An intelligent distribution load forecasting and adaptive scheduling method

Abstract

The present invention provides an intelligent distribution load forecasting and adaptive scheduling method, aiming to improve the response ability and operation efficiency of the distribution system to load fluctuations. The method collects historical load data, real-time meteorological information and user electricity consumption behavior data of the target distribution area, establishes a multi-dimensional input data set, constructs a short-term load forecasting model based on this data set, and realizes the forecasting of future load levels of multiple feeders and key nodes. According to the forecasting results, potential voltage deviation, load imbalance and equipment overload risks are identified, and then combined with the power grid topology structure and operation constraints, a rolling horizon optimization method is used to generate a dynamic scheduling control strategy. The strategy includes control measures such as transformer tap position adjustment and capacitor bank switching. By sending the scheduling strategy to the device control unit, active adjustment is completed before the load changes, realizing voltage stability, load balance and energy optimization, and providing intelligent and adaptive operation guarantee for the distribution system.

Description

Technical Field

[0001] The present invention relates to the technical field of power dispatching, and particularly to an intelligent distribution load forecasting and adaptive dispatching method. Background Art

[0002] Currently, the dispatching of distribution systems mainly relies on static rules or experience-based control strategies, usually adjusting equipment according to real-time load monitoring results or preset threshold conditions, such as transformer tap regulation, capacitor switching, and feeder load reconfiguration. In some advanced systems, short-term load forecasting technology has also been introduced, using historical load data for trend extrapolation to provide certain auxiliary references for dispatching. However, such forecasting is mostly based on a single data source, with low spatial resolution and no closed-loop linkage with equipment dispatching. In addition, existing dispatching strategies often make adjustments at fixed time intervals and are difficult to respond quickly to sudden load changes.

[0003] With the increasing access of distributed energy, the increasingly complex user load behavior, and the enhanced impact of meteorological conditions on power grid operation, the existing technologies are unable to cope with high-frequency and large-amplitude load fluctuations. Specifically, the current distribution system lags in voltage control, load balancing, and energy storage dispatching, resulting in problems such as voltage over-limit, equipment overload, or uneven load distribution during peak hours, affecting the safety and economy of power supply. In addition, the lack of a response mechanism for the dynamic behavior on the user side also makes the demand-side resources unable to effectively participate in system regulation, further restricting the flexibility and accuracy of dispatching.

[0004] Therefore, there is an urgent need for an intelligent dispatching method that can integrate multi-source data, achieve prediction-driven, and have adaptive capabilities to improve the proactive response ability and operation optimization level of the distribution system under load fluctuation scenarios. Summary of the Invention

[0005] The present application provides an intelligent distribution load forecasting and adaptive dispatching method to improve the response ability and operation efficiency of the distribution system to load fluctuations.

[0006] The present application provides an intelligent distribution load forecasting and adaptive dispatching method, including:

[0007] Obtain historical load data, real-time meteorological information, and user electricity consumption behavior data within the target distribution area, and establish a comprehensive load forecasting input data set;

[0008] Based on the comprehensive load forecasting input data set, construct a short-term load forecasting model to predict the load levels of multiple feeders and key nodes in a preset future period, and obtain the predicted load time series at the node level and feeder level;

[0009] According to the predicted load time series, judge the possible voltage deviation, load imbalance or equipment overload risks of each node and feeder during the prediction period;

[0010] Based on the judgment results of the power grid topology structure, electrical parameter constraints and load prediction time series, adopt a rolling time-domain optimization method to generate a scheduling control strategy, and the scheduling control strategy includes the advance adjustment of the transformer tap position, the pre-switching plan of the capacitor bank, the reclosing sequence and load distribution logic of the feeder tie switch, the charge and discharge time and power setting of the distributed energy storage device, or the demand response execution instruction for the user side;

[0011] Send the scheduling control strategy to the corresponding device control unit, and perform voltage control, load reconstruction, energy regulation or user load response before the prediction period according to the scheduling control strategy.

[0012] The beneficial effects of the technical solution provided by this application include:

[0013] (1) Through short-term prediction by integrating multi-source information such as historical load, real-time weather and user behavior, grasp the load trend in the future period in advance, so that the system can take active control measures before the occurrence of abnormal load, and avoid the operation risks caused by passive response. (2) Based on the prediction results, judge the possible operation risks, and combine with the distribution network structure and equipment constraints to generate an operable optimization scheduling strategy, which helps to stabilize the node voltage, relieve the feeder congestion, and improve the operation safety and reliability of the system. (3) The scheduling strategy covers various resources such as transformers, capacitors, feeder tie switches, energy storage devices and user load responses, supports multi-objective and cross-level comprehensive control, and significantly improves the refined level of the distribution system in terms of resource allocation and energy management. (4) Through the rolling optimization and dynamic execution mechanism, realize the continuous matching of the scheduling strategy with the real-time load change, reduce the energy loss caused by the peak-valley difference, enhance the load balance and energy efficiency management ability of the system, and thus provide a solid support for the construction of the smart grid. Brief Description of the Drawings

[0014] Figure 1 is a flowchart of an intelligent distribution load prediction and adaptive scheduling method provided by the first embodiment of this application. Detailed Embodiment

[0015] Many specific details are set forth in the following description in order to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of this application. Therefore, this application is not limited by the specific embodiments disclosed below.

[0016] The first embodiment of this application provides an intelligent distribution load prediction and adaptive scheduling method. Please refer toFigure 1 , this figure is a schematic diagram of the first embodiment of this application. The following will combine Figure 1 to elaborate in detail on an intelligent distribution load forecasting and adaptive scheduling method provided by the first embodiment of this application.

[0017] Step S101: Obtain historical load data, real-time meteorological information, and user electricity consumption behavior data within the target distribution area, and establish a comprehensive load forecasting input data set.

[0018] Step S101 involves the acquisition and fusion of historical load data, real-time meteorological information, and user electricity consumption behavior data within the target distribution area, aiming to establish a multi-dimensional, temporally continuous, and semantically complete comprehensive load forecasting input data set, providing a high-quality input basis for subsequent short-term load forecasting and optimal scheduling.

[0019] First, for the acquisition of historical load data, the system retrieves the load data recorded by distributed measurement devices (such as smart meters, distribution terminal units DTU, or feeder monitoring units FTU) from various key feeders and nodes in the target distribution area. The load data at least includes active power, reactive power, voltage, current, and their corresponding timestamps. To ensure the spatial accuracy of the data, each set of data needs to be bound with corresponding physical location coding information, such as feeder numbers, node numbers, or geographical coordinates. The acquired data should cover a continuous period in the recent past (such as the past 7 days, 30 days, or 90 days) to reflect the load change patterns under different climates and user behaviors. In the data preprocessing stage, the system cleans the historical data, including removing invalid records generated during communication failures, equipment anomalies, or power outage maintenance, and uses interpolation, moving average, or repair algorithms based on time series models to correct data points with missing or abnormal fluctuations, so as to construct a historical load sample set with good temporal continuity.

[0020] Secondly, the acquisition of real-time meteorological information can be carried out by accessing the national meteorological platform, third-party meteorological service interfaces, or locally deployed environmental monitoring devices. The collected meteorological parameters should at least include ambient temperature, relative humidity, wind speed, wind direction, atmospheric pressure, solar radiation intensity, and weather conditions (such as sunny, cloudy, rainy, snowy, etc.). Since the power supply area of the distribution network may span multiple microclimate sub-regions, to ensure the geographical consistency between meteorological data and load forecasting targets, the system performs spatial interpolation processing on meteorological data, and uses geostatistical methods such as Kriging interpolation and inverse distance weighting to map sparse meteorological observation point data to the service ranges of each feeder or node in the distribution network, thereby obtaining a set of meteorological factors with spatial labels. During the processing, meteorological change trend indicators (such as the temperature change slope in the next three hours) can also be introduced as auxiliary features for load forecasting to enhance the model's perception ability of load fluctuations under sudden weather events.

