Power load prediction method, system and product considering industry coupling correlation characteristics

Abstract

The invention discloses a power load prediction method, system and product considering industry coupling correlation characteristics, and belongs to the field of power load prediction, and the method comprises the steps: determining the disturbance quantity of the power load data of all industries according to the obtained power load data of all industries and a data disturbance analysis model; performing variance decomposition on the power load data of each industry according to the disturbance quantity of the power load data among the industries to obtain a total disturbance variance of the power load data of each industry influenced by all other industries and a disturbance variance of the power load data of each industry influenced by all other industries; for each industry, calculating the ratio of each disturbance variance of the industry to the total disturbance variance of the industry, obtaining the coupling correlation characteristics among the industries, and further determining the correlation industry of each industry; and for each industry, carrying out power load prediction by using the associated industry of the industry. The problem of inaccurate determination of the associated industry at present is solved, and accurate prediction of the industry power load is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power load forecasting, and in particular to a power load forecasting method, system and product taking into account industry coupling correlation characteristics. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Due to the mutual coupling between industries, the electricity demand of a single industry is not only affected by its own development, but also restricted by other industries. Therefore, how to quantify and analyze the coupling relationship between industries and support high-precision industry load forecasting has become an urgent problem that needs to be solved.

[0004] Currently, the Pearson correlation coefficient method is mainly used to calculate and determine the correlation coefficient of power load data between industries, and the related industries are determined based on the correlation coefficient.

[0005] However, the correlation coefficient of power load data between industries determined by the Pearson correlation coefficient method only considers the positive and negative correlations between power load data of different industries, but does not consider the impact of errors between power load data of different industries. As a result, the correlation coefficient of power load data between industries determined by the Pearson correlation coefficient method cannot accurately reflect the coupling relationship between power load data of different industries, which in turn leads to inaccurate determination of related industries. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a power load forecasting method, system and product that take into account the coupling correlation characteristics of industries, thereby improving the accuracy of determining the coupling correlation characteristics between industries and the accuracy of dividing strongly related industries.

[0007] To achieve the above object, the present invention adopts the following technical solution:

[0008] First, a power load forecasting method considering the industry coupling correlation characteristics is proposed, including:

[0009] Obtain power load data for multiple industries;

[0010] According to the acquired power load data of all industries and the data disturbance analysis model, the disturbance amount of the power load data among the industries is determined; wherein the data disturbance analysis model is constructed by using the vector error correction model;

[0011] According to the disturbance amount of power load data among industries, variance decomposition is performed on the power load data of each industry to obtain the total disturbance variance of each industry's power load data affected by all other industries and the disturbance variance of each industry's power load data affected by other industries;

[0012] For each industry, calculate the proportion of the disturbance variance of the power load data of the industry affected by other industries to the total disturbance variance of the power load data of the industry affected by all other industries, and obtain the coupling correlation characteristics between the industries;

[0013] According to the coupling characteristics between industries, determine the related industries of each industry;

[0014] For each industry, the power load forecast is performed using the industry's associated industries.

[0015] Furthermore, according to the disturbance amount of the power load data among various industries, the influence degree of the white noise vector on the power load data of various industries is determined;

[0016] According to the influence of the white noise vector on the power load data of each industry, the total disturbance variance of the power load data of each industry affected by all other industries and the disturbance variance of the power load data of each industry affected by other industries are determined.

[0017] Furthermore, the power load data of each industry is taken as a separate time series data, and a vector error correction model between the power load data of multiple industries is constructed;

[0018] Convert the constructed vector error correction model into a vector autoregressive model of the load data of each industry;

[0019] The vector autoregression model of the load data of each industry is transformed to obtain the data disturbance analysis model.

[0020] Furthermore, the power load data of various industries obtained are data that have passed the stationarity test, cointegration test and Granger causality test.

[0021] Furthermore, for each industry, the process of determining the related industries of the industry is as follows:

[0022] Arrange all the coupling-related characteristics of the industry in descending order, and take the first few industries whose sum of coupling-related characteristics is greater than the set threshold as the related industries of the industry.

[0023] Furthermore, for each industry, a power load forecasting model is used to forecast power load; wherein the power load forecasting model is obtained by constructing a long- and short-time network.

