Residential power consumption load prediction method and device of power system, computer equipment, storage medium and computer program product

Abstract

The invention relates to an electric power system residential electricity load prediction method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: determining a target power utilization load prediction model of current residential power utilization load data according to the power utilization stability of the current residential power utilization load data of a to-be-analyzed power system; extracting a target feature vector of the current residential power consumption load data, and inputting the target feature vector into a target power consumption load prediction model and other power consumption load prediction models to obtain first residential power consumption load data output by the target power consumption load prediction model, and second residential power consumption load data output by other power consumption load prediction models. And performing fusion processing on the first residential electricity load data and the second residential electricity load data to obtain predicted residential electricity load data of the to-be-analyzed power system. By adopting the method, the prediction accuracy of the residential electricity load can be improved.

Description

Technical Field

[0001] The present application relates to the field of power grid technology, and in particular to a method, device, computer equipment, computer-readable storage medium and computer program product for predicting residential power load in a power system. Background Art

[0002] In a smart grid environment, how to accurately predict residential electricity load is crucial for optimal scheduling and energy management of the power system.

[0003] In traditional technology, manual prediction is generally used in the process of predicting residential electricity load; however, this manual prediction method has subjective factors and is prone to errors, resulting in low accuracy in predicting residential electricity load. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for predicting residential electricity load in a power system, which can improve the prediction accuracy of residential electricity load in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for predicting residential power load in a power system, comprising:

[0006] Obtain current residential electricity load data of the power system to be analyzed;

[0007] Determining the power consumption stability of the current residential power load data;

[0008] Selecting a power load forecasting model corresponding to the power stability from a plurality of trained power load forecasting models as a target power load forecasting model for the current residential power load data;

[0009] Performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, and inputting the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain first residential electricity load data output by the target electricity load prediction model and second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models;

[0010] According to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are fused to obtain the predicted residential power load data of the power system to be analyzed.

[0011] In one embodiment, obtaining current residential electricity load data of the power system to be analyzed includes:

[0012] Obtain candidate residential electricity load data of the power system to be analyzed at the current time;

[0013] Preprocessing the candidate residential electricity load data to obtain preprocessed residential electricity load data;

[0014] Inputting each of the pre-processed residential electricity load data into a plurality of pre-trained importance prediction models to obtain a plurality of predicted importances of each of the pre-processed residential electricity load data;

[0015] fusing the plurality of predicted importances of each of the preprocessed residential electricity load data to obtain the target predicted importance of each of the preprocessed residential electricity load data;

[0016] From each of the pre-processed residential electricity load data, the pre-processed residential electricity load data whose target prediction importance is greater than the preset importance is screened out as the current residential electricity load data.

[0017] In one embodiment, determining the power consumption stability of the current residential power load data includes:

[0018] Obtaining the average value and standard deviation of the current residential electricity load data;

[0019] Determining the coefficient of variation of the current residential electricity load data according to the mean value and the standard deviation;

[0020] The power consumption stability of the current residential power load data is determined based on the coefficient of variation and the coefficient of variation threshold.

[0021] In one embodiment, the feature extraction process is performed on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, including:

[0022] Acquiring current environmental data of the power system to be analyzed;

[0023] Determine a first feature extraction model corresponding to the current residential electricity load data and a second feature extraction model corresponding to the current environmental data;

[0024] Performing feature extraction processing on the current residential electricity load data through the first feature extraction model to obtain a first feature vector of the current residential electricity load data, and performing feature extraction processing on the current environment data through the second feature extraction model to obtain a second feature vector of the current environment data;

[0025] The first feature vector and the second feature vector are concatenated in a preset concatenation order to obtain a concatenated feature vector as the target feature vector.

[0026] In one embodiment, after the first residential power load data and the second residential power load data are fused according to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting model to obtain the predicted residential power load data of the power system to be analyzed, the method further includes:

[0027] Acquire the power system type of the power system to be analyzed;

[0028] Determining a power dispatch instruction prediction model corresponding to the power system type as a target power dispatch instruction prediction model corresponding to the power system to be analyzed;

[0029] Inputting the predicted residential electricity load data into the target power dispatch instruction prediction model to obtain the predicted probability of the power system to be analyzed under each predicted power dispatch instruction;

[0030] From each of the predicted power dispatching instructions, select the predicted power dispatching instruction with the largest predicted probability as the target power dispatching instruction of the power system to be analyzed;

[0031] According to the target power dispatching instruction, power dispatching processing is performed on the power system to be analyzed.

[0032] In one embodiment, the plurality of trained power load prediction models are trained in the following manner:

[0033] Obtain sample residential electricity load data of a sample power system;

[0034] Determining the sample electricity consumption stability of the sample residential electricity load data;

[0035] According to the sample electricity consumption stability, the sample residential electricity load data is divided and processed to obtain a plurality of training data sets corresponding to the sample residential electricity load data;

[0036] According to each training data set, the to-be-trained power load prediction model corresponding to each training data set is iteratively trained to obtain a plurality of trained power load prediction models;

[0037] The hyperparameters of each trained power load forecasting model are fused to obtain the global model parameters;

[0038] According to the global model parameters, each trained power load forecasting model is iteratively updated to obtain each iteratively updated power load forecasting model as the multiple trained power load forecasting models.

[0039] In one embodiment, the iterative training of the to-be-trained power load prediction model corresponding to each training data set is performed according to each training data set to obtain a plurality of trained power load prediction models, including:

[0040] Inputting the training data in each training data set into the power load prediction model to be trained corresponding to each training data set respectively, and obtaining the predicted value of the training data in each training data set output by the power load prediction model to be trained corresponding to each training data set;

[0041] Obtaining the true value of the training data in each training data set, and obtaining the current loss value of the to-be-trained power load prediction model corresponding to each training data set according to the difference between the predicted value and the true value of the training data in each training data set;

[0042] When the current loss value of the power load prediction model to be trained corresponding to each training data set is greater than the preset loss value, the gradient value of the power load prediction model to be trained corresponding to each training data set is determined according to the difference between the predicted value and the true value of the training data in each training data set;

[0043] Performing perturbation processing on the gradient value of the power load prediction model to be trained corresponding to each training data set to obtain the perturbation gradient value of the power load prediction model to be trained corresponding to each training data set;

[0044] According to the disturbance gradient value of the power load prediction model to be trained corresponding to each training data set, the model parameters of the power load prediction model to be trained corresponding to each training data set are adjusted to obtain multiple power load prediction models after the model parameters are adjusted as the multiple trained power load prediction models.

[0045] In a second aspect, the present application also provides a device for predicting residential power load in a power system, comprising:

[0046] A data acquisition module, used to acquire current residential electricity load data of the power system to be analyzed;

[0047] A stability determination module, used to determine the power stability of the current residential power load data;

[0048] A model screening module, used to screen out the power load forecasting model corresponding to the power stability from a plurality of trained power load forecasting models as the target power load forecasting model for the current residential power load data;

[0049] A model prediction module, used for performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, and inputting the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain the first residential electricity load data output by the target electricity load prediction model and the second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models;

[0050] A data fusion module is used to fuse the first residential electricity load data and the second residential electricity load data according to the first weight corresponding to the target electricity load prediction model and the second weight corresponding to the other electricity load prediction models to obtain the predicted residential electricity load data of the power system to be analyzed.

[0051] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0052] Obtain current residential electricity load data of the power system to be analyzed;

[0053] Determining the power consumption stability of the current residential power load data;

[0054] Selecting a power load forecasting model corresponding to the power stability from a plurality of trained power load forecasting models as a target power load forecasting model for the current residential power load data;

[0055] Performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, and inputting the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain first residential electricity load data output by the target electricity load prediction model and second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models;

[0056] According to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are fused to obtain the predicted residential power load data of the power system to be analyzed.

[0057] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0058] Obtain current residential electricity load data of the power system to be analyzed;

[0059] Determining the power consumption stability of the current residential power load data;

[0060] Selecting a power load forecasting model corresponding to the power stability from a plurality of trained power load forecasting models as a target power load forecasting model for the current residential power load data;

[0061] Performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, and inputting the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain first residential electricity load data output by the target electricity load prediction model and second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models;

[0062] According to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are fused to obtain the predicted residential power load data of the power system to be analyzed.

[0063] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0064] Obtain current residential electricity load data of the power system to be analyzed;

[0065] Determining the power consumption stability of the current residential power load data;

[0066] Selecting a power load forecasting model corresponding to the power stability from a plurality of trained power load forecasting models as a target power load forecasting model for the current residential power load data;

[0067] Performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, and inputting the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain first residential electricity load data output by the target electricity load prediction model and second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models;

[0068] According to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are fused to obtain the predicted residential power load data of the power system to be analyzed.

