A power selling decision method and system for incremental distribution network users

Abstract

The application discloses a kind of incremental distribution network user-oriented electricity selling decision method and system, including the following steps: user operation data is collected and preprocessed, and standardized data is obtained;Feature parameters are extracted and feature vectors are constructed, input into improved self-organizing mapping network hierarchical model, and user hierarchical result is obtained;User grouping is carried out based on user hierarchical result, and user set and group characteristic parameters are generated;Predictive input parameters, user hierarchical result related parameters are input into joint prediction decision model, and load prediction result and electricity selling decision parameter are obtained;Electricity selling execution scheme is generated based on electricity selling decision parameter and execution feedback data is collected;Improved self-organizing mapping network hierarchical model and joint prediction decision model are updated based on execution feedback data.The application adopts improved self-organizing mapping network hierarchical technology, solves the problem of rough user layering and inaccurate electricity selling decision.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and in particular to a method and system for electricity sales decision-making for incremental distribution network users. Background Technology

[0002] With the continuous advancement of power market reforms, incremental distribution networks are gradually becoming important power supply and sales carriers in industrial parks, industrial clusters, and integrated energy scenarios. Compared with traditional distribution networks, incremental distribution networks have more diverse user types, including not only regular electricity users but also users connected to distributed power sources, energy storage devices, and those participating in demand response. Therefore, electricity sales decisions need to comprehensively consider user operating data, electricity price changes, and the operating status of the distribution network.

[0003] In existing technologies, electricity sales decision-making methods typically begin by collecting user electricity consumption data, electricity price data, and distribution network operation data, followed by user classification, load forecasting, and electricity sales optimization. While these methods can achieve a certain degree of automated decision-making, they still have shortcomings. Existing user classification methods rely heavily on a limited number of basic indicators, making it difficult to simultaneously reflect load characteristics, price response characteristics, demand response characteristics, and distributed power generation usage characteristics, resulting in insufficiently refined classification results. Furthermore, existing classification results usually only output category labels, lacking further utilization of category centers and inter-category relationships, making it difficult to support more accurate subsequent electricity sales decisions.

[0004] Furthermore, in existing technologies, user segmentation, load forecasting, and electricity sales optimization are often independent of each other, and once the classification results are determined, there is usually a lack of a continuous update mechanism based on execution feedback. When user operating status, electricity price environment, or distributed generation operation changes, the original classification results may no longer accurately reflect the current user characteristics, thus affecting the accuracy of load forecasting and electricity sales decisions. Therefore, existing technologies still have shortcomings in terms of fine-grained user segmentation, utilization of classification results, and feedback updates, and further improvements are needed.

[0005] Therefore, how to provide a method and system for electricity sales decision-making for incremental distribution network users is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method and system for electricity sales decision-making for incremental distribution network users. This invention comprehensively adopts an improved self-organizing mapping network hierarchical method and a joint prediction decision-making method to process the operating data of incremental distribution network users, realizing fine-grained user hierarchical classification, load forecasting, electricity sales decision generation, and execution feedback updates. It can complete dynamic decision-making on electricity sales prices, electricity sales allocation, and energy storage coordination strategies for different user groups, and has the advantages of high user hierarchical accuracy, strong targeting of electricity sales decisions, good adaptability to changes in operating status, and high overall decision-making intelligence.

[0007] According to an embodiment of the present invention, a method and system for electricity sales decision-making for incremental distribution network users includes the following steps:

[0008] S1. Collect the operation data of each user in the incremental distribution network, and preprocess the operation data to obtain standardized data;

[0009] S2. Based on standardized data, extract user feature parameters, construct user feature vectors, and input the user feature vectors of each user into the improved self-organizing map network hierarchical model to perform feature weighted mapping, competitive mapping, dynamic neighborhood update, topology relationship extraction and category boundary correction to obtain user hierarchical results.

[0010] S3. Based on the user segmentation results, group each user and generate the user set and group feature parameters corresponding to each user group;

[0011] S4. Input the prediction input parameters, user stratification results, group characteristic parameters and electricity sales decision-related parameters of each user group into the joint prediction decision model to obtain the load prediction results of each user group, the total load prediction results of the incremental distribution network and the electricity sales decision parameters of each user group during the target electricity sales period.

[0012] S5. Generate an electricity sales execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters, and send the electricity sales execution plan to the corresponding execution terminal, and collect the execution feedback data corresponding to the electricity sales execution plan;

[0013] S6. A self-organizing mapping network hierarchical model and joint prediction decision model based on updated execution feedback data.

[0014] Optionally, step S1 includes the following steps:

[0015] S11. Collect the operation data of each user in the incremental distribution network. The operation data includes historical electricity consumption data, real-time load data, time-of-use electricity price data, market electricity purchase data, meteorological data, demand response data, distributed power generation operation data, energy storage operation data, and distribution network operation data.

[0016] S12. Perform missing value completion and outlier removal processing on the running data;

[0017] S13. Perform timestamp alignment and dimension normalization on the processed running data to obtain standardized data;

[0018] S14. Organize and associate standardized data according to user ID, collection period, and data category.

[0019] Optionally, step S2 includes the following steps:

[0020] S21. Extract load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, and distributed power usage characteristic parameters based on the standardized data;

[0021] S22. Load characteristic parameters include load level parameters, fluctuation degree parameters and peak-valley change parameters; price response characteristic parameters include price change response parameters; demand response characteristic parameters include demand response participation parameters; distributed power usage characteristic parameters include distributed power access parameters and distributed power usage parameters.

[0022] S23. Configure corresponding feature weight parameters according to the degree of influence of various feature parameters on user segmentation;

[0023] S24. Input the user feature vectors of each user into the improved self-organizing map network hierarchical model and perform competitive mapping processing;

[0024] S25. Perform dynamic neighborhood update processing based on the competition mapping results;

[0025] S26. Extract topological proximity relationships, topological boundary relationships, and topological hierarchy relationships based on the updated neuron distribution results;

[0026] S27. Perform category boundary correction processing based on topological boundary relationships;

[0027] S28. Generate user segmentation results, which include category labels, category centers, and topological mapping relationships.

[0028] Optionally, the improved self-organizing mapping network hierarchical model includes an input layer, a competition layer, a dynamic neighborhood update unit, a topology relationship extraction unit, a category boundary correction unit, and a hierarchical result output unit.

[0029] The input layer receives the user feature vectors from each user and transmits them to the competition layer in the order of features.

[0030] The competition layer consists of multiple competing neurons arranged according to a preset topology. It performs competitive mapping processing on the input user feature vector and determines the winning neuron corresponding to each user based on the degree of mapping matching between each competing neuron and the user feature vector.

[0031] The dynamic neighborhood update unit determines the set of neighboring neurons centered on the winning neuron, and adjusts the neighborhood update range according to the user feature distribution density, category clustering degree, mapping distance between adjacent neurons, and boundary changes between adjacent categories, and then iteratively updates the winning neuron and its neighboring neurons.

[0032] The topology extraction unit extracts topology proximity relationships, topology boundary relationships, and topology hierarchy relationships based on the relative positional relationships of each winning neuron in the competition layer, and forms topology relationship data corresponding to the user's hierarchical results;

[0033] The category boundary correction unit corrects the boundary region between adjacent categories based on topological boundary relationships, and performs boundary correction on user samples located in the transition region between adjacent categories;

[0034] The hierarchical result output unit generates category labels based on the winning neuron positions for each user, generates category centers based on the aggregation results of user feature vectors within the same category, and generates topological mapping relationships based on the corrected relative positional relationships of neurons.