[0024] Secondly, a power load forecasting system considering the industry coupling correlation characteristics is proposed, including:

[0025] Data acquisition module, used to obtain power load data of multiple industries;

[0026] The disturbance quantity acquisition module is used to determine the disturbance quantity of the power load data between industries based on the acquired power load data of all industries and the data disturbance analysis model; wherein the data disturbance analysis model is constructed by using a vector error correction model;

[0027] The variance decomposition module is used to perform variance decomposition on the power load data of each industry according to the disturbance amount of the power load data between industries, and obtain the total disturbance variance of the power load data of each industry affected by all other industries and the disturbance variance of the power load data of each industry affected by other industries;

[0028] The coupling correlation feature module is used to calculate, for each industry, the proportion of the disturbance variance of the power load data of the industry affected by other industries to the total disturbance variance of the power load data of the industry affected by all other industries, and obtain the coupling correlation characteristics between the industries;

[0029] A strongly related industry determination module is used to determine the related industries of each industry based on the coupling correlation characteristics between industries;

[0030] The power load forecasting unit is used to forecast the power load for each industry by using the related industries of the industry.

[0031] Furthermore, the variance decomposition module is used to determine the degree of influence of the white noise vector on the power load data of each industry according to the disturbance amount of the power load data between each industry;

[0032] According to the influence of the white noise vector on the power load data of each industry, the total disturbance variance of the power load data of each industry affected by all other industries and the disturbance variance of the power load data of each industry affected by other industries are determined.

[0033] Furthermore, the construction process of the data disturbance analysis model in the disturbance quantity acquisition module includes:

[0034] The power load data of each industry is taken as a separate time series data, and a vector error correction model between the power load data of multiple industries is constructed;

[0035] Convert the constructed vector error correction model into a vector autoregressive model of the load data of each industry;

[0036] The vector autoregression model of the load data of each industry is transformed to obtain the data disturbance analysis model.

[0037] In a third aspect, a computer program product is proposed, which includes a computer program. When the computer program is executed by a processor, it implements the power load forecasting method considering industry coupling correlation characteristics proposed in the first aspect.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention proposes a method, system and product for predicting electric loads considering the coupling and correlation characteristics of industries. The method determines the disturbance amount of electric load data between industries through a vector error correction model, and then performs variance decomposition on the electric load data of each industry according to the disturbance amount of the electric load data between industries, so as to obtain the total disturbance variance of the electric load data of each industry affected by all other industries and the disturbance variance of the electric load data of each industry affected by other industries after considering the coupling and correlation characteristics between industries; for each industry, the proportion of the disturbance variance of the electric load data of the industry affected by other industries to the total disturbance variance of the electric load data of the industry affected by all other industries is calculated, so as to obtain the coupling and correlation characteristics between industries; the coupling and correlation characteristics can fully and accurately reflect the correlation between the electric load data of each industry; when the associated industries of each industry are determined through the coupling and correlation characteristics, the accuracy of determining the associated industries is improved, thereby improving the accuracy of electric load prediction for each industry.

[0040] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0042] Figure 1 A flowchart of a method for predicting power load considering industry coupling correlation characteristics disclosed in an embodiment;

[0043] Figure 2 A structural block diagram of a power load forecasting model in a power load forecasting method considering industry coupling correlation characteristics disclosed in an embodiment;

[0044] Figure 3 The variance decomposition results of the chemical raw materials and chemical products manufacturing industries disclosed in the examples;

[0045] Figure 4 This is the power load prediction result disclosed in the embodiment. DETAILED DESCRIPTION

[0046] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0047] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0048] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0049] Example 1

[0050] Due to the mutual coupling between industries, the electricity demand of a single industry is not only affected by its own development, but also restricted by other industries. Therefore, how to quantify and analyze the coupling relationship between industries and support high-precision industry load forecasting has become an urgent problem that needs to be solved.

[0051] In this context, the patent with publication number CN118316033A proposed a power load forecasting method based on CNN-GRU and ARIMA models, which obtains power load data and performs data preprocessing, and then constructs GRU-CNN modules and ARIMA modules to forecast power load. This helps to more accurately extract the time domain and frequency domain features of power load data and solve the shortcomings of traditional load forecasting; the patent with publication number CN118316019A proposed a short-term load forecasting method based on the TCN-BiLSTM-PSO combined model, which decomposes historical load data to obtain the input of the TCN model and the BiLSTM model, and then predicts the power load.