[0069] The above-mentioned power system residential power load prediction method, device, computer equipment, storage medium and computer program product first obtain the current residential power load data of the power system to be analyzed, then determine the power stability of the current residential power load data, and then select the power load prediction model corresponding to the power stability from multiple trained power load prediction models as the target power load prediction model of the current residential power load data, and then perform feature extraction processing on the current residential power load data to obtain the target feature vector of the current residential power load data, and respectively input the target feature vector into the target power load prediction model. The first residential electricity load data output by the target electricity load prediction model and the second residential electricity load data output by the other electricity load prediction models are obtained; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models; finally, according to the first weight corresponding to the target electricity load prediction model and the second weight corresponding to the other electricity load prediction models, the first residential electricity load data and the second residential electricity load data are fused to obtain the predicted residential electricity load data of the power system to be analyzed. In this way, in the process of forecasting residential electricity load, after obtaining the current residential electricity load data of the power system to be analyzed, the electricity stability of the current residential electricity load data is determined, and the corresponding target electricity load forecasting model is screened out, and combined with other electricity load forecasting models, so that the prediction results of multiple models can be integrated, and the predicted residential electricity load data of the power system to be analyzed can be made more accurate, which is conducive to improving the determination accuracy of the predicted residential electricity load data of the power system to be analyzed, thereby improving the prediction accuracy of the residential electricity load; moreover, the entire process does not require human intervention, avoiding the subjective factors in the manual prediction method, which is prone to errors and leads to the defect of low prediction accuracy of the residential electricity load, and further improves the prediction accuracy of the residential electricity load. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 A schematic diagram of a flow chart of a method for predicting residential power load in a power system in one embodiment;

[0072] Figure 2A schematic diagram of a flow chart of a method for predicting residential power load in a power system in another embodiment;

[0073] Figure 3 1 is a flow chart of a method for predicting residential electricity load based on differential privacy and federated learning algorithm in one embodiment;

[0074] Figure 4 1 is a schematic diagram of the structure of a residential electricity load forecasting system based on differential privacy and federated learning algorithm in one embodiment;

[0075] Figure 5 is a flow chart of a residential electricity load forecasting method based on differential privacy and federated learning algorithm in another embodiment;

[0076] Figure 6 is a schematic diagram of four random user load curves in one embodiment;

[0077] Figure 7 A schematic diagram of comparison of prediction results in one embodiment;

[0078] Figure 8 A schematic diagram of comparison of prediction results in another embodiment;

[0079] Fig. 9 A schematic diagram of comparison of prediction results in yet another embodiment;

[0080] Fig.10 A schematic diagram of a comparison of prediction performance between a CAT (Critical Activation Threshold) method and a FedAvg (Federated Averaging Algorithm) algorithm in an embodiment;

[0081] Fig.11 A schematic diagram showing a comparison of prediction performance between the CAT method and the FedAvg algorithm in another embodiment;

[0082] Fig.12 A schematic diagram showing a comparison of prediction performance between the CAT method and the FedAvg algorithm in yet another embodiment;

[0083] Fig.13 A structural block diagram of a device for predicting residential power load in a power system in one embodiment;

[0084] Fig.14 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0086] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0087] In an exemplary embodiment, Figure 1 As shown, a method for predicting residential power load in a power system is provided. This embodiment uses the method applied to a server as an example for illustration; it can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be but is not limited to various personal computers, laptops, smart phones and tablets; the server can be implemented with an independent server or a server cluster composed of multiple servers. In this embodiment, the method includes the following steps:

[0088] Step S101, obtaining current residential power load data of the power system to be analyzed.

[0089] The power system to be analyzed refers to the power system for which power load forecasting is required.

[0090] Among them, the current residential electricity load data refers to various data related to the power load consumed by residential users in the power system to be analyzed at the current time, including power data (such as active power, reactive power, apparent power, etc.) and electricity data (such as instantaneous electricity, accumulated electricity, etc.).

[0091] Exemplarily, the server obtains the current number of users of the candidate power system, and screens out the candidate power systems whose current number of users is greater than the preset number of users from each candidate power system as the power system to be analyzed; then, the server obtains the power system identifier of the power system to be analyzed, and obtains the current residential electricity load data corresponding to the power system identifier from the database as the current residential electricity load data of the power system to be analyzed.

[0092] Step S102, determining the power consumption stability of the current residential power load data.

[0093] Among them, electricity consumption stability is used to indicate the degree to which the current residential electricity load data remains stable within a certain period of time.

[0094] Exemplarily, the server determines the coefficient of variation of the current residential power load data, and based on the correspondence between the coefficient of variation and power stability, obtains the power stability corresponding to the coefficient of variation of the current residential power load data as the power stability of the current residential power load data.

[0095] Step S103, selecting a power load prediction model corresponding to power stability from a plurality of trained power load prediction models as a target power load prediction model for current residential power load data.

[0096] Among them, the electricity load forecasting model refers to a network model that can predict the residential electricity load data of the power system in the future, including the LSTM (Long Short-Term Memory) model, the ARMA (AutoRegressive Moving Average) model, the MA (Moving Average) model, etc.

[0097] Among them, the target electricity load prediction model is used to represent the electricity load prediction model corresponding to the current residential electricity load data.

[0098] Exemplarily, the server selects a power load prediction model corresponding to power stability from multiple trained power load prediction models based on the correspondence between power stability and power load prediction models, and uses the power load prediction model as the target power load prediction model for the current residential power load data.

[0099] For example, when the power consumption stability is high, the corresponding power load forecasting model is the MA model; when the power consumption stability is medium, the corresponding power load forecasting model is the ARMA model; when the power consumption stability is low, the corresponding power load forecasting model is the LSTM model.

[0100] Step S104, perform feature extraction processing on the current residential electricity load data to obtain the target feature vector of the current residential electricity load data, and input the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain the first residential electricity load data output by the target electricity load prediction model, and the second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among multiple trained electricity load prediction models.

[0101] Among them, the target feature vector is used to represent the comprehensive characterization vector of the current residential electricity load data.

[0102] Among them, other power load forecasting models are used to represent power load forecasting models other than the target power load forecasting model among multiple trained power load forecasting models. For example, when multiple trained power load forecasting models are MA models, ARMA models, and LSTM models, and the target power load forecasting model is an LSTM model, other power load forecasting models are MA models and ARMA models.

[0103] The first residential electricity load data is used to represent the predicted value of the residential electricity load data output by the target electricity load prediction model.

[0104] The second residential electricity load data is used to represent the predicted value of the residential electricity load data output by other electricity load prediction models.

[0105] Exemplarily, the server determines the data type of the current residential electricity load data; then, the server queries the correspondence between the data type and the feature extraction model, obtains the feature extraction model corresponding to the data type of the current residential electricity load data, and uses it as the feature extraction model corresponding to the current residential electricity load data; then, the server inputs the current residential electricity load data into the feature extraction model corresponding to the current residential electricity load data for feature extraction processing, and obtains the target feature vector of the current residential electricity load data; then, the server uses the electricity load prediction models other than the target electricity load prediction model among multiple trained electricity load prediction models as other electricity load prediction models; then, the server inputs the target feature vector into the target electricity load prediction model and the other electricity load prediction models, respectively, and obtains the first residential electricity load data output by the target electricity load prediction model, and the second residential electricity load data output by the other electricity load prediction models.

[0106] Step S105, according to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are merged to obtain the predicted residential power load data of the power system to be analyzed.

[0107] Among them, the first weight is used to represent the model weight corresponding to the target power load prediction model.

[0108] The second weight is used to represent the model weights corresponding to other power load prediction models.

[0109] The fusion processing may refer to weighted summation processing.

[0110] The predicted residential electricity load data is used to represent the comprehensive predicted value of the residential electricity load data of the power system to be analyzed in the future.

[0111] Exemplarily, the server determines the prediction accuracy corresponding to the target power load prediction model, and the prediction accuracy corresponding to other power load prediction models; then, the server determines the model weight corresponding to the target power load prediction model according to the prediction accuracy corresponding to the target power load prediction model, as the first weight, and determines the model weights corresponding to other power load prediction models according to the prediction accuracy corresponding to other power load prediction models, as the second weight; then, the server fuses the first residential power load data and the second residential power load data according to the first weight and the second weight, and obtains the fused residential power load data as the predicted residential power load data for the power system to be analyzed.