[0021] For the acquisition of users' electricity consumption behavior data, the system obtains high-frequency electricity consumption records at the user end by connecting to smart meters, building energy consumption management systems, home energy management systems (HEMS), or commercial customer energy consumption collection platforms. The recorded content includes electricity consumption, voltage levels, electrical appliance operating status, demand data, etc. every 15 minutes or every hour. To achieve behavior modeling, the system divides users into multiple categories, such as residential users, commercial users, small industrial users, etc., and then extracts typical load characteristics based on time patterns in historical data, including peak and valley periods during the day, differences between weekdays and holidays, load curve stability, response to electricity price changes, etc., and uses clustering algorithms (such as K-means, DBSCAN, etc.) to form user portraits with similar behavior characteristics. In addition, external variables can be introduced, such as whether the user participates in demand response, the installation of photovoltaic power generation or energy storage devices, etc., to enhance the semantic level of the input data.

[0022] After the above three types of data are preprocessed and feature-extracted, they are uniformly formatted and fused into a structured comprehensive input data set. This data set is constructed into a sliding window structure of continuous time periods with time as the main axis. Each prediction period generates a composite sample containing load characteristics, meteorological characteristics, and user behavior characteristics, which is used to drive the subsequent short-term load prediction model. The system has the ability to dynamically update, that is, as real-time data continues to be accessed, it can update this data set at a fixed time step (such as every 5 minutes or 15 minutes) to ensure that the prediction input is consistent with the current system state.

[0023] In summary, step S101 provides a complete path for data acquisition, cleaning, spatial matching, behavior modeling, and input structure construction, forming a load prediction input data preparation mechanism that fuses multi-source heterogeneous information, providing accurate, real-time, and highly context-related data support for the intelligent power distribution scheduling strategy of the present invention.

[0024] Furthermore, the acquisition of historical load data, real-time meteorological information, and users' electricity consumption behavior data in the target distribution area, and the establishment of a comprehensive load prediction input data set, includes:

[0025] Obtain the historical active and reactive load data of multiple monitoring points within the time window, classify and group them according to the feeder and node structure, and perform missing value filling, anomaly elimination, and trend stability analysis on them to form a basic load sample set with complete time series;

[0026] Receive and fuse real-time meteorological information from multiple sources, including temperature, humidity, wind speed, sunshine intensity, and weather type, map the meteorological data to the coverage area of each feeder through a spatial interpolation algorithm, and construct a meteorological feature set with geographical annotations;

[0027] Collect the electricity consumption behavior information on the user side, including the periodic electricity consumption pattern, peak-valley electricity consumption response habit, electricity price sensitivity, and device operation logs, and perform clustering modeling based on user categories to generate a behavior label feature matrix;

[0028] Based on the load sample set, meteorological feature set, and behavior label feature matrix, construct a unified multi-dimensional input data structure, and dynamically update the data set according to the rolling prediction period to drive the subsequent short-term load prediction model.

[0029] In this embodiment, through means such as data cleaning, feature extraction, spatial mapping, and behavior modeling, a multi-source fusion input data set with time continuity, spatial consistency, and semantic identifiability is constructed to support the training and real-time deduction of the short-term load prediction model.

[0030] First, in terms of load data processing, the system collects historical load data within a specified time window from multiple monitoring points in the target distribution area. This data should include two dimensions of active power (P) and reactive power (Q), and have a unified timestamp mark. To ensure the spatial resolvability of the data, the system classifies each monitoring point according to its location in the feeder and the node in the network topology, thus forming a preliminary load database for hierarchical management. On this basis, perform data cleaning operations on the collected raw data. First, identify and fill in missing values. Preferred filling methods include moving average based on time series, interpolation algorithm, or weighted mapping of adjacent node data. Secondly, the system uses statistical detection and machine learning models to identify outliers in the data, such as load jumps caused by sensor failures, communication packet losses, or human misoperations, and perform deletion or replacement processing. Finally, perform trend stability analysis on the processed load data to judge the fluctuation characteristics, periodicity, and abnormal mutation frequency of the data within the selected time window, ensuring that the constructed basic sample has a stable time series structure and can be effectively adopted by downstream models.

[0031] Secondly, in terms of meteorological information acquisition and processing, the system needs to synchronously receive real-time meteorological data from multiple sources, which can include national meteorological services, third-party commercial API interfaces, or locally deployed micro-meteorological stations. The meteorological parameters to be collected include but are not limited to ambient temperature, relative humidity, wind speed, sunshine intensity, and weather type (such as sunny, rainy, snowy, cloudy, etc.). Since the spatial distribution of meteorological observation points does not necessarily coincide with the nodes of the distribution network, the system maps the meteorological data to the service areas of each feeder and around key nodes through geospatial interpolation algorithms (such as inverse distance weighting, Kriging interpolation, or Voronoi diagram method), forming a meteorological feature set with spatial identifiers. Each prediction unit (such as a feeder segment or transformer node) will be associated with a set of meteorological features after interpolation and smoothing, ensuring the geographical consistency of the prediction model input.

[0032] Third, in terms of modeling users' electricity consumption behavior, the system needs to collect fine-grained electricity consumption data from the user side. The data sources include smart meters, building automation systems, industrial control systems, or user energy consumption management platforms. The collected content covers periodic electricity consumption patterns (such as daily electricity consumption time rules), peak-valley response behaviors (such as whether there is load curtailment), sensitivity to electricity price fluctuations (for example, whether there is the ability to automatically postpone non-critical loads when the electricity price changes), and operation logs of terminal devices (such as the opening frequency and duration of devices like air conditioners, water heaters, and charging piles). The system conducts behavior clustering according to the user type (such as residential, commercial, light industrial, agricultural users, etc.), and uses algorithms such as K-means, DBSCAN, or GMM to build category models, further extracts representative behavior labels of users in load forecasting, and forms a structured behavior label feature matrix to describe the behavior driving force behind the user load change.

[0033] The behavior label feature matrix refers to a structured data set organized in the form of a two-dimensional array or tensor, which is used to quantitatively describe the key features of different types of users in electricity consumption behavior. Each row of this matrix corresponds to a user individual or user group, and each column corresponds to a behavior feature dimension, which is used to reflect the performance value or category label of the user in a certain behavior attribute.

[0034] The behavior feature columns in the behavior label feature matrix can at least include the following dimensions:

[0035] Periodic electricity consumption pattern labels, such as "daytime electricity consumption dominant", "nighttime electricity consumption dominant", "uniform throughout the day", or "significant weekend load increase", and the corresponding values can be discrete categorical encodings, such as 0, 1, 2, 3;

[0036] Peak-valley response sensitivity index, which represents the degree of response of users to electricity price fluctuations or dispatching instructions, and can be a normalized floating point number between 0 and 1. The higher the value, the stronger the response ability;

[0037] Electricity price elasticity coefficient, which is a parameter describing the load change trend of users under different electricity prices. For example, it can be obtained by fitting a regression model of historical electricity prices and electricity consumption, and reflects the influence intensity of electricity prices on user behavior;

[0038] Equipment load composition vector, which reflects the usage frequency of various high-energy-consuming equipment and its proportion in the total load, such as air conditioners, water heaters, charging piles, etc. The proportion of each type of equipment constitutes a dimension in the vector;

[0039] The load fluctuation stability coefficient reflects the volatility of the user's load, such as the standard deviation or coefficient of variation calculated based on the historical load sequence, and is used to describe the difficulty of predicting the user's load; and a Boolean tag indicating whether to participate in the demand response program, which is used to distinguish whether the user has the ability to actively adjust the load and can take values of 1 (participate) or 0 (not participate).

[0040] After obtaining and extracting the data of the above three dimensions, the system performs multi-dimensional fusion on the load sample set, meteorological feature set, and user behavior label matrix to construct an input data structure in a unified format. This structure is usually represented in the form of a tensor or a table, and each sample contains multiple time steps (such as the past 24 hours), multiple feature dimensions (such as power, temperature, behavior label, etc.), and a spatial index (such as node number, feeder ID). To adapt to the continuous change of data in actual operation, this data structure is dynamically updated according to the rolling prediction period, for example, updated every 5 minutes, 15 minutes, or 30 minutes, so that the prediction model always receives the latest and complete input information during operation.