[0052] The above two patents only rely on the industry's own power load data when forecasting the industry's power load, without considering the impact of coupled industries on their own industries, resulting in low accuracy of the industry's power load forecast results.

[0053] The patent with publication number CN113962456A proposes a medium- and long-term load forecasting method that takes into account industry correlation. It analyzes and quantifies industry correlation based on the Shapiro-Wilk test and the Pearson correlation coefficient, selects strongly correlated industries based on correlation indicators, and uses long- and short-term memory neural networks to predict industry medium- and long-term loads based on industry load data, temperature data, and industry correlation. Although the correlation between industries is taken into account; however, the correlation coefficient of power load data between industries determined by the Pearson correlation coefficient method only considers the positive and negative correlation between power load data of different industries, but does not consider the influence of errors between power load data of different industries. As a result, the correlation coefficient of power load data between industries determined by the Pearson correlation coefficient method cannot accurately reflect the coupling relationship between power load data of various industries, which in turn leads to inaccurate determination of the associated industries, resulting in low accuracy of the obtained industry power load forecast results.

[0054] In order to accurately measure the coupling relationship between the power load data of various industries, this embodiment discloses a power load forecasting method considering the industry coupling correlation characteristics, such as Figure 1-Figure 3 As shown, including:

[0055] S1: Obtain power load data for multiple industries.

[0056] Taking into account the large gap between power load data of different industries and the similar characteristics of power load data among some industries, the industries are classified and summarized, resulting in a total of eleven major industry categories, as shown in Table 1.

[0057] Table 1 List of eleven major industry categories

[0058] Industry No. Industry Name I1 1. Mining I2 2. Food processing and manufacturing industry I3 3. Chemical raw materials and chemical products manufacturing industry I4 4. Pharmaceutical Manufacturing I5 5. Non-metallic mineral products industry I6 6. Ferrous metal smelting and rolling processing industry I7 7. Nonferrous metal smelting and rolling processing industry I8 8. Metal products industry I9 9. Equipment manufacturing industry I10 10. Other manufacturing industries I11 11. Electricity, heat, gas and water production and supply industry

[0059] The power load data of various industries obtained in this embodiment are data that have passed the stationarity test, the cointegration relationship test, and the Granger causality test.

[0060] The power load data of various industries obtained in this embodiment include the power load data of the 11 industries in the above table.

[0061] The process of obtaining the power load data of each industry in this embodiment includes:

[0062] Obtain the original load sequence of each industry, which is the original power load sequence;

[0063] Preprocess the original load sequences of each industry to obtain the preprocessed sequences;

[0064] Perform a stationary test on the preprocessed series;

[0065] If there are unsteady sequences in the preprocessed sequence, the unsteady sequence is processed by difference method so that all the preprocessed sequences are stable sequences;

[0066] The Akaike information criterion (AIC), Schwarz Information Criterion (SIC) and Hannan-Quinn Criterion (HQ) criteria are used to process the stationary preprocessed series to obtain the optimal lag order p.

[0067] Afterwards, the Johansen test method is used to test the cointegration relationship of each stable pre-processed series to determine the cointegration relationship between power load series in different industries;

[0068] Granger causality is used to determine the causal relationship of each stable preprocessed series, and the causal relationship between the power load series of each industry is obtained.

[0069] Only the data that have passed the stationarity test, cointegration test and Granger causality test are used to obtain the power load data of various industries.

[0070] Preferably, the process of preprocessing the original load sequences of various industries includes using the moving average method to identify anomalies in the original load sequences, using the interpolation method to replace and fill the identified abnormal values ​​and missing values, and then taking the logarithm of the data processed by the interpolation method to narrow the gap between the load sequences of different industries and obtain the preprocessed sequences.

[0071] Specifically, the results of the stationarity test (ADF test) of the preprocessed series of each industry are shown in Table 2.