[0112] In the above-mentioned method for predicting residential electricity load of the power system, the current residential electricity load data of the power system to be analyzed is first obtained, and then the electricity stability of the current residential electricity load data is determined. Then, from multiple trained electricity load prediction models, the electricity load prediction model corresponding to the electricity stability is screened out as the target electricity load prediction model of the current residential electricity load data. Then, the current residential electricity load data is subjected to feature extraction processing to obtain the target feature vector of the current residential electricity load data, and the target feature vector is respectively input into the target electricity load prediction model and other electricity load prediction models to obtain the first residential electricity load data output by the target electricity load prediction model and the second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among multiple trained electricity load prediction models. Finally, according to the first weight corresponding to the target electricity load prediction model and the second weight corresponding to the other electricity load prediction models, the first residential electricity load data and the second residential electricity load data are fused to obtain the predicted residential electricity load data of the power system to be analyzed. In this way, in the process of forecasting residential electricity load, after obtaining the current residential electricity load data of the power system to be analyzed, the electricity stability of the current residential electricity load data is determined, and the corresponding target electricity load forecasting model is screened out, and combined with other electricity load forecasting models, so that the prediction results of multiple models can be integrated, and the predicted residential electricity load data of the power system to be analyzed can be made more accurate, which is conducive to improving the determination accuracy of the predicted residential electricity load data of the power system to be analyzed, thereby improving the prediction accuracy of the residential electricity load; moreover, the entire process does not require human intervention, avoiding the subjective factors in the manual prediction method, which is prone to errors and leads to the defect of low prediction accuracy of the residential electricity load, and further improves the prediction accuracy of the residential electricity load.

[0113] In an exemplary embodiment, the above-mentioned step S101, obtaining the current residential electricity load data of the power system to be analyzed, specifically includes the following contents: obtaining candidate residential electricity load data of the power system to be analyzed at the current time; preprocessing the candidate residential electricity load data to obtain preprocessed residential electricity load data; inputting each preprocessed residential electricity load data into multiple pre-trained importance prediction models to obtain multiple predicted importances of each preprocessed residential electricity load data; fusing the multiple predicted importances of each preprocessed residential electricity load data respectively to obtain the target predicted importance of each preprocessed residential electricity load data; from each preprocessed residential electricity load data, screening out the preprocessed residential electricity load data whose target predicted importance is greater than the preset importance as the current residential electricity load data.

[0114] The candidate residential electricity load data is used to represent the residential electricity load data to be selected.

[0115] The pre-processed residential electricity load data refers to the candidate residential electricity load data after pre-processing.

[0116] Among them, the importance prediction model refers to a network model that can predict the importance of pre-processed residential electricity load data, such as a random forest model.

[0117] The prediction importance is used to indicate the importance of the pre-processed residential electricity load data output by the pre-trained importance prediction model.

[0118] Among them, the target prediction importance is used to represent the comprehensive importance of the pre-processed residential electricity load data.

[0119] The preset importance refers to a preset importance threshold. It should be noted that the preset importance depends on the situation.

[0120] Exemplarily, the server obtains the active power, reactive power, apparent power, instantaneous power and accumulated power of the power system to be analyzed at the current time, and fuses the active power, reactive power and apparent power to obtain the power data of the power system to be analyzed at the current time, and fuses the instantaneous power and accumulated power to obtain the power data of the power system to be analyzed at the current time; then, the server uses the power data and power data of the power system to be analyzed at the current time as candidate residential power load data of the power system to be analyzed at the current time; then, the server identifies the noise data in the candidate residential power load data, and denoises the candidate residential power load data according to the noise data in the candidate residential power load data to obtain the pre-processed The server then inputs each pre-processed residential electricity load data into a plurality of pre-trained importance prediction models, and obtains a plurality of predicted importances of each pre-processed residential electricity load data through the plurality of pre-trained importance prediction models; then, the server fuses the plurality of predicted importances of each pre-processed residential electricity load data according to the model weight of each pre-trained importance prediction model, and obtains a target predicted importance of each pre-processed residential electricity load data; then, the server selects the pre-processed residential electricity load data whose target predicted importance is greater than the preset importance from each pre-processed residential electricity load data, and uses the pre-processed residential electricity load data as the current residential electricity load data.

[0121] In this embodiment, by performing a series of processes such as preprocessing and importance screening on the candidate residential electricity load data of the power system to be analyzed at the current time, more accurate current residential electricity load data can be obtained, which is beneficial to improving the accuracy of determining the current residential electricity load data and provides a good data foundation for subsequent data processing processes.

[0122] In an exemplary embodiment, the above step S102 determines the power consumption stability of the current residential power load data, which specifically includes the following contents: obtaining the mean value and standard deviation of the current residential power load data; determining the coefficient of variation of the current residential power load data based on the mean value and standard deviation; determining the power consumption stability of the current residential power load data based on the coefficient of variation and the coefficient of variation threshold.

[0123] The coefficient of variation is used to represent the numerical value of the dispersion degree of the current residential electricity load data, which is obtained by dividing the standard deviation of the current residential electricity load data by the average value of the current residential electricity load data.

[0124] The coefficient of variation threshold refers to a preset coefficient of variation. It should be noted that the coefficient of variation threshold depends on the situation.

[0125] Exemplarily, the server obtains the mean and variance of the current residential electricity load data, and determines the standard deviation of the current residential electricity load data based on the variance of the current residential electricity load data; then, the server divides the standard deviation of the current residential electricity load data by the mean of the current residential electricity load data, and uses the quotient obtained as the coefficient of variation of the current residential electricity load data; then, based on the coefficient of variation and the coefficient of variation threshold, determines the electricity stability corresponding to the coefficient of variation as the electricity stability of the current residential electricity load data.

[0126] For example, when the coefficient of variation is small, such as less than 0.1, the electricity consumption stability of the current residential electricity load data is determined to be high; when the coefficient of variation is within a certain range, such as greater than 0.1 and less than 0.2, the electricity consumption stability of the current residential electricity load data is determined to be medium; when the coefficient of variation is large, such as greater than 0.2, the electricity consumption stability of the current residential electricity load data is determined to be low.

[0127] In this embodiment, the coefficient of variation is calculated by obtaining the mean value and standard deviation of the current residential electricity load data, so that the fluctuation of the residential electricity load can be converted into a more intuitive numerical indicator, and then the electricity stability of the current residential electricity load data can be more accurately determined, which is conducive to improving the accuracy of determining the electricity stability of the current residential electricity load data.

[0128] In an exemplary embodiment, the above-mentioned step S104, performing feature extraction processing on the current residential electricity load data to obtain the target feature vector of the current residential electricity load data, specifically includes the following contents: obtaining the current environmental data of the power system to be analyzed; determining the first feature extraction model corresponding to the current residential electricity load data, and the second feature extraction model corresponding to the current environmental data; performing feature extraction processing on the current residential electricity load data through the first feature extraction model to obtain the first feature vector of the current residential electricity load data, and performing feature extraction processing on the current environmental data through the second feature extraction model to obtain the second feature vector of the current environmental data; splicing the first feature vector and the second feature vector according to a preset splicing order to obtain a spliced ​​feature vector as the target feature vector.

[0129] Among them, the current environmental data refers to the external environmental information of the power system to be analyzed at the current time, including temperature data, humidity data, wind speed data, precipitation data, etc.

[0130] Among them, the first feature extraction model refers to the feature extraction model corresponding to the current residential electricity load data.

[0131] The second feature extraction model refers to the feature extraction model corresponding to the current environment data.

[0132] Among them, the first eigenvector is used to represent the characterization vector corresponding to the current residential electricity load data.

[0133] The second eigenvector is used to represent the characterization vector corresponding to the current environment data.

[0134] The preset splicing order refers to a pre-set splicing order, such as splicing from left to right. It should be noted that the preset splicing order depends on the situation.

[0135] The concatenated feature vector refers to a feature vector obtained by concatenating the first feature vector and the second feature vector.

[0136] Exemplarily, the server obtains temperature data, humidity data, wind speed data, and precipitation data of the power system to be analyzed at the current time, all of which are used as current environmental data of the power system to be analyzed; then, the server determines a first data type corresponding to the current residential electricity load data, and a second data type corresponding to the current environmental data; then, based on the first data type, the server queries the correspondence between the data type and the feature extraction model, obtains the feature extraction model corresponding to the first data type, as the first feature extraction model corresponding to the current residential electricity load data, and based on the second data type, queries the correspondence between the data type and the feature extraction model, obtains the feature extraction model corresponding to the second data type, as the second feature extraction model corresponding to the current environmental data feature extraction model; then, the server inputs the current residential electricity load data into the first feature extraction model, performs feature extraction processing on the current residential electricity load data through the first feature extraction model, and obtains the first feature vector of the current residential electricity load data, and inputs the current environment data into the second feature extraction model, performs feature extraction processing on the current environment data through the second feature extraction model, and obtains the second feature vector of the current environment data; then, the server splices the first feature vector and the second feature vector according to a preset splicing order to obtain a spliced ​​feature vector; then, the server performs dimensionality reduction processing on the spliced ​​feature vector to obtain a spliced ​​feature vector after dimensionality reduction, and uses the spliced ​​feature vector after dimensionality reduction as the target feature vector.