[0035] When user feature distribution changes, the improved self-organizing map network hierarchical model synchronously adjusts the topological proximity, topological boundary, and topological hierarchy relationships based on the updated neuron competition results, and generates category labels, category centers, and topological mapping relationships corresponding to the current user feature distribution.

[0036] Optionally, step S3 includes the following steps:

[0037] S31. Read the category labels, category centers, and topological mapping relationships from the user stratification results;

[0038] S32. Group users according to their category labels and generate a user set for each user group.

[0039] S33. Based on the load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, distributed power usage characteristic parameters, and topology mapping relationship of users in each user group, generate the group characteristic parameters corresponding to each user group.

[0040] S34. Write the topological proximity parameters, topological boundary parameters, and topological hierarchy parameters corresponding to the topological mapping relationship into the group feature parameters;

[0041] S35. Based on the group feature parameters after writing, determine the degree of association, adjacency relationship and hierarchical difference between different user groups, and use them as input to the joint prediction decision model.

[0042] Optionally, step S4 includes the following steps:

[0043] S41. Construct a joint forecasting and decision-making model, which includes a forecasting input construction unit, a load forecasting unit, a state set construction unit, and an electricity sales optimization unit.

[0044] S42. Input the prediction input parameters of each user group into the load prediction unit to obtain the load prediction results of each user group and the total load prediction results of the incremental distribution network.

[0045] S43. Input the user stratification results, group characteristic parameters, load forecasting results, and electricity sales decision-related parameters into the state set construction unit to form a joint decision state set;

[0046] S44. Input the joint decision state set into the electricity sales optimization unit to obtain the electricity sales decision parameters for each user group.

[0047] Optionally, the forecast input parameters include historical load sequences, time stamps, meteorological characteristics, electricity price signals, and demand response status;

[0048] Parameters relevant to electricity sales decisions include market electricity purchase prices, predicted output of distributed power sources, energy storage status of charge, line load factor, and node voltage.

[0049] The joint decision-making state set includes category labels, category centers, topology mapping relationships, group characteristic parameters, load forecast results for each user group, total load forecast results for the incremental distribution network, and parameters related to electricity sales decisions.

[0050] The electricity sales optimization unit uses the soft actor commentator algorithm to generate electricity sales decision parameters for each user group;

[0051] Electricity sales decision parameters include electricity price, electricity sales volume allocation results, energy storage charging and discharging power, and demand response incentive parameters.

[0052] Optionally, step S5 includes the following steps:

[0053] S51. Generate an electricity sales execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters;

[0054] S52. Send the electricity sales execution plan to the electricity sales management platform, energy storage control platform and user interaction terminal;

[0055] S53. Collect execution feedback data corresponding to the electricity sales execution plan. The feedback data includes the user's actual electricity consumption, electricity purchase and sales deviation, electricity sales revenue, demand response completion rate, energy storage execution status, and distribution network operation status.

[0056] S54. Collect and organize the execution feedback data according to user groups and execution time periods to form a feedback dataset corresponding to user-level results and electricity sales execution plans.

[0057] Optionally, step S6 includes the following steps:

[0058] S61. A self-organizing mapping network hierarchical model and joint prediction decision model based on updated execution feedback data;

[0059] S62. Based on the execution feedback data, the user characteristic parameters, characteristic weight parameters, neighborhood update parameters, and topology relationship parameters are corrected. The correction of user characteristic parameters includes the correction of load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, and distributed power source usage characteristic parameters. The correction of characteristic weight parameters includes the correction of the weights corresponding to different characteristic parameters. The correction of neighborhood update parameters includes the correction of the neighborhood update range. The correction of topology relationship parameters includes the correction of topology proximity relationship, topology boundary relationship, and topology hierarchy relationship.

[0060] S63. The improved self-organizing mapping network hierarchical model regenerates class labels, class centers, and topological mapping relationships based on updated user feature parameters, feature weight parameters, neighborhood update parameters, and topological relationship parameters.

[0061] S64. Re-input the updated category labels, category centers, and topological mapping relationships into the joint prediction decision model;

[0062] S65. Based on the updated joint forecasting decision model, output new load forecasting results and electricity sales decision parameters.

[0063] The photovoltaic power plant unmanned aerial vehicle (UAV) autonomous inspection system based on digital twin according to an embodiment of the present invention includes the following modules:

[0064] The data acquisition and preprocessing module collects operational data from each user within the incremental distribution network and preprocesses the operational data to obtain standardized data.

[0065] The user feature extraction and improved self-organizing map hierarchical module extracts user feature parameters based on standardized data, constructs user feature vectors, and inputs the user feature vectors of each user into the improved self-organizing map network hierarchical model to perform feature weighted mapping, competitive mapping, dynamic neighborhood update, topology relationship extraction and category boundary correction to obtain user hierarchical results.

[0066] The user grouping module groups users based on the user segmentation results and generates a user set and group feature parameters for each user group.

[0067] The joint forecasting and decision-making module inputs the forecasting input parameters, user stratification results, group characteristic parameters, and electricity sales decision-related parameters of each user group into the joint forecasting and decision-making model to obtain the load forecasting results of each user group, the total load forecasting results of the incremental distribution network, and the electricity sales decision parameters of each user group during the target electricity sales period.

[0068] The execution feedback module generates an execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters, sends the execution plan to the corresponding execution terminal, and collects the execution feedback data corresponding to the electricity sales execution plan.

[0069] The model update module updates the improved self-organizing map network hierarchical model and joint prediction decision model based on execution feedback data.

[0070] The beneficial effects of this invention are:

[0071] Compared with existing technologies, this invention forms a complete closed-loop processing flow from data input to execution feedback updates by uniformly collecting, preprocessing, extracting features, stratifying users, jointly predicting, and making electricity sales decisions for operational data of all users within the incremental distribution network. By introducing an improved self-organizing mapping network hierarchical model, this invention can perform more refined stratification of users based on load characteristics, price response characteristics, demand response characteristics, and distributed power generation usage characteristics. It not only generates user category labels but also further generates category centers and topological mapping relationships, thus enabling user classification results to go beyond simple grouping and possess stronger structural expressive capabilities, providing a more sufficient basis for subsequent electricity sales decisions.

[0072] This invention further integrates user stratification results, group characteristic parameters, load forecasting results, and electricity sales decision-related parameters into a unified joint forecasting and decision-making model. This eliminates the separation between load forecasting and electricity sales decision-making, allowing them to be completed collaboratively within the same processing chain. Consequently, the generation of electricity sales prices, electricity allocation, energy storage charging and discharging strategies, and demand response incentive parameters becomes more aligned with the actual operating characteristics of different user groups, improving the relevance and adaptability of electricity sales decisions. Compared to existing technologies that underutilize classification results and where forecasting and optimization are independent, this invention is better suited to the complex user types, significant load fluctuations, and diverse energy access methods in incremental distribution networks.