[0072] Table 2 Stationarity test results of preprocessed series in various industries

[0073]

[0074]

[0075] According to Table 2, the stationarity test results of the eleven industries are all "yes", indicating that the pre-processed sequences of these eleven industries are all stationary and no difference is required to proceed to the subsequent analysis steps.

[0076] When determining the optimal lag order: first establish a general vector autoregressive model (VAR) model, then use the AIC criterion, SIC criterion and HQ criterion to find the optimal lag order of the model. The results of the three criteria analysis are shown in Table 3.

[0077] Table 3. Judgment of the optimal lag order

[0078]

[0079]

[0080] According to Table 3, “*” indicates the optimal lag order selected by the criterion. It can be seen that the AIC criterion, SIC criterion and HQ criterion all consider the 16th order to be the optimal lag order. The selected order is relatively high, but considering that the load is 96-point daily load data, the 16th order is equivalent to a lag of 4 hours, so the 16th order is reasonable.

[0081] The results of the cointegration test of the pre-processed series of each industry using the Johansen test method are shown in Tables 4 and 5. Among them, Table 4 is the test result of the trace statistic, and Table 5 is the test result of the maximum eigenvalue statistic.

[0082] Table 4 Trance statistic test results

[0083]

[0084] Table 5 Maximum Eigenvalue statistic test results

[0085]

[0086]

[0087] As can be seen from Table 4 and Table 5, the Trance statistic test results and the Maximum Eigenvalue statistic test results both indicate that there are 10 cointegration relationships in the preprocessed sequences of the eleven industries, which meet the basic conditions for building a Vector Error Correction Model (VECM) model, and a VECM model can be built.

[0088] Taking the chemical raw materials and chemical products manufacturing industry (I3) as an example, the Granger causality test results between the chemical raw materials and chemical products manufacturing industry (I3) and the other ten industries are shown in Table 6.

[0089] Table 6 Granger causality test results for chemical raw materials and chemical products manufacturing industries

[0090] Industry No. Chi-square statistic Degrees of Freedom p-value I1 313.451 16 0 I2 475.254 16 0 I4 213.241 16 0 I5 354.448 16 0 I6 101.510 16 0 I7 16.286 16 0.433 I8 138.383 16 0 I9 178.362 16 0 I10 173.609 16 0 I11 240.892 16 0 All 313.451 16 0

[0091] As can be seen from Table 6, although there is no Granger causality between the chemical raw materials and chemical products manufacturing industry and the I7 industry, namely the non-ferrous metal smelting and rolling processing industry, it has a Granger causal relationship with the other ten industries as a whole. Therefore, it can be considered that this industry has a causal relationship with other industries.

[0092] S2: Determine the disturbance amount of the power load data among industries based on the acquired power load data of all industries and the data disturbance analysis model; wherein the data disturbance analysis model is constructed using a vector error correction model.

[0093] Among them, the construction process of the data perturbation analysis model is:

[0094] S21: The power load data of each industry is taken as a separate time series data, and a vector error correction model between the power load data of multiple industries is constructed.

[0095] The Vector Autoregressive Model (VAR) model is a model used to analyze the relationship between multiple time series variables. It can effectively reveal the dynamic relationship between variables and quantitatively express the degree of influence between variables. The mathematical expression of the VAR model is:

[0096] y t =A 1 y t-1 +A 2 y t-2 +…+A p y t-p +Bx t +ε t (1)

[0097] In the formula, y t is an endogenous variable; x t is an exogenous variable; A 1 ,A 2 ,…,A p is the coefficient matrix; p is the lag order; B is x t The coefficient to be estimated; ε t is a random disturbance.

[0098] Vector Error Correction Model (VECM). As an extension of the VAR model, the VECM model introduces cointegration analysis and error correction terms, so that the VECM model can capture short-term fluctuations while taking into account the impact of long-term relationships, thereby solving the problem of the VAR model ignoring cointegration relationships when dealing with non-stationary time series data. Cointegration relationships are used to represent the long-term equilibrium relationship between variables, that is, if two or more time series variables are non-stationary, but some linear combination of them is stationary, then there is a cointegration relationship between these variables.

