An Industry Power Situation Awareness Method and System Based on Multi-Model Fusion

Abstract

The present invention relates to the field of electric power, and provides an industry electric power situation awareness method and system based on multi-model fusion, including the steps of: respectively predicting the future electricity sales volume by using a first base model, a second base model, and a third base model; using the output results predicted by the three base models and the historical electricity sales volume as the features of the fusion model to perform multi-model fusion prediction; in the feature matrix of the fusion model, the first, second, and third column features are the output result data predicted by the first, second, and third base models respectively, and the fourth column feature is the historical electricity sales volume data; shifting the indexes after standardizing each column feature in the feature matrix; obtaining the information utility values of each index and calculating the weights of each index; and performing weighted summation of the weights of each index, the output results predicted by the three base models, and the historical electricity sales volume to obtain the prediction result of the fusion model. The present invention improves the weighting algorithm of the fusion model, reduces the subjective intervention of artificial weighting, and makes the prediction of electricity sales volume more scientific.

Description

Technical Field

[0001] The present invention relates to the field of electric power, and particularly to an industry electric power situation awareness method and system based on multi-model fusion. Background Art

[0002] The electricity sales volume is an important economic indicator for power grid enterprises. Accurate prediction of the electricity sales volume can help make early economic plans for the enterprises, carry out work on power demand side management, power supply expansion, and power consumption management in a targeted manner, which has important practical significance for reasonably determining the total quota of electricity sales volume, decomposing the electricity sales volume indicators, and improving the operation of power enterprises. At the same time, the prediction of electricity sales volume is a basic task in the power market. Correctly predicting the level of regional electricity sales volume is of great significance for guiding the reasonable operation of power plants and transmission and distribution grids and promoting the development and construction of the power market.

[0003] Due to the influence of various uncertain factors on the prediction of electricity sales volume, traditional prediction methods can no longer achieve satisfactory prediction results. Summary of the Invention

[0004] In view of this, the present invention proposes an industry electric power situation awareness method and system based on multi-model fusion. In the process of multi-model fusion, an improved entropy method is introduced to assign weights to each basic model in the fusion model in real time, and the weight assignment algorithm of the fusion model is improved to reduce the subjective intervention of manual weight assignment and make the prediction of electricity sales volume more scientific.

[0005] The method of the present invention is implemented by the following technical solutions: An industry electric power situation awareness method based on multi-model fusion, comprising the following steps:

[0006] Feature selection, where the selected features include industry electricity sales volume, industry information factors, weather data, and resident consumption information;

[0007] Perform data extraction and cleaning on the selected features;

[0008] Respectively use the first basic model, the second basic model, and the third basic model to predict the future electricity sales volume;

[0009] Take the output results predicted by the first basic model, the second basic model, and the third basic model and the historical electricity sales volume as the features of the fusion model for multi-model fusion prediction;

[0010] In the feature matrix of the fusion model, the first column of features is the output result data predicted by the first base model, the second column of features is the output result data predicted by the second base model, the third column of features is the output result data predicted by the third base model, and the fourth column of features is the historical electricity sales data; the indicators after standardizing each column of features in the feature matrix are shifted to obtain the improved positive-processed indicators; then the information utility values of each indicator are obtained, and the weights of each indicator are calculated; the weights of each indicator are weighted and summed with the output results predicted by the first base model, the second base model, and the third base model and the historical electricity sales to obtain the prediction result of the fusion model.

[0011] Preferably, the method further includes the following steps:

[0012] Calculate the dynamic deviation: Calculate the average error of the prediction results of the fusion model in the past period of time, divide it into positive deviation and negative deviation, and dynamically adjust the current prediction result so that the current single-point prediction value is converted into an interval prediction value.

[0013] Preferably, the first base model is the support vector machine algorithm, the second base model is the XGBOOST algorithm, the third base model is the random forest algorithm; the historical electricity sales is the electricity sales of last month.

[0014] Preferably, the process of multi-model fusion prediction specifically includes the steps:

[0015] Perform standardization processing on the output results predicted by the first base model, the second base model, the third base model and the historical electricity sales to obtain a standardized matrix; shift the indicators with extreme values in the standardized matrix to obtain the improved positive-processed indicators; perform index normalization processing on the positive-processed indicators;

[0016] According to the normalized index data, calculate the information entropy of each index in the feature matrix of the fusion model; according to the information entropy of each index, calculate the information utility value of each index in the feature matrix of the fusion model;

[0017] According to the information utility values of each index, calculate the weights of each index;

[0018] Weight and sum the weights of each index with the output result data predicted by the first base model, the second base model, the third base model, and the historical electricity sales data to obtain the prediction result of the fusion model.