[0041] In summary, the described dataset construction method not only fully considers the spatial structure, electrical characteristics, and user differences of the distribution system, but also ensures the high quality and timeliness of the input data through data processing and modeling means, providing a solid foundation for constructing a short-term load prediction model with high precision, interpretability, and strong robustness, and is a key link to achieve intelligent prediction and dispatching linkage.

[0042] Step S102: Based on the comprehensive load prediction input dataset, construct a short-term load prediction model to predict the load levels of multiple feeders and key nodes in a preset future period, and obtain the predicted load time series at the node level and feeder level.

[0043] Step S102 involves constructing a short-term load prediction model based on the comprehensive load prediction input dataset to accurately predict the future load levels of multiple feeders and key nodes in the distribution system, and output the load time series prediction results with node-level and feeder-level resolutions.

[0044] Before executing this step, it is first necessary to confirm that the comprehensive load prediction input dataset has been constructed. This dataset contains multi-dimensional feature information from multiple sources, including historical active and reactive load data, real-time meteorological factors such as temperature, humidity, wind speed, sunlight, etc., and the time series characteristics of user electricity consumption behavior, such as load change patterns, response capabilities, electricity consumption categories, etc. All data are processed through time alignment, spatial mapping, feature normalization, etc. to form a unified time series sample, which constitutes the input of the prediction model.

[0045] In the model construction phase, the system needs to select a prediction algorithm suitable for multi-variable time series modeling according to the time limit requirement and resolution requirement of load forecasting. Preferably, the long short-term memory network (LSTM) or gated recurrent unit (GRU) in the deep learning model can be used. Such models can effectively capture the non-linear dynamic change trend and mutation behavior of historical load and are suitable for processing load data with seasonality, suddenness or periodicity. If the model calculation complexity and interpretability are considered, traditional statistical regression methods such as ARIMA or support vector regression (SVR) can also be used as the baseline comparison scheme.

[0046] In the specific implementation, the input of each model includes a data window of multiple time steps, and each time step contains information of different feature dimensions. The model predicts the load levels of each feeder and key node within a certain number of future time steps according to the continuous feature sequence in the historical window. To improve the prediction accuracy, the system can introduce an attention mechanism to dynamically adjust the attention weights of the model to different features, or adopt an integrated prediction strategy that fuses the results of multiple models to generate the final prediction output through weighted average or voting mechanism.

[0047] The model training phase is usually executed based on historical data. The sliding window strategy is adopted to construct the training sample set, and the actual load data is used as the supervision label to minimize the error between the prediction result and the true value, such as the mean square error (MSE) or mean absolute percentage error (MAPE). After the model is deployed, the system will continuously input the latest data in a rolling prediction manner, update the model state in real time and output new prediction results to ensure the timeliness and adaptability of the prediction results.

[0048] The prediction output needs to cover different spatial levels, specifically including the prediction of the total load curve at the feeder level and the local load prediction at key nodes. The prediction duration can be set to 15 minutes, 30 minutes, 1 hour or longer in the future according to the actual application. The prediction result format is time series data with timestamps, and the output structure includes load values, confidence intervals, trend directions and mutation warning information.

[0049] It should be emphasized that the prediction model in this step is not statically constructed and used for a long time, but a model that is dynamically updated in coordination with the changes in the operation environment of the distribution network. The system supports periodic fine-tuning or incremental learning of the model based on real-time feedback data, so as to ensure the accuracy and robustness of the long-term operation of the model.

[0050] To sum up, step S102 realizes the active perception of the future load state of the distribution system by constructing a high-precision and updatable short-term load prediction model, providing high-resolution and high-confidence data support for subsequent risk identification and dispatching strategy formulation, which is the key basis for the intelligent and forward-looking regulation of the present invention.

[0051] Furthermore, the short-term load forecasting model includes a topology encoding sub-module, a feature fusion sub-module, a time series forecasting sub-module, and an output reconstruction sub-module;

[0052] Among them, the topology encoding sub-module takes the power grid topology structure of the target distribution area as input, constructs node vector embeddings with electrical attributes using a graph neural network, and encodes the connection relationships, impedance parameters, equipment constraints, and physical location relationships of each feeder and node in the topology into low-dimensional continuous vectors, serving as the explicit expression of the impact of the topology structure on load forecasting;

[0053] The feature fusion sub-module takes the topology encoding result, historical load time series, spatially interpolated meteorological features, and behavior label matrix as joint input, adopts a multi-layer perception fusion mechanism, extracts key features based on the attention mechanism and feature gating mechanism, eliminates the scale differences and semantic conflicts between data, and outputs a multi-dimensional dynamic feature vector sequence of unified length as the comprehensive driving factor for load changes;

[0054] The time series forecasting sub-module takes the fused multi-dimensional dynamic feature vector sequence as input, uses a bidirectional gated recurrent unit network enhanced by time series attention to predict node-level and feeder-level loads step by step in time. By introducing an adaptive time window mechanism and an abnormal state dynamic correction structure, the model is capable of identifying load mutation events in the short term and actively suppressing prediction errors;

[0055] The output reconstruction sub-module maps the predicted values back to the physical numbers of each node or feeder according to the time series forecasting results and the actual configuration relationships of the power grid nodes, and attaches confidence indicators, upper and lower bounds of prediction errors, and typical electricity consumption scenario labels to provide multi-level prediction information support for subsequent scheduling strategy optimization and execution control.

[0056] The overall model includes a topology encoding sub-module, a feature fusion sub-module, a time series forecasting sub-module, and an output reconstruction sub-module. The modules are connected in sequence and cooperate to form a prediction system with a clear structure and logical closed-loop.

[0057] The topological encoding sub-module takes the power grid topological structure of the target distribution area as the basic input. This topological structure consists of transformers, feeders, tie switches, nodes, and connection relationships, and is accompanied by the electrical parameters of each line segment, such as impedance, conductance, capacitance, voltage level, and the boundary parameters of adjustable devices. To effectively extract the constraints and coupling effects of the topology on load evolution, this module is modeled based on the graph neural network (GNN). When implementing, first, the power grid structure is represented as a graph, where the nodes represent physical entities (such as distribution transformers, load nodes, switch positions), the edges represent electrical connection relationships, and each edge is accompanied by electrical attributes. Subsequently, the graph convolutional neural network (GCN) or graph attention network (GAT) is used to learn vector embedding for each node, so that the output low-dimensional node representation vector can capture both its local electrical characteristics and global topological context at the same time. These node vectors will serve as the topological-aware representation form and participate in the subsequent feature fusion process.

[0058] The feature fusion sub-module starts with the node vectors output by the topological encoding sub-module and combines historical load time series data, meteorological feature data processed by spatial interpolation, and user behavior label matrices. The historical load data provides the power curve information of each node or feeder over a past period of time. The meteorological feature data includes temperature, humidity, wind speed, sunshine intensity, etc., and has been mapped to the specific feeder or node service area through geographical interpolation methods (such as Kriging interpolation or inverse distance weighting method). The behavior label matrix is used to describe the typical electricity consumption characteristics of each user or load node, such as periodic load, response ability, price sensitivity, etc.

[0059] In terms of the fusion method, the feature fusion module uses a multi-layer perception mechanism to construct a multi-source joint representation. The specific implementation is to perform embedding encoding on each type of feature input, unify their dimensions and then splice them, and automatically learn the weight distribution of each type of feature for the prediction target through the attention mechanism. At the same time, to avoid the scale inconsistency, statistical distribution differences, and information redundancy existing between different data sources, this module also introduces a feature gating mechanism, such as a gating unit or a channel-wise attention mechanism, to achieve the functions of feature selection and suppression. Finally, the fusion module outputs a multi-dimensional dynamic feature vector sequence with a unified length and unified dimension as the direct input for time series prediction.

[0060] The goal of the time series prediction sub-module is to perform step-by-step prediction of the load levels of each node or feeder within a future time period using the fused dynamic feature sequence. This module adopts a bidirectional gated recurrent unit network (Bi-GRU) structure and combines a time series attention mechanism to enhance the model's ability to focus on key time segments. The model not only fuses information before and after in time but also dynamically determines the required historical depth for prediction by introducing an adaptive time window mechanism. For example, when the load volatility is large or meteorological mutations are frequent, the system automatically increases the input window length to improve the model's ability to grasp the context; while in the stable operation stage, the window length is shortened to improve the calculation efficiency. In addition, this module also integrates an abnormal state dynamic correction structure, which can perform feature perturbation adjustment or confidence weight adjustment on the detected sudden anomalies (such as outliers caused by equipment jumps or extreme weather) in the input sequence, so as to avoid the model overfitting abnormal data and improve the overall prediction robustness. The final output is the future load time series prediction sequences at the node level and feeder level, covering multiple future time steps.

