Power grid load prediction method

Abstract

The invention relates to load prediction, in particular to a power grid load prediction method, which comprises the following steps of: decomposing a power grid load prediction problem into different levels, constructing a judgment matrix, and calculating a preliminary weight of each factor based on the judgment matrix; calculating a variation coefficient of each factor, and adjusting the initial weight of each factor by using the variation coefficient to obtain a weight coefficient of each factor; constructing a grey incidence matrix of each factor according to the historical data of the power grid load, and calculating the correlation degree between each factor and the power grid load based on the grey incidence matrix of each factor; combining the weight coefficient and the correlation degree of each factor to construct a fuzzy relation matrix of each factor, and obtaining an electric quantity prediction matrix based on the fuzzy relation matrix of each factor; inputting the electric quantity prediction matrix into a pre-trained power grid load prediction model to obtain a first power grid load prediction result; according to the technical scheme provided by the invention, the defect that the power grid load is difficult to accurately predict in the prior art can be effectively overcome.

Description

Technical Field

[0001] The present invention relates to load forecasting, and more particularly to a power grid load forecasting method. Background Art

[0002] The overall power grid composed of substations and transmission and distribution lines at various voltages in the power system is called the power grid, which includes three components: transformation, transmission, and distribution. At present, traditional power grid load forecasting methods generally train and verify a neural network model for forecasting through historical load data, and use the non-linear characteristics of the neural network model to output the power grid load forecasting result.

[0003] However, there are many and complex external influencing factors in power grid load forecasting, and not all factors can play a role in load forecasting. Relying solely on the neural network model cannot fully consider the categories of influencing factors and the importance of their influence on load forecasting, resulting in the inability to guarantee the forecasting accuracy. Summary of the Invention

[0004] In view of the above-mentioned drawbacks of the prior art, the present invention provides a power grid load forecasting method, which can effectively overcome the defect of the prior art that it is difficult to accurately forecast the power grid load.

[0005] To achieve the above object, the present invention is realized through the following technical solutions: A power grid load forecasting method includes the following steps: S1. Decompose the power grid load forecasting problem into different levels, construct a judgment matrix, and calculate the preliminary weights of each factor based on the judgment matrix; S2. Calculate the coefficient of variation of each factor, and use the coefficient of variation to adjust the preliminary weights of each factor to obtain the weight coefficients of each factor; S3. Construct a grey correlation matrix of each factor according to the historical data of the power grid load, and calculate the correlation degree between each factor and the power grid load based on the grey correlation matrix of each factor; S4. Combine the weight coefficients and correlation degrees of each factor to construct a fuzzy relation matrix of each factor, and obtain a power quantity forecasting matrix based on the fuzzy relation matrix of each factor; S5. Input the power quantity forecasting matrix into a pre-trained power grid load forecasting model to obtain a first power grid load forecasting result; S6. Define the type-2 fuzzy sets of the input variables and output variables according to the historical data of the power grid load, set the membership function for the type-2 fuzzy sets, and formulate fuzzy rules; S7. Fuzzify each factor based on the membership function, combine the fuzzified factors and the fuzzy rules, and perform fuzzy reasoning using the type-2 fuzzy inference mechanism to obtain the type-2 fuzzy set corresponding to the power grid load; S8. Defuzzify the type-2 fuzzy set corresponding to the grid load to obtain the second grid load prediction result; S9. Conduct comprehensive analysis on the first grid load prediction result and the second grid load prediction result to obtain the final grid load prediction result.

[0006] Preferably, in S1, the grid load prediction problem is decomposed into different levels, and a judgment matrix is constructed. Based on the judgment matrix, the preliminary weights of each factor are calculated, including: S11. Decompose the grid load prediction problem into an objective layer, a criterion layer, and an index layer. The objective layer is the grid load prediction result, the criterion layer is the main factors affecting the grid load, and the index layer is the historical data of each factor; S12. Conduct pairwise comparison of each factor in the criterion layer to construct a judgment matrix; S13. Calculate the eigenvector of the judgment matrix and perform normalization processing on the eigenvector to obtain the preliminary weights of each factor; Among them, the elements in the judgment matrix represent the relative importance between each factor.

