AI-based power selling enterprise power declaration dynamic allocation method and system

Abstract

The application relates to the technical field of dynamic allocation of declaration, and provides an AI-based power selling enterprise power declaration dynamic allocation method and system, which solves the problems of low accuracy and efficiency of power selling enterprise power declaration dynamic allocation. The application periodically collects active power, reactive power and harmonic parameters of users, extracts power fluctuation characteristics through waveform transformation, and constructs a user power image library; the feature data in the image library is compressed to obtain power feature compression data; a decentralized demand prediction model based on federated learning is used to aggregate the compression data of all users to obtain overall power prediction; based on the overall prediction, a dynamic programming is used to set a time-of-use electricity price and an electricity allocation strategy, a real-time market electricity price is combined, a peak-valley period power declaration ratio is dynamically adjusted, power selling enterprise power declaration dynamic allocation is realized, and the accuracy and efficiency of power selling enterprise power declaration dynamic allocation can be improved.

Description

Technical Field

[0001] This application relates to the field of dynamic allocation technology, and in particular to an AI-based method and system for dynamic allocation of electricity declared by electricity sales companies. Background Technology

[0002] In AI-based scenarios for dynamic allocation of electricity consumption declarations by power retailers, the technological demands primarily stem from the volatility of electricity load and the dynamic nature of the electricity market. With the increasing integration of distributed energy resources and the widespread adoption of smart devices on the user side, the electricity consumption behavior of residential and commercial users exhibits multi-period, high-frequency fluctuations. Meanwhile, electricity market prices change in real time according to supply and demand. Power retailers need to accurately capture user electricity consumption trends and dynamically adjust electricity declarations during peak and off-peak periods to minimize electricity purchase and sales costs, balance supply and demand, and improve user satisfaction. Simultaneously, they must consider both data privacy protection and the real-time nature of allocation strategies.

[0003] Currently, to address the aforementioned needs, existing solutions mostly employ a centralized AI model-based electricity reporting and allocation method. This solution involves deploying a unified data acquisition platform to aggregate user-side electricity consumption data (such as smart meter data and historical load curves) and market electricity price data. A load prediction model (such as an LSTM neural network) is trained in the cloud, and an electricity reporting and allocation strategy is generated based on the prediction results and fixed threshold rules. The strategy is then distributed to the electricity sales company's business system for execution.

[0004] The existing solution has significant drawbacks: on the one hand, the centralized data collection and model training mode requires uploading a large amount of user electricity consumption data to the cloud, which poses a risk of user privacy leakage and makes it difficult to meet data security regulations; on the other hand, the fixed threshold rule lacks dynamic adaptability. When user electricity consumption behavior fluctuates suddenly or market electricity prices change drastically, the model prediction deviation will be amplified, resulting in a mismatch between the electricity declaration and the actual demand, which in turn leads to increased electricity purchase costs or power supply shortages. Summary of the Invention

[0005] The purpose of this application is to provide an AI-based method, system, electronic device, and storage medium for dynamic allocation of electricity consumption reported by electricity retailers, in order to solve the problems of accuracy and efficiency in dynamic allocation of electricity consumption reported by electricity retailers caused by privacy leaks, poor adaptability, and supply-demand mismatch in the existing technology.

[0006] To address the aforementioned technical issues, firstly, this application provides an AI-based method for dynamic allocation of electricity consumption reported by electricity retailers, comprising:

[0007] The active power parameters, reactive power parameters, and harmonic parameters of each user are periodically collected. Waveform transformation is used to extract the user's electricity consumption fluctuation characteristics from the active power parameters, reactive power parameters, and harmonic parameters in order to construct a user electricity consumption profile database.

[0008] The feature data in the user electricity consumption profile database is subjected to feature compression processing to obtain compressed electricity consumption feature data;

[0009] Using a distributed demand forecasting model based on a federated learning framework, the compressed electricity consumption data corresponding to all users are aggregated to obtain the overall electricity consumption forecast of the electricity sales company.

[0010] Based on the overall electricity consumption forecast, a time-of-use pricing rule and an electricity allocation strategy are set using a dynamic programming algorithm. Based on the time-of-use pricing rule and the electricity allocation strategy, and in conjunction with the real-time market electricity price, the allocation ratio of the electricity declaration volume of electricity sales companies during peak and off-peak periods is dynamically adjusted to achieve dynamic allocation of electricity declarations by electricity sales companies.

[0011] Optionally, the distributed demand forecasting model includes a local forecasting model and a global forecasting model;

[0012] The decentralized demand forecasting model based on a federated learning framework aggregates the compressed electricity consumption data corresponding to all users to obtain the overall electricity consumption forecast for the electricity sales company, including:

[0013] Each user is assigned to a corresponding local node, and a local prediction model is generated based on the local electricity consumption characteristic data of the corresponding user in each local node.

[0014] The parameters of the local prediction model are input to the central node of the federated learning framework, and the parameters of all the local prediction models are merged by the central node to obtain the global prediction model.

[0015] The global prediction model is used to aggregate the compressed electricity consumption characteristic data of each user to obtain the predicted electricity consumption of each user.

[0016] By summing up the electricity consumption forecasts of all users, the overall electricity consumption forecast of the electricity sales company can be obtained.

[0017] Optionally, the step of merging the parameters of all the local prediction models through the central node to obtain a global prediction model includes:

[0018] The central node calculates the merging weight of the local electricity consumption feature data based on the overlap of the coverage area of ​​the corresponding user of each local node and the number of data records.

[0019] The parameters of the local prediction model corresponding to each local node are proportionally adjusted according to the merging weights to obtain the adjusted parameters of the local prediction model corresponding to each local node.

[0020] Integrate similar parameters from all adjusted local prediction models to form unified parameters for the global model;

[0021] All the unified parameters of the global model are loaded into a preset model structure to obtain a global prediction model.

[0022] Optionally, the step of performing feature compression processing on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data includes:

[0023] Extract the feature data of each user from the user electricity consumption profile database, and determine the multiple feature items contained in the feature data of each user;

[0024] Assess the degree of influence of each of the aforementioned features on reflecting the fluctuation characteristics of user electricity consumption, and rank each of the aforementioned features from high to low according to the degree of influence.

[0025] Based on the ranking results and a pre-defined number, features with high influence are selected as retained features.

[0026] Extract the values ​​corresponding to the retained feature items from the feature data of each user, combine the values ​​to form simplified feature data, and determine the electricity consumption feature compressed data based on the simplified feature data of all users.

[0027] Optionally, the evaluation of the influence of each of the aforementioned features on reflecting the fluctuation characteristics of user electricity consumption, and the ranking of the aforementioned features from high to low according to the degree of influence, includes:

[0028] Count the number of times each of the aforementioned feature items is associated with the user's electricity consumption fluctuation characteristics in the feature data of all users;

[0029] Based on the number of times the association occurs, an influence value is assigned to each of the aforementioned features, wherein the number of times the association occurs is directly proportional to the influence value.

[0030] Compare the magnitudes of the influence values ​​of each feature item, and sort the feature items in descending order of influence values ​​to obtain the sorting result.

[0031] Optionally, the step of extracting the user's electricity consumption fluctuation characteristics from the active power parameters, reactive power parameters, and harmonic parameters using waveform transformation to construct a user electricity consumption profile database includes:

[0032] The active power parameters, reactive power parameters and harmonic parameters are respectively subjected to waveform transformation to obtain waveform change information of each parameter at multiple consecutive time points;

[0033] Extract the magnitude of increase, magnitude of decrease, and number of times the corresponding parameters change per unit time from the waveform change information.

[0034] The power consumption fluctuation characteristics of the user are generated by combining the numerical increase and decrease of the active power parameters, reactive power parameters and harmonic parameters corresponding to the same user, respectively, as well as the number of numerical changes per unit time.

[0035] All users' electricity consumption fluctuation characteristics are associated and recorded according to each user's unique identifier, and a set containing each user's electricity consumption fluctuation characteristics is constructed. This set is then defined as the user electricity consumption profile library.

[0036] Optionally, the step of setting time-of-use pricing rules and power allocation strategies based on the overall electricity consumption forecast using a dynamic programming algorithm, and dynamically adjusting the allocation ratio of electricity sales companies' declared power volume during peak and off-peak periods based on the time-of-use pricing rules and the power allocation strategy, combined with real-time market electricity prices, includes:

[0037] Based on the overall electricity consumption forecast, the total predicted electricity consumption for each peak and valley period and the electricity consumption percentage for each period are calculated.

[0038] Using a dynamic programming algorithm, based on the predicted total electricity consumption and the electricity consumption ratio, time-of-use pricing rules and electricity allocation strategies are determined. The time-of-use pricing rules are different electricity price standards for different time periods, and the electricity allocation strategy is that each time period has its own electricity allocation limit.

[0039] The current electricity price during each peak and valley period in the market is obtained in real time, and the current electricity price is compared with the electricity price standard for the corresponding period.

[0040] When the comparison result indicates that the current electricity price is higher than the electricity price standard for the corresponding time period, the proportion of electricity declaration for the corresponding time period is reduced in conjunction with the electricity allocation limit; or, when the comparison result indicates that the current electricity price is lower than the electricity price standard for the corresponding time period, the proportion of electricity declaration for the corresponding time period is increased in conjunction with the electricity allocation limit.

