Electrical load prediction method, system, equipment and medium

Abstract

The invention discloses an electrical load prediction method, system, device and medium, and the method comprises the steps: building a load prediction model, setting the initial parameters and hyper-parameters of the load prediction model, carrying out the additive feature attribution based on the initial output data of the model, and obtaining the contribution degree of each feature of each sample to a prediction value; the initial parameters and the hyper-parameters are reversely optimized, the learning efficiency and prediction accuracy of the model can be improved, it is ensured that the model is better in performance on different data sets, the explanatory property and performance of the model are improved, abnormal conditions in model prediction can be found in time by supervising the output result, potential problems can be recognized, the direction can be improved, and the method is suitable for popularization and application. According to the method, the reliability of a prediction result is ensured, potential problems and improvement directions can be identified, and by evaluating and re-optimizing the model, the performance of the model can be continuously improved, the method adapts to continuously changing power demands and environmental conditions, and the stability and accuracy of prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power monitoring, and in particular to an electric load forecasting method, system, equipment and medium. Background Art

[0002] With the acceleration of industrialization and urbanization, the demand for electricity is growing, and grid load management has become a key issue in the operation of power systems. In addition to being related to historical load, power load is also affected by many factors such as policy macroeconomic regulation and economic factors. These factors are random and changeable, and can significantly affect the power load of smart grids. Therefore, it is necessary to predict the power load to ensure the stable operation of electricity. In the existing technology, mathematical statistical methods are usually used, mainly including regression analysis, Kalman filtering, time series analysis, etc., which mainly adopt the idea of ​​linear analysis. It has certain limitations when dealing with load and other factors (nonlinear relationship), but the prediction effect is poor. When using machine learning methods such as support vector machines and artificial neural networks, there is a general problem of non-convergence when there are many data samples. Traditional load forecasting only considers historical load data, temperature, rainfall and other structured data, which has certain limitations and does not fully consider the impact of factors such as policy macroeconomic regulation on power load. Summary of the invention

[0003] The technical problem to be solved by the present invention is that it does not combine comprehensive consideration of multiple factors, and the prediction accuracy is low. The purpose is to provide a method, system, equipment and medium for electricity load prediction. By constructing a load prediction model, setting the initial parameters and hyperparameters of the load prediction model, and performing additive feature attribution based on the initial output data of the model, the contribution of each feature of each sample to the predicted value is obtained, and the initial parameters and hyperparameters are reversely optimized. This can improve the learning efficiency and prediction accuracy of the model, ensure that the model performs better on different data sets, and improve the interpretability and performance of the model. By supervising the output results, abnormal situations in the model prediction can be discovered in time, helping to identify potential problems and improvement directions, ensuring the reliability of the prediction results, helping to identify potential problems and improvement directions, and by evaluating and re-optimizing the model, the performance of the model can be continuously improved to adapt to the changing power demand and environmental conditions, and to improve the stability and accuracy of the prediction.

[0004] The present invention is achieved through the following technical solutions:

[0005] A first aspect of the present invention provides a method for predicting power load, comprising the following specific steps:

[0006] Obtain power data in the target monitoring area and build a load forecasting model;

[0007] Obtain the initial parameters and hyperparameters that meet the prediction model, and use grid search to find the optimal hyperparameter combination;

[0008] Perform additive feature attribution on the initial output data of the load forecasting model to obtain the contribution of each feature of each sample to the predicted value;

[0009] Based on the contribution, the initial parameters and hyperparameters of the load forecasting model are adjusted, the weights of the features are changed, and the optimized load forecasting model is obtained;

[0010] Obtain the output results of the optimized load forecasting model, and conduct load forecasting supervision on the output results based on fitting matching analysis and residual analysis between the forecast data and the actual monitoring data;

[0011] Based on the load forecasting supervision results, the load forecasting model is evaluated and re-optimized to carry out load forecasting.

[0012] Furthermore, the acquisition of power data in the target monitoring area and the construction of a load forecasting model specifically include:

[0013] Obtain the historical data of power load collected in the target monitoring area, and determine the data fields that need to be collected based on the monitoring indicators;

[0014] Extract target fields from the collected historical data of power load to obtain power data;

[0015] Perform unique constraint check on a single field or a combination of multiple fields of the collected power data, verify by repeating record rows or other attribute values ​​that violate the unique constraint, and use the power data that passes the check as sample data;

[0016] Construct a load forecasting framework, train the load forecasting framework based on sample data, and generate a load forecasting model.