[0007] Preferably, in S13, calculate the eigenvector of the judgment matrix and perform normalization processing on the eigenvector to obtain the preliminary weights of each factor, including: S131. Perform normalization processing on each column of the judgment matrix, and add the normalized judgment matrix row by row to obtain the eigenvector of the judgment matrix; S132. Divide each element in the eigenvector by the sum of all elements to obtain the normalized eigenvector; Among them, the elements in the normalized eigenvector represent the preliminary weights of each factor.

[0008] Preferably, in S2, calculate the coefficient of variation of each factor, and use the coefficient of variation to adjust the preliminary weights of each factor to obtain the weight coefficients of each factor, including: S21. For each factor, obtain the corresponding historical data in the index layer, and calculate the average value and standard deviation of the data; S22. Divide the standard deviation by the average value to calculate the coefficient of variation for measuring the relative dispersion degree of each factor's data; S23. Divide the coefficient of variation of each factor by the sum of all coefficients of variation to obtain the coefficient of variation weights of each factor based on the degree of data change; S24. Perform weighted summation on the coefficient of variation weights and preliminary weights of each factor to obtain the fusion weights of each factor; S25. Perform normalization processing on the fusion weights of each factor to obtain the weight coefficients of each factor.

[0009] Preferably, in S3, a grey correlation matrix of each factor is constructed based on the historical data of the grid load, and the correlation degree between each factor and the grid load is calculated based on the grey correlation matrix of each factor, including: S31. For each factor, obtain the corresponding historical data in the index layer, and perform feature extraction to obtain the historical data features; S32. Perform intrinsic correlation processing on the historical data features of each factor to obtain the grey correlation matrix of each factor; S33. Calculate the correlation degree between each factor and the grid load based on the grey correlation matrix of each factor; Among them, the greater the correlation degree, the greater the influence of the factor on the grid load.

[0010] Preferably, in S4, a fuzzy relation matrix of each factor is constructed by combining the weight coefficient and the correlation degree of each factor, and an electricity quantity prediction matrix is obtained based on the fuzzy relation matrix of each factor, including: S41. Combine the weight coefficient and the correlation degree of each factor to construct a correlation matrix between each factor and the grid load; S42. Fuzzify the correlation matrix of each factor to obtain the fuzzy correlation matrix of each factor; S43. Perform multiple linear regression on the fuzzy correlation matrix of each factor using the multiple linear regression algorithm to obtain the prediction matrix of each factor; S44. Perform matrix fitting on the prediction matrix of each factor using the quadratic function fitting algorithm to obtain the electricity quantity prediction matrix.

[0011] Preferably, in S5, the electricity quantity prediction matrix is input into a pre-trained grid load prediction model to obtain the first grid load prediction result, including: S51. Select a suitable grey prediction model as the grid load prediction model according to the electricity quantity prediction matrix; S52. Screen and construct a historical data set from the historical data of the grid load, and divide the historical data set into a training set, a validation set, and a test set according to a preset ratio; S53. Set the loss function and optimizer of the grid load prediction model; S54. Input the training set into the grid load prediction model for model training; S55. Calculate the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information; S56. If the loss value is less than the preset threshold, the model training ends, and the current grid load prediction model is the trained grid load prediction model; otherwise, return to S54 and continue to use the training set for model training; S57. Input the validation set into the trained grid load prediction model, evaluate the generalization ability of the model by observing the performance on the validation set, and tune the hyperparameters and structure of the model; S58. Input the test set into the optimized power grid load forecasting model for model performance evaluation to obtain a pre-trained power grid load forecasting model; S59. Input the power consumption forecasting matrix into the pre-trained power grid load forecasting model to obtain the first power grid load forecasting result.