[0099] Let the time series vector P be t =[p 1,t ,…,p i,t ,…,p n,t ], where each component p i,t are all time series. In this embodiment, pi ,t is the power load data of industry i. t Linear regression can be used to obtain the VAR model, which is expressed as

[0100] P t =A 1 P t-1 +A 2 P t-2 +…+A p P t-p +ε t (2)

[0101] Where P t is the current value of industry load; P t-p is the p-order lag of the current value of the industry load; A p is the parameter matrix, p is the optimal lag order determined by AIC criterion, SIC criterion and HQ criterion; ε t is a random disturbance.

[0102] Performing differential transformation on equation (2), we can get

[0103]

[0104] Where ∏ = αβ′, the α matrix is ​​the adjustment coefficient matrix, each column vector of which corresponds to a set of weights of a cointegrating combination, and the β′ matrix is ​​the cointegrating vector matrix, each row vector of which is a cointegrating vector, which is determined by performing a cointegration test on the preprocessed sequence.

[0105] Therefore, ∏P t-1 It can be expressed as αβ′P t-1 , and because β′P t-1 is the error correction term e t-1, so the vector error correction model between the power load data of multiple industries can be obtained as:

[0106]

[0107] Error correction term e t-1 It reflects the long-term equilibrium relationship between variables. The coefficient matrix α reflects the adjustment speed of the error correction term to adjust the equilibrium relationship between industries to the equilibrium state when the equilibrium relationship between industries deviates from the long-term equilibrium state. k It reflects the impact of short-term fluctuations in various industries on short-term changes in the target industry.

[0108] S22: Convert the constructed vector error correction model into a vector autoregression model of the load data of each industry.

[0109] Compared with the common methods for analyzing the correlation between variables, such as gray correlation analysis and causal analysis, VECM can be used to analyze both long-term and short-term relationships between variables and can reflect the dynamic change rules. However, the gray correlation analysis method is weak in capturing the dynamic relationship between variables, and the causal analysis rule may not be able to provide such a comprehensive analysis. Therefore, this embodiment selects VECM to analyze the coupling correlation characteristics between industries.

[0110] According to the above analysis, it can be seen that the coupling correlation characteristics between industries are mainly affected by three parameters, namely α, β′, ∏ k For these three parameters, the variance decomposition method is usually used in VECM to reorganize the information of the three parameters and extract key information. The variance decomposition method decomposes the overall fluctuation (variance) of a complex variable into the contribution of each component, thereby quantifying the impact of different factors on the overall fluctuation.

[0111] For formula (4), it is transformed into the VAR model shown in formula (1), and the lag operator L is introduced to obtain the vector autoregression model of the load data of each industry:

[0112]

[0113] S23: Transform the vector autoregression model of the load data of each industry to obtain a data disturbance analysis model.

[0114] The data perturbation analysis model is:

[0115]

[0116] Where, L k is the k-order lag operator, I n is the identity matrix.

[0117] In this embodiment, the power load data of each industry is substituted into formula (6) to calculate the disturbance amount ε of the power load data between industries: t .

[0118] S3: According to the disturbance amount of power load data between industries, the power load data of each industry is decomposed with variance to obtain the total disturbance variance of each industry's power load data affected by all other industries and the disturbance variance of each industry's power load data affected by other industries. The process includes:

[0119] S31: According to the disturbance amount of power load data among various industries, determine the influence degree C of white noise vector on power load data of various industries k .

[0120] Since the model corresponding to equation (6) is stable, equation (6) can be expressed as an infinite moving average process of a white noise vector to obtain equation (7), namely:

[0121]

[0122] In the formula, C k is the coefficient matrix, which represents the i-th order lagged white noise vector pair P t The degree of influence, C 0 =I n .

[0123] The component form corresponding to formula (7) is as follows:

[0124]

[0125] In the formula, is the coefficient matrix C k The specific element of is the jth component ε of the k-order lagged disturbance vector j,t-k P i,t impact.

[0126] In this embodiment, the disturbance amount ε of the power load data between industries t and the time series vector P composed of the power load data of various industries t Substituting into equation (8), we can calculate the influence of the white noise vector on the power load data of each industry.

[0127] S32: The influence of white noise vector on the power load data of various industries Calculate and determine the total disturbance variance of each industry's power load data affected by all other industries and the disturbance variance of each industry's power load data affected by other industries.