[0073] Furthermore, this invention continuously updates the improved self-organizing mapping network hierarchical model and joint prediction decision model by executing feedback data, enabling user characteristic parameters, feature weight parameters, neighborhood update parameters, and topological relationship parameters to be dynamically corrected according to changes in actual operation. This ensures that user category labels, category centers, and topological mapping relationships remain consistent with the current user characteristic distribution, reducing hierarchical distortion and decision bias caused by changes in user behavior, electricity price fluctuations, or distributed power source status changes, thus improving the dynamic adaptability and long-term stability of the entire electricity sales decision-making process. Therefore, this invention has the beneficial effects of high user hierarchical accuracy, strong targeting of electricity sales decisions, good model update capability, and high adaptability to the complex operating environment of incremental distribution networks. Attached Figure Description

[0074] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0075] Figure 1 This is an overall flowchart of a power sales decision-making method for incremental distribution network users proposed in this invention;

[0076] Figure 2 This is a schematic diagram of a user hierarchical processing flow based on an improved self-organizing map network hierarchical model proposed in this invention.

[0077] Figure 3 This is a structural block diagram of an electricity sales decision-making system for incremental distribution network users proposed in this invention. Detailed Implementation

[0078] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0079] refer to Figures 1-3 A method and system for electricity sales decision-making for incremental distribution network users, comprising the following steps:

[0080] S1. Collect the operation data of each user in the incremental distribution network, and preprocess the operation data to obtain standardized data;

[0081] S2. Based on standardized data, extract user feature parameters, construct user feature vectors, and input the user feature vectors of each user into the improved self-organizing map network hierarchical model to perform feature weighted mapping, competitive mapping, dynamic neighborhood update, topology relationship extraction and category boundary correction to obtain user hierarchical results.

[0082] S3. Based on the user segmentation results, group each user and generate the user set and group feature parameters corresponding to each user group;

[0083] S4. Input the prediction input parameters, user stratification results, group characteristic parameters and electricity sales decision-related parameters of each user group into the joint prediction decision model to obtain the load prediction results of each user group, the total load prediction results of the incremental distribution network and the electricity sales decision parameters of each user group during the target electricity sales period.

[0084] S5. Generate an electricity sales execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters, and send the electricity sales execution plan to the corresponding execution terminal, and collect the execution feedback data corresponding to the electricity sales execution plan;

[0085] S6. A self-organizing mapping network hierarchical model and joint prediction decision model based on updated execution feedback data.

[0086] In this embodiment, incremental distribution network refers to the distribution network built and operated in a park, industrial zone, commercial zone or integrated energy area;

[0087] Electricity sales decision-making refers to the process of determining electricity sales prices, electricity volume allocation, energy storage operation arrangements, and demand response incentive arrangements for different users or different user groups.

[0088] Operational data refers to a set of data that reflects user status, price status, energy status, and distribution network status;

[0089] Standardized data refers to data in a unified format that has undergone missing value completion, outlier removal, timestamp alignment, and unit normalization.

[0090] Timestamp alignment refers to unifying data from different sources and with different sampling frequencies under the same time reference.

[0091] Dimensional normalization refers to converting data with different units and magnitudes into a unified numerical range so that subsequent models can process them uniformly.

[0092] In this embodiment, S1 includes:

[0093] S11 collects operational data from each user within the incremental distribution network. The operational data includes historical electricity consumption data, real-time load data, time-of-use electricity price data, market electricity purchase data, meteorological data, demand response data, distributed power generation operational data, energy storage operational data, and distribution network operational data.

[0094] S12, perform missing value completion and outlier removal processing on the running data;

[0095] S13, perform timestamp alignment and dimension normalization on the processed running data to obtain standardized data;

[0096] S14. The standardized data is linked and organized according to user number, collection period and data category to form standardized input data for subsequent model calls.

[0097] In this embodiment, S2 includes:

[0098] S21, based on standardized data, extract load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, and distributed power usage characteristic parameters;

[0099] S22, load characteristic parameters include load level parameters, fluctuation degree parameters and peak-valley change parameters, price response characteristic parameters include price change response parameters, demand response characteristic parameters include demand response participation parameters, distributed power usage characteristic parameters include distributed power access parameters and distributed power usage parameters;

[0100] S23, Configure the corresponding feature weight parameters according to the degree of influence of various feature parameters on user segmentation;

[0101] S24, weighted combination of various feature parameters and corresponding feature weight parameters to form user feature vectors for each user;

[0102] S25, input the user feature vectors of each user into the improved self-organizing map network hierarchical model and perform competitive mapping processing;

[0103] S26, Perform dynamic neighborhood update processing based on the competition mapping result;

[0104] S27, Extract topological proximity relationships, topological boundary relationships, and topological hierarchy relationships based on the updated neuron distribution results;

[0105] S28, generate user segmentation results, which include category labels, category centers, and topological mapping relationships.

[0106] In this embodiment, user characteristic parameters refer to a set of data parameters used to characterize user operation behavior and electricity sales-related behavior;

[0107] A user feature vector is a fixed-length multidimensional data sequence formed by combining multiple feature parameters in a preset order.

[0108] Feature weight parameters refer to numerical parameters that indicate the proportion of different categories of features in the user segmentation process.

[0109] The improvement here is that instead of roughly classifying users based on a single electricity consumption or load indicator, it simultaneously introduces load characteristics, price response characteristics, demand response characteristics, and distributed power generation usage characteristics, and configures characteristic weight parameters to distinguish the contribution of different types of characteristics in user stratification, thereby making the user stratification results closer to the actual operating status of incremental distribution network users.

[0110] In this embodiment, the improved self-organizing mapping network hierarchical model includes an input layer, a feature weighting layer, a competitive mapping layer, a dynamic neighborhood update layer, a topology relationship extraction layer, a category boundary correction layer, and a hierarchical result output layer.

[0111] The input layer receives the user feature vectors of each user and combines the user feature vectors of multiple users into batch input data according to the user order. The batch input data adopts a two-dimensional data table format, with each row corresponding to a user and each column corresponding to a feature parameter.

[0112] The feature weighting layer receives batch input data and feature weight parameters, performs weighting processing on each feature in each user feature vector, and outputs the weighted user feature vector.

[0113] The competitive mapping layer receives the weighted user feature vectors and compares them with the weights of the competing neurons, outputting the position of the winning neuron for each user;

[0114] The dynamic neighborhood update layer receives the location of the winning neuron and the current weight distribution of the competing layer, and updates the neighboring neurons accordingly.

[0115] The topology extraction layer receives the updated neuron distribution results and generates topological proximity relationships, topological boundary relationships, and topological hierarchy relationships.

[0116] The category boundary correction layer receives topological boundary relationships and performs boundary correction on user samples located in category transition regions;

[0117] The layered output layer receives the corrected mapping results and generates category labels, category centers, and topological mapping relationships.

[0118] In this embodiment, the self-organizing map network hierarchical model refers to a model that aggregates and hierarchizes user samples through the mapping relationship between competing neurons;

[0119] Competitive mapping refers to the process by which multiple competing neurons match the same user feature vector, and the neuron with the highest matching degree is determined as the winning neuron for that user.

[0120] Dynamic neighborhood update refers to not using a fixed update range, but rather adaptively adjusting the update region around the winning neuron based on the current sample distribution, class clustering degree, and boundary changes, and then synchronously updating neighboring neurons.