[0012] Preferably, in S6, define the type-2 fuzzy sets of the input variables and output variables according to the historical power grid load data, set the membership functions for the type-2 fuzzy sets, and formulate fuzzy rules, including: S61. According to the historical power grid load data and domain knowledge, screen the main factors affecting the power grid load as input variables, regard the power grid load as the output variable, and define the type-2 fuzzy sets of the input variables and output variables; S62. Set appropriate membership functions for each type-2 fuzzy set to determine the degree to which the variable belongs to each type-2 fuzzy set; S63. According to the historical power grid load data and domain knowledge, formulate a series of fuzzy rules to describe the fuzzy relationship between the input variables and the output variable; Among them, the fuzzy rules are formulated in the form of "if - then".

[0013] Preferably, in S62, set appropriate membership functions for each type-2 fuzzy set to determine the degree to which the variable belongs to each type-2 fuzzy set, including: S621. According to the historical power grid load data and domain knowledge, determine the ranges of the input variables and output variables; S622. According to the variable characteristics and prediction requirements, set appropriate membership functions for each type-2 fuzzy set; S623. According to the historical power grid load data and domain knowledge, determine the parameters in each membership function, and at the same time adjust the shape and parameters of each membership function considering the fuzziness of the variable; Among them, the membership functions of the type-2 fuzzy sets include: 1) Linear membership function: ; Among them, f ( x ) is the membership degree of the variable x , a’ , b’ are constants; 2) Triangular membership function: ; Among them, f ( x ) is the membership degree of the variable x , a , b ,c are the three vertices of a triangle; 3) Gaussian membership function: ; wherein, f ( x ) is the membership degree of variable x , is the mean value, is the standard deviation.

[0014] Preferably, in S7, each factor is fuzzified based on the membership function. Combining the fuzzified factors and fuzzy rules, a type-2 fuzzy inference mechanism is used for fuzzy inference to obtain a type-2 fuzzy set corresponding to the grid load, including: S71. Calculate the membership degree of a certain factor belonging to each type-2 fuzzy set based on the membership function, and take the type-2 fuzzy set with the largest membership degree as the type-2 fuzzy set corresponding to the factor; S72. Repeat S71 until the type-2 fuzzy sets corresponding to all factors are obtained, completing the fuzzification of each factor; S73. Combining the type-2 fuzzy sets corresponding to all factors and fuzzy rules, a type-2 fuzzy inference mechanism is used for fuzzy inference to obtain a type-2 fuzzy set corresponding to the grid load to describe the grid load prediction situation.