[0041] By integrating the adjusted electricity declaration proportions corresponding to each peak and valley period, the electricity declaration allocation ratio of electricity sales companies is obtained.

[0042] Secondly, this application provides an AI-based dynamic allocation system for electricity sales companies based on electricity demand declarations, including:

[0043] The module is used to periodically collect the active power parameters, reactive power parameters and harmonic parameters of each user, and to extract the user's power consumption fluctuation characteristics from the active power parameters, reactive power parameters and harmonic parameters by waveform transformation, so as to build a user power consumption profile library.

[0044] The processing module is used to perform feature compression processing on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data.

[0045] The aggregation module is used to aggregate the compressed electricity consumption data corresponding to all users using a distributed demand forecasting model based on a federated learning framework, so as to obtain the overall electricity consumption forecast of the electricity sales company.

[0046] The adjustment module is used to set time-of-use pricing rules and power allocation strategies based on the overall electricity consumption forecast using a dynamic programming algorithm. Based on the time-of-use pricing rules and the power allocation strategies, and combined with the real-time market electricity price, the module dynamically adjusts the allocation ratio of the power sales enterprise's declared power volume during peak and valley periods, so as to realize the dynamic allocation of the power sales enterprise's declared power volume.

[0047] Thirdly, this application provides an electronic device, comprising:

[0048] Memory, used to store computer programs;

[0049] A processor is configured to execute the computer program to implement the steps of the AI-based dynamic allocation method for electricity sales enterprise declaration as described in the first aspect above.

[0050] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the AI-based dynamic allocation method for electricity sales enterprise declaration as described in the first aspect above.

[0051] This application provides an AI-based method for dynamic allocation of electricity consumption declarations by electricity retailers. The method includes: periodically collecting active power parameters, reactive power parameters, and harmonic parameters from each user; extracting user electricity consumption fluctuation characteristics from these parameters using waveform transformation to construct a user electricity consumption profile database; performing feature compression processing on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data; aggregating the compressed electricity consumption feature data corresponding to all users using a distributed demand forecasting model based on a federated learning framework to obtain the overall electricity consumption forecast for the electricity retailers; based on the overall electricity consumption forecast, setting time-of-use pricing rules and electricity allocation strategies using a dynamic programming algorithm; and dynamically adjusting the allocation ratio of the electricity consumption declarations of the electricity retailers during peak and off-peak periods based on the time-of-use pricing rules and electricity allocation strategies, combined with real-time market prices, to achieve dynamic allocation of electricity consumption declarations by the electricity retailers.

[0052] The beneficial effects of this application are:

[0053] The AI-based dynamic allocation method for electricity sales enterprise electricity reporting provided in this application periodically collects active power parameters, reactive power parameters, and harmonic parameters from each user. It then uses waveform transformation to extract electricity consumption fluctuation characteristics from these parameters to construct a user electricity consumption profile database. This method accurately captures differences in user electricity consumption patterns, providing structured and characteristic foundational data for subsequent data processing. By performing feature compression processing on the characteristic data in the user electricity consumption profile database to obtain compressed electricity consumption feature data, it can reduce the data volume while retaining key electricity consumption characteristics, improving data transmission and computation efficiency. Furthermore, by utilizing a federated... The distributed demand forecasting model of the learning framework aggregates compressed data of electricity consumption characteristics corresponding to all users to obtain the overall electricity consumption forecast of electricity sales companies. It can achieve accurate aggregation of electricity consumption trends of each user while protecting user data privacy, and provide a reliable forecast basis for global allocation. By setting time-of-use pricing rules and electricity allocation strategies based on the overall electricity consumption forecast, and dynamically adjusting the allocation ratio of electricity sales companies' electricity declarations during peak and valley periods in combination with real-time market electricity prices, the electricity declaration strategy can adapt to the differences in peak and valley electricity consumption and market price fluctuations, optimize the cost of purchasing and selling electricity and ensure supply and demand balance.

[0054] Furthermore, the distributed demand forecasting model includes local and global forecasting models. First, each user is assigned to a corresponding local node, and a local forecasting model is generated based on local electricity consumption characteristic data. Then, the parameters of the local models are input into the central node of the federated learning framework. The central node calculates the merging weights based on the overlap of user electricity consumption characteristic coverage and the number of data records at each local node. After adjusting the local model parameters according to the weights, similar parameters are integrated to form unified parameters for the global model. This unified parameter is then loaded into a preset structure to obtain the global forecasting model. Finally, the global forecasting model is used to predict the electricity consumption of each user, and the results are aggregated to obtain the overall electricity consumption forecast. By constructing local and global models in a hierarchical manner and optimizing parameter integration through weight adjustment, the adaptability and accuracy of the overall electricity consumption forecast to different user electricity consumption characteristics are improved. Furthermore, the federated learning framework avoids the privacy risks caused by centralized processing of user data, enhancing the model's practicality and security.

[0055] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1A flowchart illustrating an AI-based dynamic allocation method for electricity sales enterprise electricity declaration provided in this application embodiment;

[0058] Figure 2 A schematic diagram illustrating a specific implementation of an AI-based dynamic allocation method for electricity sales enterprise declarations, provided in this application embodiment;

[0059] Figure 3 This is a schematic diagram of the structure of an AI-based dynamic allocation system for electricity sales enterprise electricity reporting, provided as an embodiment of this application. Detailed Implementation

[0060] To address the privacy leaks, poor adaptability, and supply-demand mismatch issues in existing technologies, this application provides an AI-based method for dynamic allocation of electricity consumption declarations by power retailers. This method employs the following design concept: periodically collecting electricity consumption-related data from users, extracting features reflecting user electricity consumption fluctuations, and forming user electricity consumption profiles; compressing the feature data in these profiles to obtain simplified electricity consumption feature data; integrating the simplified electricity consumption feature data of all users using a distributed prediction model to obtain the overall electricity consumption prediction result for the power retailer; and then, based on this prediction result, setting time-of-use pricing rules and electricity allocation strategies using a flexible calculation method, and dynamically adjusting the electricity consumption declaration ratio during peak and off-peak periods in conjunction with real-time market prices to achieve dynamic allocation of electricity consumption declarations.

[0061] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] The core of this application is to provide an AI-based method for dynamic allocation of electricity consumption reported by power sales companies. A flowchart illustrating one specific implementation is shown below. Figure 1 As shown, the method includes:

[0063] S11. Periodically collect the active power parameters, reactive power parameters, and harmonic parameters of each user, and use waveform transformation to extract the user's power consumption fluctuation characteristics from the active power parameters, reactive power parameters, and harmonic parameters in order to build a user power consumption profile database.

[0064] Among them, active power parameters are the actual power consumption data used to drive electrical appliances when users use electricity; reactive power parameters are the power data used to maintain the operation of electrical appliances but not to directly perform work; harmonic parameters are irregular fluctuation data of current or voltage waveforms that deviate from the standard shape; waveform transformation is a processing method that decomposes complex electricity consumption data into fluctuation components of different time periods and extracts key change characteristics; electricity consumption fluctuation characteristics are the changing patterns of user electricity consumption (such as the difference in electricity consumption during peak and valley periods); and the user electricity consumption profile database is a dataset that reflects user electricity consumption habits after integrating all user electricity consumption fluctuation characteristics.

[0065] In this embodiment, active power parameters, reactive power parameters, and harmonic parameters of each user are collected at fixed intervals. For example, the user's three types of power data are recorded at 8:00 AM, 12:00 PM, 7:00 PM, and 12:00 AM every day. These parameters are processed using waveform transformation. For example, the data of a user for a day is divided into fluctuation components of different time periods, and the characteristics of power consumption changes during the morning peak and the nighttime trough are extracted. Finally, the power consumption fluctuation characteristics of all users are integrated to build a user power consumption profile database. That is, the system stores information such as the peak and valley power consumption patterns and fluctuation magnitudes of each user, forming the basic data for subsequent processing.

[0066] S12. Perform feature compression processing on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data.

[0067] Among them, feature compression processing is a method of reducing the total amount of data while retaining key information on power allocation; power consumption feature compressed data is simplified data that can still accurately reflect the key characteristics of users' peak and off-peak power consumption after compression.

[0068] In this embodiment of the application, key features related to peak-valley scheduling are selected from the user electricity consumption profile database, such as peak fluctuation amplitude and valley stability duration; duplicate or secondary information is removed through feature compression processing, such as records of minor fluctuations that have minimal impact on scheduling; finally, compressed electricity consumption feature data is obtained, that is, data with reduced data volume but clear core electricity consumption features.

[0069] S13. Using a distributed demand forecasting model based on a federated learning framework, the compressed electricity consumption data corresponding to all users are aggregated to obtain the overall electricity consumption forecast of the electricity sales company.

[0070] Among them, the federated learning framework is a technical framework that allows data to be processed locally and only uploads model parameters without leaking the original data; the distributed demand forecasting model is a forecasting model under the federated learning framework in which the data is processed separately by each user and then aggregated; the overall electricity consumption forecast is the estimated total electricity demand that the electricity sales company needs to cope with after combining the electricity consumption forecasts of all users.