[0017] Furthermore, the construction of the load forecasting framework, training based on the sample data load forecasting framework, and generating a load forecasting model specifically includes:

[0018] Construct a load forecasting framework to extract features from sample data and obtain initial features;

[0019] Calculate the correlation between each initial feature and the target variable, and select the feature with the highest correlation with the target variable as the feature subset;

[0020] The remaining features are combined with the historical load data to form a new data set, and the features with high correlation with the target variable and a set threshold and low correlation with the selected features and a set threshold are selected in the new data set and merged into the feature subset;

[0021] Repeat the above steps until the required number of features is reached to generate feature data;

[0022] The load forecasting framework is trained based on the characteristic data to obtain the load forecasting model.

[0023] Furthermore, the initial output data of the load forecasting model is subjected to additive feature attribution to obtain the contribution of each feature of each sample to the predicted value, specifically including:

[0024] In obtaining the initial output data of the load forecasting model, a gradient-based attribution method is used to calculate the gradient of the initial output data relative to each input feature, wherein the gradient represents the degree of change of the predicted value when the feature is changed;

[0025] The partial derivative of the gradient of the initial output data with respect to each input feature is calculated to estimate the contribution of each feature of each sample to the predicted value, where the contribution of each feature can be interpreted as the direct contribution of the feature to the model prediction value. A positive contribution indicates that an increase in the feature value will lead to an increase in the predicted value, while a negative contribution indicates that an increase in the feature value will lead to a decrease in the predicted value.

[0026] Furthermore, the load forecasting supervision on the output results is performed based on the fitting matching analysis and residual analysis of the predicted data and the actual monitoring data, specifically including:

[0027] Fit and match the predicted data with the actual monitoring data to obtain the matching accuracy;

[0028] If the matching accuracy is not within the set threshold, residual analysis is performed on the prediction results;

[0029] Adjust the optimized load forecasting model according to the residual analysis results.

[0030] Furthermore, the residual analysis of the prediction results specifically includes:

[0031] Obtain the non-random distribution of the forecast result data, perform residual processing on the non-random distribution, and obtain the residual time series;

[0032] Based on the residual time series, the autocorrelation of the time series corresponding to the residual is calculated to determine the data capture capability of the optimized load forecasting model:

[0033] When the residual is higher than the set threshold, the load forecast of the optimized load forecasting model does not meet the standard. The model is retrained to obtain a new load forecasting model for prediction until the residual of the prediction result reaches the set threshold and the final prediction result is output.

[0034] Furthermore, the threshold interval setting step for residual determination includes:

[0035] Calculate the standardized residual of each observation and determine the distribution characteristics of the residual through white noise test;

[0036] According to the distribution characteristics of the residuals, the absolute value ±3σ is used as the threshold of the standardized residuals;

[0037] If the absolute value of the standardized residual exceeds the threshold, the corresponding observation is considered an outlier.

[0038] A second aspect of the present invention provides a power load forecasting system, which is applied to a power load forecasting method including:

[0039] A data acquisition module is used to obtain power data of the target monitoring area;

[0040] The model building module is used to build a load forecasting framework, train the load forecasting framework based on sample data, and generate a load forecasting model;

[0041] The model optimization module is used to obtain the initial parameters and hyperparameters that meet the prediction model, adjust the feature weights, and obtain the optimized load prediction model;

[0042] The data processing module is used to pre-process the power data and generate sample data; it is also used to obtain the output results of the optimized load forecasting model and perform load forecasting supervision on the output results;

[0043] Output module, used to output the final load forecast results.

[0044] A third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for predicting power load when executing the program.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements a method for predicting power load when executed by a processor.