[0015] Compared with the prior art, a grid load prediction method provided by the present invention has the following beneficial effects: 1) Decompose the grid load prediction problem into different levels, construct a judgment matrix, calculate the preliminary weights reflecting the relative importance between factors based on the judgment matrix, then calculate the coefficient of variation of each factor, and use the coefficient of variation to adjust the preliminary weights of each factor, so that the relative importance between factors can be corrected by the coefficient of variation reflecting the data dispersion degree of each factor, and the obtained weight coefficients of each factor can better reflect the relative importance between factors; 2) Construct a grey correlation matrix of each factor according to the historical data of the grid load, calculate the correlation degree reflecting the influence of each factor on the grid load, that is, the absolute importance of each factor, and construct a fuzzy relation matrix of each factor by combining the weight coefficients of each factor (reflecting the relative importance between factors) and the correlation degree (reflecting the absolute importance of each factor), ensuring that the factors involved in the fuzzy relation matrix are all elements that have a substantial impact on the load prediction, so that the power prediction matrix obtained based on the fuzzy relation matrix of each factor is more accurate. Inputting the power prediction matrix into a pre-trained grid load prediction model, a relatively accurate first grid load prediction result can be obtained; 3) Define the type-2 fuzzy sets of the input variables and output variables according to the historical data of the grid load, set the membership functions for the type-2 fuzzy sets, and formulate fuzzy rules. Fuzzify each factor based on the membership functions. Combine the fuzzified factors and the fuzzy rules, and use the type-2 fuzzy inference mechanism to perform fuzzy inference to obtain the type-2 fuzzy set corresponding to the grid load. Defuzzify the type-2 fuzzy set corresponding to the grid load. Through the adjustable membership function, it can better handle the influencing factors of the grid load with fuzziness and uncertainty, and is applicable to power systems with severe nonlinearity and random interference at the same time. It can use fuzzy rules and fuzzy inference to capture the nonlinear characteristics of the grid load changes, so as to realize the accurate prediction of the grid load, obtain the second grid load prediction result, and through the comprehensive analysis of the first grid load prediction result and the second grid load prediction result, the prediction accuracy of the finally obtained grid load prediction result is fully guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 is a schematic flowchart of the present invention; Figure 2 is a schematic flowchart of obtaining the second grid load prediction result by using the type-2 fuzzy inference mechanism in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0019] A method for predicting grid load, as Figure 1 shown, S1. Decompose the grid load prediction problem into different levels, construct a judgment matrix, and calculate the preliminary weights of each factor based on the judgment matrix. Specifically, it includes: S11. Decompose the grid load prediction problem into an objective layer, a criterion layer, and an index layer. The objective layer is the grid load prediction result, the criterion layer is the main factors affecting the grid load, and the index layer is the historical data of each factor; S12. Pairwise compare each factor in the criterion layer to construct a judgment matrix; S13. Calculate the eigenvector of the judgment matrix and perform normalization processing on the eigenvector to obtain the preliminary weights of each factor; Among them, the elements in the judgment matrix represent the relative importance between each factor.

[0020] Specifically, in S13, calculating the eigenvector of the judgment matrix and performing normalization processing on the eigenvector to obtain the preliminary weights of each factor includes: S131. Perform normalization processing on each column of the judgment matrix, and add the normalized judgment matrix row by row to obtain the eigenvector of the judgment matrix; S132. Divide each element in the eigenvector by the sum of all elements to obtain the normalized eigenvector; Among them, the elements in the normalized eigenvector represent the preliminary weights of each factor.

[0021] S2. Calculate the coefficient of variation of each factor, and use the coefficient of variation to adjust the preliminary weights of each factor to obtain the weight coefficients of each factor, specifically including: S21. For each factor, obtain the corresponding historical data in the index layer and calculate the average value and standard deviation of the data; S22. Divide the standard deviation by the average value to calculate the coefficient of variation that measures the relative dispersion degree of the data of each factor; S23. Divide the coefficient of variation of each factor by the sum of all coefficients of variation to obtain the coefficient of variation weights of each factor based on the degree of data change; S24. Perform weighted summation on the coefficient of variation weights and preliminary weights of each factor to obtain the fusion weights of each factor; S25. Perform normalization processing on the fusion weights of each factor to obtain the weight coefficients of each factor.

[0022] In the above technical solution, the power grid load forecasting problem is decomposed into different levels, and a judgment matrix is constructed. Based on the judgment matrix, the preliminary weights reflecting the relative importance between each factor are calculated. Then, the coefficient of variation of each factor is calculated, and the preliminary weights of each factor are adjusted using the coefficient of variation. Thus, the relative importance between each factor can be corrected using the coefficient of variation reflecting the data dispersion degree of each factor, so that the obtained weight coefficients of each factor can better reflect the relative importance between each factor.

[0023] As Figure 1 shown, S3. Construct a grey relational matrix of each factor based on the historical data of the power grid load, and calculate the correlation degree between each factor and the power grid load based on the grey relational matrix of each factor, specifically including: S31. For each factor, obtain the corresponding historical data in the index layer, and perform feature extraction to obtain the historical data features; S32. Perform intrinsic correlation processing on the historical data features of each factor to obtain the grey correlation matrix of each factor; S33. Calculate the correlation degree between each factor and the grid load based on the grey correlation matrix of each factor; Among them, the greater the correlation degree indicates that the factor has a greater impact on the grid load.