[0071] In this embodiment of the application, based on the federated learning framework, the compressed data of each user's electricity consumption characteristics are distributed to the corresponding local processing nodes. Each node processes the data using the local part of the distributed demand forecasting model. Each local node only uploads the model parameters to the central node of the electricity sales company. The central node aggregates all parameters to form a global forecasting model. The global model is used to integrate the electricity consumption forecasts of all users to obtain the overall electricity consumption forecast of the electricity sales company.

[0072] S14. Based on the overall electricity consumption forecast, use dynamic programming algorithm to set time-of-use pricing rules and electricity allocation strategies. Based on the time-of-use pricing rules and electricity allocation strategies, combined with real-time market electricity prices, dynamically adjust the allocation ratio of electricity sales companies' declared electricity volume during peak and valley periods to achieve dynamic allocation of electricity sales companies' declared electricity volume.

[0073] Among them, dynamic programming algorithm is a calculation method that optimizes decision-making in multiple time periods by flexibly adjusting strategies; time-of-use pricing rules are rules that set different electricity prices according to peak and valley periods; electricity allocation strategy is a scheme to determine the proportion of electricity declared during different time periods; peak and valley period allocation ratio is the proportion of electricity declared during peak and valley periods to the total declared amount.

[0074] In this embodiment, based on the overall electricity consumption forecast, a dynamic programming algorithm is used to analyze the proportion of electricity demand in different time periods; time-of-use pricing rules (higher price during peak hours and lower price during off-peak hours) and electricity allocation strategies are set according to the proportion; and the peak-valley declaration ratio is adjusted in conjunction with the real-time market electricity price (such as lowering the declaration ratio when the peak electricity price increases) to achieve dynamic allocation of electricity declarations.

[0075] This application provides the following specific example: First, electricity consumption data for users A, B, and C is collected periodically at different times, such as morning and evening. This includes the actual power used to drive appliances, the power required to maintain appliance operation, and irregular fluctuations in current and voltage. This data is then processed to identify the characteristics of electricity consumption changes at different times. For example, it is found that A's electricity consumption fluctuates more significantly in the evening, while B's electricity consumption is more stable at night. These characteristics are then integrated to form a profile database reflecting the electricity consumption habits of community users. Next, key information related to peak and off-peak electricity allocation is selected from this profile database, such as the fluctuations in each user's electricity consumption during peak hours and the duration of stable electricity consumption during off-peak hours. Redundant or minor information that has little impact on allocation is removed, reducing the data volume to obtain compressed data that still clearly reflects the characteristics of peak and off-peak electricity consumption. Then, using a corresponding technical framework, each user's compressed data is processed on their respective local nodes, and their electricity demand is predicted based on their peak and off-peak electricity consumption characteristics. Only the model parameters used in the processing are then uploaded to the central node of the community electricity sales company. The central node integrates these parameters to form a unified prediction model, and then aggregates the electricity demand prediction results for all users to obtain the overall electricity consumption forecast for the community. Finally, based on the overall electricity consumption forecast, the proportion of electricity demand in different time periods is analyzed, such as relatively high demand during peak hours and relatively low demand during off-peak hours. Based on this, different electricity pricing rules and electricity declaration ratio schemes are set for different time periods, such as slightly higher prices and appropriately adjusted declaration ratios during peak hours, and vice versa during off-peak hours. Combined with real-time market electricity prices, the electricity declaration ratios for peak and off-peak periods are flexibly adjusted to achieve dynamic allocation of electricity declarations by the community electricity sales company.

[0076] By executing S11~S14, this embodiment of the application periodically collects multiple types of power parameters and extracts electricity consumption characteristics to construct an accurate user electricity consumption profile database, providing a foundation for subsequent processing; data is simplified through feature compression, reducing transmission and processing pressure while retaining key information; with the help of a federated learning framework and a distributed model, accurate overall electricity consumption forecasts are obtained while protecting user privacy; finally, by combining dynamic programming algorithms with real-time electricity price adjustments to the peak-valley declaration ratio, precise matching of electricity declarations with actual demand is achieved, adapting to market fluctuations, reducing electricity purchase costs, and balancing user electricity consumption experience with the profitability of electricity sales companies.

[0077] In one possible embodiment, S13, using a distributed demand forecasting model based on a federated learning framework, the compressed electricity consumption data corresponding to all users are aggregated to obtain the overall electricity consumption forecast of the electricity sales company, including:

[0078] Step 131: Assign each user to the corresponding local node, and generate a local prediction model based on the local electricity consumption characteristic data of the corresponding user on each local node.

[0079] Among them, local nodes are devices deployed on the user side or near the user location to process electricity consumption data of a single user or a group of users; local electricity consumption characteristic data are information stored in local nodes that reflects the electricity consumption patterns of the corresponding user, including electricity consumption fluctuations during peak hours and electricity consumption stability during off-peak hours; local prediction models are models generated based on these data that are specifically designed to predict the electricity demand of the corresponding user.

[0080] In this embodiment, each user is first assigned to a corresponding local node based on their residential area or power line. For example, user A in community A is assigned to node 1, and user B is assigned to node 2, ensuring dedicated processing. Secondly, each local node uses a preset method to analyze stored local electricity consumption characteristic data and generate a local prediction model applicable only to the corresponding user. For example, node 1 generates a local model predicting user A's electricity demand based on user A's large fluctuations in evening electricity consumption and stable consumption at night.

[0081] Step 132: Input the parameters of the local prediction model into the central node of the federated learning framework. The central node merges the parameters of all local prediction models to obtain the global prediction model.

[0082] The central node of the federated learning framework is the central equipment deployed by the electricity sales company, which is used to aggregate and process data from each local node; the parameters of the local prediction model are the key settings for calculating the prediction results in the model, including coefficients that reflect the correlation between electricity consumption characteristics and prediction results; the global prediction model is a prediction model that is generated by merging all local model parameters and is applicable to the entire user group.

[0083] In this embodiment, firstly, each local node uploads the parameters (not raw data) of its local prediction model to the federated learning center node. For example, node 1 uploads the model parameters of user A, and node 2 uploads the parameters of user B. Secondly, the central node integrates these parameters, such as adjusting the weights of similar parameters according to their correlation with electricity consumption characteristics and then merging them to generate a global prediction model. For example, by integrating the parameters of A, B, and C, a global model reflecting the overall electricity consumption patterns of the community is obtained.

[0084] Step 133: Use the global prediction model to aggregate the compressed data of electricity consumption characteristics for each user to obtain the predicted electricity consumption for each user.

[0085] Among them, the global prediction model is the model generated in step 132 that integrates all local model parameters, taking into account both individual user and overall electricity consumption patterns; the electricity consumption feature compressed data is data that still reflects the core electricity consumption characteristics of users after simplification (result of step 12); the electricity consumption forecast is the user's future electricity demand estimate obtained through model calculation.

[0086] In this embodiment, firstly, the central node distributes the global prediction model to each local node. Secondly, each local node inputs the compressed electricity consumption characteristic data of the corresponding user into the global model for calculation to obtain the user's predicted electricity consumption. For example, node 1 inputs the compressed data of user A into the global model to obtain A's estimated electricity consumption.

[0087] Step 134: Summarize the electricity consumption forecasts of all users to obtain the overall electricity consumption forecast for the electricity sales company.

[0088] Among them, the overall electricity consumption forecast is an estimated result reflecting the total electricity demand within the service area of ​​the electricity sales company, obtained by summarizing the electricity consumption forecasts of all users. It is the core basis for electricity declaration and allocation.

[0089] In this embodiment, firstly, each local node sends its predicted user electricity consumption to the central node; for example, node 1 sends the predicted amount for user A, and node 2 sends the predicted amount for user B. Secondly, the central node sums up all the predicted amounts to obtain the overall predicted electricity consumption of the electricity sales company; for example, by summing up the predicted amounts for users A, B, and C, the total demand estimate for the community is obtained.

[0090] This application provides the following specific example: In community A, the system first assigns users A, B, and C to corresponding local nodes 1, 2, and 3, respectively. Each node generates a specialized prediction model based on the electricity consumption characteristics of its corresponding user. For example, node 1 analyzes user A's fluctuating electricity consumption in the evening and stable electricity consumption at night, generating a model that can predict A's electricity demand. Nodes 2 and 3 also generate similar specialized models for B and C. Afterward, these local nodes only upload the key settings used for calculation in the model to the central node of the community's electricity sales company, without transmitting the raw electricity consumption data. The central node collects these settings and integrates them with the correlation of each user's electricity consumption characteristics to form a global prediction model that reflects the electricity consumption patterns of the entire community. After the global model is generated, it is distributed to each local node. Each node inputs the simplified electricity consumption characteristic data of its corresponding user into the global model to obtain its own electricity demand estimate. For example, node 1 calculates A's electricity demand, while nodes 2 and 3 obtain the estimates for B and C, respectively. Finally, each local node sends these forecast results to the central node, which aggregates all the results to obtain the overall electricity demand forecast for the entire community, providing a basis for electricity sales companies to submit and allocate electricity.

[0091] By executing steps 131 to 133, this embodiment of the application ensures that the predictions match individual electricity consumption characteristics by allocating dedicated local nodes to users and generating local models, while avoiding centralized transmission of raw data and protecting user privacy; by merging local model parameters to generate a global model, it takes into account both individual characteristics and overall patterns, thereby improving prediction accuracy; finally, the overall electricity consumption forecast obtained by summarizing provides electricity sales companies with a reliable basis for total demand, which helps to formulate accurate electricity declaration and allocation strategies and reduce supply and demand imbalance problems.