[0061] The output reconstruction sub-module restores the above prediction results to a structured output recognizable by the dispatching. This module decodes the vectors or sequences output by the model according to the physical numbers of each prediction unit in the original power grid topology and their mapping relationships, and restores them to the feeder IDs, node numbers at the physical level, and their load prediction values at each prediction moment. To enhance the interpretability and practicality of the prediction results, this module also calculates the confidence of each prediction value (such as based on Bayesian estimation or Monte Carlo Dropout), the error upper and lower bound intervals (such as 95% confidence interval), and automatically attaches typical electricity consumption scenario labels, such as "high load during high temperature", "holiday cycle load", "meteorological mutation impact load", etc. labels, to indicate which typical mode the current prediction belongs to. These additional information helps the dispatching control module to make early warning judgments, priority settings, and policy redundancy designs when generating control strategies in the follow-up.

[0062] Through the collaborative work of the above sub-modules, this short-term load prediction model realizes the whole process modeling link from structural modeling, multi-source fusion, dynamic prediction to executable output, fully combines the physical structure characteristics, load behavior laws, and operating environment changes of the distribution network, and provides an accurate, interpretable, and controllable prediction basis for subsequent adaptive dispatching. This structure has good scalability and model migration ability, and can adapt to the intelligent distribution dispatching requirements under different regions and different network structures.

[0063] The abnormal state dynamic correction structure in the present invention belongs to the compensation mechanism in the time series prediction sub-module, aiming to identify and correct the prediction deviations caused by sudden meteorological changes, drastic fluctuations in user behavior, or other external disturbances. The core of this structure lies in making an adjustable and interpretable secondary correction based on the initial prediction results according to the current environmental characteristics, behavior change characteristics, and node historical response characteristics, so as to enhance the robustness and accuracy of the model under complex operating conditions.

[0064] Furthermore, the abnormal state dynamic correction structure in the time series prediction sub-module dynamically adjusts the initial prediction value at the node level through the following formula 1:

[0065] ;

[0066] where represents the predicted load value of the th node at time , which is the prediction result obtained by weighted compensation of the initial prediction value combined with multiple correction factors. This predicted value will be used for subsequent power grid dispatching judgment.

[0067] is the initial predicted value of the th node at time , which is output by the main model (such as a bidirectional GRU structure) without considering abnormal disturbance factors. This value provides the main trend of the prediction, but there may be deviations when the load fluctuates violently.

[0068] is the individual correction coefficient of node , which is jointly trained according to the load fluctuation degree of this node in historical prediction tasks and the prediction confidence of the model, and is defined as:

[0069] ;

[0070] where is the variance of the historical load of node ; is the mean value of the historical load ;

[0071] is the average prediction confidence of node , which is used to quantify the credibility of the prediction results given by the main model (i.e., the short-term load prediction model) for this node during historical prediction. The purpose is to introduce uncertainty factors as part of the personalized correction coefficient. In other words, the smaller it is, the greater the uncertainty of the model's prediction for this node, and stronger post-processing correction is required; while when Approaching 1 indicates that the model's prediction for this node is highly stable and reliable, and the correction coefficient should be appropriately reduced.

[0072] In implementation, this confidence can obtain the prediction distribution through a neural network structure based on Bayesian methods or the Monte Carlo sampling type Dropout method (MC Dropout), rather than a single predicted point value. The degree of dispersion of the prediction distribution can be used to represent uncertainty, and the "bandwidth of the prediction interval" is commonly used to measure it.

[0073] Assume that the Bayesian method or MC Dropout mechanism is adopted for the node At time point Perform times of prediction sampling to obtain:

[0074] ;

[0075] Among them, Represents the th prediction result of the th node at time Each

[0076] The confidence interval bandwidth of this group of predictions can be calculated, for example:

[0077] ;

[0078] The meaning of this formula is used to represent the uncertainty range of the prediction result at a certain node and a certain time point. It is obtained by statistically analyzing multiple predicted values, so as to obtain the width of a confidence interval covering most possible situations.

[0079] Specifically, for the prediction of the node at time Instead of only taking a single value output by the model, a set of predicted values is obtained through multiple samplings or model uncertainty modeling (such as MC Dropout, Bayesian neural network, etc.). This set of predicted values reflects the output distribution of the model under the current input. After sorting this set of predicted values, the difference between the 97.5th percentile and the 2.5th percentile is taken, and this difference represents the fluctuation range of the predicted value at the 95% confidence level, that is, the width of the prediction interval.

[0080] Therefore, this width reflects the model's prediction for the node at time The stability of the prediction results. The narrower the interval, the more concentrated and stable the model prediction is, and the more confident it is; the wider the interval, the more divergent the prediction distribution is, the higher the uncertainty is, and the lower the model's confidence in the prediction point is. This width will eventually be used to construct the prediction confidence index of the node, further affecting the response strength of the anomaly correction mechanism.

[0081] In order to unify the dimensions and range of each node, it is also necessary to normalize them. For example, the normalized width is defined as:

[0082] ;

[0083] In this way, the nodes with more unstable predictions and wider intervals are The smaller it is; while the prediction is concentrated, the nodes with small fluctuations, Approaching 1.

[0084] Final That is, the average confidence of the node in several historical periods (such as the past week or an operating cycle):

[0085] ;

[0086] in Indicates the number of sampling time points. It is recommended to take at least (i.e. 15 minutes of data per day) to ensure statistical stability.

[0087] It represents the overall load variability index at the current moment, which is dynamically calculated based on the standard deviation of the network-wide forecast load sequence; it is defined as:

[0088] ;

[0089] in is the total number of nodes participating in the prediction in the system; It is Nodes at time The initial predicted load value is the original prediction result output by the short-term load forecasting model (such as the bidirectional gated recurrent neural network Bi-GRU) before the abnormal correction mechanism is applied.

[0090] It's time The average initial prediction value of all nodes. This parameter is used to amplify the effect of the abnormal correction term when the overall system fluctuates violently.

[0091] Is a node The abnormal response sensitivity factor is used to characterize the load response intensity of the node under external disturbance conditions such as meteorology. It is obtained by the following regression statistics:

[0092] ;

[0093] where is the average load change of the node during the meteorological anomaly period ; is the change range of the meteorological index (such as temperature, humidity, etc.) within its corresponding time period.

[0094] is the behavior perturbation index of the node at the moment , which comes from the "volatility" type feature dimension in the behavior label matrix. For example, whether the user enters the demand response state at this moment, enables high-load devices, participates in energy-saving strategies, etc. It can be extracted through the following formula:

[0095] ;

[0096] where represents the th perturbation dimension of the node in the behavior feature matrix; is the perturbation importance coefficient of the corresponding dimension. represents the total number of perturbation dimensions, usually taking ; can be automatically learned through training or assigned by experience (such as taking 0.5 for load switch behavior, 0.3 for response state, etc.). This item and jointly control the exponential function part , realizing a non-linear response function. When the perturbation intensity is small, the exponential term approaches zero and no correction is made; when the perturbation is strong and the sensitivity is high, this term approaches 1 and obvious correction is generated.

[0097] is the topological embedding vector of the node , which is generated by the topological coding sub-module (such as a graph neural network) according to the electrical connection structure, physical attributes and location relationship of the node, and is usually a dense vector with a dimension between 16 and 64, reflecting the structural role of this node in the distribution network.

[0098] is the dynamic weight vector of the behavior feature at the moment , which is calculated by the feature fusion sub-module. This vector represents the influence of the current behavior feature in the global prediction through the attention mechanism, and the dimension matches , and is used to perform an inner product with the topological vector to form the final perturbation response weight term.

