigenvalue statistical difference values; The n first load data sets correspond one-to-one to the n electrical devices; Inputting the n first load data sets into a load prediction model to obtain a target first load data set for each of the n electrical devices in a future time period, thereby obtaining n target first load data sets; Before inputting the n first load data sets into the load forecasting model, the processing unit is further used to: Acquire multiple load association matrices from the coupling analysis matrix; the load association matrix includes load data corresponding to any two electrical devices among the n electrical devices; Determine an eigenvalue set corresponding to each load association matrix in the plurality of load association matrices to obtain a plurality of third eigenvalue sets; Any electric device corresponding to each load association matrix is ​​used as the associated electric device corresponding to the load association matrix, and the load data corresponding to the associated electric device of each load association matrix is ​​used as the associated load data of the load association matrix; Filling the associated load data of each load association matrix in the multiple load association matrices with preset white noise to obtain multiple associated reference matrices; Determine an eigenvalue set corresponding to each correlation reference matrix in the plurality of correlation reference matrices to obtain a plurality of fourth eigenvalue sets; Determining, according to the plurality of third eigenvalue sets and the plurality of fourth eigenvalue sets, a correlation degree between any two of the n electrical devices; Based on the correlation between any two of the n electrical devices, determine the sum of the correlations corresponding to each of the n electrical devices to obtain the total correlation corresponding to each of the n electrical devices; based on the total correlation of each of the n electrical devices, determine the load correlation amount of each of the n electrical devices in the future time period; based on the load correlation amount corresponding to each of the n electrical devices, adjust the n first load data sets to obtain adjusted n first load data sets.

8. An electronic device, characterized in that: include: A processor and a memory, the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device executes the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method according to any one of claims 1 to 6.

Images