[0121] Compared with existing ordinary self-organizing map network models, this invention adds feature weight settings during user hierarchical processing, allowing the role of different types of features in hierarchical calculation to be adjusted according to actual needs. During the mapping update process, a dynamic neighborhood update method is adopted, where the neighborhood range is no longer fixed or reduced according to a single rule, but is adjusted based on sample distribution and changes in class boundaries, thereby improving the hierarchical adaptability to different distribution areas. In the hierarchical result output process, in addition to class labels, class centers and topological mapping relationships are also output, and boundary correction is performed on samples located in the transition area between adjacent classes, thus giving the hierarchical results better discriminability and structural integrity.

[0122] In this embodiment, topological mapping relationship refers to the structural relationship between different categories or different users in the mapping space;

[0123] Topological proximity refers to the degree of proximity between different users, user groups, or categories in the mapping space. The higher the degree of proximity, the more similar their operational characteristics are.

[0124] Topological boundary relationships refer to the boundary states between adjacent categories in the mapping space, used to represent boundary locations, transition regions, and boundary changes.

[0125] Topological hierarchy refers to the hierarchical distribution of different categories, user groups, or users in the mapping space, used to indicate their position in the upper, lower, or adjacent layers of the mapping structure.

[0126] The category center refers to the central location data of a sample within a certain category in the mapping space.

[0127] By introducing the above three types of topological relationships, this invention not only provides the result of "which category it belongs to", but also provides a structured result of "which category it is close to", "where the boundary is located", and "what level it is at", thereby enhancing the usability of the hierarchical results. This is also the important significance of this invention compared with the existing ordinary self-organizing map network model.

[0128] In this embodiment, the training data for the improved self-organizing mapping network hierarchical model comes from the historical operation database, which consists of user operation data, price operation data, demand response data, distributed power source operation data, energy storage operation data and distribution network operation data from multiple historical periods.

[0129] During model training, standardized data for the corresponding time period in the historical operation database are converted into user feature vectors and input into the model in batches.

[0130] Since the self-organizing map network hierarchical model is an unsupervised hierarchical model, there is no need to manually label each category during the training phase. After training, the generated categories can be interpreted by combining the category center and the feature parameters of each group. For example, one category can be interpreted as a high-load, high-volatility user group, another category as a price-sensitive user group, and yet another category as a high-response, high-participation user group.

[0131] In this embodiment, the training objectives of the improved self-organizing map network hierarchical model include competitive mapping error control, neighborhood smoothing control, and boundary stability control.

[0132] Among them, competition mapping error control is used to reduce the difference between the user feature vector and its winning neuron, neighborhood smoothing control is used to reduce the degree of mutation between adjacent neurons, and boundary stability control is used to reduce the irregular fluctuations of adjacent class boundaries in continuous training cycles.

[0133] Training parameters include training batch size, number of training rounds, initial learning step size, initial range of neighborhood update, and weight decay parameters. During training, when the change in neuron distribution is lower than the preset change threshold for multiple consecutive training cycles, and the change in class center and boundary position is lower than the preset stability threshold, the model is deemed to have reached the convergence condition, training is stopped, and the model parameters are saved.

[0134] Through the training method described above, the improved self-organizing map network hierarchical model can form a stable and interpretable user hierarchical structure.

[0135] In this embodiment, S3 includes:

[0136] S31, Read the category labels, category centers, and topological mapping relationships from the user stratification results;

[0137] S32, group users according to their category labels and generate a user set for each user group;

[0138] S33, based on the load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, distributed power usage characteristic parameters and topology mapping relationship of users in each user group, generate the group characteristic parameters corresponding to each user group.

[0139] S34, write the topological proximity parameters, topological boundary parameters, and topological hierarchy parameters corresponding to the topological mapping relationship into the group feature parameters;

[0140] S35, based on the written group feature parameters, determine the degree of association, adjacency relationship, and hierarchical differences between different user groups, and use them as input to the joint prediction decision model. Here, the user set refers to the set of users under the same category label; the group feature parameters refer to the comprehensive parameter set formed for the entire user group. The improvement of this invention in this step is that it not only uses category labels to complete user grouping, but also writes the topological mapping relationship into the group feature parameters, so that the adjacency relationship and boundary relationship between user groups are also included in the subsequent decision-making process, thereby improving the continuity and adaptability of subsequent prediction and decision-making.

[0141] In this embodiment, S4 includes:

[0142] S41, Construct a joint forecasting and decision-making model, which includes a forecasting input construction unit, a load forecasting unit, a state set construction unit, and an electricity sales optimization unit;

[0143] S42, input the prediction input parameters of each user group into the load prediction unit to obtain the load prediction results of each user group and the total load prediction results of the incremental distribution network;

[0144] S43, input the user stratification results, group characteristic parameters, load forecast results and electricity sales decision-related parameters into the state set construction unit to form a joint decision state set;

[0145] S44 inputs the joint decision state set into the electricity sales optimization unit to obtain the electricity sales decision parameters for each user group.

[0146] In this embodiment, the internal structure of the joint prediction decision model includes a prediction input construction unit, a load prediction unit, a state set construction unit, and an electricity sales optimization unit.

[0147] The prediction input building unit receives historical load sequences, time stamps, meteorological characteristics, and demand response status, and organizes the above data into time-series input data and static input data. The time-series input data adopts a sequence data format arranged in chronological order, and the static input data adopts a vector data format arranged by user group.

[0148] The load forecasting unit receives time-series input data and static input data. Internally, it includes an input mapping sublayer, a time-series feature extraction sublayer, a feature integration sublayer, and a forecast output sublayer. The input mapping sublayer maps data from different sources into a uniform-length input representation. The time-series feature extraction sublayer extracts the time variation patterns in historical load sequences. The feature integration sublayer fuses time-series features with static input features. The forecast output sublayer outputs the load forecast results for each user group.

[0149] The state set construction unit receives user hierarchical results, group characteristic parameters, load forecasting results, and electricity sales decision-related parameters, and splices and organizes them into a joint decision state set. The joint decision state set adopts a data table format arranged in a fixed order, with each row corresponding to a user group and each column corresponding to a state parameter.

[0150] The electricity sales optimization unit receives the joint decision state set and outputs the electricity sales price, electricity sales allocation results, energy storage charging and discharging power, and demand response incentive parameters.

[0151] In this embodiment, the joint forecasting decision model refers to a model that completes load forecasting and electricity sales optimization within the same processing framework;

[0152] The forecast input parameters include historical load series, time stamps, meteorological characteristics, electricity price signals, and demand response status;

[0153] Parameters relevant to electricity sales decisions include market electricity purchase prices, predicted output of distributed power sources, energy storage status of charge, line load factor, and node voltage.

[0154] The joint decision state set refers to the data set formed by combining user hierarchical information, group characteristic information, prediction results, and operational constraint information.

[0155] The electricity sales optimization unit uses the soft actor commentator algorithm to generate electricity sales decision parameters for each user group. The soft actor commentator algorithm is a reinforcement learning solution algorithm.

[0156] Electricity sales decision parameters include electricity price, electricity sales volume allocation results, energy storage charging and discharging power, and demand response incentive parameters.

[0157] Compared with existing methods that separate prediction and optimization, the improvement of this invention lies in inputting user stratification results, group characteristic parameters and load prediction results into a joint prediction decision model. This makes prediction and optimization no longer separate, but are completed collaboratively in the same model, thereby improving the consistency between the electricity sales decision results of different user groups and the actual operating status.