[0128] In ε tUnder the premise that the components are not serially correlated and the covariance matrix is ​​a diagonal matrix, the power load data p of industry i is i,t The total disturbance variance δ(p i,t )as follows:

[0129]

[0130] In the formula, σ j is the variance of the disturbance corresponding to the jth industry, according to the disturbance ε t Obtained by calculation.

[0131] S4: For each industry, calculate the proportion of the disturbance variance of the power load data of the industry affected by other industries to the total disturbance variance of the power load data of the industry affected by all other industries, and obtain the coupling correlation characteristics between industries.

[0132] p i,t The total disturbance variance can be decomposed into n unrelated influencing components according to p i,t The variance of disturbances affected by other industries is p i,t The proportion of the total disturbance variance affected by all other industries is used to observe the impact of other industries j on industry i. In other words, by calculating R i,j The size of the coupling correlation characteristics between industries can be quantitatively analyzed. i,j The calculation formula is as follows:

[0133]

[0134] The coupling correlation characteristics between industry j and industry i can also be called the contribution rate of industry j to industry i.

[0135] S5: Determine the related industries of each industry based on the coupling correlation characteristics between industries.

[0136] For each industry, the process of determining the industry's associated industries is:

[0137] Arrange all the coupling-related characteristics of the industry in descending order, and take the first few industries whose sum of coupling-related characteristics is greater than the set threshold as the related industries of the industry.

[0138] Taking the chemical raw materials and chemical products manufacturing industry as an example, according to the above analysis, the variance decomposition of the constructed VECM error term is carried out. The specific results of the variance decomposition of the chemical raw materials and chemical products manufacturing industry are shown in Figure 3Each color in the figure represents the coupling correlation characteristics between an industry and the chemical raw materials and chemical products manufacturing industry. The coupling correlation characteristics are arranged in descending order and accumulated from front to back. When the cumulative coupling correlation characteristics with the chemical raw materials and chemical products manufacturing industry exceed 90%, the selected industry is considered to be an associated industry of the chemical raw materials and chemical products manufacturing industry.

[0139] according to Figure 3 , the chemical raw materials and chemical products manufacturing industry is mainly affected by its own historical development and other industries, among which the impact of the industry's own historical development on the industry load is expected to reach 40%. From the perspective of its own historical development, the impact is the largest in the first period, with a contribution rate of nearly 68% to the load, and then the impact of its own historical development gradually weakens, and the final contribution rate is expected to be about 40%; from the perspective of the impact of other industries, mining (I1), food processing manufacturing (I2) and pharmaceutical manufacturing (I4) are the main externally affected industries. As time goes by, along with the weakening of the impact of the historical development of the chemical raw materials and chemical products manufacturing industry, the impact of the three industries on the load of the chemical raw materials and chemical products manufacturing industry increases, and it is expected to reach more than 50%. Therefore, mining (I1), food processing manufacturing (I2) and pharmaceutical manufacturing (I4) can be selected as related industries of the chemical raw materials and chemical products manufacturing industry.

[0140] S6: For each industry, use the industry's associated industries to perform power load forecasting.

[0141] In this embodiment, for each industry, power load forecasting model is used to perform power load forecasting; wherein, the power load forecasting model takes the historical load data of a certain industry and its related industries as input, takes the power load forecasting result of the industry as output, and is constructed using a long short-term memory (LSTM).

[0142] Mainstream load forecasting technologies are concentrated in machine learning and deep learning. Since the Long Short-Term Memory (LSTM) network can capture and retain long-term memory well, can selectively forget previous memory, and can adapt to data at different time points, this embodiment selects this algorithm for load forecasting in typical industries.

[0143] In this embodiment, mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) are selected as evaluation indicators of load forecasting results.

[0144] The MAE calculation formula is as follows:

[0145]

[0146] The RMSE calculation formula is as follows:

[0147]

[0148] The MAPE calculation formula is as follows:

[0149]

[0150] In the formula, n is the number of samples, Y i is the actual value, is the predicted value.

[0151] The hyper parameters of the power load prediction model constructed in this embodiment are shown in Table 6.