[0019] The perception system of the present invention is implemented by the following technical solution: A multi-model fusion industry power situation perception system includes the following modules:

[0020] A feature selection module, used for feature selection, and the selected features include industry electricity sales, industry information factors, weather data and resident consumption information;

[0021] A preprocessing module for data extraction and cleaning of the selected features;

[0022] A single base model prediction module for predicting future electricity sales using the first base model, the second base model, and the third base model respectively;

[0023] A fusion prediction module for using the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales as features of the fusion model for multi-model fusion prediction;

[0024] Among them, in the feature matrix of the fusion model, the first column of features is the output result data predicted by the first base model, the second column of features is the output result data predicted by the second base model, the third column of features is the output result data predicted by the third base model, and the fourth column of features is historical electricity sales data; translate the indicators after standardizing each column of features in the feature matrix to obtain the improved positive indicators; then calculate the information utility value of each indicator and calculate the weight of each indicator; perform weighted summation of the weights of each indicator and the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales to obtain the prediction result of the fusion model.

[0025] The present invention has the following advantages and effects compared with the prior art:

[0026] 1. The present invention is based on power data such as industry electricity sales and industry information factors, combines meteorological factors and economic factors such as temperature, weather data, GDP, and household consumption, and adopts an electricity sales prediction algorithm based on multi-model fusion learning; in this prediction algorithm, support vector machine, XGBOOST, and random forest prediction methods are selected as the basic prediction methods, and data fusion technology is used to fuse the prediction results obtained by the basic prediction methods to obtain the final prediction result; in the fusion process, an improved entropy weight method is introduced to perform real-time weighting on each basic model in the fusion model. It solves the deficiency of artificial weighting of the basic model or directly applying another machine learning model in the multi-model fusion process; the introduced information entropy improves the weighting algorithm of the fusion model, making the multi-model fusion prediction more objective and scientific.

[0027] 2. Further, in the model fusion process, an improved entropy weight method is introduced to limit the range of the entropy weight method or perform a translation operation to avoid the situation of 0 entropy value, thereby effectively improving the objectivity and accuracy of model prediction.

[0028] 3. In addition, based on the final prediction result of the fusion model, the concept of dynamic deviation is introduced, the average error of the prediction results in the past six months is calculated, and it is divided into positive deviation and negative deviation, and the current prediction result is dynamically adjusted, so that the current single-point prediction value is converted into an interval prediction value, improving the prediction accuracy. Description of the Drawings

[0029] Figure 1 It is a flowchart of the electricity sales prediction by the industry power situation perception method of multi-model fusion in the embodiment of the present invention;

[0030] Figure 2 It is a structural block diagram of the industry power situation perception system of multi-model fusion in the embodiment of the present invention. Detailed implementation manners

[0031] The present invention will be further described in detail below in conjunction with embodiments and the accompanying drawings, but the implementation manners of the present invention are not limited thereto.

[0032] Embodiment 1

[0033] Please refer to Figure 1 , this embodiment is an industry power situation perception method of multi-model fusion, which uses a fusion model to predict electricity sales, improves the weight assignment algorithm of the fusion model through information entropy, reduces the subjective intervention of manual weight assignment, and makes the prediction more scientific; at the same time, calculates the dynamic deviation of the prediction results in the past six months, converts the single-point prediction value into an interval prediction value, and improves the prediction accuracy. The perception method of this embodiment specifically includes the following steps:

[0034] Step S1, feature selection. The selected features include industry electricity sales, industry information factors (including industry category, increase and decrease in capacity, historical electricity consumption ranking of the industry, classification of industry electricity consumption characteristics, historical number of users in the industry, historical user categories in the industry, etc.), weather data (including monthly average temperature, monthly maximum temperature, monthly minimum temperature, as well as rainfall, number of cloudy days, number of sunny days, number of rainy days, proportion of sunny days, proportion of cloudy days, proportion of rainy days), GDP (including gross domestic product, gross domestic product of the primary industry, gross domestic product of the secondary industry, gross domestic product of the tertiary industry), resident consumption information (including CPI, year-on-year growth rate of CPI, month-on-month growth rate of CPI), etc.