[0158] In this embodiment, the training data for the joint prediction decision model comes from the historical operation database and the historical execution database;

[0159] The historical operation database provides load sequences, electricity price data, meteorological data, demand response status, distributed power generation status, energy storage status, and distribution network status.

[0160] The historical execution database provides historical electricity sales prices, actual electricity sales volume, energy storage execution results, user response results, revenue results, and deviation results.

[0161] The load forecasting unit in the joint forecasting decision model is trained using a supervised training method, with its training labels being the actual future load values.

[0162] The electricity sales optimization unit in the joint forecasting decision-making model employs reinforcement learning for training. Its training feedback is comprised of electricity sales revenue, electricity purchase and sales deviation, demand response completion rate, energy storage execution deviation, and distribution network constraint satisfaction. In other words, while the "labeling method" of the load forecasting part uses the actual future load as the target value, the "labeling method" of the electricity sales optimization part does not use manually provided standard answers, but rather uses the reward and penalty feedback generated by the execution results as the training basis.

[0163] In this embodiment, the training objectives of the joint forecasting decision model include load forecasting error control and electricity sales decision revenue optimization; wherein, load forecasting error control is used to reduce the difference between the load forecast results and the actual load.

[0164] Electricity sales decision-making revenue optimization is used to improve electricity sales revenue, reduce the deviation between electricity purchase and sale, improve the demand response completion rate, and control energy storage execution deviation and distribution network operation risks.

[0165] Training parameters include training batch size, number of training epochs, learning rate, experience replay capacity, target network update cycle, and policy update cycle;

[0166] When the load prediction error decreases below a preset threshold over multiple consecutive training periods, and the average decision-making benefit stabilizes over multiple consecutive training periods, while the power purchase / sale deviation and the number of constraint violations are below a preset upper limit, the joint prediction decision model is deemed to have reached the convergence condition, training is stopped, and the model parameters are saved.

[0167] In this embodiment, S5 includes:

[0168] S51, Generate an electricity sales execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters;

[0169] S52 sends the electricity sales execution plan to the electricity sales management platform, energy storage control platform, and user interaction terminal;

[0170] S53 collects execution feedback data corresponding to the electricity sales execution plan. The execution feedback data includes the user's actual electricity consumption, electricity purchase and sales deviation, electricity sales revenue, demand response completion rate, energy storage execution status, and distribution network operation status.

[0171] S54. The execution feedback data is collected and organized according to user groups and execution periods to form a feedback dataset corresponding to the user-level results and the electricity sales execution plan. Here, the electricity sales execution plan refers to the price arrangement, power arrangement, energy storage arrangement, and demand response arrangement that are actually issued and executed within the target electricity sales period; the execution feedback dataset refers to the set of execution result data organized according to user groups and execution periods.

[0172] In this embodiment, S6 includes:

[0173] S61, a self-organizing mapping network hierarchical model and joint prediction decision model based on updated execution feedback data;

[0174] S62, based on the execution feedback data, correct the user characteristic parameters, characteristic weight parameters, neighborhood update parameters and topology relationship parameters. Among them, the correction of user characteristic parameters includes the correction of load characteristic parameters, price response characteristic parameters, demand response characteristic parameters and distributed power usage characteristic parameters; the correction of characteristic weight parameters includes the correction of the weights corresponding to different characteristic parameters; the correction of neighborhood update parameters includes the correction of the neighborhood update range; and the correction of topology relationship parameters includes the correction of topology proximity relationship, topology boundary relationship and topology hierarchy relationship.

[0175] S63, the improved self-organizing map network hierarchical model regenerates class labels, class centers and topological mapping relationships based on updated user feature parameters, feature weight parameters, neighborhood update parameters and topological relationship parameters;

[0176] S64, re-input the updated category labels, category centers, and topological mappings into the joint prediction decision model;

[0177] S65, based on the updated joint forecasting decision model, outputs new load forecasting results and electricity sales decision parameters. The core improvement of this invention at this step is that it not only updates the joint forecasting decision model, but also simultaneously updates the improved self-organizing mapping network hierarchical model. In other words, this invention does not only correct the decision results, but also further applies the execution feedback to the user hierarchical structure itself, so that the category labels, category centers, and topology mapping relationships can be dynamically adjusted according to changes in user operating status, thereby enhancing the system's adaptability to price changes, load fluctuations, distributed power output changes, and changes in demand response behavior.

[0178] The photovoltaic power plant drone autonomous inspection system based on digital twins includes the following modules:

[0179] The data acquisition and preprocessing module collects operational data from each user within the incremental distribution network and preprocesses the operational data to obtain standardized data.

[0180] The user feature extraction and improved self-organizing map hierarchical module extracts user feature parameters based on standardized data, constructs user feature vectors, and inputs the user feature vectors of each user into the improved self-organizing map network hierarchical model to perform feature weighted mapping, competitive mapping, dynamic neighborhood update, topology relationship extraction and category boundary correction to obtain user hierarchical results.

[0181] The user grouping module groups users based on the user segmentation results and generates a user set and group feature parameters for each user group.

[0182] The joint forecasting and decision-making module inputs the forecasting input parameters, user stratification results, group characteristic parameters, and electricity sales decision-related parameters of each user group into the joint forecasting and decision-making model to obtain the load forecasting results of each user group, the total load forecasting results of the incremental distribution network, and the electricity sales decision parameters of each user group during the target electricity sales period.

[0183] The execution feedback module generates an execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters, sends the execution plan to the corresponding execution terminal, and collects the execution feedback data corresponding to the electricity sales execution plan.

[0184] The model update module updates the improved self-organizing map network hierarchical model and joint prediction decision model based on execution feedback data.

[0185] Example 1: To verify the feasibility of this invention in practice, it was applied to the electricity sales operation scenario of a certain industrial park-level incremental distribution network. This park connects various types of electricity users, including continuous production manufacturing users, shift-based processing users, commercial office users, and integrated energy users connected to distributed photovoltaic and energy storage devices. The park's load fluctuates significantly, with large differences between peak and off-peak periods. Distributed power output exhibits periodic changes, and energy storage devices participate in peak shaving and valley filling, as well as demand response regulation. The original electricity sales method mainly relied on industry categories and historical average electricity consumption for rough grouping, combined with fixed rules to generate electricity sales prices and purchase plans. This resulted in problems such as imprecise user segmentation, large load forecasting errors, high deviations in electricity purchase and sales, unstable demand response execution, and untimely energy storage coordination.

[0186] In this scenario, the park's power distribution network connects to over 100 users, with a distributed photovoltaic installed capacity of nearly 7 MW, an energy storage system with a rated power of approximately 4 MW, and a rated energy storage capacity of approximately 8 MWh. The park's peak load during high-load periods approaches 30 MW, while the off-peak load is approximately 11-12 MW, resulting in a significant peak-to-valley difference. Throughout a complete operating cycle, the system continuously collects historical electricity consumption data, real-time load data, time-of-use pricing data, market electricity purchase data, meteorological data, demand response data, distributed power generation operation data, energy storage operation data, and distribution network operation data from each user within the park, generating a large number of operation records. These operation data are then processed through missing value completion, outlier removal, timestamp alignment, and dimension normalization to obtain standardized data. Based on this standardized data, load level parameters, fluctuation parameters, peak-to-valley variation parameters, price change response parameters, demand response participation parameters, distributed power generation access parameters, and distributed power generation usage parameters are extracted to construct user feature vectors. These user feature vectors are then input into an improved self-organizing map network hierarchical model to obtain user hierarchical results.