[0137] In this embodiment, in the process of determining the target characteristic vector of the current residential electricity load data, by combining the current environmental data, it is possible to comprehensively reflect the internal and external influencing factors of the power system operation, and thus obtain a more comprehensive target characteristic vector, which is beneficial to improving the accuracy of determining the target characteristic vector of the current residential electricity load data.

[0138] In an exemplary embodiment, the above-mentioned step S105, after fusing the first residential electricity load data and the second residential electricity load data according to the first weight corresponding to the target electricity load forecasting model and the second weight corresponding to the other electricity load forecasting models to obtain the predicted residential electricity load data of the power system to be analyzed, specifically includes the following contents: obtaining the power system type of the power system to be analyzed; determining the power dispatching instruction prediction model corresponding to the power system type as the target power dispatching instruction prediction model corresponding to the power system to be analyzed; inputting the predicted residential electricity load data into the target power dispatching instruction prediction model to obtain the predicted probability of the power system to be analyzed under each predicted power dispatching instruction; from each predicted power dispatching instruction, screening out the predicted power dispatching instruction with the largest predicted probability as the target power dispatching instruction of the power system to be analyzed; and performing power dispatching processing on the power system to be analyzed according to the target power dispatching instruction.

[0139] Among them, the power system type is used to indicate the system type of the power system to be analyzed, including centralized power generation system, distributed power generation system, and microgrid power system.

[0140] Among them, the power dispatch instruction prediction model refers to a network model that can predict the power dispatch instructions of the power system, such as the constant incremental rate model, the distributed model predictive control model, the mixed integer linear programming model, etc. It should be noted that different types of power systems correspond to different power dispatch instruction prediction models; for example, the centralized power generation system corresponds to the constant incremental rate model, the distributed power generation system corresponds to the distributed power generation system, and the microgrid power system corresponds to the mixed integer linear programming model.

[0141] Among them, the target power dispatching instruction prediction model is used to represent the power dispatching instruction prediction model corresponding to the power system to be analyzed.

[0142] The predicted power dispatch instruction is used to represent the predicted value of the power dispatch instruction output by the target power dispatch instruction prediction model based on the predicted residential power load data.

[0143] The prediction probability is used to indicate the possibility that the target power dispatch instruction prediction model determines that the predicted power dispatch instruction is correct.

[0144] The target power dispatch instruction is used to represent the predicted power dispatch instruction with the highest predicted probability among each predicted power dispatch instruction.

[0145] Exemplarily, the server determines the power system type of the power system to be analyzed based on the power supply characteristics of the power system to be analyzed; for example, centralized power generation systems usually use large power plants as the main power supply, the power supply locations in distributed power generation systems are relatively dispersed, and microgrid power systems contain multiple types of power supplies; then, the server queries the correspondence between the power system type and the power dispatching instruction prediction model, obtains the power dispatching instruction prediction model corresponding to the power system type, and uses the power dispatching instruction prediction model as the target power dispatching instruction prediction model corresponding to the power system to be analyzed; then, the server inputs the predicted residential electricity load data into the target power dispatching instruction prediction model to obtain the predicted probability of the power system to be analyzed under each predicted power dispatching instruction; then, the server screens out the predicted power dispatching instruction with the largest predicted probability from each predicted power dispatching instruction, and uses the predicted power dispatching instruction as the target power dispatching instruction for the power system to be analyzed; then, the server performs power dispatching processing on the power resources in the power system to be analyzed according to the target power dispatching instruction.

[0146] In this embodiment, by accurately adapting the target power dispatch instruction prediction model corresponding to the power system to be analyzed and combining the prediction probability under each predicted power dispatch instruction, the target power dispatch instruction of the power system to be analyzed is made more accurate, thereby improving the quality and effect of power dispatch and achieving optimal allocation of power resources.

[0147] In an exemplary embodiment, the method for predicting residential electricity load in an electric power system provided in the present application also includes the training steps of multiple trained electricity load prediction models, specifically including the following contents: obtaining sample residential electricity load data of a sample power system; determining the sample electricity stability of the sample residential electricity load data; dividing and processing the sample residential electricity load data according to the sample electricity stability to obtain multiple training data sets corresponding to the sample residential electricity load data; iteratively training the to-be-trained electricity load prediction model corresponding to each training data set according to each training data set, to obtain multiple trained electricity load prediction models; fusing the hyperparameters of each trained electricity load prediction model to obtain global model parameters; iteratively updating each trained electricity load prediction model according to the global model parameters to obtain each iteratively updated electricity load prediction model as multiple trained electricity load prediction models.

[0148] The sample power system refers to the power system used to train the power load prediction model.

[0149] Among them, the sample residential electricity load data refers to the residential electricity load data used to train the electricity load prediction model.

[0150] Among them, the sample electricity consumption stability refers to the electricity consumption stability of the sample residential electricity load data.

[0151] The training data set refers to the data set obtained by dividing and processing the sample residential electricity load data.

[0152] The power load prediction model to be trained refers to the power load prediction model that needs to be trained.

[0153] The trained power load prediction model refers to a power load prediction model obtained by iteratively training the to-be-trained power load prediction model corresponding to each training data set.

[0154] Among them, the hyperparameters include the number of neural network layers, the number of neurons, and the size and number of convolution kernels of each trained power load forecasting model.

[0155] The global model parameters refer to the fused hyperparameters obtained by fusion processing the hyperparameters of each trained power load forecasting model.

[0156] The iteratively updated power load forecasting model refers to a power load forecasting model obtained by iteratively updating each trained power load forecasting model.

[0157] Exemplarily, the server determines an associated power system of the sample power system; then, the server obtains historical residential power load data of the sample power system and historical residential power load data of the associated power system, both of which are used as sample residential power load data of the sample power system; then, the server determines the sample power stability of the sample residential power load data; then, the server divides and processes the sample residential power load data according to the sample power stability, and obtains multiple training data sets corresponding to the sample residential power load data; then, the server iteratively trains the to-be-trained power load prediction model corresponding to each training data set according to each training data set, A plurality of trained electricity load prediction models are obtained; then, the server fuses the hyperparameters of each trained electricity load prediction model to obtain the fused hyperparameters as global model parameters; then, the server updates the model parameters of each trained electricity load prediction model according to the global model parameters to obtain the electricity load prediction model after each model parameter is updated, and iteratively trains each electricity load prediction model after each model parameter is updated according to each training data set to obtain each iteratively updated electricity load prediction model, and uses these electricity load prediction models as a plurality of trained electricity load prediction models.

[0158] In this embodiment, by pre-training multiple trained electricity load prediction models, it is convenient to predict the predicted residential electricity load data of the power system to be analyzed after obtaining the current residential electricity load data of the power system to be analyzed in actual applications; moreover, the electricity load prediction model receives new data in each round of iteration, and performs internal improvements and optimizations on the model, so as to facilitate more effective predictions, which is beneficial to improving the prediction accuracy of the electricity load prediction model.

[0159] In an exemplary embodiment, according to each training data set, the to-be-trained power load prediction model corresponding to each training data set is iteratively trained to obtain multiple trained power load prediction models, which specifically includes the following contents: the training data in each training data set is input into the to-be-trained power load prediction model corresponding to each training data set, and the predicted value of the training data in each training data set output by the to-be-trained power load prediction model corresponding to each training data set is obtained; the true value of the training data in each training data set is obtained, and according to the difference between the predicted value and the true value of the training data in each training data set, the current loss value of the to-be-trained power load prediction model corresponding to each training data set is obtained; in each training data set, When the current loss value of the corresponding power load prediction model to be trained is greater than the preset loss value, the gradient value of the power load prediction model to be trained corresponding to each training data set is determined according to the difference between the predicted value and the true value of the training data in each training data set; the gradient value of the power load prediction model to be trained corresponding to each training data set is perturbed to obtain the perturbed gradient value of the power load prediction model to be trained corresponding to each training data set; according to the perturbed gradient value of the power load prediction model to be trained corresponding to each training data set, the model parameters of the power load prediction model to be trained corresponding to each training data set are adjusted to obtain multiple power load prediction models after the model parameters are adjusted, as multiple trained power load prediction models.

[0160] The training data refers to the data in the training data set.

[0161] The predicted value is used to represent the predicted result corresponding to the training data.

[0162] Among them, the true value is used to represent the true result corresponding to the training data.

[0163] The current loss value refers to the loss value obtained based on the difference between the predicted value and the true value of the training data in each training data set.

[0164] The preset loss value refers to a preset loss value threshold. It should be noted that the preset loss value depends on the situation.

[0165] Among them, the gradient value is used to represent the rate of change of the model's loss function at the current parameter point.