[0024] S4. Combine the weight coefficients and correlation degrees of each factor to construct the fuzzy relation matrix of each factor, and obtain the power prediction matrix based on the fuzzy relation matrix of each factor, specifically including: S41. Combine the weight coefficients and correlation degrees of each factor to construct the correlation matrix between each factor and the grid load; S42. Fuzzify the correlation matrix of each factor to obtain the fuzzy correlation matrix of each factor; S43. Use the multiple linear regression algorithm to perform multiple linear regression on the fuzzy correlation matrix of each factor to obtain the prediction matrix of each factor; S44. Use the quadratic function fitting algorithm to perform matrix fitting on the prediction matrix of each factor to obtain the power prediction matrix.

[0025] S5. Input the power prediction matrix into the pre-trained grid load prediction model to obtain the first grid load prediction result, specifically including: S51. Select a suitable grey prediction model as the grid load prediction model according to the power prediction matrix; S52. Screen and construct a historical data set from the grid load historical data, and divide the historical data set into a training set, a validation set, and a test set according to a preset ratio; S53. Set the loss function and optimizer of the grid load prediction model; S54. Input the training set into the grid load prediction model for model training; S55. Calculate the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information; S56. If the loss value is less than the preset threshold, the model training ends, and the current grid load prediction model is the trained grid load prediction model, otherwise return to S54 and continue to use the training set for model training; S57. Input the validation set into the trained grid load prediction model, evaluate the generalization ability of the model by observing the performance on the validation set, and tune the hyperparameters and structure of the model; S58. Input the test set into the tuned grid load prediction model to perform model performance evaluation to obtain the pre-trained grid load prediction model; S59. Input the power consumption prediction matrix into the pre-trained power grid load prediction model to obtain the first power grid load prediction result.

[0026] In the above technical solution, the grey relational matrix of each factor is constructed based on the historical data of the power grid load. The correlation degree reflecting the influence of each factor on the power grid load, that is, the absolute importance of each factor, is calculated based on the grey relational matrix of each factor. The fuzzy relation matrix of each factor is constructed by combining the weight coefficients of each factor (reflecting the relative importance between factors) and the correlation degree (reflecting the absolute importance of each factor), ensuring that the factors involved in the fuzzy relation matrix are all elements that have a substantial impact on load prediction. Thus, the power consumption prediction matrix obtained based on the fuzzy relation matrix of each factor is more accurate. Inputting the power consumption prediction matrix into the pre-trained power grid load prediction model can obtain a relatively accurate first power grid load prediction result.

[0027] As Figure 1 and Figure 2 shown, S6. Define the type-2 fuzzy sets of the input variables and output variables according to the historical data of the power grid load, set the membership functions for the type-2 fuzzy sets, and formulate fuzzy rules, specifically including: S61. According to the historical data of the power grid load and domain knowledge, screen the main factors affecting the power grid load as input variables, regard the power grid load as the output variable, and define the type-2 fuzzy sets of the input variables and output variables; S62. Set appropriate membership functions for each type-2 fuzzy set to determine the degree to which the variable belongs to each type-2 fuzzy set; S63. According to the historical data of the power grid load and domain knowledge, formulate a series of fuzzy rules to describe the fuzzy relationship between the input variables and the output variable; Among them, the fuzzy rules are formulated in the form of "if - then".