[0099] Inner product term Structural regulation is achieved. That is, only when the topological structure of the node is significantly correlated with the current behavioral characteristics, the correction term will be amplified, thus realizing the dynamic regulation that combines locality and context relevance.

[0100] Overall, this formula forms an interpretable, learnable, and personalized load prediction correction mechanism, which integrates system-level fluctuations, node-level historical behaviors, topological positions, and disturbance events, thereby providing highly credible prediction support for the intelligent distribution dispatch in the present invention, and is particularly applicable to distribution scenarios with unstable loads, significant behavior driving, and tight topological coupling.

[0101] In the short-term load prediction model of the present invention, the feature fusion sub-module is used to uniformly integrate information from different sources into a set of structured dynamic input features to drive the subsequent time series prediction process. Considering the significant differences in behavior driving and meteorological sensitivity among different nodes, the system does not adopt a unified linear splicing method, but designs a feature fusion method based on a behavior and meteorology joint decoupling attention mechanism. Its core is to proportionally weight and combine the mapping results of behavior features and meteorological features in different semantic spaces to achieve personalized fusion expression for each node.

[0102] Furthermore, the feature fusion sub-module includes a feature construction structure based on a behavior and meteorology joint decoupling attention mechanism, and generates the fusion input representation for each node through the following formula 2:

[0103] ;

[0104] where represents the fusion input vector of node at time , which is used as the input of the time series prediction sub-module (such as a Bi-GRU network) to predict the load evolution trend in the next several moments. Its essence is a structured expression of multi-source heterogeneous features, taking into account the different influence mechanisms of behavior features and meteorological features on the load.

[0105] represents the behavior driving vector of node extracted from the behavior label feature matrix at time , reflecting the characteristic state of this node based on the user's electricity consumption behavior at the current moment. It can include information such as the user's periodic load type (such as night-dominated), whether it is in the demand response cycle, the electricity price sensitivity index, the proportion of controllable load, the user fluctuation frequency, etc. Usually, it is represented in the form of a numerical vector, and the dimension range is between 5 and 20, depending on the feature selection strategy.

[0106] represents the meteorological features after spatial interpolation at node The local environmental factor vector at the node is the meteorological observation data (such as temperature, humidity, wind speed, sunshine intensity, weather category, etc.) mapped to the node through spatial interpolation method The local meteorological vector is obtained after the geographical location of the node. Meteorological interpolation can use Kriging, inverse distance weighted (IDW) or geographic mapping algorithm based on Voronoi polygons. This vector expresses the physical impact of the current environment of the node on the load change, and the dimension is generally between 5 and 10.

[0107] are trainable weight matrices used for decoupling transformation of behavior and meteorological features, with the goal of mapping the original behavior or meteorological vector into a unified target feature space. The matrix sizes are and ,in and are the dimensions of original behavioral characteristics and meteorological characteristics, is the dimension of the target fusion vector, the recommended value is or 64. The behavior vector is obtained by the hyperbolic tangent function Nonlinear compression is performed to retain detail changes and control their numerical range; the meteorological vector maintains positive sensitivity through the modified linear rectification function ReLU to reflect the incentive effect of weather changes on load.

[0108] For Node The individual behavior weight coefficient is calculated using the following formula 3:

[0109] ;

[0110] in, Representation Node The historical behavior dominance index is obtained by statistical learning based on the explanatory strength of user behavior on load changes in the past scheduling cycle. The specific calculation method is to use linear regression or feature importance scoring (such as SHAP value) to evaluate the contribution of behavioral input features to prediction accuracy in the historical prediction process.

[0111] Representation Node The meteorological sensitivity parameter is estimated based on the partial derivative of the node load change during historical meteorological drastic events. This value can be estimated by analyzing the load change curve during historical meteorological drastic events (such as high temperature, typhoon, rainstorm), and taking the derivative of the load change and the meteorological variable change or fitting the linear slope, so as to estimate the response sensitivity of the node.

[0112] In the formula It realizes personalized and continuous control of behavior and meteorological weights, rather than through static weighting or fixed policy fusion. This enables the system to dynamically determine the dominant weights of behavior and environmental factors in feature fusion according to the physical attributes and user types of different nodes, thereby improving the accuracy and generalization ability of prediction.

[0113] Finally, the fused input vector constructed through the above mechanism has the advantages of strong interpretability and flexible fusion strategy, provides a high-quality input basis for the time series prediction model, and has wide adaptability in multi-source heterogeneous input scenarios.

[0114] Step S103: According to the predicted load time series, judge the possible voltage deviation, load imbalance or equipment overload risks of each node and feeder during the prediction period.

[0115] The core of step S103 is: after obtaining the load prediction results of each feeder and key node in the future period of time, further evaluate the operation risks of the distribution system during this prediction period, including various potential abnormal states such as voltage deviation, load imbalance and equipment overload, so as to provide a quantitative basis and target constraint for subsequent scheduling optimization.

[0116] This step first takes the node-level and feeder-level predicted load time series data output by step S102 as the basis, combines the static structure parameters and dynamic state variables of the power grid operation, and simulates and deduces the voltage and current change trends in the future time period. The judgment of voltage deviation is based on the deviation value between the calculated node voltage value and the rated voltage after the node predicted load is injected. At the implementation level, through power flow calculation algorithms such as the Newton-Raphson method or the fast decoupling method, combined with factors such as the current system topology, power supply side output, and voltage regulation equipment settings, simulate the electrical distribution situation in the future period, and compare the normalized node voltages with the upper and lower allowable values. When the predicted voltage value of a certain node continuously deviates from the normal range and the deviation exceeds the set threshold (such as ±5% or ±7%), it can be determined that there is a potential voltage deviation risk at this node.

[0117] In the process of judging the load imbalance of the feeder or node, the system needs to compare the phases according to the predicted injection amounts of the three-phase loads. If the system adopts a three-phase asymmetric modeling method, each feeder segment or node will have three-phase power prediction values. The system can calculate the difference or imbalance rate between the powers of each phase (such as the load deviation rate or the zero-sequence current estimation value), and compare it with the safe operation threshold, so as to identify the areas where serious inter-phase imbalance may occur during the prediction period, and prompt the subsequent scheduling system to perform phase reconstruction or phase shift adjustment on the load.

[0118] The judgment of equipment overload risk is usually carried out based on the relationship between the predicted load and the rated capacity of the equipment. For key equipment such as transformers, feeders, switches, capacitor banks, etc., the system records their rated capacity and overload tolerance, and summarizes them in combination with the predicted values of the loads connected thereto. When a certain equipment continuously bears a power exceeding its rated capacity or exceeding the thermal stability limit value (which can be dynamically set according to the equipment thermal characteristic curve) during the prediction period, it can be regarded that there is an overload potential hazard for this equipment. Such judgment can be assisted by tools such as equipment thermal models and load-temperature rise relationships, so as to improve the recognition accuracy.

[0119] In practical applications, this judgment process is usually executed in a rolling manner, that is, after the prediction results are generated, risk scans are performed on future time periods at a minute-level or hour-level update frequency to ensure that any potential anomalies can be identified in advance. In addition, the system can also introduce a risk level assessment mechanism to classify the risks according to factors such as the deviation amplitude, duration, equipment level, etc., so as to provide a basis for priority ranking for subsequent dispatching optimization strategies.

[0120] This step does not stop at detecting the risk itself. More importantly, it outputs the risk identification results in a structured manner, including potential risk types, risk locations, occurrence time periods, severity levels, and influence ranges, etc., so that the dispatching module can clarify the intervention objects and target time periods and achieve targeted optimization.

[0121] To sum up, step S103 realizes the closed-loop connection from data-driven prediction to physical constraint identification through the quantitative analysis and risk identification of the predicted load results, significantly enhancing the forward-looking, accuracy, and reliability of the dispatching control strategy in the present invention.