[0166] The perturbation gradient value refers to the gradient value after the perturbation process.

[0167] Exemplarily, the server inputs the training data in each training data set into the power load prediction model to be trained corresponding to each training data set, and obtains the predicted value of the training data in each training data set output by the power load prediction model to be trained corresponding to each training data set; then, the server obtains the true value of the training data in each training data set, and according to the difference between the predicted value and the true value of the training data in each training data set, based on the loss function of the power load prediction model to be trained, obtains the current loss value of the power load prediction model to be trained corresponding to each training data set; then, the server judges the current loss value of the power load prediction model to be trained corresponding to each training data set; when the current loss value of the power load prediction model to be trained corresponding to each training data set is greater than the preset loss value, the server judges the current loss value of the power load prediction model to be trained corresponding to each training data set according to the difference between the predicted value and the true value of the training data in each training data set. , determine the gradient value of the power load prediction model to be trained corresponding to each training data set; then, the server generates random noise according to the Gaussian distribution, and according to the random noise, perturbs the gradient value of the power load prediction model to be trained corresponding to each training data set to obtain the perturbed gradient value of the power load prediction model to be trained corresponding to each training data set; then, the server adjusts the model parameters of the power load prediction model to be trained corresponding to each training data set according to the perturbed gradient value of the power load prediction model to be trained corresponding to each training data set, obtains multiple power load prediction models after the model parameters are adjusted, and re-judges the multiple power load prediction models after the model parameters are adjusted; when the current loss values ​​of the multiple power load prediction models after the model parameters are adjusted are less than or equal to the preset loss values, the server uses these power load prediction models as multiple trained power load prediction models.

[0168] In this embodiment, differential privacy technology is used to perturb the gradient value and introduce randomness into model training, so that the model can jump out of the local optimal solution and explore a wider parameter space, avoiding overfitting of noise or specific patterns in the training data, which is beneficial to improving the model's adaptability to new data and thus improving the model's prediction accuracy.

[0169] In an exemplary embodiment, Figure 2 As shown, another method for predicting residential power load in a power system is provided, which is described by taking the method applied to a server as an example, and includes the following steps:

[0170] Step S201, obtaining candidate residential power load data of the power system to be analyzed at the current time; preprocessing the candidate residential power load data to obtain preprocessed residential power load data.

[0171] Step S202: input each pre-processed residential electricity load data into a plurality of pre-trained importance prediction models to obtain a plurality of predicted importances of each pre-processed residential electricity load data.

[0172] Step S203 , fusing the multiple prediction importances of each preprocessed residential electricity load data to obtain the target prediction importance of each preprocessed residential electricity load data.

[0173] Step S204 , from each pre-processed residential electricity load data, the pre-processed residential electricity load data with a target prediction importance greater than a preset importance is selected as the current residential electricity load data.

[0174] Step S205, obtaining the mean value and standard deviation of the current residential power load data; determining the coefficient of variation of the current residential power load data based on the mean value and standard deviation; determining the power stability of the current residential power load data based on the coefficient of variation and the coefficient of variation threshold.

[0175] Step S206, selecting a power load prediction model corresponding to the power stability from a plurality of trained power load prediction models as a target power load prediction model for the current residential power load data.

[0176] Step S207, obtaining current environmental data of the power system to be analyzed; determining a first feature extraction model corresponding to the current residential power load data, and a second feature extraction model corresponding to the current environmental data.

[0177] Step S208, performing feature extraction processing on the current residential electricity load data through the first feature extraction model to obtain the first feature vector of the current residential electricity load data, and performing feature extraction processing on the current environmental data through the second feature extraction model to obtain the second feature vector of the current environmental data.

[0178] Step S209: concatenate the first feature vector and the second feature vector according to a preset concatenation order to obtain a concatenated feature vector as a target feature vector.

[0179] Step S210, respectively input the target feature vector into the target power load prediction model and other power load prediction models to obtain the first residential power load data output by the target power load prediction model and the second residential power load data output by the other power load prediction models; the other power load prediction models are used to represent the power load prediction models other than the target power load prediction model among multiple trained power load prediction models.

[0180] Step S211, according to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are merged to obtain the predicted residential power load data of the power system to be analyzed.

[0181] In the above-mentioned method for predicting residential electricity load of the power system, in the process of predicting residential electricity load, after obtaining the current residential electricity load data of the power system to be analyzed, by determining the electricity stability of the current residential electricity load data, the corresponding target electricity load prediction model is screened out, and combined with other electricity load prediction models, the prediction results of multiple models can be integrated, so that the predicted residential electricity load data of the power system to be analyzed can be made more accurate, which is conducive to improving the determination accuracy of the predicted residential electricity load data of the power system to be analyzed, thereby improving the prediction accuracy of the residential electricity load; moreover, the whole process does not require human intervention, avoiding the defects of subjective factors and easy errors in the manual prediction method, resulting in low prediction accuracy of the residential electricity load, and further improving the prediction accuracy of the residential electricity load.

[0182] In an exemplary embodiment, in order to more clearly illustrate the method for predicting residential electricity load in a power system provided by an embodiment of the present application, the method for predicting residential electricity load in a power system is specifically described below with a specific embodiment. In one embodiment, the present application also provides a method and system for predicting residential electricity load based on differential privacy and federated learning algorithms. In the process of predicting residential electricity load, the current residential electricity load data of the power system to be analyzed is first obtained, and then the electricity stability of the current residential electricity load data is determined. Then, from multiple trained electricity load prediction models, the electricity load prediction model corresponding to the electricity stability is selected as the target electricity load prediction model for the current residential electricity load data. Then, the current residential electricity load data is subjected to feature extraction processing to obtain the target feature vector of the current residential electricity load data, and the target feature vectors are respectively converted into the target feature vectors. Input into the target power load forecasting model and other power load forecasting models, obtain the first residential power load data output by the target power load forecasting model, and the second residential power load data output by other power load forecasting models; other power load forecasting models are used to represent the power load forecasting models other than the target power load forecasting model among multiple trained power load forecasting models. Finally, according to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are fused and processed to obtain the predicted residential power load data of the power system to be analyzed. Specifically include the following contents:

[0183] In one embodiment, a method for predicting residential electricity load based on a federated learning algorithm specifically includes the following steps: Figure 3 As shown:

[0184] 1. Obtain the historical data of residents' load in a period of time before the day to be predicted, and divide the historical data of residents' load through a preset clustering algorithm to obtain several historical load data sets.

[0185] Among them, this step mainly includes: obtaining the historical data of residents' load in a period of time before the day to be predicted from the data set, including historical load parameters and environmental parameters. Historical load parameters include residents' historical electricity consumption information and electricity consumption habits, such as electricity consumption time, electricity load, etc.; environmental parameters include smart meter data, electrical equipment energy consumption, temperature, humidity, etc. By cleaning and feature engineering these data, the prediction ability of the data is further improved. Then, the k-mediods (a clustering algorithm) clustering algorithm is used to divide the data with similar residents' electricity consumption habits into several load history data sets.

[0186] 2. Train the initial neural network model according to each data set in the load history data set, and encrypt the training process through differential privacy technology.

[0187] This step mainly includes: first, input each load history data set into the corresponding initial neural network model, and use the gradient descent algorithm for optimization and update. During the optimization process, each obtained gradient value will be interfered with by differential privacy technology to protect personal data privacy. The goal of training is to minimize the loss function of the model to obtain the corresponding neural network model.

[0188] 3. Obtain the hyperparameters of each neural network model and aggregate these hyperparameters through the federated learning algorithm to obtain the global model parameters.

[0189] This step mainly includes: aggregating the hyperparameters of each neural network model by weighted averaging to obtain the initial global model parameters. Then, iteratively updating each neural network model according to these global parameters until the preset number of iterations is reached. Then, retraining to obtain the second round of neural network models, and reaggregating the hyperparameters of these models to finally obtain the optimized global model parameters.

[0190] 4. Update the initial load forecasting model according to the global model parameters, and optimize it through the optimization algorithm to obtain the final load forecasting model.

[0191] Among them, this step mainly includes: obtaining the loss function used by the neural network model and constructing a global loss function. Through the optimization algorithm, according to the sampling ratio of the data set and the model parameter update, the loss value is calculated until the loss value is less than the preset threshold. At this time, the updated load forecasting model is the final forecasting model.

[0192] 5. Input the historical data of residential load into the load forecasting model, and use the model to output the predicted value of residential electricity load on the day to be predicted.

[0193] The final load forecasting model will accept input from the historical data set, generate and output the power load forecast results for the day to be predicted.

[0194] In one embodiment, a residential electricity load forecasting system based on a federated learning algorithm, the specific structure of the system is as follows: Figure 4 As shown, it includes a data acquisition module, a model training module, a parameter aggregation module, a model optimization module and a load forecasting module.