[0046] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0047] By constructing a load forecasting model, setting the initial parameters and hyperparameters of the load forecasting model, and performing additive feature attribution based on the initial output data of the model, the contribution of each feature of each sample to the predicted value can be obtained, and the initial parameters and hyperparameters can be reversely optimized. This can improve the learning efficiency and prediction accuracy of the model, ensure that the model performs better on different data sets, and improve the interpretability and performance of the model. By supervising the output results, anomalies in the model prediction can be discovered in a timely manner, helping to identify potential problems and improvement directions, ensuring the reliability of the prediction results, and helping to identify potential problems and improvement directions. By evaluating and re-optimizing the model, the performance of the model can be continuously improved, adapting to changing power demand and environmental conditions, and improving the stability and accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0049] Figure 1 This is a load monitoring method in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0051] As a possible implementation, Figure 1 As shown, this embodiment provides a method for predicting electricity load, including the following specific steps: obtaining power data in a target monitoring area and constructing a load prediction model; obtaining initial parameters and hyperparameters that conform to the prediction model, and using grid search to find the optimal hyperparameter combination; performing additive feature attribution on the initial output data of the load prediction model to obtain the contribution of each feature of each sample to the predicted value; adjusting the initial parameters and hyperparameters of the load prediction model based on the contribution, changing the weight of the feature, and obtaining an optimized load prediction model; obtaining the output result of the optimized load prediction model, and performing load prediction supervision on the output result based on fitting matching analysis and residual analysis of the predicted data and the actual monitoring data; evaluating and re-optimizing the load prediction model based on the load prediction supervision result to perform load prediction. This embodiment constructs a load forecasting model, sets the initial parameters and hyperparameters of the load forecasting model, performs additive feature attribution based on the initial output data of the model, obtains the contribution of each feature of each sample to the predicted value, and reversely optimizes the initial parameters and hyperparameters. This can improve the learning efficiency and prediction accuracy of the model, ensure that the model performs better on different data sets, and improve the interpretability and performance of the model. By supervising the output results, abnormal situations in the model prediction can be discovered in a timely manner, helping to identify potential problems and improvement directions, ensuring the reliability of the prediction results, helping to identify potential problems and improvement directions. By evaluating and re-optimizing the model, the performance of the model can be continuously improved to adapt to the changing power demand and environmental conditions, and to improve the stability and accuracy of the prediction.

[0052] In some possible implementations, the power data is preprocessed, specifically including: obtaining the historical power load data collected in the target monitoring area, and determining the data fields to be collected according to the monitoring indicators; this step ensures the pertinence and comprehensiveness of data collection, and provides necessary data support for subsequent data analysis and model training. By clarifying the monitoring indicators, the data fields most relevant to the power load forecast can be collected more targeted, thereby improving the accuracy and efficiency of the forecast model. The target field is extracted from the collected historical power load data to obtain power data; by extracting the target field, irrelevant data can be removed, data redundancy can be reduced, and data processing efficiency can be improved. At the same time, this also helps to improve the quality of model training, because the model will only use the most relevant data for learning, which may improve the accuracy of the prediction. Perform unique constraint verification on a single field or a combination of multiple fields of the collected power data, verify through duplicate record rows or other unique constraint attribute values, and use the power data that passes the verification as sample data. The unique constraint verification helps to identify and remove duplicate records or erroneous data in the data, ensuring the uniqueness and accuracy of the data. This is crucial to improving data quality, because the accuracy of the data directly affects the effect of model training and the reliability of the prediction results. In this way, noise interference in the model training process can be reduced and the generalization ability of the model can be improved. By preprocessing the power data, not only the quality of the power data is improved, laying a solid foundation for building an accurate load forecasting model, but also the robustness of the model and the reliability of the prediction are improved by ensuring the uniqueness and accuracy of the data.

[0053] In some possible implementations, a load forecasting framework is constructed, and the load forecasting framework is trained based on sample data to generate a load forecasting model, specifically including: constructing a load forecasting framework, the constructed load forecasting framework can not only identify and utilize key features, but also optimize feature sets through an iterative feature selection process, and finally generate an efficient and accurate load forecasting model, extracting features from sample data based on the load forecasting framework to obtain initial features; calculating the correlation between each initial feature and the target variable, and selecting the feature with the highest correlation with the target variable as the feature subset; combining the remaining features with the historical load data to form a new data set, and selecting features in the new data set that have a high correlation with the target variable and a set threshold and a low correlation with the selected features and a set threshold, and merging them into the feature subset; repeating the above steps until the required number of features is reached to generate feature data; training the load forecasting framework based on the feature data to obtain a load forecasting model. By identifying the factors that have the greatest impact on power load, the dimension of the feature space is reduced, the efficiency of model training and the accuracy of prediction are improved, and combining historical load data can help the model capture patterns and trends in time series data. At the same time, by setting correlation thresholds, features that have a significant impact on the prediction results can be further screened out, reducing the problem of multicollinearity between features. Through the iterative feature selection process, feature subsets can be continuously optimized to ensure that the final feature set contains both the features that have the greatest impact on the prediction results and avoids redundancy and collinearity between features, thereby improving the generalization ability of the model and prediction accuracy.