[0035] Step S2, data extraction and cleaning of the selected features, extract data for a period of time (for example, from January 2020 to December 2021), and identify and process abnormal data that does not meet business requirements such as outliers and missing values.

[0036] Step S3, single base model prediction, use the support vector machine, XGBOOST algorithm, and random forest algorithms respectively to predict the future electricity sales.

[0037] Step S4, add the output result of the single base model prediction in step S3 to the electricity sales of the previous month as the feature of the fusion model for multi-model fusion prediction.

[0038] In this embodiment, the ferrous metal smelting and rolling processing industry is taken as an example. The data for 12 months in 2021 of this industry are shown in Table 1, which shows the prediction results of three single-base models and the electricity consumption of the previous month.

[0039] Table 1 Example of the characteristic matrix of the ferrous metal smelting and rolling processing industry for 12 months

[0040] Month Support Vector Machine XGBOOST Random Forest Last Month's Electricity Consumption 1 36862.11 32525.71 37989.15 37767.49 2 37408.7 33058.81 37068.59 36657.72 3 36123.16 35222.3 37819.99 39536.8 4 38346.81 34820.19 38527.96 38969.3 5 35207.04 31908.7 35614.34 38275.44 6 34786.69 30582.22 37517.28 36551.22 7 39148.19 35073.04 36584.56 37444.89 8 36059.88 33365.62 40678.73 35698.44 9 35526.76 30208.03 36196.98 37209.88 10 36248.46 30854.57 37974.3 39054.2 11 34446.81 30437.06 40967.86 39741.76 12 36514.31 30589.42 37133.7 37499.93

[0041] The process of multi-model fusion prediction specifically includes the following steps:

[0042] P1. Index standardization: Standardize the output result data predicted by the single-base model and the electricity sales data of the previous month to obtain a standardized matrix.

[0043] In this embodiment, formula 1 is used for index positive transformation, and the obtained conventional positive transformation indexes are shown in Table 2.

[0044] Formula 1:

[0045] Among them, Z ij is the index after positive transformation; x ij is the characteristic matrix of the fusion model, which has n rows and m columns. One row represents a sample, and one column represents an index. The row subscript is denoted by i, and the range of the row subscript i is 1...n. The column subscript is denoted by j, and the range of the column subscript j is 1...m. In the characteristic matrix of the fusion model, the first column feature is the output result data predicted by the support vector machine, the second column feature is the output result data predicted by the XGBOOST algorithm, the third column feature is the output result data predicted by the random forest, and the fourth column feature is the electricity sales data of the previous month.

[0046] Table 2 Conventional positive transformation indexes

[0047]

[0048]

[0049] It can be seen from the conventional positive transformation indexes shown in Table 2 that extreme values of 0 appear after the standardization processing of each column feature in the characteristic matrix. If the extreme value of 0 is not processed, it will cause the corresponding value after normalization to also be 0. According to the evaluation formula of information entropy (see formula 4 for details), the entropy value corresponding to the index cannot be calculated. In order to avoid the influence of extreme values on electricity sales prediction and ensure the objectivity of weight assignment, this embodiment uses the translation method to translate the indexes with extreme values in the standardized matrix and improves the standardization formula shown in formula 1, specifically as formula 2.

[0050] Formula 2:

[0051] Among them, Z ij is the index after improved positive transformation; d is the translation amount, and its value is flexibly adjusted according to the data of specific projects. In this embodiment, m is the number of column features of the feature matrix of the fusion model, which is 4. It should be noted that the translation amount d is an objective value. When the number of features of the fusion model is larger, the value of d will be smaller, the translation amount will also be smaller, and the influence on the dimension will be smaller, thereby amplifying the differences between the features of the fusion model. The improved standardization results are shown in Table 3, avoiding the influence that extreme value 0 appears in the conventional standardization process and entropy cannot be calculated.