[0187] In this embodiment, the improved self-organizing mapping network hierarchical model does not simply divide users based on average electricity consumption. Instead, it simultaneously considers load characteristics, price response characteristics, demand response characteristics, and distributed power generation usage characteristics, and further distinguishes the degree of influence of different characteristics by combining feature weight parameters. This makes the user segmentation results closer to the actual operation of the park. After model processing, the system divides park users into multiple user groups, including a high-load stable production group, a price-sensitive processing group, an office and commercial fluctuation group, a photovoltaic-storage synergy group, and an active response group. Compared with the original coarse classification by industry, multiple enterprises that were previously uniformly classified into the same category of manufacturing users are reclassified into different user groups. This allows the system to identify which users are more sensitive to price signals, which users are more suitable for participating in demand response, and which users are more suitable for prioritizing the consumption of local electricity during periods of high photovoltaic output. At the same time, the model also generates category centers and topological mapping relationships, enabling the park's electricity sales system not only to know which group a user belongs to, but also to know the adjacency, boundary positions, and hierarchical relationships between that user group and other user groups, thus providing a richer data foundation for subsequent joint forecasting and electricity sales decisions.

[0188] After completing user segmentation and grouping, the system inputs the historical load sequence, time stamps, meteorological characteristics, and demand response status of each user group into the load forecasting part of the joint forecasting decision model, forming the load forecasting results for each user group and the total load forecasting results for the park. Then, the user segmentation results, group characteristic parameters, forecasting results, market electricity purchase price, distributed power generation forecast output, energy storage state of charge, line load factor, and node voltage are combined to construct a joint decision-making state set, which is input into the electricity sales optimization part of the joint forecasting decision model. This outputs the corresponding electricity sales price, electricity sales allocation results, energy storage charging and discharging power, and demand response incentive parameters for each user group, forming an electricity sales execution plan. This plan is sent to the electricity sales management platform, energy storage control platform, and user interaction terminal. In actual operation, it can make differentiated arrangements based on the behavioral characteristics of different user groups. For example, it can increase the intensity of peak-valley price differentiation guidance for price-sensitive processing groups, provide more suitable local consumption pricing strategies for photovoltaic-storage collaborative groups during the midday high-output period of distributed power generation, and set more matching incentive parameters for responsive groups, making the behavior of various users more consistent with the overall operation goals of the park.

[0189] To verify the beneficial effects of this invention, a comparison was made between the original rule-based electricity sales scheme and the intelligent electricity sales scheme after adopting this invention. Under the same operating cycle and similar load conditions, the average absolute percentage error of the total load forecast in the park under the original scheme remained at around 9.6%. After adopting this invention, this indicator decreased to around 4.7%, a reduction of more than half; the average absolute deviation of peak load forecast decreased from over 2 MW to around 1 MW; the electricity purchase and sale deviation rate decreased from over 6% to around 2%; the gross revenue from electricity sales in the park increased by approximately 8% to 10%; the average completion rate of demand response tasks increased from around 70% to nearly 90%; the energy storage execution deviation rate decreased by more than half; the local consumption ratio of distributed photovoltaic power increased by approximately 7 percentage points to 9 percentage points; and the number of node voltage overruns significantly decreased. For park operators, this means that under the same load basis, the electricity purchase plan is more accurate, the electricity sales arrangement is more reasonable, the energy storage coordination is more timely, and the overall operational risk is lower.

[0190] Results from typical high-load operation days show that under the original scheme, peak-hour electricity purchases exceeded the planned amount by more than 10 MWh, while after adopting this invention, this figure decreased to about one-third of the original. Under the original scheme, the park's peak load was close to 30 MW, while after adopting this invention, the peak load decreased by about 1.5 MW to 2 MW, demonstrating a significant peak-shaving effect. The number of users participating in demand response increased by more than 10 during the same period, and the daily electricity sales revenue increased by more than 10,000 yuan compared to the original scheme. For price-sensitive processing groups, after adopting this invention, their peak-hour electricity transfer ratio increased from less than 6% to more than 12%; for photovoltaic-storage synergy groups, the midday local consumption ratio increased from more than 70% to nearly 90%; and for active response groups, the demand response incentive fulfillment rate increased from less than 70% to more than 90%. Therefore, this invention does not simply increase prices to generate revenue, but rather through more refined user segmentation and more targeted electricity sales decisions, enabling different user groups to participate in park operation optimization in a way more suitable to their own operating characteristics.

[0191] During continuous operation, the system updates the improved self-organizing mapping network hierarchical model and joint prediction decision model based on execution feedback data. Specifically, the system continuously collects users' actual electricity consumption, electricity purchase and sale deviation, electricity sales revenue, demand response completion rate, energy storage execution status, and distribution network operation status. It then corrects user characteristic parameters, characteristic weight parameters, neighborhood update parameters, and topology relationship parameters to regenerate category labels, category centers, and topology mapping relationships. After a period of continuous operation, some users who originally belonged to the office / commercial fluctuation group are reassigned to the price-sensitive processing group or the boundary area near the active response group due to changes in their operating patterns. The system re-assigns these users through a category boundary correction mechanism. After the model update, the average absolute percentage error of the park's total load forecast further decreases, the electricity purchase and sale deviation rate continues to decrease, and electricity sales revenue continues to increase. This demonstrates that the present invention not only solves the problems of coarse user segmentation and insufficient decision adaptability during initial deployment but also continuously corrects the hierarchical and decision results through execution feedback, ensuring the system maintains a high level of adaptability.

[0192] As can be seen from this embodiment, in incremental distribution network scenarios with complex user types, high proportion of distributed power sources and energy storage, significant price fluctuations, and frequent demand response participation, this invention can effectively solve problems in existing technologies such as imprecise user segmentation, insufficient utilization of classification results, separation of prediction and optimization, and weak feedback update capabilities. The final effects are reflected in several aspects, including user segmentation results that are more consistent with actual operating characteristics, significantly reduced load forecasting errors, significantly reduced power purchase and sales deviations, significantly improved demand response completion rate, enhanced energy storage coordination capabilities, improved local absorption capacity of distributed power sources, continuous improvement in power sales revenue, and more stable distribution network operation. Therefore, this invention has strong engineering application value.

[0193] Table 1. Comparison of the effects of adopting this invention on electricity sales decisions in park-level incremental distribution networks.

[0194]

[0195] As can be seen from the table, after adopting this invention, the effectiveness of electricity sales decisions in the park-level incremental distribution network has significantly improved across multiple key indicators. This indicates that this invention is not simply a modification of the original rule-based scheme, but rather establishes a more complete closed-loop mechanism between user segmentation, prediction, decision-making, and execution feedback. The original scheme only roughly categorized users based on industry attributes and historical average electricity consumption, ultimately resulting in only three user groups. While this approach is simple to implement, it cannot fully reflect the differences among users in terms of load fluctuations, price response, demand response participation, and the use of distributed power sources. In contrast, this invention divides users into five user groups, significantly improving the granularity of grouping. This demonstrates that this invention can more accurately identify the actual operating characteristics of different users, which is also an important foundation for subsequent improvements in prediction accuracy and optimization of electricity sales decisions.