[0092] In one possible embodiment, step 132, merging the parameters of all local prediction models through the central node to obtain a global prediction model, includes:

[0093] a1. The central node calculates the merging weight of local electricity consumption characteristic data based on the overlap of the coverage area of ​​the corresponding user's electricity consumption characteristics and the number of data records for each local node.

[0094] Among them, the overlap of electricity consumption feature coverage is the similarity of users' electricity consumption characteristics across different local nodes, such as whether the peak electricity consumption periods of different users are consistent, and whether the electricity consumption change patterns are similar; the number of data records is the total amount of electricity consumption feature information of the corresponding user stored in each local node; the merging weight is a value calculated by combining the overlap and the number of data records, used to reflect the importance of each local electricity consumption feature data in the global model.

[0095] In this embodiment, the central node first collects electricity consumption characteristic data of users corresponding to each local node and analyzes the overlap of user electricity consumption characteristic coverage between different nodes. For example, it checks whether users A and B both consume more electricity in the evening and whether their electricity consumption fluctuations are similar to determine the overlap. Secondly, it counts the number of data records for each local node, i.e., how many user electricity consumption characteristic information records the node stores. Finally, it combines these two factors to calculate the merging weight. Generally, the higher the overlap and the more data records a node has, the greater its merging weight. For example, in community A, the central node finds that the electricity consumption characteristics of users at node 1 and node 2 have a high degree of overlap. Node 1 has 100 data records, node 2 has 80, and node 3 has 60 with low overlap. After comprehensive calculation, the weight of node 1 is 0.4, node 2 is 0.3, and node 3 is 0.3.

[0096] a2. Adjust the parameters of the local prediction model corresponding to each local node according to the merging weight to obtain the adjusted parameters of the local prediction model corresponding to each local node.

[0097] The scaling adjustment is an operation that amplifies or reduces the parameters of the local prediction model based on the combined weights; the adjusted parameters of the local prediction model are those that, after scaling adjustment, better reflect their importance in the global model.

[0098] In this embodiment, the central node first obtains the merge weights of each local node calculated in step a1, such as node 1 with a weight of 0.4, node 2 with a weight of 0.3, and node 3 with a weight of 0.3. Then, the parameters of the local prediction model corresponding to each local node are adjusted proportionally according to these weights, that is, each parameter is multiplied by the corresponding merge weight. For example, if a parameter of node 1 is 5, it is adjusted to 5 multiplied by 0.4 equals 2; if the same parameter of node 2 is 6, it is adjusted to 6 multiplied by 0.3 equals 1.8; and if the same parameter of node 3 is 4, it is adjusted to 4 multiplied by 0.3 equals 1.2. Through such adjustments, the parameters of nodes with higher weights have a greater impact on the global model.

[0099] a3. Integrate similar parameters from all adjusted local prediction models to form unified parameters for the global model.

[0100] Among them, the same type of parameter refers to the parameter that reflects the same electricity consumption characteristics in different local prediction models, such as the "peak-hour electricity consumption coefficient" and "off-peak-hour electricity consumption stability coefficient" that are present in each model; the unified parameter of the global model is the parameter that is obtained by integrating all the adjusted same type parameters and is applicable to the global model.

[0101] In this embodiment, the central node first classifies the adjusted parameters of each local node obtained in step a2, identifying parameters of the same type. For example, the "peak-hour electricity consumption coefficient" of all nodes is grouped into one category, and the "off-peak-hour electricity consumption stability coefficient" into another. Secondly, each type of parameter is integrated, typically by addition. For instance, all adjusted "peak-hour electricity consumption coefficients" are summed to obtain a unified value for the parameter in the global model. This integration unifies the scattered parameters into parameters usable in the global model. For example, in community A, the adjusted "peak-hour electricity consumption coefficients" are 2, 1.8, and 1.2, which, when added together, result in a globally unified parameter of 5.

[0102] a4. Load all the unified parameters of the global model into the preset model structure to obtain the global prediction model.

[0103] The preset model structure is a pre-designed framework for the global prediction model, including the arrangement of parameters and the calculation logic; the global prediction model is a complete model that can be used for overall electricity consumption prediction after all the uniform parameters are loaded into the preset structure.

[0104] In this embodiment, the central node first retrieves a preset global prediction model structure, which specifies the location, function, and computational flow of various parameters. Next, all unified parameters of the global model obtained in step a3 are filled in according to the requirements of the preset structure; for example, unified parameters such as "peak-hour electricity consumption coefficient" and "off-peak-hour electricity consumption stability coefficient" are placed in their corresponding positions within the structure. Finally, after parameter loading, the preset structure and unified parameters combine to form a complete global prediction model, which can be used to predict and analyze the electricity consumption characteristics of all users. For example, the central node of community A fills in parameters such as "peak-hour electricity consumption coefficient 5" and "off-peak-hour electricity consumption stability coefficient 3.7" into the preset structure to form a global model capable of predicting the electricity demand of the entire community.

[0105] This application provides the following specific example: The central node first analyzes the electricity consumption characteristics of users A, B, and C corresponding to local nodes 1, 2, and 3: A and B both consume more electricity in the evening with similar fluctuation patterns, showing a high degree of overlap; C mainly consumes electricity in the morning, with a low degree of overlap with A and B. Meanwhile, node 1 has 100 data points, node 2 has 80, and node 3 has 60. When calculating the merge weight, the central node first scores based on overlap and data volume, each accounting for 50%. Node 1 scores 0.9, node 2 scores 0.8, and node 3 scores 0.45. Then, normalization is performed, resulting in a weight of 0.4 for node 1, 0.3 for node 2, and 0.3 for node 3. Next, the central node adjusts the parameters of each node according to the weight: the "peak-hour electricity consumption coefficient" of node 1 is 5, adjusted to 5 × 0.4 = 2; the same coefficient of node 2 is 6, adjusted to 6 × 0.3 = 1.8; the same coefficient of node 3 is 4, adjusted to 4 × 0.3 = 1.2. Then, the central node integrates the adjusted parameters of the same type. The "peak-hour electricity consumption coefficient" is added together to get 2 + 1.8 + 1.2 = 5. The "off-peak-hour electricity consumption stability coefficient" is adjusted to 1.6, 1.2, and 0.9 respectively, and added together to get 3.7, forming a globally unified parameter. Finally, the central node loads these unified parameters into a preset model structure (including parameter positions and calculation logic), combining them to form a global prediction model that can comprehensively analyze the electricity consumption characteristics of all users in the community and predict the overall electricity demand.

[0106] By executing a1~a4, this embodiment of the application calculates the merging weight by combining the overlap of electricity consumption characteristics and the number of data records, ensuring that the importance of data from each local node is reasonably measured; adjusting parameters according to weight highlights the influence of key nodes, making the parameters more suitable for the global model; integrating similar parameters eliminates dispersion and forms a unified standard; finally, loading the global prediction model generated by the preset structure can not only integrate the electricity consumption characteristics of all users, but also maintain the consistency and integrity of the model, providing a reliable tool for accurately predicting overall electricity demand, while avoiding the centralized transmission of raw data and protecting user privacy.

[0107] In one possible embodiment, such as Figure 2 As shown, S12, feature compression processing is performed on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data, including:

[0108] Step 121: Extract the feature data of each user from the user electricity consumption profile database and determine the multiple feature items contained in the feature data of each user.

[0109] Among them, the user electricity consumption profile database is a dataset formed by integrating the fluctuation characteristics of user electricity consumption, which contains various information that can reflect the user's electricity consumption habits; the feature data is detailed electricity consumption feature information of a single user extracted from the profile database; the feature items are the specific items that make up the feature data, such as peak-hour electricity consumption duration, off-peak-hour electricity consumption fluctuation value, average electricity consumption intensity, etc. Each user's feature data consists of multiple such specific items.

[0110] In this embodiment, the first step is to extract feature data for each user from a user electricity consumption profile database. This data consists of previously constructed detailed information reflecting the user's electricity consumption patterns. For example, the electricity consumption feature data for users A, B, and C are extracted from the profile database of community A. Secondly, the composition of this feature data is analyzed to determine which specific feature items each user's feature data includes. For instance, it is checked whether user A's feature data includes items such as evening electricity consumption duration, nighttime electricity consumption fluctuation value, morning electricity consumption intensity, and voltage stability duration, thus clarifying the feature item list for each user. For example, in community A, after extracting user A's feature data, it is found that it includes four feature items: evening electricity consumption duration, nighttime electricity consumption fluctuation value, morning electricity consumption intensity, and voltage stability duration.

[0111] Step 122: Evaluate the degree of influence of each feature item in reflecting the fluctuation characteristics of user electricity consumption, and rank each feature item from high to low according to the degree of influence.

[0112] The degree of influence refers to the importance of each feature in reflecting the fluctuation characteristics of user electricity consumption, that is, the extent to which the feature plays a role in reflecting the peak and trough change patterns of user electricity consumption; ranking is the process of arranging the feature items from high to low according to the degree of influence.