[0122] Furthermore, the judging of the possible voltage deviation, load imbalance, or equipment overload risk of each node and feeder during the prediction period according to the predicted load time series includes:

[0123] Mapping the predicted load time series to the distribution network topology structure, and combining the node voltage regulation ability, electrical connection relationship, and load conduction path to deduce the predicted voltage response trend; based on the prediction results of the three-phase load distribution, calculating the inter-phase imbalance rate of the feeder, and combining the historical asymmetric operation records to identify the continuous imbalance evolution trend;

[0124] According to the rated capacity of the equipment and the load evolution rate, perform time series analysis on the load rates of each node and feeder within the prediction interval to judge whether they cross the equipment operation safety threshold;

[0125] Cross-compare the judgment results with the load surge characteristics in the behavior tags, screen out the highly sensitive nodes that are significantly affected by the behavior drive and have overload risks, and construct a node-level risk list.

[0126] Based on the prediction results of node - level and feeder - level loads in future time periods, this embodiment constructs a dynamic, data - driven power grid operation risk identification mechanism. First, the system precisely maps the predicted load time series to the topological structure of the power grid. The load change trend of each node is embedded in its specific position in the topological network, and a sensitivity model of node voltage response to load is established by combining key factors such as the voltage regulation ability of the branch where the node is located, the electrical connection mode with the upper - level node, and the active and reactive power conduction paths from the main substation to the node. By analyzing the change amplitude and rate in the load time series and combining with the power flow distribution rules of the power grid, the system deduces the voltage response trend of the node during the prediction period, thereby identifying potential voltage deviation risk areas.

[0127] When judging the load imbalance risk, the system deconstructs the predicted load by phase according to the three - phase structure of each feeder, forming time - serialized A, B, and C - phase load distribution data. The system calculates the imbalance degree of the current or power between phases and establishes a trend - tracking model within a continuous time window based on the inter - phase deviation rate and imbalance rate indicators. If it is identified that a certain feeder has a systematic imbalance growth trend in multiple consecutive prediction time steps and cannot be corrected by normal load transfer methods, it is determined that the feeder has a potential load imbalance risk.

[0128] For the analysis of equipment overload risk, the system introduces the operating characteristics of the equipment itself, especially parameters such as rated capacity, temperature rise limit, and short - term overload tolerance, on the basis of load prediction. Combining with the time - series evolution trend of the predicted load, the system calculates the load rate curves of each equipment (including distribution transformers, feeders, tie switches, etc.) in real - time and determines whether they will exceed the safe operating range in the future time period. Especially for equipment with a relatively fast load growth rate or operating close to the capacity boundary, the system will mark them as key monitoring targets.

[0129] To further enhance the interpretability and accuracy of risk identification, the system also conducts correlation analysis between the above - mentioned various physical indicators and user - side behavior tags. By using the behavior tag matrix, it identifies user types with a high probability of high - load surges in the current prediction cycle (such as centralized charging at night, centralized startup of high - power equipment, participation in demand response to release load, etc.), and analyzes whether these behaviors are concentrated on certain nodes or branches, thus forming behavior - driven risk indicators. If a node has both electrical operation risks (such as voltage over - limit or overload tendency) and signs of behavior - driven sudden load increase, it is identified as a high - sensitivity node and included in the node - level risk list for priority intervention in subsequent dispatching strategies.

[0130] In summary, in this step, a multi-class risk judgment mechanism that can be updated in real time is constructed by fusing prediction data with electrical structure, equipment capabilities, and user behavior in multiple dimensions. It does not rely on cumbersome static simulation processes and has high adaptability and prediction accuracy.

[0131] Step S104: Based on the judgment results of the power grid topological structure, electrical parameter constraints, and load prediction time series, a rolling horizon optimization method is used to generate a dispatching control strategy. The dispatching control strategy includes the advance adjustment of transformer tap positions, the pre-switching plan of capacitor banks, the reclosing sequence and load distribution logic of feeder tie switches, the charging and discharging time and power setting of distributed energy storage devices, or the demand response execution instructions for the user side.

[0132] This step is one of the cores of the present invention. Its function is to generate a dynamically executable dispatching control strategy based on the physical structure and operation constraints of the current distribution network after obtaining the load prediction results and potential risk assessment results, so that the system can actively adjust its operating state before load mutations or operating pressures occur and achieve adaptive coordination of the entire network load.

[0133] In this step, three key information sources need to be input first, namely the topological structure of the distribution network, the electrical parameter constraint conditions, and the load prediction results and risk determination information output by step S103. The distribution network topological structure includes the positions, connection relationships, and physical parameters (such as impedance, conductivity, etc.) of substations, transformers, feeders, nodes, tie switches, and the terminal loads or distributed power sources connected thereto. This information is usually stored in the form of a topological diagram or matrix and loaded into the optimization engine before dispatching calculations. The electrical parameter constraints include the actual operation limitations such as the rated capacity, voltage level, adjustment range, switching time delay, energy storage power boundary, charge and discharge rate, and operation cost of each device. These constraints are crucial for ensuring the physical feasibility of the dispatching strategy.

[0134] The rolling horizon optimization method is the main technical means used in the present invention for generating the dispatching strategy. The core idea of this method is to divide the load prediction period into multiple consecutive prediction time windows, perform an optimization calculation within each time window, and recalculate the dispatching strategy according to the latest load prediction and actual feedback in the next period to achieve a closed-loop control of time series advancement and real-time correction. The optimization model usually adopts an algorithm framework based on mixed integer linear programming (MILP), dynamic programming (DP), or model predictive control (MPC). The objective function can be set according to actual needs to minimize system energy consumption, reduce adjustment costs, suppress load fluctuations, or improve voltage qualification rates. The constraint conditions include power flow balance equations, equipment state boundaries, action interval times, and user response capabilities.

[0135] The generated scheduling control strategy should be refined into executable specific control instructions and be able to achieve multi-level coordination in the time and space dimensions. The strategy may include the following types of operations:

[0136] First, it is the advance adjustment of the transformer tap position. The dispatching system calculates and sets the tap changing plan of the transformer voltage regulator in advance according to the load forecast and voltage deviation trend, so that the tap conversion is completed before the load rises or falls, thereby stabilizing the downstream voltage level. The tap adjustment needs to consider the minimum time interval and action cost of tap changes to avoid frequent switching from impacting the equipment.

[0137] Second, it is the pre-switching of capacitor banks. Capacitor banks are used to regulate reactive power and improve voltage quality. The system determines the combination of capacitor banks to be switched in or out in advance according to the reactive load forecast results before the low-voltage risk appears. The action instructions include the specific group number, action time, and control node information. To achieve regional coordination, a grouped control and bus voltage closed-loop correction mechanism can also be adopted.

[0138] Third, it is the reconfiguration strategy of feeder tie switches. The system dynamically generates the closing and opening sequences of tie switches by optimizing the load distribution relationship between feeders, realizing the active transfer of load from overloaded feeders to feeders with spare capacity, and alleviating the phenomenon of concentrated load. The operation of tie switches needs to consider the power flow direction, line protection coordination, and fault isolation strategy at the same time to ensure that the stability of the system after reconfiguration is not damaged.

[0139] In addition, it also includes the charge and discharge scheduling of distributed energy storage devices. The system arranges to release energy before the load peak and absorb excess electric energy during the load valley according to the load forecast curve of the whole network, the remaining capacity of the energy storage, and its health status, so as to achieve peak shaving and valley filling. The scheduling strategy needs to specify the start and end times of charge and discharge, power magnitude, and the numbers of participating units, and consider technical details such as battery efficiency, maximum cycle times, and converter limitations.

[0140] Finally, it is the issuance of demand response instructions for the user side. The system selects users with response capabilities according to the overloaded or voltage-risk areas that may appear in the forecast, and temporarily adjusts their load levels through price signals, incentive compensation, or direct control methods. For example, some users can be instructed to reduce air-conditioning loads, delay the start of high-energy-consuming equipment, or dispatch industrial loads into standby operation modes, thereby reducing the peak pressure of the system.

[0141] It should be noted that all scheduling instructions need to be compatible with the real-time execution system and can be distributed to field devices through an automated platform. After the strategy is generated, the system retains the execution time points, object identifiers, operation parameters, and expected impacts of each control operation as the basis for subsequent strategy execution confirmation and effect evaluation.

[0142] Through this step, the power distribution system can intervene in potential operation risks in advance, actively optimize various control resources, and thus significantly improve the intelligent level and adaptive ability of the overall operation.