[0028] Specifically, in S62, setting appropriate membership functions for each type-2 fuzzy set to determine the degree to which the variable belongs to each type-2 fuzzy set includes: S621. According to the historical data of the power grid load and domain knowledge, determine the ranges of the input variables and output variables; S622. According to the variable characteristics and prediction requirements, set appropriate membership functions for each type-2 fuzzy set; S623. According to the historical data of the power grid load and domain knowledge, determine the parameters in each membership function, and at the same time adjust the shape and parameters of each membership function considering the fuzziness of the variable; Among them, the membership functions of the type-2 fuzzy sets include: 1) Linear membership function: ; Among them,f ( x ) is the membership degree of the variable x , a’ , b’ are constants; 2) Triangular membership function: ; Among them, f ( x ) is the membership degree of the variable x , a , b , c are the three vertices of the triangle; 3) Gaussian membership function: ; Among them, f ( x ) is the membership degree of the variable x , is the mean value, is the standard deviation.

[0029] S7. Fuzzify each factor based on the membership function, combine the fuzzified factors and fuzzy rules, and use the type-2 fuzzy inference mechanism to perform fuzzy inference to obtain the type-2 fuzzy set corresponding to the grid load, specifically including: S71. Calculate the membership degree of a certain factor belonging to each type-2 fuzzy set based on the membership function, and take the type-2 fuzzy set with the largest membership degree as the type-2 fuzzy set corresponding to this factor; S72. Repeat S71 until the type-2 fuzzy sets corresponding to all factors are obtained, and complete the fuzzification of each factor; S73. Combine the type-2 fuzzy sets corresponding to all factors and the fuzzy rules, and use the type-2 fuzzy inference mechanism to perform fuzzy inference to obtain the type-2 fuzzy set corresponding to the grid load to describe the grid load prediction situation.

[0030] S8. Defuzzify the type-2 fuzzy set corresponding to the grid load to obtain the second grid load prediction result.

[0031] S9. Conduct a comprehensive analysis of the first grid load prediction result and the second grid load prediction result to obtain the final grid load prediction result.

[0032] In the above technical solution, type-2 fuzzy sets of input variables and output variables are defined according to the historical data of the power grid load, membership functions are set for the type-2 fuzzy sets, and fuzzy rules are formulated. Based on the membership functions, each factor is fuzzified. Combining the fuzzified factors and the fuzzy rules, a type-2 fuzzy inference mechanism is used for fuzzy inference to obtain the type-2 fuzzy set corresponding to the power grid load. The type-2 fuzzy set corresponding to the power grid load is defuzzified. Through the adjustable membership function, it is possible to better handle the influencing factors of the power grid load with fuzziness and uncertainty, and it is applicable to power systems with severe nonlinearity and random interference at the same time. It can use fuzzy rules and fuzzy inference to capture the nonlinear characteristics of the power grid load change, so as to achieve accurate prediction of the power grid load, obtain the second power grid load prediction result, and through comprehensive analysis of the first power grid load prediction result and the second power grid load prediction result, the prediction accuracy of the finally obtained power grid load prediction result is fully guaranteed.

[0033] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting power grid load, characterized in that: The following steps are involved: S1. Decompose the power grid load forecasting problem into different levels, construct a judgment matrix, and calculate the preliminary weight of each factor based on the judgment matrix; S2. Calculate the coefficient of variation of each factor, and use the coefficient of variation to adjust the initial weight of each factor to obtain the weight coefficient of each factor; S3. constructing a grey correlation matrix of each factor according to the historical data of power grid load, and calculating the correlation between each factor and power grid load based on the grey correlation matrix of each factor; S4. Construct a fuzzy relationship matrix of each factor by combining the weight coefficient and correlation degree of each factor, and obtain an electricity prediction matrix based on the fuzzy relationship matrix of each factor; S5. Inputting the power prediction matrix into a pre-trained power grid load prediction model to obtain a first power grid load prediction result; S6. Define type 2 fuzzy sets of input variables and output variables according to historical data of power grid load, set membership functions for type 2 fuzzy sets, and formulate fuzzy rules; S7, fuzzifying each factor based on the membership function, combining the fuzzified factors and fuzzy rules, and using the type 2 fuzzy reasoning mechanism to perform fuzzy reasoning to obtain the type 2 fuzzy set corresponding to the power grid load; S8, defuzzifying the type 2 fuzzy set corresponding to the power grid load to obtain a second power grid load forecast result; S9. Comprehensively analyze the first power grid load forecast result and the second power grid load forecast result to obtain a final power grid load forecast result.