[0054] In some possible implementations, the main purpose of feature selection is to select features that are highly correlated with the target variable and have low correlation with each other from a given feature set to construct an effective feature subset. This method is commonly used in machine learning and data mining tasks to improve model performance and prediction accuracy. Combining the remaining feature quantities and historical load data to form a new data set is to more comprehensively examine the relationship between the feature quantities and the target variable. In the new data set, features that have a high correlation with the target variable and a set threshold and a low correlation with the selected features and a set threshold are selected and merged into the feature subset: select those features that have a high correlation with the target variable (exceeding a certain set threshold), and at the same time ensure that the correlation between these features is low (below a certain set threshold), so as to avoid information redundancy and improve the generalization ability of the model.

[0055] For example, suppose we are building a model to predict electricity demand, and the features may include temperature, humidity, wind speed, historical electricity demand, etc. First, we may calculate the correlation between each feature and the target variable (electricity demand) and set a threshold, such as 0.7. Only those features with a correlation greater than 0.7 will be considered. Then, we check the correlation between these highly correlated features to ensure that the correlation between them is below a certain threshold, such as 0.5. In this way, we can get a subset of features that are highly correlated with electricity demand and have low information redundancy with each other, which can be used to build the final prediction model.

[0056] In some possible implementations, the initial output data of the load forecasting model is subjected to additive feature attribution to obtain the contribution of each feature of each sample to the predicted value, specifically including: obtaining the initial output data of the load forecasting model, using a gradient-based attribution method to calculate the gradient of the initial output data relative to each input feature, the gradient indicating the degree of change of the predicted value when the feature is changed; calculating the partial derivative of the gradient of the initial output data relative to each input feature to estimate the contribution of each feature of each sample to the predicted value, wherein the contribution of each feature can be interpreted as the direct contribution of the feature to the model prediction value, a positive contribution indicates that an increase in the feature value will lead to an increase in the predicted value, and a negative contribution indicates that an increase in the feature value will lead to a decrease in the predicted value. The above content can help understand the specific impact of each feature on the model prediction, and by calculating the feature contribution, it can be used to analyze the importance of features in the machine learning model, thereby optimizing the model performance.

[0057] Obtain the initial parameters and hyperparameters that meet the prediction model, adjust the feature weights, and obtain the optimized load prediction model, including:

[0058] Get the initial parameters and hyperparameters that meet the prediction model, use grid search to find the optimal hyperparameter combination, define a candidate list of hyperparameters, use grid search (GridSearchCV) to cross-validate each hyperparameter combination, and find the hyperparameter combination that performs best on the validation set. Although this method is computationally intensive, it can systematically traverse multiple possibilities and find the optimal hyperparameter combination. Grid search ensures that we will not miss any possible hyperparameter combination, thereby improving the performance and generalization ability of the model;

[0059] Perform additive feature attribution on the initial output data of the load forecasting model to obtain the contribution of each feature of each sample to the forecast value: Use methods such as SHAP (SHapley Additive exPlanations) to analyze the model's forecast output and calculate the contribution of each feature to the forecast result. This method can help us understand the model's decision-making process and identify the features that have the greatest impact on the forecast results. Through feature attribution, we can better understand the model's behavior, optimize the use of features, and improve the model's interpretability and forecasting accuracy.

[0060] Adjust the initial parameters and hyperparameters of the load forecasting model based on the contribution, change the weight of the features, and obtain the optimized load forecasting model: According to the analysis results of the feature contribution, adjust the weight of the features in the model, which may involve increasing the weight of some features or reducing the weight of other features, or even removing some unimportant features. At the same time, the initial parameters and hyperparameters of the model, such as the learning rate and regularization coefficient, can also be adjusted according to the feature contribution to optimize the performance of the model, so that the model can pay more attention to those features that have a significant impact on the prediction results, reduce the interference of noise features, and thus improve the prediction accuracy and robustness of the model.