[0113] In this embodiment, the influence of each feature item is first assessed by analyzing its correlation with electricity consumption fluctuations. For example, the duration of evening electricity consumption is observed to determine whether it significantly reflects the peak evening electricity consumption fluctuations, and whether the duration of voltage stability has a relatively small effect on reflecting electricity consumption fluctuations. This helps to determine the importance of each feature item. Secondly, based on the assessment results, the feature items are ranked from highest to lowest influence. For example, among the feature items for users in Community A, the duration of evening electricity consumption has the greatest impact on reflecting electricity consumption fluctuations, followed by nighttime electricity consumption fluctuations, then morning electricity consumption intensity, and the duration of voltage stability has the least impact. These feature items are then arranged in this order.

[0114] Step 123: Based on the sorting results and the pre-set quantity, select the feature items with high influence as the retained feature items.

[0115] The sorting result is a list of features obtained in step 122, arranged from highest to lowest level of influence; the pre-set quantity is the number of features to be retained that is determined in advance based on actual needs; the retained features are those selected from the sorting result that have a high degree of influence and whose quantity meets the preset requirements.

[0116] In this embodiment, the sorting result obtained in step 122 is first obtained, which is a list of feature items arranged from high to low according to their degree of influence. For example, the sorting result for community A is evening electricity consumption duration, nighttime electricity consumption fluctuation value, morning electricity consumption intensity, and voltage stability duration. Secondly, according to a preset number, such as 3, the top 3 feature items with the highest degree of influence are selected from the sorting result as retained feature items. For example, if the preset is to retain 3, evening electricity consumption duration, nighttime electricity consumption fluctuation value, and morning electricity consumption intensity are selected as retained feature items.

[0117] Step 124: Extract the values ​​corresponding to the retained feature items from the feature data of each user, combine the values ​​to form simplified feature data, and determine the electricity consumption feature compression data based on the simplified feature data of all users.

[0118] Among them, the value corresponding to the retained feature item is the specific data value related to the retained feature item in the feature data of each user. For example, the value corresponding to the evening electricity consumption of user A is 2 hours. The simplified feature data is the simplified feature data formed by combining the values ​​corresponding to the retained feature items of each user. The electricity consumption feature compressed data is the overall simplified electricity consumption feature dataset obtained by integrating the simplified feature data of all users.

[0119] In this embodiment, firstly, the values ​​corresponding to the retained features are extracted from each user's feature data. For example, the retained features of community A are evening electricity consumption duration, nighttime electricity consumption fluctuation value, and morning electricity consumption intensity. The values ​​corresponding to these three features for user A are extracted as follows: evening electricity consumption duration is calculated to be 2 hours by recording the start and end times of A's daily evening electricity consumption; nighttime electricity consumption fluctuation value is calculated to be 0.5 by the change in nighttime electricity consumption data; and morning electricity consumption intensity is calculated to be 3.2 by the electricity consumption per unit time in the morning. The corresponding values ​​for user B are calculated using the same method, resulting in evening electricity consumption duration of 1.5 hours, nighttime electricity consumption fluctuation value of 0.3, and morning electricity consumption intensity of 2.8. Secondly, these values ​​for each user are combined to form simplified feature data for that user. For example, the simplified feature data for user A is 2 hours, 0.5, and 3.2, and for user B it is 1.5 hours, 0.3, and 2.8. Finally, the simplified feature data of all users are integrated to form compressed electricity consumption feature data for the entire group, i.e., a set containing the simplified feature data of all users.

[0120] This application provides the following specific example: First, feature data of users A, B, and C are extracted from the user electricity consumption profile database. Analysis determines that user A's feature items include evening electricity consumption duration, nighttime electricity consumption fluctuation value, morning electricity consumption intensity, and voltage stability duration; user B's feature items include evening electricity consumption duration, nighttime electricity consumption fluctuation value, midday electricity consumption peak frequency, and current fluctuation frequency; and user C's feature items include morning electricity consumption intensity, nighttime electricity consumption fluctuation value, voltage stability duration, and total daily electricity consumption. Next, the impact of these feature items is evaluated. It is found that evening electricity consumption duration is most significantly correlated with evening peak electricity consumption fluctuation, nighttime electricity consumption fluctuation value directly reflects nighttime electricity consumption changes, morning electricity consumption intensity has a weaker reflection of morning fluctuations, and voltage stability duration and other feature items have the least correlation with electricity consumption fluctuations. Therefore, the ranking is: evening electricity consumption duration, nighttime electricity consumption fluctuation value, morning electricity consumption intensity, and voltage stability duration (and other minor feature items). Then, according to the preset requirement of retaining 3 feature items, evening electricity consumption duration, nighttime electricity consumption fluctuation value, and morning electricity consumption intensity are selected as the retained feature items. Finally, the corresponding values ​​for each user were extracted: User A's evening electricity usage duration was calculated as 2 hours from 18:00 to 20:00; the nighttime electricity fluctuation value was calculated as 0.5 by dividing the difference between the maximum and minimum values ​​of the data from 22:00 to 6:00 by the average value; and the morning electricity intensity was calculated as 3.2 by dividing the total electricity consumption from 8:00 to 10:00 by the time. The combination is (2 hours, 0.5, 3.2); User B's value was calculated as (1.5 hours, 0.3, 2.8); and User C's value was (1 hour, 0.4, 3.5). These data were integrated to form compressed electricity usage characteristic data for Community A.

[0121] By executing steps 121 to 124, this embodiment of the application extracts feature data and clarifies feature items, providing a clear analytical object for subsequent processing; by assessing the degree of influence and ranking them, the importance of feature items is distinguished; based on the ranking and a preset number, feature items are selected for retention, achieving precise screening of key information; finally, by extracting values, combining and simplifying feature data, and integrating them, compressed electricity consumption feature data is formed. This process, while retaining the core information reflecting electricity consumption fluctuations, significantly reduces the amount of data, lowers the pressure on data transmission and processing, ensures that subsequent analysis can focus on key features, and improves the efficiency and accuracy of the overall process.

[0122] In one possible embodiment, step 122, evaluating the influence of each feature item on reflecting the fluctuation characteristics of user electricity consumption, and ranking each feature item from high to low according to the degree of influence, includes:

[0123] b1. Count the number of times each feature item is associated with the user's electricity consumption fluctuation characteristics in all users' feature data.

[0124] Among them, the feature items are the specific items that make up the user's electricity consumption feature data, such as the duration of electricity consumption during peak hours and the fluctuation value of electricity consumption during off-peak hours; the user's feature data is detailed information reflecting the electricity consumption characteristics of an individual user; the user's electricity consumption fluctuation feature is the changing pattern that the user exhibits during the electricity consumption process, such as the difference in electricity consumption between peak and off-peak hours; the number of associations is the total number of times that each feature item is associated with the user's electricity consumption fluctuation feature in all users' feature data.

[0125] In this embodiment, the feature data of all users are first examined one by one to determine whether each feature is associated with the user's electricity consumption fluctuation characteristics. For example, if a user's "evening electricity consumption duration" increases and their evening electricity consumption fluctuation also increases significantly, it is considered that the feature has been associated with the electricity consumption fluctuation characteristics. Secondly, for each feature, the total number of times this association occurs among all users is counted. For example, among the three users in community A, the number of times "evening electricity consumption duration" is associated with the electricity consumption fluctuation characteristics is: user A has 2 times, user B has 3 times, and user C has 2 times. The total number of associations is 2 + 3 + 2 = 7 times.

[0126] b2. Assign an influence value to each feature based on the number of times the association occurs. The number of times the association occurs is directly proportional to the influence value.

[0127] The number of associations is the total number of times each feature item is associated with the user's electricity consumption fluctuation characteristics as counted in step b1; the influence value is a numerical value assigned to each feature item based on the number of associations, used to reflect the importance of the feature item in reflecting electricity consumption fluctuations; the proportional relationship means that the more associations there are, the greater the influence value, and the fewer associations there are, the smaller the influence value.

[0128] In this embodiment, the first step is to determine the correspondence rule between the number of associations and the degree of influence. For example, the number of associations can be directly used as the degree of influence, or it can be converted according to a fixed ratio (such as multiplying the number of associations by 0.5). Secondly, a degree of influence value is assigned to each feature item according to this rule. For example, in Community A, the number of associations for "evening electricity consumption duration" is 7, so if the direct correspondence rule is used, its degree of influence value is 7; the number of associations for "nighttime electricity consumption fluctuation value" is 5, so its degree of influence value is 5; and the number of associations for "morning electricity consumption intensity" is 3, so its degree of influence value is 3.

[0129] b3. Compare the magnitude of the influence values ​​of each feature item, and sort the feature items in descending order of influence value to obtain the sorting result.

[0130] The influence value is a numerical value assigned to each feature item in step b2, reflecting its importance to electricity fluctuations; the sorting result is a list formed by arranging all feature items in descending order of influence value.

[0131] In this embodiment, the influence values ​​of all feature items are first compared to clarify their relative magnitudes. For example, in community A, the influence value of "evening electricity usage duration" is 7, "nighttime electricity usage fluctuation value" is 5, and "morning electricity usage intensity" is 3. Obviously, 7 is greater than 5, and 5 is greater than 3. Secondly, the feature items are arranged in descending order of influence value to form a ranking result. For example, the ranking result of the above three feature items is "evening electricity usage duration", "nighttime electricity usage fluctuation value", and "morning electricity usage intensity".