[0195] The data acquisition module is used to obtain the historical data of residents' load in a period of time before the predicted date, and divide the historical data of residents' load through a preset clustering algorithm to obtain several historical load data sets.

[0196] The data acquisition module includes a data preprocessing unit and a data clustering unit. The data preprocessing unit is used to obtain the initial historical data of residents' loads, which includes historical load parameters and environmental parameters; the historical load parameters include residents' historical electricity consumption information (such as electricity consumption time, electricity load) and electricity consumption habit information (such as electricity consumption characteristics during the day and night); environmental parameters include smart meter data, equipment energy consumption monitoring, temperature, humidity, etc. After data cleaning and feature engineering processing, the data clustering unit divides the data through the k-mediods clustering algorithm, classifies similar historical electricity consumption data into the same category, and forms multiple load history data sets.

[0197] Among them, the model training module is used to train the pre-built initial neural network model according to each load history data set, and encrypt the model training process through differential privacy technology.

[0198] The model training module includes a parameter optimization unit, a privacy protection unit, and a training unit. The parameter optimization unit inputs each load history data set into the initial neural network model and optimizes and updates the model parameters through the gradient descent algorithm; during the parameter update process, the privacy protection unit uses differential privacy technology to perturb the gradient obtained each time to ensure the privacy of user data; the training unit calculates the loss function, optimizes the difference between the model prediction value and the true value, and finally generates a neural network model corresponding to each data set.

[0199] Among them, the parameter aggregation module is used to obtain the hyperparameters of each neural network model, and aggregate these hyperparameters through the federated learning algorithm to obtain the global model parameters.

[0200] The parameter aggregation module includes a parameter update unit and an iterative aggregation unit. The parameter update unit performs weighted averaging on the hyperparameters of each neural network model to obtain the initial global model parameters, and performs multiple iterative updates based on the global parameters until the preset number of iterations is met. The iterative aggregation unit further trains and updates the hyperparameters based on the multiple neural network models obtained in the first round of training, and aggregates these hyperparameters into the final global model parameters through weighted averaging.

[0201] Among them, the model optimization module is used to update the initial neural network model according to the global model parameters, obtain the load forecasting model, and optimize the model through the threshold optimization algorithm.

[0202] Among them, the model optimization module updates the initial neural network model according to the global model parameters and optimizes it in combination with the global loss function. By gradually optimizing the training data of the model and using the optimization algorithm with a preset threshold, the final load forecasting model is generated when the loss value reaches the set standard.

[0203] The load forecasting module is used to input the historical data of residents' load into the load forecasting model, and output the predicted value of residents' electricity load on the day to be predicted through the model.

[0204] Among them, the load forecasting module applies the optimized load forecasting model to the historical data input and generates the predicted value of the residential electricity load on the day to be predicted.

[0205] The residential electricity load forecasting system based on the federated learning algorithm can systematically improve the accuracy of electricity load forecasting through modules such as data collection, data clustering, neural network training, privacy protection, parameter aggregation and model optimization, and effectively use data for global forecasting while protecting user data privacy. Through different stages of model training, different categories of historical residential load data are specially modeled and predicted, which not only improves the accuracy of the forecasting model, but also greatly reduces the risk of privacy leakage during data transmission, ensuring the efficiency and security of the model.

[0206] In one embodiment, a method for predicting residential electricity load based on a federated learning algorithm specifically includes the following steps: Figure 5 As shown:

[0207] 1. Obtain the historical data of residential load in a period of time before the forecast date, and classify the data through a preset clustering algorithm to form several historical data sets.

[0208] Among them, this step mainly includes: first, obtaining residential load parameters and environmental parameters; among them, residential load parameters include residential electricity consumption information and electricity consumption habit information; residential electricity consumption information includes electricity consumption time and electricity load; electricity consumption habit information includes characteristics such as significant electricity consumption during the day, significant electricity consumption at night, and stable fluctuations in electricity consumption throughout the day; environmental parameters include smart meters, electrical equipment energy consumption data, daily electricity consumption curves, electricity consumption time, temperature, humidity and other related information.

[0209] Data is collected through smart meters, sensors or other power monitoring equipment to record residents' electricity consumption, such as electricity consumption, power consumption period, power, etc. After obtaining the data, data cleaning and feature engineering are carried out. Feature engineering involves exploratory analysis of the original data, mining the structure, characteristics and missing conditions of the data, including data visualization and missing value processing. The features are then transformed and standardized, and the most predictive features are selected through methods such as principal component analysis (PCA).

[0210] Next, the clustering algorithm is used to divide the residential electricity consumption data, and residents with similar electricity consumption characteristics are grouped together to form several clusters. Each cluster represents a group of residents with similar electricity consumption patterns. These clusters will serve as clients in federated learning to ensure that data privacy is protected and each client retains local data.

[0211] Specifically, to obtain residential electricity information, for example, based on statistical data, we use energy usage data and meteorological data of 22 households in the past three years, with a time granularity of 24 points per day. The example selects houses with ID (identity) 3-14, 18-20 within a specific time period, with a total of 15 users' data. 80% of the data set of each responding user is used for training, 10% for testing, and 10% for verification.

[0212] After selecting appropriate data, the data needs to be preprocessed. For missing data, this article uses the average value or interpolation method to process it. The specific method is as follows:

[0213]

[0214] Among them, γ i It is the power load value in a certain period of time.

[0215] In addition, in order to speed up the convergence speed of the model training process and improve the training efficiency, the continuous value data will be normalized:

[0216]

[0217] Among them, γ is the original data; γ nom is the normalized data; γ max and γ min are the maximum and minimum values ​​in the series data, respectively.

[0218] The feature selection and feature transformation of the dataset are shown in Table 1.

[0219] Table 1 Feature selection and feature transformation of data sets

[0220]

[0221] 2. Perform local model training on each historical data set and use differential privacy technology to protect the training process.

[0222] This step mainly includes: inputting each historical data set into the pre-built initial LSTM model and updating the model parameters through a local optimization algorithm (such as gradient descent). At the same time, differential privacy technology is introduced to interfere with the gradient value each time the parameters are updated to ensure that the privacy of user data will not be leaked. The error between the model prediction value and the true value is calculated through the loss function, and the optimization is continued until the preset accuracy is reached.

[0223] Specifically, the proposed improved algorithm DCScaffold (a federated learning algorithm) is based on the SCAFFOLD (a federated learning algorithm) algorithm, which comprehensively considers the heterogeneity of user power load data, privacy security, and communication bandwidth during model training. In the rth round of training, first, the client samples mini-batch data to calculate the gradient, and then adds Gaussian noise and correction terms to update the local model parameters. Gaussian noise that meets the privacy budget can further protect the privacy security of client data, and the introduction of correction terms can effectively alleviate the client drift caused by data heterogeneity. After the local model parameters are updated K times, the local control variables are updated.

[0224] 3. Aggregate global model parameters through federated learning algorithm.

[0225] In this step, the model parameters uploaded by each client are aggregated through the central server, and the hyperparameters of each client are aggregated using methods such as weighted average to generate global model parameters. The aggregated parameters will be sent to each client for the next round of training. To ensure the accuracy and generalization ability of the model, iterative updates will continue until the predetermined number of training times or convergence conditions are reached.

[0226] Specifically, based on the CAT strategy, the update parameters are compared with the trigger threshold. If communication is triggered, local updates are uploaded, which can effectively save communication bandwidth. After receiving the client update, the server uses gradient optimization to perform global aggregation updates to generate the next round of global models. If the client update received by the server is empty, the client parameters of the previous round are reused for global updates.

[0227] 4. Update the load forecasting model according to the global model parameters and perform load forecasting.

[0228] In this step, the central server uses the aggregated global model parameters to update the initial load forecasting model, and uses the optimization algorithm to further adjust the model to reduce the forecast error. Finally, the load forecasting model inputs the residents' historical data and outputs the power load forecast value for the forecasted day.

[0229] 5. Evaluate the performance of the model.

[0230] The prediction results are evaluated using evaluation indicators such as mean squared error (MSE), root mean square error (RMSE) and mean absolute percentage error (MAPE) to ensure the accuracy and reliability of the prediction results.

[0231] Through the above steps, this embodiment provides a prediction method based on a federated learning algorithm that can effectively protect data privacy and improve load prediction accuracy. It not only ensures the security of individual data, but also enhances the accuracy of prediction results through the aggregation of global models.

[0232] Specifically, the MSE is calculated as follows:

[0233]

[0234] Specifically, RMSE is calculated as follows:

[0235]

[0236] Specifically, MAPE is calculated as follows:

[0237]

[0238] The current mainstream algorithms involved in the comparison are as follows:

[0239] (1) FedAvg (a federated averaging algorithm): a standard federated averaging algorithm.