[0196] The improvements made by this invention are particularly significant in load forecasting. The average absolute percentage error of the total load forecast decreased from 9.6% to 4.7%, a reduction of over 50%, indicating that the invention more closely approximates the actual operating conditions in judging the overall load trend. Furthermore, the average absolute deviation of peak load forecasting decreased from 2.38 MW to 1.04 MW, and the average absolute deviation of valley load forecasting decreased from 1.42 MW to 0.73 MW, demonstrating that the invention not only improves overall forecast accuracy but also provides more accurate characterization of peak and valley periods. For incremental distribution network electricity sales scenarios, the forecast accuracy of peak and valley periods directly affects power purchase plans, energy storage dispatch, and electricity pricing strategies; therefore, these improvements have strong practical significance.

[0197] In terms of controlling power purchase and sales deviations, this invention also demonstrates strong advantages. The power purchase and sales deviation rate decreased from 6.4% to 2.2%, and the peak-hour over-planned power purchase decreased from 14.2 MWh to 4.9 MWh, indicating a better match between the power purchase plan formulated by the park's power sales side and the actual operating results. In other words, this invention can significantly reduce power purchase and sales deviations caused by inaccurate forecasts or unreasonable grouping, reducing the need for temporary adjustments to power purchase plans, thereby reducing operational risks and additional costs. At the same time, the park's peak load decreased from 29.1 MW to 27.4 MW, a reduction of 1.7 MW, indicating that this invention has also achieved significant results in peak shaving, which has a positive effect on the stability of distribution network operation and the safety margin of equipment.

[0198] From an economic perspective, the improvements brought about by this invention are also quite significant. Gross revenue from electricity sales increased from 3.815 million yuan to 4.148 million yuan, an increase of 333,000 yuan, representing a revenue improvement of 8.73%. This demonstrates that this invention not only improves technical indicators but also directly translates into economic benefits for electricity sales operations. The table also shows that the maximum daily revenue increase on a typical high-load day exceeds 14,000 yuan, indicating that the advantages of this invention are even more pronounced in scenarios with significant load fluctuations, high electricity price sensitivity, or large variations in distributed power generation output. In other words, the revenue improvement effect of this invention under complex operating conditions is more representative.

[0199] In terms of demand response and energy storage coordination, this invention has also achieved significant improvements. The average completion rate of demand response increased from 71.8% to 89.6%, and the number of users participating in demand response increased from 46 to 63. This indicates that the invention can more accurately identify suitable users for participation and improve user participation through more reasonable pricing and incentive mechanisms. The energy storage execution deviation rate decreased from 12.1% to 5.3%, indicating that energy storage dispatch is closer to actual operational needs, and the role of energy storage in peak shaving and valley filling and coordinating with electricity sales strategies is more fully realized. This shows that the invention does not simply optimize electricity prices, but achieves more effective linkage between electricity sales, power allocation, energy storage, and demand response.

[0200] From the perspective of distributed power generation and distribution network operation, this invention also demonstrates good overall performance. The local consumption rate of distributed photovoltaic power increased from 83.4% to 91.2%, indicating that this invention can better guide local consumption and reduce the pressure of curtailment or external transmission. The number of node voltage over-limit incidents decreased from 19 to 6, a reduction of nearly 70%, indicating that while increasing electricity sales revenue, this invention did not sacrifice the stability of distribution network operation; on the contrary, it improved the node voltage operation and network operation quality, which is of great significance for the safe operation of the industrial park distribution network.

[0201] The refined advantages of this invention are more evident when considering the segmented effects on different types of users. The peak-shifting rate for price-sensitive users increased from 5.6% to 12.8%, indicating that this invention is more effective in guiding price signals. The midday local consumption rate for photovoltaic-storage synergy users increased from 76.3% to 88.7%, demonstrating that this invention can more effectively utilize the synergy between photovoltaics and energy storage. The incentive fulfillment rate for responsive users increased from 68.9% to 92.4%, indicating that the incentive parameters are designed to better suit users' actual responsiveness. Meanwhile, the number of work orders regarding unreasonable electricity prices decreased from 23 to 9, indicating that this invention, while improving operational efficiency, also increases user acceptance and recognition of the electricity sales plan.

[0202] In summary, the data reflected in the table above demonstrates that this invention effectively addresses the problems of coarse user segmentation, large forecasting deviations, mismatched power purchase and sales plans, insufficient demand response execution, and weak energy storage coordination capabilities in traditional rule-based power sales schemes. This invention achieves fine-grained user segmentation through an improved self-organizing mapping network hierarchical model, coordinates load forecasting and power sales optimization through a joint forecasting decision model, and continuously updates model parameters and hierarchical structure based on execution feedback. Therefore, it has achieved good implementation results in multiple dimensions, including technical indicators, economic benefits, operational stability, and user acceptance.

[0203] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for electricity sales decision-making for incremental distribution network users, characterized in that, Includes the following steps: S1. Collect the operation data of each user in the incremental distribution network, and preprocess the operation data to obtain standardized data; S2. Based on standardized data, extract user feature parameters, construct user feature vectors, and input the user feature vectors of each user into the improved self-organizing map network hierarchical model to perform feature weighted mapping, competitive mapping, dynamic neighborhood update, topology relationship extraction and category boundary correction to obtain user hierarchical results. S3. Based on the user segmentation results, group each user and generate the user set and group feature parameters corresponding to each user group; S4. Input the prediction input parameters, user stratification results, group characteristic parameters and electricity sales decision-related parameters of each user group into the joint prediction decision model to obtain the load prediction results of each user group, the total load prediction results of the incremental distribution network and the electricity sales decision parameters of each user group during the target electricity sales period. S5. Generate an electricity sales execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters, and send the electricity sales execution plan to the corresponding execution terminal, and collect the execution feedback data corresponding to the electricity sales execution plan; S6. A self-organizing mapping network hierarchical model and joint prediction decision model based on updated execution feedback data.

2. The electricity sales decision-making method for incremental distribution network users according to claim 1, characterized in that, S1 includes the following steps: S11. Collect the operation data of each user in the incremental distribution network. The operation data includes historical electricity consumption data, real-time load data, time-of-use electricity price data, market electricity purchase data, meteorological data, demand response data, distributed power generation operation data, energy storage operation data, and distribution network operation data. S12. Perform missing value completion and outlier removal processing on the running data; S13. Perform timestamp alignment and dimension normalization on the processed running data to obtain standardized data; S14. The standardized data is linked and organized according to user number, collection period and data category to obtain standardized data.

3. The electricity sales decision-making method for incremental distribution network users according to claim 1, characterized in that, S2 includes the following steps: S21. Extract load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, and distributed power usage characteristic parameters based on the standardized data; S22. Load characteristic parameters include load level parameters, fluctuation degree parameters and peak-valley change parameters; price response characteristic parameters include price change response parameters; demand response characteristic parameters include demand response participation parameters; distributed power usage characteristic parameters include distributed power access parameters and distributed power usage parameters. S23. Configure corresponding feature weight parameters according to the degree of influence of various feature parameters on user segmentation; S24. Input the user feature vectors of each user into the improved self-organizing map network hierarchical model, perform competitive mapping processing, and obtain the competitive mapping result; S25. Perform dynamic neighborhood update processing based on the competition mapping results to obtain the updated neuron distribution results; S26. Extract topological proximity relationships, topological boundary relationships, and topological hierarchy relationships based on the updated neuron distribution results; S27. Perform category boundary correction processing based on topological boundary relationships to generate user stratification results, which include category labels, category centers, and topological mapping relationships.