[0132] This application provides the following specific example: First, step b1 is performed: the number of associations between each feature and the electricity consumption fluctuation feature is counted. "Evening electricity consumption duration" shows association on 2 days in user A's electricity consumption data, 3 days in user B's, and 2 days in user C's, for a total of 2+3+2=7 associations; "Nighttime electricity consumption fluctuation value" shows association on 2 days in user A's data, 2 days in user B's data, and 1 day in user C's data, for a total of 2+2+1=5 associations; "Morning electricity consumption intensity" shows association on 1 day in each of the three users' data, for a total of 1+1+1=3 associations. Next, step b2 is performed: using the rule that the number of associations directly serves as the influence value, an influence value is assigned to each feature. The influence value for "Evening electricity consumption duration" is 7, for "Nighttime electricity consumption fluctuation value" it is 5, and for "Morning electricity consumption intensity" it is 3. Finally, step b3 is performed: the magnitude of the influence values ​​are compared, 7 > 5 > 3, so they are sorted from largest to smallest, and the results are "Evening electricity consumption duration", "Nighttime electricity consumption fluctuation value" and "Morning electricity consumption intensity".

[0133] By executing steps b1 to b3, this embodiment of the application objectively quantifies the correlation between feature items and electricity consumption fluctuations by statistically analyzing the number of associations between feature items and electricity consumption fluctuations, thus avoiding subjective judgment bias. Based on the number of associations, an influence value is assigned, which clearly quantifies the importance of feature items, and the proportional relationship ensures that feature items with closer associations receive higher attention. Finally, a clear order of importance is formed by sorting, providing an intuitive basis for subsequent screening of key feature items, improving the targeting and efficiency of data processing, and ensuring that features that are more important to reflecting electricity consumption fluctuations are retained first.

[0134] In one possible embodiment, S11, waveform transformation is used to extract the user's electricity consumption fluctuation characteristics from active power parameters, reactive power parameters, and harmonic parameters to construct a user electricity consumption profile database, including:

[0135] Step 111: Perform waveform transformation on the active power parameters, reactive power parameters, and harmonic parameters respectively to obtain the waveform change information of each parameter at multiple continuous time points.

[0136] Among them, active power parameters are the actual power consumed by users to drive electrical appliances; reactive power parameters are the power data that maintains the normal operation of electrical appliances but does not directly produce work; harmonic parameters are the irregular fluctuation data generated when the current or voltage waveform deviates from the standard shape; waveform transformation is a processing method that decomposes the continuous curve of power parameters changing over time into values ​​at multiple specific time points to present the details of parameter changes; waveform change information is the detailed data obtained after waveform transformation that reflects the changes in the values ​​of each parameter at different time points.

[0137] In this embodiment, the raw data of a user's active power parameters, reactive power parameters, and harmonic parameters over time are first obtained, such as the hourly records of these three types of power parameters for a user throughout the day. Next, waveform transformation is performed on each type of parameter, breaking down the continuously changing curve into specific values ​​at multiple time points. For example, the active power parameters for a day are broken down into values ​​for 24 hours, obtaining the waveform change information for each parameter at each time point. For instance, the active power parameters of user A in community A, after waveform transformation, yield values ​​of 20 at 8:00 AM, 30 at 12:00 PM, and 40 at 7:00 PM, forming the waveform change information for this parameter. The reactive power parameters and harmonic parameters are processed in the same way to obtain their respective waveform change information.

[0138] Step 112: Extract the numerical increase rate, numerical decrease rate, and number of numerical changes per unit time of the corresponding parameters from the waveform change information.

[0139] Among them, the waveform change information is the specific values ​​of each parameter at multiple consecutive time points obtained in step 111; the value increase rate refers to the amount by which the value of the parameter increases from the value of the previous value at adjacent time points; the value decrease rate refers to the amount by which the value of the parameter decreases from the value of the previous value; the number of value changes per unit time refers to the total number of times the parameter value increases or decreases within a set time interval (such as 1 hour).

[0140] In this embodiment, the numerical changes of each parameter are first extracted from the waveform change information. The values ​​at adjacent time points are compared to calculate the increase and decrease. The increase is the value at the later time point minus the value at the previous time point (a positive result indicates an increase), and the decrease is the value at the previous time point minus the value at the later time point (a positive result indicates a decrease). For example, if the active power parameter changes from 20 to 30, the increase is 30 minus 20 = 10; if it changes from 30 to 25, the decrease is 30 minus 25 = 5. Next, the number of increases or decreases in value per unit time (e.g., 1 hour) is counted. For example, if the parameter value changes from 20 to 30, then to 25, and finally back to 30 within 1 hour, there are 3 changes in total, so the number of changes per unit time is 3. For example, in the active power parameter waveform change information of user A in community A, from 8:00 AM to 9:00 AM, the value changes from 20 to 25 and then to 22. The increase is 25 minus 20 = 5, and the decrease is 25 minus 22 = 3. The number of changes in value within 1 hour is 2.

[0141] Step 113: Combine the positive power parameters, reactive power parameters, and harmonic parameters corresponding to the same user, respectively, along with the magnitude of their increase and decrease, and the number of times their values ​​change per unit time, to generate the user's power consumption fluctuation characteristics.

[0142] Among them, the three types of parameters corresponding to the same user refer to the active power parameters, reactive power parameters and harmonic parameters belonging to the same user; the magnitude of the increase in value, the magnitude of the decrease in value and the number of changes in value per unit time are the change characteristics of various parameters extracted in step 112; the electricity consumption fluctuation characteristics are comprehensive information that reflects the overall fluctuation pattern of the user's electricity consumption after combining the change characteristics of the three types of parameters of the same user.

[0143] In this embodiment, the following steps are first taken: The active power parameters, reactive power parameters, and harmonic parameters of the same user are collected, including their respective magnitudes of increase and decrease, as well as the number of changes per unit time. For example, user A's active power parameters show an increase of 5, a decrease of 3, and changes twice per hour; reactive power parameters show an increase of 2, a decrease of 1, and changes once per hour; and harmonic parameters show an increase of 0.1, a decrease of 0.2, and changes once per hour. Next, these features are combined according to parameter type to form the user's power consumption fluctuation characteristics. For example, user A's power consumption fluctuation characteristics include active power parameters showing an increase of 5, a decrease of 3, and changes twice; reactive power parameters showing an increase of 2, a decrease of 1, and changes once; and harmonic parameters showing an increase of 0.1, a decrease of 0.2, and changes once.

[0144] Step 114: Associate and record the electricity consumption fluctuation characteristics of all users according to each user's unique identifier, construct a set containing the electricity consumption fluctuation characteristics of each user, and determine the set as the user electricity consumption profile library.

[0145] Among them, the user's electricity consumption fluctuation characteristics are comprehensive information generated in step 113 that reflects the electricity consumption fluctuation pattern of a single user; the unique identifier is a unique symbol or number used to distinguish different users; the associated record is the operation of storing each user's unique identifier in a one-to-one correspondence with its electricity consumption fluctuation characteristics; the user electricity consumption profile database is a dataset containing the electricity consumption fluctuation characteristics of all users after integrating the associated records of all users.

[0146] In this embodiment, each user is first assigned a unique identifier, such as ID001 for user A and ID002 for user B. Next, each user's unique identifier is associated with and stored in relation to their corresponding electricity consumption fluctuation characteristics; that is, ID001 stores the electricity consumption fluctuation characteristics of user A, and ID002 stores the electricity consumption fluctuation characteristics of user B. Finally, all user association records are integrated to form a set containing the electricity consumption fluctuation characteristics of all users; this set is the user electricity consumption profile library. For example, the user electricity consumption profile library of community A stores the electricity consumption fluctuation characteristics and their corresponding unique identifiers for all users, such as ID001 (user A) and ID002 (user B).

[0147] This application provides the following specific examples:

[0148] By executing steps 111 to 114, this embodiment of the application transforms the waveforms of three types of power parameters, converting continuously changing parameters into numerical information at specific time points, thus providing a detailed data foundation for subsequent extraction of fluctuation characteristics. By extracting the magnitude of increase, decrease, and number of changes per unit time, the dynamic change patterns of various parameters are accurately captured. The change characteristics of the three types of parameters are combined to form electricity consumption fluctuation characteristics, comprehensively reflecting users' electricity consumption habits and fluctuation patterns. Finally, by associating unique user identifiers, a user electricity consumption profile database is constructed. The system integrates the electricity consumption characteristics of all users, providing structured and targeted basic data for subsequent power allocation and other work, ensuring that subsequent processing can accurately adapt to the electricity consumption characteristics of different users.

[0149] In one possible embodiment, S14, based on the overall electricity consumption forecast, a time-of-use pricing rule and an electricity allocation strategy are set using a dynamic programming algorithm. Based on the time-of-use pricing rule and the electricity allocation strategy, and in conjunction with the real-time market electricity price, the allocation ratio of the electricity sales enterprise's declared electricity volume during peak and off-peak periods is dynamically adjusted, including:

[0150] Step 141: Based on the overall electricity consumption forecast, calculate the total predicted electricity consumption for each peak and valley period and the electricity consumption percentage for each period.