2. The power grid load forecasting method according to claim 1, characterized in that: In S1, the grid load forecasting problem is decomposed into different levels, and a judgment matrix is ​​constructed. The preliminary weights of each factor are calculated based on the judgment matrix, including: S11, decomposing the power grid load forecasting problem into a target layer, a criterion layer and an indicator layer, wherein the target layer is the power grid load forecasting result, the criterion layer is the main factors affecting the power grid load, and the indicator layer is the historical data of each factor; S12, compare the factors in the criterion layer in pairs and construct a judgment matrix; S13, calculating the eigenvector of the judgment matrix, and normalizing the eigenvector to obtain the preliminary weight of each factor; Among them, the elements in the judgment matrix represent the relative importance of each factor.

3. The power grid load forecasting method according to claim 2, characterized in that: In S13, the eigenvector of the judgment matrix is ​​calculated and normalized to obtain the preliminary weight of each factor, including: S131, normalizing each column of the judgment matrix, and adding the normalized judgment matrix row by row to obtain a eigenvector of the judgment matrix; S132, dividing each element in the feature vector by the sum of all elements to obtain a normalized feature vector; Among them, the elements in the normalized feature vector represent the preliminary weights of each factor.

4. The power grid load forecasting method according to claim 2, characterized in that: In S2, the coefficient of variation of each factor is calculated, and the initial weight of each factor is adjusted using the coefficient of variation to obtain the weight coefficient of each factor, including: S21. For each factor, obtain the corresponding historical data in the indicator layer, and calculate the mean and standard deviation of the data; S22. Divide the standard deviation by the mean to calculate the coefficient of variation, which measures the relative dispersion of the data for each factor; S23, the coefficient of variation of each factor is added to the sum of all coefficients of variation to obtain the weight of the coefficient of variation of each factor based on the degree of data variation; S24, performing weighted summation on the coefficient of variation weight and the preliminary weight of each factor to obtain the fusion weight of each factor; S25. Normalize the fusion weights of each factor to obtain a weight coefficient of each factor.

5. The power grid load forecasting method according to claim 4, characterized in that: In S3, the grey correlation matrix of each factor is constructed according to the historical data of power grid load, and the correlation between each factor and power grid load is calculated based on the grey correlation matrix of each factor, including: S31. For each factor, obtain the corresponding historical data in the indicator layer, and perform feature extraction to obtain the historical data features; S32, performing intrinsic correlation processing on the historical data characteristics of each factor to obtain the grey correlation matrix of each factor; S33, calculating the correlation between each factor and the power grid load based on the grey correlation matrix of each factor; Among them, the greater the correlation, the greater the impact of this factor on the grid load.

6. The power grid load forecasting method according to claim 5, characterized in that: In S4, the fuzzy relationship matrix of each factor is constructed by combining the weight coefficient and correlation degree of each factor. Based on the fuzzy relationship matrix of each factor, the power forecast matrix is ​​obtained, including: S41, combining the weight coefficient and correlation degree of each factor, constructing a correlation matrix between each factor and the power grid load; S42, fuzzifying the correlation matrix of each factor to obtain a fuzzy correlation matrix of each factor; S43, using a multiple linear regression algorithm to perform multiple linear regression on the fuzzy correlation matrix of each factor to obtain a prediction matrix of each factor; S44. Use a quadratic function fitting algorithm to perform matrix fitting on the prediction matrix of each factor to obtain a power prediction matrix.