4. The electricity sales decision-making method for incremental distribution network users according to claim 3, characterized in that, The improved self-organizing map network hierarchical model includes an input layer, a competition layer, a dynamic neighborhood update unit, a topology relationship extraction unit, a category boundary correction unit, and a hierarchical result output unit. The input layer receives the user feature vectors from each user and transmits them to the competition layer in the order of features. The competition layer consists of multiple competing neurons arranged according to a preset topology. It performs competitive mapping processing on the input user feature vector and determines the winning neuron corresponding to each user based on the degree of mapping matching between each competing neuron and the user feature vector. The dynamic neighborhood update unit determines the set of neighboring neurons centered on the winning neuron, and adjusts the neighborhood update range according to the user feature distribution density, category clustering degree, mapping distance between adjacent neurons, and boundary changes between adjacent categories, and then iteratively updates the winning neuron and its neighboring neurons. The topology extraction unit extracts topology proximity relationships, topology boundary relationships, and topology hierarchy relationships based on the relative positional relationships of each winning neuron in the competition layer, and forms topology relationship data corresponding to the user's hierarchical results; The category boundary correction unit corrects the boundary region between adjacent categories based on topological boundary relationships, and performs boundary correction on user samples located in the transition region between adjacent categories; The hierarchical result output unit generates category labels based on the winning neuron positions for each user, generates category centers based on the aggregation results of user feature vectors within the same category, and generates topological mapping relationships based on the corrected relative positional relationships of neurons. When user feature distribution changes, the improved self-organizing map network hierarchical model synchronously adjusts the topological proximity, topological boundary, and topological hierarchy relationships based on the updated neuron competition results, and generates category labels, category centers, and topological mapping relationships corresponding to the current user feature distribution.

5. The electricity sales decision-making method for incremental distribution network users according to claim 1, characterized in that, S3 includes the following steps: S31. Read the category labels, category centers, and topological mapping relationships from the user stratification results; S32. Group users according to their category labels and generate a user set for each user group. S33. Based on the load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, distributed power usage characteristic parameters, and topology mapping relationship of users in each user group, generate the group characteristic parameters corresponding to each user group. S34. Write the topological proximity parameters, topological boundary parameters, and topological hierarchy parameters corresponding to the topological mapping relationship into the group feature parameters; S35. Based on the group feature parameters after writing, determine the degree of association, adjacency relationship and hierarchical difference between different user groups, and use them as input to the joint prediction decision model.

6. The electricity sales decision-making method for incremental distribution network users according to claim 1, characterized in that, S4 includes the following steps: S41. Construct a joint forecasting and decision-making model, which includes a forecasting input construction unit, a load forecasting unit, a state set construction unit, and an electricity sales optimization unit. S42. Input the prediction input parameters of each user group into the load prediction unit to obtain the load prediction results of each user group and the total load prediction results of the incremental distribution network. S43. Input the user stratification results, group characteristic parameters, load forecasting results, and electricity sales decision-related parameters into the state set construction unit to form a joint decision state set; S44. Input the joint decision state set into the electricity sales optimization unit to obtain the electricity sales decision parameters for each user group.

7. The electricity sales decision-making method for incremental distribution network users according to claim 6, characterized in that, The prediction input parameters include historical load sequences, time stamps, meteorological characteristics, electricity price signals, and demand response status. Parameters relevant to electricity sales decisions include market electricity purchase prices, predicted output of distributed power sources, energy storage status of charge, line load factor, and node voltage. The joint decision-making state set includes category labels, category centers, topology mapping relationships, group characteristic parameters, load forecast results for each user group, total load forecast results for the incremental distribution network, and parameters related to electricity sales decisions. The electricity sales optimization unit uses the soft actor commentator algorithm to generate electricity sales decision parameters for each user group; Electricity sales decision parameters include electricity price, electricity sales volume allocation results, energy storage charging and discharging power, and demand response incentive parameters.

8. The electricity sales decision-making method for incremental distribution network users according to claim 1, characterized in that, S5 includes the following steps: S51. Generate an electricity sales execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters; S52. Send the electricity sales execution plan to the electricity sales management platform, energy storage control platform and user interaction terminal; S53. Collect execution feedback data corresponding to the electricity sales execution plan. The feedback data includes the user's actual electricity consumption, electricity purchase and sales deviation, electricity sales revenue, demand response completion rate, energy storage execution status, and distribution network operation status. S54. Collect and organize the execution feedback data according to user groups and execution time periods to form a feedback dataset corresponding to user-level results and electricity sales execution plans.

9. The electricity sales decision-making method for incremental distribution network users according to claim 1, characterized in that, S6 includes the following steps: S61. A self-organizing mapping network hierarchical model and joint prediction decision model based on updated execution feedback data; S62. Based on the execution feedback data, the user characteristic parameters, characteristic weight parameters, neighborhood update parameters, and topology relationship parameters are corrected. The correction of user characteristic parameters includes the correction of load characteristic parameters, price response characteristic parameters, demand response characteristic parameters, and distributed power source usage characteristic parameters. The correction of characteristic weight parameters includes the correction of the weights corresponding to different characteristic parameters. The correction of neighborhood update parameters includes the correction of the neighborhood update range. The correction of topology relationship parameters includes the correction of topology proximity relationship, topology boundary relationship, and topology hierarchy relationship. S63. The improved self-organizing mapping network hierarchical model regenerates class labels, class centers, and topological mapping relationships based on updated user feature parameters, feature weight parameters, neighborhood update parameters, and topological relationship parameters. S64. Re-input the updated category labels, category centers, and topological mapping relationships into the joint prediction decision model; S65. Based on the updated joint forecasting decision model, output new load forecasting results and electricity sales decision parameters.

10. A power sales decision-making system for incremental distribution network users, characterized in that, The method for implementing the electricity sales decision-making method for incremental distribution network users as described in any one of claims 1 to 9 includes: The data acquisition and preprocessing module collects operational data from each user within the incremental distribution network and preprocesses the operational data to obtain standardized data. The user feature extraction and improved self-organizing map hierarchical module extracts user feature parameters based on standardized data, constructs user feature vectors, and inputs the user feature vectors of each user into the improved self-organizing map network hierarchical model to perform feature weighted mapping, competitive mapping, dynamic neighborhood update, topology relationship extraction and category boundary correction to obtain user hierarchical results. The user grouping module groups users based on the user segmentation results and generates a user set and group feature parameters for each user group. The joint forecasting and decision-making module inputs the forecasting input parameters, user stratification results, group characteristic parameters, and electricity sales decision-related parameters of each user group into the joint forecasting and decision-making model to obtain the load forecasting results of each user group, the total load forecasting results of the incremental distribution network, and the electricity sales decision parameters of each user group during the target electricity sales period. The execution feedback module generates an electricity sales execution plan for the corresponding target electricity sales period based on the electricity sales decision parameters, sends the electricity sales execution plan to the corresponding execution terminal, and collects the execution feedback data corresponding to the electricity sales execution plan. The model update module updates the improved self-organizing map network hierarchical model and joint prediction decision model based on execution feedback data.

Images