[0151] Among them, the overall electricity consumption forecast is the total electricity demand of all users within the service area of ​​the electricity sales company; peak and valley periods are the periods of high and low electricity consumption, such as the peak period of high electricity consumption during the day and the valley period of low electricity consumption at night; the total predicted electricity consumption is the total predicted electricity consumption within each peak and valley period; and the electricity consumption ratio is the proportion of the total predicted electricity consumption in each peak and valley period to the overall electricity consumption forecast.

[0152] In this embodiment, the electricity consumption is first divided into peak and off-peak periods based on the overall electricity consumption forecast. For example, 8:00-22:00 is designated as the peak period, and 22:00-8:00 the next day is designated as the off-peak period. Next, the total predicted electricity consumption for each period is calculated, which is the sum of the predicted electricity consumption of all users within that period. Finally, the electricity consumption percentage for each period is calculated, which is the total predicted electricity consumption for that period divided by the overall electricity consumption forecast. For example, if the overall electricity consumption forecast for Community A is 1000 kWh, the predicted electricity consumption during the peak period (8:00-22:00) is 600 kWh, and the percentage is 600 divided by 1000, which equals 0.6; the predicted electricity consumption during the off-peak period (22:00-8:00 the next day) is 400 kWh, and the percentage is 400 divided by 1000, which equals 0.4.

[0153] Step 142: Using dynamic programming algorithm, determine the time-of-use pricing rules and power allocation strategy based on the predicted total electricity consumption and electricity consumption ratio. The time-of-use pricing rules are different electricity price standards for different time periods, and the power allocation strategy is that each time period has its own power allocation limit.

[0154] Among them, dynamic programming algorithm is a method to flexibly plan strategies for different time periods to achieve the overall optimal result; the predicted total electricity consumption is the total electricity consumption of each time period counted in step 141; the electricity consumption ratio is the proportion of electricity consumption of each time period to the total; the time-of-use electricity price rule is the different electricity price standard set for different peak and valley time periods, and the electricity price may be higher during the time period with more electricity consumption; the electricity allocation strategy is the electricity allocation limit set for each time period, that is, the maximum amount of electricity that can be declared during that time period.

[0155] In this embodiment, a dynamic programming algorithm is first used to analyze the predicted total electricity consumption and electricity consumption ratio for each peak and valley period, determining the adjustment direction of electricity price and allocation limit for different periods. For example, the electricity price can be set higher and the allocation limit can be correspondingly larger for periods with a higher electricity consumption ratio. Secondly, time-of-use pricing rules are formulated based on the analysis results, setting different electricity price standards for peak and valley periods, such as higher prices during peak periods and lower prices during valley periods. Finally, an electricity allocation strategy is determined, setting a corresponding electricity allocation limit for each period, i.e., the maximum electricity volume allowed to be declared during that period, such as a larger limit during peak periods and a smaller limit during valley periods. For example, if the peak electricity consumption ratio for Community A is 0.6%, its electricity price standard is set at 1.2 yuan / kWh, and the allocation limit is 600 kWh; if the valley period ratio is 0.4%, the electricity price standard is 0.8 yuan / kWh, and the allocation limit is 400 kWh.

[0156] Step 143: Obtain the current electricity price for each peak and valley period in the market in real time, and compare the current electricity price with the electricity price standard for the corresponding period.

[0157] Among them, real-time acquisition is to collect current electricity price information in the market in a timely manner; the current electricity price for each peak and valley period in the market is the electricity price for the corresponding peak and valley periods in the current actual transaction in the electricity market; the electricity price standard for the corresponding period is the electricity price in the time-of-use electricity price rules formulated in step 142; comparison is to compare the current electricity price with the electricity price standard to determine the relationship between the two.

[0158] In this embodiment, the current electricity price for each peak and off-peak period is first collected in real time through market information channels, such as obtaining the actual transaction electricity price for the peak and off-peak periods of the day through an electricity trading platform. Secondly, the collected current electricity price is compared with the corresponding electricity price standard set in step 142 to determine whether the current electricity price is higher, lower, or equal to the standard price. For example, the standard electricity price for peak periods in Community A is 1.2 yuan / kWh, and the real-time current electricity price is 1.3 yuan / kWh; after comparison, it is found that the current electricity price is higher than the standard. The standard for off-peak periods is 0.8 yuan / kWh, and the current electricity price is 0.7 yuan / kWh; after comparison, it is lower than the standard.

[0159] Step 144: If the comparison result indicates that the current electricity price is higher than the electricity price standard for the corresponding period, reduce the proportion of electricity declared for the corresponding period in conjunction with the electricity allocation limit; or, if the comparison result indicates that the current electricity price is lower than the electricity price standard for the corresponding period, increase the proportion of electricity declared for the corresponding period in conjunction with the electricity allocation limit.

[0160] Among them, the comparison result is the comparison result between the current electricity price and the electricity price standard in step 143; the current electricity price is higher than the electricity price standard for the corresponding period when the actual market electricity price is higher than the set standard; the electricity allocation limit is the maximum amount of electricity that can be declared in each period as set in step 142; the electricity declaration volume ratio is the proportion of the electricity declared in this period to the total declaration volume; the current electricity price is lower than the electricity price standard for the corresponding period when the actual market electricity price is lower than the set standard.

[0161] In this embodiment, the comparison results from step 143 are used to determine whether the current electricity price for each time period is higher or lower than the standard price. Secondly, if the current electricity price is higher than the standard price, the proportion of electricity declared for that time period is reduced based on the electricity allocation limit for that period. This means reducing the proportion of electricity declared during that time period to avoid increased costs due to high prices. For example, during peak hours when the current price is high, the proportion is reduced from 60% to 55%, and it does not exceed the limit of 600 kWh. Finally, if the current electricity price is lower than the standard price, the proportion of electricity declared for that time period is increased based on the allocation limit. This means increasing the proportion of electricity declared during that time period to utilize the lower price to reduce costs. For example, during off-peak hours when the current price is low, the proportion is increased from 40% to 45%, and it does not exceed the limit of 400 kWh.

[0162] Step 145: Integrate the adjusted electricity declaration ratios corresponding to each peak and valley period to obtain the electricity declaration allocation ratio for electricity sales companies.

[0163] Among them, the adjusted electricity declaration ratio corresponding to each peak and valley period is the proportion of the declared electricity in each period after adjustment in step 144 to the total declared electricity; the electricity declaration allocation ratio of the electricity sales company is the final proportional allocation scheme of the declared electricity in each period after integrating the adjusted proportions of all periods.

[0164] In this embodiment, the adjusted electricity declaration percentages for each peak and off-peak period in step 144 are first collected. For example, the adjusted percentage for peak periods is 55%, and for off-peak periods it is 45% (if there are no quota restrictions). These percentages are then integrated to ensure the sum of the percentages for each period is 100%, forming the electricity declaration allocation ratio for the electricity sales company. This clarifies the specific proportion of the declared electricity in each period within the total declaration. For example, in Community A, the peak period percentage is 55%, and the off-peak period percentage is 45%, which, when added together, equals 100%, forming the final declaration allocation ratio.

[0165] This application provides the following specific example: The overall electricity consumption forecast is 1000 kWh. In step 141, the peak period is divided into 8:00-22:00 and the off-peak period is 22:00-8:00 the next day. The predicted total electricity consumption during the peak period is 600 kWh, and during the off-peak period it is 400 kWh. The electricity consumption ratio during the peak period is calculated as 600 divided by 1000, which equals 0.6, and the electricity consumption ratio during the off-peak period is calculated as 400 divided by 1000, which equals 0.4. In step 142, a time-of-use pricing rule is formulated using a dynamic programming algorithm: RMB 1.2 / kWh during the peak period and RMB 0.8 / kWh during the off-peak period. At the same time, an electricity allocation strategy is determined: a limit of 600 kWh during the peak period and 400 kWh during the off-peak period. In step 143, the current electricity price during peak hours is obtained in real time from the power trading platform. The price is 1.3 yuan / kWh during peak hours and 0.7 yuan / kWh during off-peak hours. The peak hour price of 1.3 yuan / kWh is compared to the standard price of 1.2 yuan / kWh, and the off-peak price of 0.7 yuan / kWh is compared to the standard price of 0.8 yuan / kWh. This results in a conclusion that the current peak hour price is higher than the standard price, and the current off-peak price is lower than the standard price. In step 144, the peak hour reporting ratio is reduced from 60% to 55% (550 kWh) due to quota limitations, which does not exceed the quota. During off-peak hours, the ratio remains at 40% (400 kWh) due to quota limitations. In step 145, after consolidation, the peak hour ratio is 55% and the off-peak ratio is 45%, ensuring a total of 100%, thus forming the electricity reporting allocation ratio for the community's electricity sales companies.

[0166] By executing steps 141 to 145, this embodiment of the application clearly understands the distribution of electricity demand by statistically analyzing electricity consumption data for each time period, providing a basis for formulating electricity prices and allocation strategies; the rules and strategies formulated using dynamic programming algorithms take into account both electricity demand and rationality; real-time comparison with market electricity prices and adjustment of the declaration ratio enable electricity declarations to flexibly adapt to market changes; the final integrated allocation ratio ensures that the declaration volume for each time period is coordinated and reasonable, which not only reduces electricity purchase costs but also meets actual electricity demand and improves the electricity allocation efficiency of electricity sales companies.