[0240] (2) FedAdaGrad (a federated adaptive gradient algorithm): a federated adaptive gradient algorithm.

[0241] (3) DCScaffold: the algorithm proposed in this paper.

[0242] The power load changes of four randomly selected users within a preset time period, such as Figure 6 shown.

[0243] Table 2 Comparison of different algorithms in terms of MSE, RMSE, and MAPE

[0244] algorithm MSE RMSE MAPE FedAvg 0.141 0.362 22.9 FedAdaGrad 0.032 0.178 13.8 DCScaffold 0.019 0.138 9.47

[0245] The prediction results are compared Figure 7 , Figure 8 , Fig. 9As shown in Table 2, the DCScaffold algorithm shows the highest prediction accuracy, whether it is a single user's violently fluctuating load or a relatively gentle load curve of multiple users. Its prediction curve is most consistent with the true value curve, especially the fitting ability of complex and violent fluctuations is more prominent, which verifies that the algorithm effectively overcomes the negative impact of heterogeneous data by introducing an adaptive correction mechanism; although the prediction quality of the FedAdaGrad algorithm is better than that of the classic federated average algorithm, there is still a certain deviation at the peak, and the overall performance is between the other two algorithms; and FedAvg fails to fully deal with Non-IID (Non-Independent and Identically Distributed) data, so its prediction results are the most different from the true value, and there are obvious overestimations or underestimations at multiple time points. The experimental results prove the significant advantages of the improved algorithm DCScaffold over the classic algorithm when facing data heterogeneity.

[0246] Through comparative analysis of the three federated learning algorithms on the three evaluation indicators of MSE, RMSE and MAPE, whether it is mean square error MSE, root mean square error RMSE, or mean absolute percentage error MAPE, the DCScaffold algorithm has the best indicator values, and has the highest prediction accuracy and robustness for heterogeneous data; FedAvg still cannot deal with the problem of data heterogeneity well, and its three indicator values ​​all perform the worst; although the adaptive algorithm FedAdaGrad's various indicator values ​​are better than FedAvg, there is still room for improvement, and its overall performance is between the other two algorithms. These evaluation results are highly consistent with the previous analysis of the algorithm prediction curve, further confirming the significant advantages of DCScaffold over the classic algorithm when facing data heterogeneity.

[0247] Table 3 Comparison between CAT method and FedAvg algorithm

[0248] method MSE RMSE MAPE Save communication FedAvg 0.14 0.37 22.9 0 CAT1 0.019 0.138 9.4 16.2％ CAT2 0.19 0.43 28.3 78.9％ CAT3 0.23 0.48 31.5 83.3％ CAT4 0.45 0.67 45.3 89.7％

[0249] Table 3 Fig.10 , Fig.11 and Fig.12 A comparison between the FedAvg and CAT methods is provided, with results for four cases: CAT1, with a threshold of 0.1%; CAT2, with a threshold of 2%; CAT3, with a threshold of 4% and CAT4, with a threshold of 10%. Fig.10 , Fig.11 and Fig.12The performance of the FedAvg algorithm and the CAT method in terms of MSE, RMSE, and MAPE were compared. From the results, it can be seen that the CAT1 algorithm has the best performance, which is much stronger than the FedAvg algorithm, and saves 16.2% of the communication overhead. Due to the low threshold setting, CAT1 has good model performance but limited communication overhead savings. Although the algorithm performance of CAT2 is slightly lower than that of FedAvg, it saves nearly 80% of the communication overhead, which is very practical for models with large global models and more clients. In addition, the higher the CAT threshold, the more communication is saved, but the lower the global model performance. CAT4 can save nearly 90% of the communication overhead. The use of the CAT method can save about 80% of the communication overhead when the preset accuracy is achieved, which achieves a good compromise between communication bandwidth and model performance.

[0250] This embodiment provides a method for predicting residential electricity load based on a federated learning algorithm. The method clusters the user load data, classifies users with similar electricity usage habits into the same category, and then aggregates the load for each category and trains the prediction model in combination with federated learning. First, by clustering the historical data of residential loads, the user data with similar electricity usage characteristics are divided into several data sets to ensure that the users in each data set have a high degree of similarity in their electricity usage patterns. Then, a federated learning algorithm is used to perform local model training on each clustered category data. During the training process, differential privacy technology is introduced to protect the privacy of user data. Only model parameters that have been processed with differential privacy will be uploaded to the central server, thereby avoiding the leakage of user sensitive information. At the same time, the new loss value judgment method is used to reduce unnecessary update uploads, further saving communication resources and improving training efficiency.

[0251] During the model training process, each client only performs data training and parameter updates locally, and finally aggregates the locally updated parameters through the federated learning algorithm to generate a global load forecasting model. The aggregated model is used to predict residential electricity load, providing reliable decision support for load scheduling and energy management of smart grids. The experimental results show that the proposed method effectively protects the user's data privacy while improving the accuracy and personalization of load forecasting, fully demonstrating the application potential of federated learning in load forecasting, and can provide important technical support and practical value for the optimal scheduling and energy management of smart grids.

[0252] In the above-mentioned embodiment, in the process of predicting the residential electricity load, after obtaining the current residential electricity load data of the power system to be analyzed, the electricity stability of the current residential electricity load data is determined, and the corresponding target electricity load prediction model is screened out, and combined with other electricity load prediction models, so that the prediction results of multiple models can be integrated, and then the predicted residential electricity load data of the power system to be analyzed can be made more accurate, which is conducive to improving the determination accuracy of the predicted residential electricity load data of the power system to be analyzed, thereby improving the prediction accuracy of the residential electricity load; moreover, the whole process does not require human intervention, avoiding the defect of low prediction accuracy of residential electricity load caused by subjective factors and prone to errors in manual prediction, thereby further improving the prediction accuracy of residential electricity load.

[0253] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0254] Based on the same inventive concept, the embodiment of the present application also provides a device for predicting residential power load in a power system for implementing the above-mentioned method for predicting residential power load in a power system. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the device for predicting residential power load in a power system provided below can refer to the limitations of the method for predicting residential power load in a power system above, and will not be repeated here.

[0255] In an exemplary embodiment, Fig.13 As shown, a device for predicting residential power load in a power system is provided, comprising: a data acquisition module 1301, a stability determination module 1302, a model screening module 1303, a model prediction module 1304 and a data fusion module 1305, wherein:

[0256] The data acquisition module 1301 is used to acquire the current residential power load data of the power system to be analyzed.

[0257] The stability determination module 1302 is used to determine the power stability of the current residential power load data.

[0258] The model screening module 1303 is used to screen out a power load forecasting model corresponding to power stability from a plurality of trained power load forecasting models as a target power load forecasting model for current residential power load data.

[0259] The model prediction module 1304 is used to perform feature extraction processing on the current residential electricity load data to obtain the target feature vector of the current residential electricity load data, and input the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain the first residential electricity load data output by the target electricity load prediction model and the second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among multiple trained electricity load prediction models.

[0260] The data fusion module 1305 is used to fuse the first residential electricity load data and the second residential electricity load data according to the first weight corresponding to the target electricity load prediction model and the second weight corresponding to other electricity load prediction models to obtain the predicted residential electricity load data of the power system to be analyzed.

[0261] In an exemplary embodiment, the data acquisition module 1301 is also used to obtain candidate residential electricity load data of the power system to be analyzed at the current time; preprocess the candidate residential electricity load data to obtain preprocessed residential electricity load data; input each preprocessed residential electricity load data into multiple pre-trained importance prediction models to obtain multiple predicted importances of each preprocessed residential electricity load data; fuse the multiple predicted importances of each preprocessed residential electricity load data respectively to obtain a target predicted importance of each preprocessed residential electricity load data; and screen out the preprocessed residential electricity load data whose target predicted importance is greater than the preset importance from each preprocessed residential electricity load data as the current residential electricity load data.

[0262] In an exemplary embodiment, the stability determination module 1302 is also used to obtain the mean value and standard deviation of the current residential electricity load data; determine the coefficient of variation of the current residential electricity load data based on the mean value and standard deviation; and determine the electricity stability of the current residential electricity load data based on the coefficient of variation and the coefficient of variation threshold.

[0263] In an exemplary embodiment, the model prediction module 1304 is also used to obtain current environmental data of the power system to be analyzed; determine a first feature extraction model corresponding to the current residential electricity load data, and a second feature extraction model corresponding to the current environmental data; perform feature extraction processing on the current residential electricity load data through the first feature extraction model to obtain a first feature vector of the current residential electricity load data, and perform feature extraction processing on the current environmental data through the second feature extraction model to obtain a second feature vector of the current environmental data; splice the first feature vector and the second feature vector in accordance with a preset splicing order to obtain a spliced ​​feature vector as a target feature vector.