[0152] Table 7 Hyperparameter statistics

[0153] Parameter name size time_step (training step length) 16 time_prediction (prediction step) 16 batch_size 32 hidden_size(hidden layer) 64 num_layers (LSTM layers) 4

[0154] According to the above analysis, for the chemical raw materials and chemical products manufacturing industry, the mining industry (I1), food processing manufacturing industry (I2) and pharmaceutical manufacturing industry (I4) are taken as related industries, that is, the loads of these three industries are input into the power load forecasting model as influencing factors; for the non-metallic mineral products industry, the mining industry (I1), food processing manufacturing industry (I2), non-ferrous metal smelting and rolling processing industry (I7) and metal products industry (I8) are taken as related industries. For the ferrous metal smelting and rolling processing industry, the mining industry (I1), non-metallic mineral products industry (I5) and metal products industry (I8) are taken as related industries. The evaluation indicators of the load forecast results of the three typical industries are shown in Table 8.

[0155] Table 8 Evaluation indicators of typical industry forecast results

[0156]

[0157] According to Table 8, when only the impact of the industry's own historical development is considered, that is, in the single industry forecast, the MAPE of the three typical industries is 2.26%, 1.47%, and 2.92%, respectively. Using the method disclosed in this embodiment, after introducing the associated industries screened out by VECM variance decomposition, that is, multi-industry forecast, the MAE of the forecast results of the three typical industries is reduced by 80%, 45%, and 27%, respectively, the RMSE is reduced by 78%, 45%, and 25%, respectively, and the MAPE is reduced to 0.47%, 0.83%, and 2.16%, respectively. Figure 4The load forecast result diagram for the chemical raw materials and chemical products manufacturing industry is shown in FIG. It can be seen that when the method disclosed in this embodiment is used to forecast the load of a typical industry, the load forecast accuracy can be effectively improved. Therefore, the power load forecasting method considering the industry coupling correlation characteristics disclosed in this embodiment can reliably and effectively study the mutual influence between industries and support accurate industry load forecasting.

[0158] Example 2

[0159] In this embodiment, a power load forecasting system considering industry coupling correlation characteristics is disclosed, including:

[0160] Data acquisition module, used to obtain power load data of multiple industries;

[0161] The disturbance quantity acquisition module is used to determine the disturbance quantity of the power load data between industries based on the acquired power load data of all industries and the data disturbance analysis model; wherein the data disturbance analysis model is constructed by using a vector error correction model;

[0162] The variance decomposition module is used to perform variance decomposition on the power load data of each industry according to the disturbance amount of the power load data between industries, and obtain the total disturbance variance of the power load data of each industry affected by all other industries and the disturbance variance of the power load data of each industry affected by other industries;

[0163] The coupling correlation feature module is used to calculate, for each industry, the proportion of the disturbance variance of the power load data of the industry affected by other industries to the total disturbance variance of the power load data of the industry affected by all other industries, and obtain the coupling correlation characteristics between the industries;

[0164] A strongly related industry determination module is used to determine the related industries of each industry based on the coupling correlation characteristics between industries;

[0165] The power load forecasting unit is used to forecast the power load for each industry by using the related industries of the industry.

[0166] Furthermore, the variance decomposition module is used to determine the degree of influence of the white noise vector on the power load data of each industry according to the disturbance amount of the power load data between each industry;

[0167] According to the influence of the white noise vector on the power load data of each industry, the total disturbance variance of all other industries on the power load data of the target industry and the disturbance variance of each other industry on the power load data of the target industry are determined.

[0168] Furthermore, the construction process of the data disturbance analysis model in the disturbance quantity acquisition module includes:

[0169] The power load data of each industry is taken as a separate time series data, and a vector error correction model between the power load data of multiple industries is constructed;

[0170] Convert the constructed vector error correction model into a vector autoregressive model of the load data of each industry;

[0171] The vector autoregression model of the load data of each industry is transformed to obtain the data disturbance analysis model.

[0172] The present invention also discloses a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the power load forecasting method considering industry coupling association characteristics disclosed in Example 1.