[0167] Figure 3 This application provides a schematic diagram of the structure of an AI-based dynamic allocation system for electricity sales enterprise electricity reporting, as shown in the embodiments of this application. Figure 3 As shown, the system includes:

[0168] Module 31 is used to periodically collect the active power parameters, reactive power parameters and harmonic parameters of each user, and to extract the user's power consumption fluctuation characteristics from the active power parameters, reactive power parameters and harmonic parameters by waveform transformation, so as to build a user power consumption profile library.

[0169] The processing module 32 is used to perform feature compression processing on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data.

[0170] The aggregation module 33 is used to aggregate the compressed electricity consumption data corresponding to all users using a distributed demand forecasting model based on a federated learning framework, so as to obtain the overall electricity consumption forecast of the electricity sales company.

[0171] The adjustment module 34 is used to set time-of-use pricing rules and power allocation strategies based on the overall electricity consumption forecast using a dynamic programming algorithm. Based on the time-of-use pricing rules and power allocation strategies, and combined with the real-time market electricity price, the module dynamically adjusts the allocation ratio of the power sales enterprise's declared power volume during peak and valley periods, so as to realize the dynamic allocation of the power sales enterprise's declared power volume.

[0172] The AI-based dynamic allocation system for electricity sales enterprise declaration in this application embodiment is used to implement the aforementioned AI-based dynamic allocation method for electricity sales enterprise declaration. Therefore, the specific implementation of the AI-based dynamic allocation system for electricity sales enterprise declaration can be found in the embodiment section of the AI-based dynamic allocation method for electricity sales enterprise declaration mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0173] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the AI-based dynamic allocation method for electricity sales enterprise declaration described above.

[0174] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described AI-based dynamic allocation methods for electricity sales enterprise declarations.

[0175] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0176] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the AI-based dynamic allocation method for electricity sales enterprise declaration.

[0177] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0178] The foregoing has provided a detailed description of an AI-based dynamic allocation method, system, electronic device, and storage medium for electricity sales enterprise electricity declaration, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for dynamic allocation of electricity consumption reported by electricity retailers based on AI, characterized in that, include: The active power parameters, reactive power parameters, and harmonic parameters of each user are periodically collected. Waveform transformation is used to extract the user's electricity consumption fluctuation characteristics from the active power parameters, reactive power parameters, and harmonic parameters in order to construct a user electricity consumption profile database. The feature data in the user electricity consumption profile database is subjected to feature compression processing to obtain electricity consumption feature compressed data; Using a distributed demand forecasting model based on a federated learning framework, the compressed electricity consumption data corresponding to all users are aggregated to obtain the overall electricity consumption forecast of the electricity sales company. Based on the overall electricity consumption forecast, a time-of-use pricing rule and an electricity allocation strategy are set using a dynamic programming algorithm. Based on the time-of-use pricing rule and the electricity allocation strategy, and combined with the real-time market electricity price, the allocation ratio of the electricity declaration volume of the electricity sales enterprise during peak and valley periods is dynamically adjusted to achieve dynamic allocation of the electricity declaration volume of the electricity sales enterprise. The step of performing feature compression processing on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data includes: Extract the feature data of each user from the user electricity consumption profile database, and determine the multiple feature items contained in the feature data of each user; Count the number of times each of the aforementioned feature items is associated with the user's electricity consumption fluctuation characteristics in the feature data of all users; Based on the number of times the association occurs, an influence value is assigned to each of the aforementioned features, wherein the number of times the association occurs is directly proportional to the influence value. Compare the magnitudes of the influence values ​​of each feature item, and sort the feature items in descending order of influence values ​​to obtain a sorting result; Based on the ranking results and a pre-defined number, features with high influence are selected as retained features. Extract the values ​​corresponding to the retained feature items from the feature data of each user, combine the values ​​to form simplified feature data, and determine the electricity consumption feature compressed data based on the simplified feature data of all users. The process of extracting the user's electricity consumption fluctuation characteristics from the active power parameters, reactive power parameters, and harmonic parameters using waveform transformation to construct a user electricity consumption profile database includes: The active power parameters, reactive power parameters and harmonic parameters are respectively subjected to waveform transformation to obtain waveform change information of each parameter at multiple consecutive time points; Extract the numerical increase rate, numerical decrease rate, and number of numerical changes per unit time of the corresponding parameters from the waveform change information; The power consumption fluctuation characteristics of the user are generated by combining the numerical increase and decrease of the active power parameters, reactive power parameters and harmonic parameters corresponding to the same user, respectively, as well as the number of numerical changes per unit time. All users' electricity consumption fluctuation characteristics are associated and recorded according to each user's unique identifier, and a set containing each user's electricity consumption fluctuation characteristics is constructed. This set is then defined as the user electricity consumption profile library.

2. The AI-based dynamic allocation method for electricity sales enterprise electricity declaration according to claim 1, characterized in that, The decentralized demand forecasting model, based on a federated learning framework, aggregates compressed electricity consumption data corresponding to all users to obtain the overall electricity consumption forecast for electricity retailers, including: The decentralized demand forecasting model based on a federated learning framework aggregates the compressed electricity consumption data corresponding to all users to obtain the overall electricity consumption forecast for the electricity sales company, including: Each user is assigned to a corresponding local node, and a local prediction model is generated based on the local electricity consumption characteristic data of the corresponding user in each local node. The parameters of the local prediction model are input to the central node of the federated learning framework, and the parameters of all the local prediction models are merged by the central node to obtain the global prediction model. The global prediction model is used to aggregate the compressed electricity consumption characteristic data of each user to obtain the predicted electricity consumption of each user. By summing up the electricity consumption forecasts of all users, the overall electricity consumption forecast of the electricity sales company can be obtained.

3. The AI-based dynamic allocation method for electricity sales enterprise electricity declaration according to claim 2, characterized in that, The step of merging the parameters of all the local prediction models through the central node to obtain the global prediction model includes: The central node calculates the merging weight of the local electricity consumption feature data based on the overlap of the coverage area of ​​the corresponding user's electricity consumption feature of each local node and the number of data records. The number of data records is the total amount of electricity consumption feature data of the corresponding user stored in each local node. The parameters of the local prediction model corresponding to each local node are proportionally adjusted according to the merging weights to obtain the adjusted parameters of the local prediction model corresponding to each local node. Integrate similar parameters from all adjusted local prediction models to form unified parameters for the global model; All the unified parameters of the global model are loaded into a preset model structure to obtain a global prediction model.

4. The AI-based dynamic allocation method for electricity sales enterprise electricity declaration according to claim 1, characterized in that, The process involves setting time-of-use pricing rules and power allocation strategies based on the overall electricity consumption forecast using a dynamic programming algorithm. Then, based on these rules and strategies, and in conjunction with real-time market prices, the process dynamically adjusts the allocation ratio of electricity sales companies' declared power consumption during peak and off-peak periods. This includes: Based on the overall electricity consumption forecast, the total predicted electricity consumption for each peak and valley period and the electricity consumption percentage for each period are calculated. Using a dynamic programming algorithm, based on the predicted total electricity consumption and the electricity consumption ratio, time-of-use pricing rules and electricity allocation strategies are determined. The time-of-use pricing rules are different electricity price standards for different time periods, and the electricity allocation strategy is that each time period has its own electricity allocation limit. The current electricity price during each peak and valley period in the market is obtained in real time, and the current electricity price is compared with the electricity price standard for the corresponding period. When the comparison result indicates that the current electricity price is higher than the electricity price standard for the corresponding time period, the proportion of electricity declaration for the corresponding time period is reduced in conjunction with the electricity allocation limit; or, when the comparison result indicates that the current electricity price is lower than the electricity price standard for the corresponding time period, the proportion of electricity declaration for the corresponding time period is increased in conjunction with the electricity allocation limit. By integrating the adjusted electricity declaration proportions corresponding to each peak and valley period, the electricity declaration allocation ratio of electricity sales companies is obtained.

5. An AI-based dynamic allocation system for electricity sales companies based on electricity demand reporting, characterized in that, The method for executing the AI-based dynamic allocation method for electricity sales enterprise electricity declaration as described in claim 1 includes: The module is used to periodically collect the active power parameters, reactive power parameters and harmonic parameters of each user, and to extract the user's power consumption fluctuation characteristics from the active power parameters, reactive power parameters and harmonic parameters by waveform transformation, so as to build a user power consumption profile library. The processing module is used to perform feature compression processing on the feature data in the user electricity consumption profile database to obtain compressed electricity consumption feature data. The aggregation module is used to aggregate the compressed electricity consumption data corresponding to all users using a distributed demand forecasting model based on a federated learning framework, so as to obtain the overall electricity consumption forecast of the electricity sales company. The adjustment module is used to set time-of-use pricing rules and power allocation strategies based on the overall electricity consumption forecast using a dynamic programming algorithm. Based on the time-of-use pricing rules and the power allocation strategies, and combined with the real-time market electricity price, the module dynamically adjusts the allocation ratio of the power sales enterprise's declared power volume during peak and valley periods, so as to realize the dynamic allocation of the power sales enterprise's declared power volume.

6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the AI-based dynamic allocation method for electricity sales enterprise declaration as described in any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the AI-based dynamic allocation method for electricity sales enterprise electricity declaration as described in any one of claims 1 to 4.

Images