[0143] Furthermore, based on the judgment results of the power grid topology structure, electrical parameter constraints, and load prediction time series, a rolling horizon optimization method is used to generate a scheduling control strategy, including:

[0144] Using the graph structure modeling method to map the power grid topology of the target distribution area into a graph data structure, where the nodes represent feeder connection points, transformer positions, or load access points, the edges represent line connection relationships, and considering electrical attributes such as line impedance, voltage level, and switch status, a graph topology representation integrating topology and parameters is constructed;

[0145] Embed the node-level load prediction time series obtained according to the load prediction model into the above graph structure as the dynamic additional attribute of the time evolution feature, and extract the load response feature distribution under the whole network structure dependence through the graph neural network to capture the coupling influence between system levels;

[0146] Based on the extracted whole network node state features, introduce a rolling horizon optimization framework, and construct a multi-objective nonlinear optimization problem in each rolling period. Minimizing voltage deviation, suppressing feeder imbalance, reducing equipment overload risk, and minimizing load adjustment cost are used as the combined objective function, and integrating the operation boundaries, action frequency constraints, and interlocking action logics of various electrical equipment as the constraint conditions;

[0147] Through evolutionary strategy search, dynamically generate a joint operation sequence covering different regulation resources in each rolling period, including transformer tap adjustment, capacitor bank switching, tie switch opening and closing, energy storage unit power distribution, and user load priority ranking, and attach its execution window, prediction risk area, and priority level information to each strategy to ensure that the scheduling strategy has regional awareness, self-adaptability, and high timeliness.

[0148] In this embodiment, the core of the rolling horizon optimization method lies in establishing a set of dynamic scheduling strategy formulation mechanisms that can be adjusted in real time with the prediction update. First, in order to make the power grid structure computable and learnable in the data space, the system maps the topology structure of the target distribution area into a graph data structure. In this graph, all nodes represent feeder connection points, distribution transformer positions, or user load access points in the actual power grid, and the edges in the graph represent physical line connections, and corresponding electrical attributes are attached, such as line impedance, voltage level, maximum transmission capacity, and switch on-off status. This graph structure not only retains the spatial relationship of electrical connections but also incorporates the operation boundaries and control attributes of equipment, thus providing a highly structured power grid foundation for subsequent scheduling optimization.

[0149] After constructing the topology graph, the system embeds the node-level load prediction time series output by the load prediction model into the graph structure. The predicted load curve of each node over time will be used as a time-evolving feature, which, together with the static structural attributes of the nodes in the graph, constitutes the dynamic graph node features. The system uses a Graph Neural Network (GNN) to extract the state representation of each node in the power grid under topological constraints. This representation not only reflects the load changes of the node itself but also captures the mutual influence caused by topological coupling between it and neighboring nodes. For example, the load change of a certain line may exert pressure on the upstream transformer or adjacent branches.

[0150] Based on the graph structure, the system introduces a rolling horizon optimization framework. This framework takes each prediction period as a rolling window and constructs an optimization model for the combined regulation problem within the current window. The optimization model adopts a non-linear multi-objective function form, comprehensively considering multiple objectives such as minimizing voltage deviation, suppressing three-phase load imbalance, reducing equipment overload risk, and minimizing the cost of regulation operations. Among them, minimizing voltage deviation means ensuring that the voltages of all nodes are kept within the allowable fluctuation range. The three-phase imbalance suppression objective aims to reduce the deviation of the load distribution between phases. Minimizing the overload risk is used to prevent key equipment from being overloaded for a long time, while minimizing the operation cost ensures that the regulation measures will not act frequently, thus reducing the equipment life or causing secondary disturbances.

[0151] This optimization problem also incorporates a series of constraints related to the operation of distribution equipment. For example, for transformer tap changers, the allowable switching frequency and the upper and lower tap ranges need to be considered; the switching of capacitor banks is restricted by the number of switchings and the voltage response rate; the operation of feeder tie switches must comply with the network connectivity constraints and cannot cause islanding effects or form loops; the regulation of energy storage devices is physically limited by the amount of electricity, current rate, and charge-discharge cycle; and the response on the user load side needs to consider its adjustable label, response delay, and user priority.

[0152] In the process of generating the scheduling control strategy of the present invention, to achieve high-precision and high-reliability optimization decisions, the system designs and solves a non-linear multi-objective optimization problem. This optimization model not only focuses on the safety of the operating state and power quality but also particularly emphasizes the economy and executability of the regulation behavior. Its objective function covers multiple performance indicators with practical significance and is solved coordinately in the dynamic environment of the distribution network.

[0153] First, the voltage deviation minimization objective is used to ensure that the voltage levels at all nodes in the distribution system are maintained within the allowable fluctuation range of the rated voltage as much as possible, usually ±5% or ±10%, depending on the national electricity standard. In the model, by introducing the deviation term between the predicted value and the target value of the node voltage, the voltage deviations of all key nodes are accumulated to form a penalty amount. This index has a direct impact on the power quality of the user side and the stability of equipment, so a relatively high weight is assigned in the optimization.

[0154] Secondly, the three-phase load imbalance suppression objective is used to reduce the problem of uneven phase power or current at the feeder level. The system calculates the inter-phase difference of the predicted values of the three-phase loads in each feeder and introduces a standardized unbalance index to reflect the skewness of the load between phase A, phase B, and phase C. The introduction of this objective helps to improve the operation efficiency of the system and avoid problems such as increased zero-sequence current, transformer overload, and increased energy loss caused by inter-phase imbalance.

[0155] The third optimization objective is to reduce the risk of equipment overload, which aims to avoid the continuous overload state that may occur in important equipment in the distribution system in future periods in advance. By comparing the predicted load curve with the equipment rated capacity or allowable load curve (such as the transformer overload curve, the line temperature rise limit, etc.), the model evaluates the overload risk level of each key node or line, and introduces the degree of load exceeding of high-risk nodes into the objective function in a weighted form. This design can effectively prevent the equipment from gradually entering the dangerous working area without control intervention and improve the safety margin of the overall power supply system.

[0156] In addition, in order to avoid the scheduling strategy being executed too frequently in theory but causing mechanical fatigue of equipment or cumulative system disturbances in practice, the model introduces the minimization of the regulation operation cost as the fourth objective. In this part, each regulation action, whether it is transformer tap adjustment, capacitor bank switching, tie switch opening and closing, energy storage unit scheduling, or user response invocation, is assigned an operation cost weight constructed based on parameters such as action frequency, response delay, and recovery cost. The optimization process tends to select a strategy with a lower regulation cost and a better action path on the premise of meeting the power quality and safety requirements, so as to ensure the economy and stability of the regulation behavior.

[0157] To ensure the feasibility of the above multi-objective optimization under actual power grid conditions, multiple types of constraint conditions related to the operating boundaries of electrical equipment are also introduced into the model. For example, when adjusting the tap of a transformer, its mechanical structure limitations need to be considered, frequent switching is not allowed, and each adjustment must be carried out within the specified tap range; the capacitor bank is affected by the action response time and the number of switching operations, and its minimum action interval and maximum continuous switching times need to be controlled; the operation of the tie switch must ensure that the system topology remains connected after the operation to prevent the occurrence of power supply islands or the formation of unacceptable closed-loop structures; when scheduling energy storage devices, the physical boundary conditions of their state of charge, current change rate, and periodic charge and discharge times need to be satisfied simultaneously; the participation ability on the user load side is also restricted by behavior tags. For example, certain types of users only support low-level demand response, have obvious start-up delays, or have constraints on the non-adjustability of certain loads, and all these information needs to be dynamically incorporated into the optimization model.

[0158] Finally, in each rolling time domain, the optimization model constructs a non-linear joint solution problem based on the above objectives and constraints, and uses multi-objective optimization algorithms or evolutionary search mechanisms for iterative solution, outputting a set of joint scheduling strategies that meet the system operation objectives and physical boundary conditions. This strategy not only covers various control resources but also has attributes of execution priority and effective time period, facilitating the downstream control system to execute efficiently by region and by level, thus realizing true intelligent and adaptive distribution network scheduling control.