[0052] Table 3 Improved Index Standardization Results Table

[0053] Month Support Vector Machine XGBOOST Random Forest Last Month's Electricity Consumption 1 1.235090 1.183564 1.164945 1.233068 2 1.351352 1.289881 0.992991 0.958598 3 1.077913 1.721348 1.133347 1.670657 4 1.550891 1.641154 1.265591 1.530302 5 0.883051 1.060514 0.721348 1.358695 6 0.793641 0.795973 1.076803 0.932258 7 1.721348 1.691580 0.902578 1.153282 8 1.064453 1.351068 1.667340 0.721348 9 0.951057 0.721348 0.830181 1.095159 10 1.104565 0.850288 1.162172 1.551299 11 0.721348 0.767023 1.721348 1.721348 12 1.161112 0.797408 1.005153 1.166895

[0054] Then, use Formula 3 to perform index normalization on the index after improved positive transformation. The normalized data is shown in Table 4.

[0055] Formula 3:

[0056] Among them, y ij is the normalized index data.

[0057] Table 4 Normalization Results Table

[0058] Month Support Vector Machine XGBOOST Random Forest Last Month's Electricity Consumption 1 0.090710 0.085326 0.085383 0.081699 2 0.099249 0.092990 0.072780 0.063513 3 0.079166 0.124096 0.083067 0.110691 4 0.113904 0.118314 0.092759 0.101392 5 0.064855 0.076455 0.052870 0.090022 6 0.058288 0.057383 0.078923 0.061768 7 0.126423 0.121950 0.066153 0.076412 8 0.078178 0.097401 0.122205 0.047794 9 0.069849 0.052003 0.060847 0.072561 10 0.081124 0.061299 0.085179 0.102783 11 0.052979 0.055296 0.126163 0.114050 12 0.085277 0.057487 0.073671 0.077314

[0059] P2. According to the normalized index data, calculate the information entropy of each index in the feature matrix of the fusion model. According to the information entropy of each index, further calculate the information utility value of each index in the feature matrix of the fusion model.

[0060] Calculate the information entropy of each index, which is completed by Formula 4. The information entropy of each index obtained is shown in Table 5.

[0061] Formula 4:

[0062] Among them, the value of k is taken as e j is the information entropy of each index.

[0063] Table 5 Information Entropy of Each Index

[0064]

[0065]

[0066] Calculate the information utility value d of each index in the feature matrix of the fusion model using Formula 5 j(i.e., information entropy redundancy), and the results are shown in Table 6. The information utility value is used to reflect the influence of each column feature in the feature matrix of the fusion model on the prediction of the fusion model. It can be seen that in this embodiment, the utility value of the output result data predicted by the XGBOOST algorithm is the highest, that is, the prediction result of the XGBOOST algorithm in the single base model will have a greater impact on the prediction of the fusion model.

[0067] Formula 5: d j = 1 - e j

[0068] Table 6 Information utility values of each index

[0069] Support Vector Machine XGBOOST Random Forest Last Month's Electricity Consumption Utility Value 0.012330 0.019863 0.012489 0.011755

[0070] P3. Calculate the weight w of each index according to the information utility value of each index in the feature matrix of the fusion model j .

[0071] Calculate the weights of each index using Formula 6, and obtain that the weight of the output result data predicted by the support vector machine is 0.218472, the weight of the output result data predicted by the XGBOOST algorithm is 0.351954, the weight of the output result data predicted by the random forest is 0.221290, and the weight of the electricity sales data of the previous month is 0.208283.

[0072] Formula 6:

[0073] Table 6 Weights of each index

[0074] Support Vector Machine XGBOOST Random Forest Last Month's Electricity Consumption Weight 0.218472 0.351954 0.221290 0.208283

[0075] P4. Model fusion: Perform weighted summation of the weights of each index obtained in step P3, the output result data predicted by the three single base models, and the electricity sales data of the previous month to obtain the prediction result of the fusion model.

[0076] Step S5. Calculate the dynamic deviation: Calculate the average error of the prediction results of the fusion model in the recent six months, divide it into positive deviation and negative deviation, and dynamically adjust the current prediction result to convert the current single-point prediction value into an interval prediction value to improve the prediction accuracy.

[0077] Similarly, other industries such as the chemical fiber manufacturing industry, metal products industry, rubber and plastic products industry, textile industry, and non-metallic mineral products industry also perform data processing in the same process as the ferrous metal smelting and rolling processing industry to complete the electricity sales prediction of all industries.