[0061] In some possible implementations, load forecasting supervision is performed on the output results, specifically including:

[0062] According to the prediction data and the actual monitoring data, the model's prediction results are matched with the actual power load monitoring data, and the matching accuracy is calculated to quantify the accuracy of the prediction model. If the matching accuracy is lower than the preset threshold, it means that the model's prediction has a large deviation. At this time, residual analysis is required to identify the systematic deviation and random error in the model prediction. The load prediction model is adjusted based on the results of the residual analysis, including reselecting or adjusting features, modifying the model structure, and adjusting hyperparameters. After adjustment, the model is retrained, and fitting matching and residual analysis are performed again until the model's prediction performance reaches a satisfactory level. Through iterative model adjustment and verification, the prediction accuracy and robustness of the model can be gradually improved. This method helps to build a dynamic prediction model that can adapt to actual power load changes.

[0063] In some possible implementations, residual analysis is performed on the prediction results, specifically including:

[0064] First, obtain the non-random distribution of the prediction result data, compare the model's prediction results with the actual monitoring data, and calculate the residuals (actual values ​​and predicted values) at each time point. Then, analyze the distribution characteristics of the residual data, check whether there are non-random patterns or trends, perform residual processing on the non-random distribution to identify systematic deviations in the model prediction, and obtain the residual time series;

[0065] Based on the residual time series, the autocorrelation of the time series corresponding to the residual is calculated to judge the data capture capability of the optimized load forecasting model, that is, to evaluate whether the load forecasting model has fully captured the information in the data. The model is evaluated through autocorrelation analysis to determine whether it has fully captured the information in the data. If the residual shows significant autocorrelation, it is judged that the residual is higher than the set threshold, and the load forecast of the optimized load forecasting model does not meet the standard. The model is retrained to obtain a new load forecasting model for prediction until the residual of the prediction result reaches the set threshold, and the final prediction result is output. By performing residual analysis on the prediction results and optimizing the model based on residual feedback, the prediction results of the load forecasting model can be ensured to be both accurate and reliable. This method helps to improve the prediction accuracy of the power system, optimize resource allocation, reduce operating costs, and improve the stability and efficiency of the system.

[0066] In some possible implementations, the step of setting a threshold interval for residual determination includes:

[0067] First, the standardized residual of each observation is calculated, which is the value obtained by dividing the residual by its standard deviation, also known as Pearson residual or half-studentized residual. Standardized residuals can reflect the difference between the observed value and the predicted value, and make the residuals of different observations comparable. The distribution characteristics of the residuals can be determined by the white noise test. The white noise test is a statistical test used to determine whether a time series is random, that is, there is no identifiable pattern or structure. If the residual sequence passes the white noise test, it means that the model has captured most of the information in the data and there is no systematic information remaining in the residuals. According to the distribution characteristics of the standardized residuals, the absolute value ±3σ (i.e. ±3 times the standard deviation) is usually used as the threshold of the residuals. This threshold is determined based on the characteristics of the normal distribution, because in the normal distribution, about 99.7% of the data are within the ±3σ range of the mean. Therefore, standardized residuals outside this range are considered outliers. If the absolute value of the standardized residual exceeds the set threshold (±3σ), it indicates that the corresponding observation may not be caused by random errors, but may be an outlier caused by data input errors, model assumptions that do not hold, or other non-random factors.

[0068] As a possible implementation mode, this embodiment provides a power load forecasting system, which is applied to a power load forecasting method, including:

[0069] A data acquisition module is used to obtain power data of the target monitoring area;

[0070] The model building module is used to build a load forecasting framework, train the load forecasting framework based on sample data, and generate a load forecasting model;

[0071] The model optimization module is used to obtain the initial parameters and hyperparameters that meet the prediction model, adjust the feature weights, and obtain the optimized load prediction model;

[0072] The data processing module is used to pre-process the power data and generate sample data; it is also used to obtain the output results of the optimized load forecasting model and perform load forecasting supervision on the output results;

[0073] Output module, used to output the final load forecast results.

[0074] As a possible implementation, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for predicting power load when executing the program.

[0075] As a possible implementation, this embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, a method for predicting power load is implemented.

[0076] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting power load, characterized in that: The specific steps include: Obtain power data in the target monitoring area and build a load forecasting model; Obtain the initial parameters and hyperparameters that meet the prediction model, and use grid search to find the optimal hyperparameter combination; Perform additive feature attribution on the initial output data of the load forecasting model to obtain the contribution of each feature of each sample to the predicted value; Based on the contribution, the initial parameters and hyperparameters of the load forecasting model are adjusted, the weights of the features are changed, and the optimized load forecasting model is obtained; Obtain the output results of the optimized load forecasting model, and conduct load forecasting supervision on the output results based on fitting matching analysis and residual analysis between the forecast data and the actual monitoring data; Based on the load forecasting supervision results, the load forecasting model is evaluated and re-optimized to carry out load forecasting.