[0264] In an exemplary embodiment, the power system residential electricity load prediction device also includes a power dispatching module, which is used to obtain the power system type of the power system to be analyzed; determine the power dispatching instruction prediction model corresponding to the power system type as the target power dispatching instruction prediction model corresponding to the power system to be analyzed; input the predicted residential electricity load data into the target power dispatching instruction prediction model to obtain the prediction probability of the power system to be analyzed under each predicted power dispatching instruction; from each predicted power dispatching instruction, screen out the predicted power dispatching instruction with the largest prediction probability as the target power dispatching instruction of the power system to be analyzed; and perform power dispatching processing on the power system to be analyzed according to the target power dispatching instruction.

[0265] In an exemplary embodiment, the power system residential electricity load prediction device also includes a model training module, which is used to obtain sample residential electricity load data of a sample power system; determine the sample electricity stability of the sample residential electricity load data; divide and process the sample residential electricity load data according to the sample electricity stability to obtain multiple training data sets corresponding to the sample residential electricity load data; iteratively train the to-be-trained electricity load prediction model corresponding to each training data set according to each training data set, to obtain multiple trained electricity load prediction models; fuse the hyperparameters of each trained electricity load prediction model to obtain global model parameters; iteratively update each trained electricity load prediction model according to the global model parameters to obtain each iteratively updated electricity load prediction model as multiple trained electricity load prediction models.

[0266] In an exemplary embodiment, the model training module is further used to input the training data in each training data set into the power load prediction model to be trained corresponding to each training data set, respectively, to obtain the predicted value of the training data in each training data set output by the power load prediction model to be trained corresponding to each training data set; to obtain the true value of the training data in each training data set, and to obtain the current loss value of the power load prediction model to be trained corresponding to each training data set based on the difference between the predicted value and the true value of the training data in each training data set; and when the current loss value of the power load prediction model to be trained corresponding to each training data set is greater than the preset loss value Under the present invention, according to the difference between the predicted value and the true value of the training data in each training data set, the gradient value of the power load prediction model to be trained corresponding to each training data set is determined; the gradient value of the power load prediction model to be trained corresponding to each training data set is perturbed to obtain the perturbed gradient value of the power load prediction model to be trained corresponding to each training data set; according to the perturbed gradient value of the power load prediction model to be trained corresponding to each training data set, the model parameters of the power load prediction model to be trained corresponding to each training data set are adjusted to obtain multiple power load prediction models after the model parameters are adjusted, as multiple trained power load prediction models.

[0267] Each module in the above-mentioned power system residential power load prediction device can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0268] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.14As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store current residential power load data, power stability and other data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting residential power load in a power system is implemented.

[0269] Those skilled in the art will understand that Fig.14 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0270] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0271] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0272] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0273] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0274] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0275] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting residential power load in a power system, characterized in that: The method comprises: Obtain current residential electricity load data of the power system to be analyzed; Determining the power consumption stability of the current residential power load data; Selecting a power load forecasting model corresponding to the power stability from a plurality of trained power load forecasting models as a target power load forecasting model for the current residential power load data; Performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, and inputting the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain first residential electricity load data output by the target electricity load prediction model and second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models; According to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models, the first residential power load data and the second residential power load data are fused to obtain the predicted residential power load data of the power system to be analyzed.

2. The method according to claim 1, characterized in that: The obtaining of current residential electricity load data of the power system to be analyzed includes: Obtain candidate residential electricity load data of the power system to be analyzed at the current time; Preprocessing the candidate residential electricity load data to obtain preprocessed residential electricity load data; Inputting each of the pre-processed residential electricity load data into a plurality of pre-trained importance prediction models to obtain a plurality of predicted importances of each of the pre-processed residential electricity load data; fusing the plurality of predicted importances of each of the preprocessed residential electricity load data to obtain the target predicted importance of each of the preprocessed residential electricity load data; From each of the pre-processed residential electricity load data, the pre-processed residential electricity load data whose target prediction importance is greater than a preset importance is screened out as the current residential electricity load data.

3. The method according to claim 1, characterized in that The step of determining the power consumption stability of the current residential power load data includes: Obtaining the average value and standard deviation of the current residential electricity load data; Determining the coefficient of variation of the current residential electricity load data according to the mean value and the standard deviation; The power consumption stability of the current residential power load data is determined based on the coefficient of variation and the coefficient of variation threshold.

4. The method according to claim 1, characterized in that: The performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data includes: Acquiring current environmental data of the power system to be analyzed; Determine a first feature extraction model corresponding to the current residential electricity load data and a second feature extraction model corresponding to the current environmental data; Performing feature extraction processing on the current residential electricity load data through the first feature extraction model to obtain a first feature vector of the current residential electricity load data, and performing feature extraction processing on the current environment data through the second feature extraction model to obtain a second feature vector of the current environment data; The first feature vector and the second feature vector are concatenated in a preset concatenation order to obtain a concatenated feature vector as the target feature vector.

5. The method according to claim 1, characterized in that After fusing the first residential power load data and the second residential power load data according to the first weight corresponding to the target power load forecasting model and the second weight corresponding to the other power load forecasting models to obtain the predicted residential power load data of the power system to be analyzed, the method further includes: Acquire the power system type of the power system to be analyzed; Determining a power dispatch instruction prediction model corresponding to the power system type as a target power dispatch instruction prediction model corresponding to the power system to be analyzed; Inputting the predicted residential electricity load data into the target power dispatch instruction prediction model to obtain the predicted probability of the power system to be analyzed under each predicted power dispatch instruction; From each of the predicted power dispatching instructions, select the predicted power dispatching instruction with the largest predicted probability as the target power dispatching instruction of the power system to be analyzed; According to the target power dispatching instruction, power dispatching processing is performed on the power system to be analyzed.

6. The method according to any one of claims 1 to 5, characterized in that: The multiple trained power load prediction models are trained in the following manner: Obtain sample residential electricity load data of a sample power system; Determining the sample electricity consumption stability of the sample residential electricity load data; According to the sample electricity consumption stability, the sample residential electricity load data is divided and processed to obtain a plurality of training data sets corresponding to the sample residential electricity load data; According to each training data set, the to-be-trained power load prediction model corresponding to each training data set is iteratively trained to obtain a plurality of trained power load prediction models; The hyperparameters of each trained power load forecasting model are fused to obtain the global model parameters; According to the global model parameters, each trained power load forecasting model is iteratively updated to obtain each iteratively updated power load forecasting model as the multiple trained power load forecasting models.

7. The method according to claim 6, characterized in that The method of iteratively training the power load prediction model to be trained corresponding to each training data set according to each training data set to obtain multiple trained power load prediction models includes: Inputting the training data in each training data set into the power load prediction model to be trained corresponding to each training data set respectively, and obtaining the predicted value of the training data in each training data set output by the power load prediction model to be trained corresponding to each training data set; Obtaining the true value of the training data in each training data set, and obtaining the current loss value of the to-be-trained power load prediction model corresponding to each training data set according to the difference between the predicted value and the true value of the training data in each training data set; When the current loss value of the power load prediction model to be trained corresponding to each training data set is greater than the preset loss value, the gradient value of the power load prediction model to be trained corresponding to each training data set is determined according to the difference between the predicted value and the true value of the training data in each training data set; Performing perturbation processing on the gradient value of the power load prediction model to be trained corresponding to each training data set to obtain the perturbation gradient value of the power load prediction model to be trained corresponding to each training data set; According to the disturbance gradient value of the power load prediction model to be trained corresponding to each training data set, the model parameters of the power load prediction model to be trained corresponding to each training data set are adjusted to obtain multiple power load prediction models after the model parameters are adjusted as the multiple trained power load prediction models.

8. A device for predicting residential power load in a power system, characterized in that: The device comprises: A data acquisition module, used to acquire current residential electricity load data of the power system to be analyzed; A stability determination module, used to determine the power stability of the current residential power load data; A model screening module, used to screen out the power load forecasting model corresponding to the power stability from a plurality of trained power load forecasting models as the target power load forecasting model for the current residential power load data; A model prediction module, used for performing feature extraction processing on the current residential electricity load data to obtain a target feature vector of the current residential electricity load data, and inputting the target feature vector into the target electricity load prediction model and other electricity load prediction models respectively to obtain the first residential electricity load data output by the target electricity load prediction model and the second residential electricity load data output by the other electricity load prediction models; the other electricity load prediction models are used to represent the electricity load prediction models other than the target electricity load prediction model among the multiple trained electricity load prediction models; A data fusion module is used to fuse the first residential electricity load data and the second residential electricity load data according to the first weight corresponding to the target electricity load prediction model and the second weight corresponding to the other electricity load prediction models to obtain the predicted residential electricity load data of the power system to be analyzed.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images