7. The power grid load forecasting method according to claim 6, characterized in that: In S5, the power prediction matrix is ​​input into the pre-trained power grid load prediction model to obtain a first power grid load prediction result, including: S51, selecting a suitable grey prediction model as a power grid load prediction model according to the power prediction matrix; S52, selecting and constructing a historical data set from the historical data of power grid load, and dividing the historical data set into a training set, a validation set, and a test set according to a preset ratio; S53, setting a loss function and an optimizer of a power grid load forecasting model; S54, inputting the training set into the power grid load forecasting model for model training; S55, calculating the loss value based on the loss function, and the optimizer updates the model parameters according to the loss value and the network gradient information; S56. If the loss value is less than the preset threshold, the model training ends, and the current power grid load prediction model is the trained power grid load prediction model. Otherwise, the process returns to S54 and continues to use the training set for model training. S57, input the validation set into the trained power grid load forecasting model, evaluate the generalization ability of the model by observing the performance on the validation set, and tune the hyperparameters and structure of the model; S58, inputting the test set into the optimized power grid load forecasting model, performing model performance evaluation, and obtaining a pre-trained power grid load forecasting model; S59: Input the power prediction matrix into a pre-trained power grid load prediction model to obtain a first power grid load prediction result.

8. The power grid load forecasting method according to claim 1, characterized in that: In S6, type 2 fuzzy sets of input variables and output variables are defined according to the historical data of power grid load, membership functions are set for type 2 fuzzy sets, and fuzzy rules are formulated, including: S61. Based on the historical data of power grid load and domain knowledge, the main factors affecting the power grid load are selected as input variables, the power grid load is regarded as the output variable, and the type 2 fuzzy sets of the input variables and the output variables are defined; S62, setting a suitable membership function for each type 2 fuzzy set to determine the degree to which the variable belongs to each type 2 fuzzy set; S63. Based on the historical data of power grid load and domain knowledge, a series of fuzzy rules are formulated to describe the fuzzy relationship between input variables and output variables; Among them, fuzzy rules are formulated in the form of "if-then".

9. The power grid load forecasting method according to claim 8, characterized in that: In S62, a suitable membership function is set for each type 2 fuzzy set to determine the degree to which the variable belongs to each type 2 fuzzy set, including: S621. Determine the range of input variables and output variables based on historical grid load data and domain knowledge; S622, setting a suitable membership function for each type 2 fuzzy set according to the variable characteristics and prediction requirements; S623, determining the parameters in each membership function according to the historical data of the power grid load and the domain knowledge, and adjusting the shape and parameters of each membership function by considering the fuzziness of the variables; Among them, the membership functions of type 2 fuzzy sets include: 1) Linear membership function: ； in, f ( x ) is a variable x The membership degree of a’ , b’ is a constant; 2) Triangle membership function: ； in, f ( x ) is a variable x The membership degree of a , b , c are the three vertices of the triangle; 3) Gaussian membership function: ； in, f ( x ) is a variable x The membership degree of is the mean, is the standard deviation.

10. The power grid load forecasting method according to claim 8, characterized in that: In S7, each factor is fuzzified based on the membership function. The fuzzified factors and fuzzy rules are combined and fuzzy reasoning is performed using the type 2 fuzzy reasoning mechanism to obtain the type 2 fuzzy set corresponding to the power grid load, including: S71, calculating the membership of a factor to each type-2 fuzzy set based on the membership function, and taking the type-2 fuzzy set with the largest membership as the type-2 fuzzy set corresponding to the factor; S72, repeat S71 until the type 2 fuzzy sets corresponding to all factors are obtained, and the fuzzification of each factor is completed; S73. Combining the type-2 fuzzy sets and fuzzy rules corresponding to all factors, using the type-2 fuzzy reasoning mechanism to perform fuzzy reasoning, obtain the type-2 fuzzy set corresponding to the power grid load to describe the power grid load forecasting situation.

Images