2. The power load forecasting method according to claim 1, characterized in that: The obtaining of power data of the target monitoring area and building of a load forecasting model specifically includes: Obtain the historical data of power load collected in the target monitoring area, and determine the data fields that need to be collected based on the monitoring indicators; Extract target fields from the collected historical data of power load to obtain power data; Perform unique constraint check on a single field or a combination of multiple fields of the collected power data, verify by repeating record rows or other attribute values ​​that violate the unique constraint, and use the power data that passes the check as sample data; Construct a load forecasting framework, train the load forecasting framework based on sample data, and generate a load forecasting model.

3. The power load forecasting method according to claim 2, characterized in that: The load forecasting framework is constructed, and training is performed based on the sample data load forecasting framework to generate a load forecasting model, specifically including: Construct a load forecasting framework to extract features from sample data and obtain initial features; Calculate the correlation between each initial feature and the target variable, and select the feature with the highest correlation with the target variable as the feature subset; The remaining features are combined with the historical load data to form a new data set, and the features with high correlation with the target variable and a set threshold and low correlation with the selected features and a set threshold are selected in the new data set and merged into the feature subset; Repeat the above steps until the required number of features is reached to generate feature data; The load forecasting framework is trained based on the characteristic data to obtain the load forecasting model.

4. The method for predicting power load according to claim 1, characterized in that: The initial output data of the load forecasting model is subjected to additive feature attribution to obtain the contribution of each feature of each sample to the predicted value, specifically including: In obtaining the initial output data of the load forecasting model, a gradient-based attribution method is used to calculate the gradient of the initial output data relative to each input feature, wherein the gradient represents the degree of change of the predicted value when the feature is changed; The partial derivative of the gradient of the initial output data with respect to each input feature is calculated to estimate the contribution of each feature of each sample to the predicted value, where the contribution of each feature can be interpreted as the direct contribution of the feature to the model prediction value. A positive contribution indicates that an increase in the feature value will lead to an increase in the predicted value, while a negative contribution indicates that an increase in the feature value will lead to a decrease in the predicted value.

5. The method for predicting power load according to claim 1, characterized in that: The load forecasting supervision based on the output results of the fitting matching analysis and residual analysis based on the predicted data and the actual monitoring data specifically includes: Fit and match the predicted data with the actual monitoring data to obtain the matching accuracy; If the matching accuracy is not within the set threshold, residual analysis is performed on the prediction results; Adjust the optimized load forecasting model according to the residual analysis results.

6. The method for predicting power load according to claim 5, characterized in that: The residual analysis of the prediction results specifically includes: Obtain the non-random distribution of the forecast result data, perform residual processing on the non-random distribution, and obtain the residual time series; Based on the residual time series, the autocorrelation of the time series corresponding to the residual is calculated to determine the data capture capability of the optimized load forecasting model: When the residual is higher than the set threshold, the load forecast of the optimized load forecasting model does not meet the standard. The model is retrained to obtain a new load forecasting model for prediction until the residual of the prediction result reaches the set threshold and the final prediction result is output.

7. The method for predicting power load according to claim 6, characterized in that: The steps of setting the threshold interval for residual determination include: Calculate the standardized residual of each observation and determine the distribution characteristics of the residual through white noise test; According to the distribution characteristics of the residuals, the absolute value ±3σ is used as the threshold of the standardized residuals; If the absolute value of the standardized residual exceeds the threshold, the corresponding observation is considered an outlier.

8. A power load forecasting system, applied to the power load forecasting method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to obtain power data of the target monitoring area; The model building module is used to build a load forecasting framework, train the load forecasting framework based on sample data, and generate a load forecasting model; The model optimization module is used to obtain the initial parameters and hyperparameters that meet the prediction model, adjust the feature weights, and obtain the optimized load prediction model; A data processing module is used to pre-process the power data and generate sample data; It is also used to obtain the output results of the optimized load forecasting model and to conduct load forecasting supervision on the output results; Output module, used to output the final load forecast results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the power load forecasting method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the power load forecasting method as claimed in any one of claims 1 is implemented.

Images