[0078] Embodiment 2

[0079] Based on the same inventive concept as Embodiment 1, this embodiment provides a multi-model fusion industry power situation awareness system, such asFigure 2 As shown in the figure, it includes the following modules:

[0080] A feature selection module for performing feature selection. The selected features include industry electricity sales volume, industry information factors, weather data, and residential consumption information;

[0081] A preprocessing module for data extraction and cleaning of the selected features;

[0082] A single base model prediction module for predicting future electricity sales volume using the first base model, the second base model, and the third base model respectively;

[0083] A fusion prediction module for using the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales volume as features of the fusion model to perform multi-model fusion prediction;

[0084] Among them, in the feature matrix of the fusion model, the first column of features is the output result data predicted by the first base model, the second column of features is the output result data predicted by the second base model, the third column of features is the output result data predicted by the third base model, and the fourth column of features is historical electricity sales volume data; perform translation on the indexes after standardizing each column of features in the feature matrix to obtain the improved positive indexes; then calculate the information utility values of each index and calculate the weights of each index; perform weighted summation of each index weight with the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales volume to obtain the prediction result of the fusion model.

[0085] In this embodiment, the process of multi-model fusion prediction includes:

[0086] Perform standardization processing on the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales volume to obtain a standardized matrix; perform translation on the indexes with extreme values in the standardized matrix to obtain the improved positive indexes; perform index normalization processing on the positively processed indexes;

[0087] According to the normalized index data, calculate the information entropy of each index in the feature matrix of the fusion model; calculate the information utility value of each index in the feature matrix of the fusion model according to the information entropy of each index;

[0088] Calculate the weight of each index according to the information utility value of each index;

[0089] Perform weighted summation of each index weight with the output result data predicted by the first base model, the second base model, and the third base model, and historical electricity sales volume data to obtain the prediction result of the fusion model;

[0090] Among them, the improved positive index Z ij is:

[0091]

[0092] Among them, x ij is the feature matrix of the fusion model, which has n rows and m columns. One row represents a sample, and one column represents an index. The range of the row subscript i is 1...n, and the range of the column subscript j is 1...m; d is the translation amount, and its value is:

[0093] where m is the number of column features of the feature matrix of the fusion model.

[0094] In a preferred embodiment, the perception system of this embodiment further includes a dynamic deviation calculation module, which is used to calculate the average error of the prediction results of the fusion model in the past period of time, divide it into positive deviation and negative deviation, and dynamically adjust the current prediction result, so that the current single-point prediction value is converted into an interval prediction value.

[0095] Each of the above modules in this embodiment is used to implement each step of Embodiment 1. For the detailed implementation process, please refer to Embodiment 1 and will not be elaborated here.

[0096] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. An industry power situation awareness method based on multi-model fusion, characterized in that, It includes the following steps: Feature Selection, the selected features include industry electricity sales volume, industry information factors, weather data and residential consumption information; Perform data extraction and cleaning on the selected features; Use the first base model, the second base model, and the third base model to predict future electricity sales volume respectively; Take the output results predicted by the first base model, the second base model, and the third base model and the historical electricity sales volume as the features of the fusion model for multi-model fusion prediction; In the feature matrix of the fusion model, the first column of features is the output result data predicted by the first base model, the second column of features is the output result data predicted by the second base model, the third column of features is the output result data predicted by the third base model, and the fourth column of features is the historical electricity sales volume data; perform translation on the indicators after standardizing each column of features in the feature matrix to obtain the improved positive indicators; then calculate the information utility value of each indicator and calculate the weight of each indicator; perform weighted summation of the weights of each indicator and the output results predicted by the first base model, the second base model, and the third base model and the historical electricity sales volume to obtain the prediction result of the fusion model; The first base model is the support vector machine algorithm, the second base model is the XGBOOST algorithm, and the third base model is the random forest algorithm; the historical electricity sales volume is the electricity sales volume of last month; The process of multi-model fusion prediction specifically includes the steps: Perform standardization processing on the output results predicted by the first base model, the second base model, and the third base model and the historical electricity sales volume to obtain a standardized matrix; perform translation on the indicators with extreme values in the standardized matrix to obtain the improved positive indicators; perform index normalization processing on the positive indicators; According to the normalized index data, calculate the information entropy of each indicator in the feature matrix of the fusion model; according to the information entropy of each indicator, calculate the information utility value of each indicator in the feature matrix of the fusion model; According to the information utility values of the indicators, calculate the weights of the indicators; Perform weighted summation of the weights of the indicators and the output result data predicted by the first base model, the second base model, and the third base model, and the historical electricity sales volume data to obtain the prediction result of the fusion model; When performing translation on the indicators with extreme values in the standardized matrix, the translation amount d takes the value of: Where m is the number of column features of the feature matrix of the fusion model.