[0173] The method disclosed in Example 1 can be directly embodied as a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0174] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0175] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A power load forecasting method considering industry coupling correlation characteristics, characterized in that: include: Obtain power load data for multiple industries; According to the acquired power load data of all industries and the data disturbance analysis model, the disturbance amount of the power load data among various industries is determined; wherein the data disturbance analysis model is constructed by using a vector error correction model; According to the disturbance amount of power load data among industries, variance decomposition is performed on the power load data of each industry to obtain the total disturbance variance of each industry's power load data affected by all other industries and the disturbance variance of each industry's power load data affected by other industries; For each industry, calculate the proportion of the disturbance variance of the power load data of the industry affected by other industries to the total disturbance variance of the power load data of the industry affected by all other industries, and obtain the coupling correlation characteristics between the industries; According to the coupling characteristics between industries, determine the related industries of each industry; For each industry, the power load forecast is performed using the industry's associated industries.

2. The method for predicting power load considering industry coupling correlation characteristics according to claim 1, characterized in that: According to the disturbance amount of power load data among various industries, the influence degree of white noise vector on power load data of various industries is determined; According to the influence of the white noise vector on the power load data of each industry, the total disturbance variance of each industry's power load data affected by all other industries and the disturbance variance of each industry's power load data affected by other industries are determined.

3. The method for predicting power load considering industry coupling correlation characteristics as claimed in claim 1, characterized in that: The power load data of each industry is taken as a separate time series data, and a vector error correction model between the power load data of multiple industries is constructed; Convert the constructed vector error correction model into a vector autoregressive model of the load data of each industry; The vector autoregression model of the load data of each industry is transformed to obtain the data disturbance analysis model.

4. The method for predicting power load considering industry coupling correlation characteristics according to claim 1, characterized in that: The power load data obtained for each industry are data that have passed the stationarity test, cointegration test and Granger causality test.

5. The method for predicting power load considering industry coupling correlation characteristics as claimed in claim 1, characterized in that: For each industry, the process of determining the industry's associated industries is: Arrange all the coupling-related characteristics of the industry in descending order, and take the first few industries whose sum of coupling-related characteristics is greater than the set threshold as the related industries of the industry.

6. The method for predicting power load considering industry coupling correlation characteristics according to claim 1, characterized in that: For each industry, the power load forecasting model is used to forecast the power load; wherein, the power load forecasting model is obtained by constructing a long- and short-time network.

7. The power load forecasting system considering the industry coupling correlation characteristics is characterized by: include: Data acquisition module, used to obtain power load data of multiple industries; The disturbance quantity acquisition module is used to determine the disturbance quantity of the power load data between industries based on the acquired power load data of all industries and the data disturbance analysis model; wherein the data disturbance analysis model is constructed by using a vector error correction model; The variance decomposition module is used to perform variance decomposition on the power load data of each industry according to the disturbance amount of the power load data between industries, and obtain the total disturbance variance of the power load data of each industry affected by all other industries and the disturbance variance of the power load data of each industry affected by other industries; The coupling correlation feature module is used to calculate, for each industry, the proportion of the disturbance variance of the power load data of the industry affected by other industries to the total disturbance variance of the power load data of the industry affected by all other industries, and obtain the coupling correlation characteristics between the industries; A strongly related industry determination module is used to determine the related industries of each industry based on the coupling correlation characteristics between industries; The power load forecasting unit is used to forecast the power load for each industry by using the related industries of the industry.

8. The power load forecasting system considering industry coupling correlation characteristics as claimed in claim 7, characterized in that: The variance decomposition module is used to determine the influence of the white noise vector on the power load data of each industry according to the disturbance amount of the power load data between each industry; According to the influence of the white noise vector on the power load data of each industry, the total disturbance variance of all other industries on the power load data of the target industry and the disturbance variance of each other industry on the power load data of the target industry are determined.

9. The power load forecasting system considering industry coupling correlation characteristics as claimed in claim 7, characterized in that: The construction process of the data disturbance analysis model in the disturbance acquisition module includes: The power load data of each industry is taken as a separate time series data, and a vector error correction model between the power load data of multiple industries is constructed; Convert the constructed vector error correction model into a vector autoregressive model of the load data of each industry; The vector autoregression model of the load data of each industry is transformed to obtain the data disturbance analysis model.

10. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, it implements the power load forecasting method considering industry coupling association characteristics as described in any one of claims 1-6.

Images