2. The industry power situation awareness method according to claim 1, wherein The method further includes the following steps: Calculate the dynamic deviation: calculate the average error of the prediction results of the fusion model in the past period of time, divide it into positive deviation and negative deviation, and dynamically adjust the current prediction result so that the current single-point prediction value is converted into an interval prediction value.

3. The industrial power situation awareness method according to claim 1, characterized in that The improved index Z after the forward processing ij is as follows: where x ij is the feature matrix of the fusion model, with a total of n rows and m columns. One row represents a sample, and one column represents an index. The range of the row subscript i is 1...n, and the range of the column subscript j is 1...m; d is the translation amount.

4. The industry power situation awareness method according to claim 3, wherein The specific method for performing index normalization processing on the positive indicators is: Among them, y ij is the normalized index data; The calculation formula for the information entropy of each indicator is: Among them, the value of k is taken as e j is the information entropy of each index; The information utility value d of each indicator j The calculation formula is as follows: d j = 1 - e j The information utility value is used to reflect the influence of each column of features in the feature matrix of the fusion model on the fusion model prediction; The weight w of each indicator j The calculation formula is as follows:

5. The industrial power situation awareness method according to claim 1, wherein The industry information factors include industry categories, increased or decreased capacity, historical electricity consumption rankings of industries, classification of industry electricity consumption characteristics, historical number of industry users and historical user categories of industries; the weather data includes monthly average temperature, monthly maximum temperature, monthly minimum temperature, as well as rainfall, number of cloudy days, number of sunny days, number of rainy days, proportion of sunny days, proportion of cloudy days and proportion of rainy days.

6. An industry power situation awareness system with multi-model fusion, characterized in that, It includes the following modules: A feature selection module for performing feature selection. The selected features include industry electricity sales volume, industry information factors, weather data, and residential consumption information; A preprocessing module for data extraction and cleaning of the selected features; A single base model prediction module that uses the first base model, the second base model, and the third base model to predict future electricity sales volume respectively; A fusion prediction module for using the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales volume as features of the fusion model to perform multi-model fusion prediction; Among them, in the feature matrix of the fusion model, the first column of features is the output result data predicted by the first base model, the second column of features is the output result data predicted by the second base model, the third column of features is the output result data predicted by the third base model, and the fourth column of features is historical electricity sales volume data; the indexes after standardizing each column of features in the feature matrix are shifted to obtain indexes after improved positive processing; then the information utility values of each index are obtained, and the weights of each index are calculated; the weights of each index are weighted and summed with the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales volume to obtain the prediction result of the fusion model; The first base model is the support vector machine algorithm, the second base model is the XGBOOST algorithm, and the third base model is the random forest algorithm; the historical electricity sales volume is the electricity sales volume of last month; The process of multi-model fusion prediction includes: Performing standardization processing on the output results predicted by the first base model, the second base model, and the third base model and historical electricity sales volume to obtain a standardized matrix; shifting the indexes with extreme values in the standardized matrix to obtain indexes after improved positive processing; performing index normalization processing on the indexes after positive processing; Calculating the information entropy of each index in the feature matrix of the fusion model according to the normalized index data; calculating the information utility value of each index in the feature matrix of the fusion model according to the information entropy of each index; Calculating the weights of each index according to the information utility values of each index; Weightedly summing the weights of each index with the output result data predicted by the first base model, the second base model, and the third base model, and historical electricity sales volume data to obtain the prediction result of the fusion model; When shifting the indexes with extreme values in the standardized matrix, the shift amount d takes the value of: Where m is the number of column features of the feature matrix of the fusion model.

7. The industry power situation awareness system according to claim 6, characterized in that The improved index Z after the forwardization process ij is as follows: where x ij is the feature matrix of the fusion model, with a total of n rows and m columns. One row represents a sample, and one column represents an indicator. The range of the row subscript i is 1...n, and the range of the column subscript j is 1...m.

Images