[0159] During the optimization solution process, the system adopts evolutionary strategy search algorithms, such as reinforcement learning, genetic algorithms, or simulated annealing methods, etc., to improve the solution efficiency of non-linear multi-objective optimization and the ability to explore the solution space. The optimization of each rolling period will output a set of joint operation sequences, and these operation sequences will cover multiple types of control objects. For example, in a certain time period, to which tap position the transformer should be adjusted, whether a certain capacitor bank should be switched in advance, which tie switches should be closed to reconstruct the load path, when and with what power an energy storage system should charge or discharge, or what level of demand response instructions should be issued to which users.

[0160] In addition, to ensure the effectiveness and local adaptability of the strategy during implementation, the system attaches supplementary information such as execution window, risk-related area, and control priority to each operation strategy. The execution window defines the start and end time periods when the strategy takes effect, the risk area represents the load risk points or line segments that this operation is mainly used to alleviate, and the control priority is used for sorting decisions in case of resource conflicts or concurrent multi-strategies. Through these auxiliary attributes, the system can quickly respond to sudden changes during actual execution and avoid the spread of uncertainties caused by the rigid execution of the global strategy.

[0161] Furthermore, the graph structure modeling method includes:

[0162] Generate an initial topology map based on the GIS data of the distribution network and the device master data in the SCADA system. The node type is determined by parsing the device attribute fields, and electrical parameters such as line impedance and voltage level are complemented by real-time monitoring data and the standard parameter library.

[0163] Identify breakpoints, island nodes or loop structures in the map through the topology connectivity verification algorithm, and perform map structure closing processing in combination with electrical operation rules to ensure the integrity and decoupling of the topology map.

[0164] Embed structured parameter vectors into the topology map nodes and edges respectively. The node parameters include access capacity, node type and control ability identifier, and the edge parameters include impedance value, maximum current capacity and current switch state to realize the integrated expression of the topology structure and operation characteristics.

[0165] In the process of dispatching optimization, the present invention adopts a graph neural network-based modeling method to represent the power grid structure. Therefore, the accurate construction of the graph structure and parameter characterization are the basis of the entire intelligent distribution load forecasting and dispatching method. To realize graph structure modeling, the system first needs to structurally abstract the information of the real distribution network. This step is completed by obtaining the GIS (Geographic Information System) data of the distribution network and the device master data recorded in the SCADA system. Among them, the GIS data provides the spatial distribution and connection relationship of facilities such as distribution lines, transformers, feeder switches, and user access points, and the SCADA system provides the operation attributes and control fields of each electrical device, such as device type, voltage level, installation location, and operation status. Through parsing the fields of this information, the system can automatically identify the physical nodes in the distribution network and determine the type of each node, such as load nodes, switch nodes, transformer nodes, or feeder connection points, etc., so as to construct an initial network node set.

[0166] After the node construction is completed, the system generates the corresponding edge structure according to the physical connection relationship between the devices. The edge represents the line connection between the nodes. These edges not only represent the existence of the connection, but also carry key electrical parameter information, including the impedance value, voltage level, maximum transmission current, etc. of the line. Some parameters can be obtained from the SCADA real-time monitoring data, such as current, voltage and power factor; other static parameters come from the power design standard or the power grid archive database to complement the missing physical attributes in the graph structure.

[0167] The initially constructed topology graph may have structural errors or be incomplete. Therefore, the system needs to perform a topological connectivity check on the graph structure. Through the connectivity check algorithm, breakpoints, isolated nodes (i.e., isolated devices not connected to the main network), or loop structures in the network are identified, and these problems are closed. The closing process will refer to the grid operation regulations and safety constraints, appropriately add virtual edges, complete the disconnection information, or disconnect illegal loops according to the regulations, so as to ensure that the entire topology graph is closed, connected, and has the structural legality required for network power flow calculation and optimal scheduling.

[0168] After the topological structure is constructed, the operating parameters related to each node and edge also need to be further embedded in the graph to form an attribute representation of the graph. For nodes, the embedded structured parameter vector includes the maximum access capacity of the node (such as the allowable load upper limit or energy storage capacity), the functional type label of the node (such as load node, transformer node), and whether it supports control (such as having tap changers, circuit breakers, adjustable capacitors, etc.). For edges, the impedance value, the maximum allowable current capacity, and the current switch state of the line (closed, open, or tripped) need to be embedded. These parameters are transformed into a standardized vector form for the input of the subsequent graph neural network, enabling it to not only learn the network topology structure but also consider the actual operating physical characteristics and regulation ability boundaries simultaneously.

[0169] Through the above modeling process, the graph structure constructed by the system is no longer a single static connection graph but a comprehensive expression integrating structure, parameters, and control capabilities. The graph structure constructed in this way can not only accurately reflect the physical topology of the power grid but also support deep learning algorithms to extract local or global load response relationships, providing a solid foundation for the generation of rolling optimal scheduling strategies and ensuring the executability of regulation measures both physically and logically. This graph structure can be continuously updated and supplemented within different rolling prediction periods, thus realizing the timeliness, accuracy, and adaptive ability of the distribution network operation model.

[0170] Furthermore, the process of embedding the node-level load prediction time series into the graph structure includes:

[0171] Construct the load values of each node at multiple prediction times into a dynamic feature vector, and append the load confidence, prediction upper and lower bounds, and abnormal fluctuation flags to form a time series feature set with uncertainty;

[0172] Through the node and time double-index mechanism, this time series feature set is mapped one-to-one with the topology graph nodes, and the time window mechanism is used to segment the prediction sequence in a rolling manner to ensure the time consistency of the graph neural network input within each rolling period;

[0173] The multi - layer graph convolutional structure is used to fuse static topological attributes and dynamic load characteristics, enabling the node state representation to not only reflect the local load evolution trend but also embody the coupling effects with neighboring nodes in terms of structure and time sequence, providing context - relevant inputs for rolling optimization.

[0174] In the scheduling optimization scheme of the present invention, the graph structure not only carries the topological information of the distribution network but also needs to dynamically fuse the time - series prediction results from the short - term load prediction model. To achieve this goal, the system first structures the predicted load of each node over a future period. Specifically, the load prediction model outputs the load prediction values of each node at multiple future moments, such as 96 15 - minute values within the next hour or hourly predictions for several hours. These prediction values are organized into a feature vector in the form of a time series to express the evolution trend of the node's future load.

[0175] To enhance the identification and modeling of uncertainty factors, the system also attaches prediction confidence, upper and lower bounds, and anomaly markers to each prediction point. The prediction confidence reflects the credibility of the model's current prediction value for the node, usually calculated from the standard deviation or confidence interval width output by the prediction model; the upper and lower bounds are used to depict the possible maximum fluctuation range, such as the upper and lower limits of a 95% confidence interval; the anomaly fluctuation marker is used to mark nodes that exhibit mutations, sharp jumps, or high fluctuations during the prediction period. These additional information together constitute a "time - series feature set with uncertainty", which has higher expressiveness and stronger model robustness compared to traditional point - prediction sequences.

[0176] To fuse the above - mentioned time - series feature set into the graph structure, the mapping problem between nodes and time must be solved. The present invention adopts a dual - index mechanism of node index and time index to clearly bind the features of each node at each prediction time point to the corresponding node in the graph. This means that in the input of the graph neural network, instead of only providing a static attribute vector for each node, for each rolling period, the prediction sequence and its confidence index within the current time window of the node are dynamically loaded. To ensure the consistency of the time - series structure, the system introduces a fixed - length sliding time - window mechanism to segment the load prediction sequence according to the rolling period. For example, if the rolling period is 15 minutes and the prediction window is 2 hours, only 96 prediction points within the current 2 - hour period are loaded as part of the graph input each time it slides, thus ensuring that the time window used in each round of network training or inference is consistent and improving the stability of the model in the time - series dimension.

[0177] In the part of the graph neural network, the system adopts a multi-layer graph convolution structure to update and fuse the node states in the graph. These graph convolution operations not only act on the static electrical topology, such as the connection relationship between nodes, the impedance of edges, the voltage level, etc., but also process the dynamically loaded time-series load characteristics at the same time. Through convolution operations, the model can aggregate the time-series evolution information from adjacent nodes and learn the coupling effect between nodes in the time dimension while maintaining the structural information. For example, an abnormal increase in the future load of a node may cause a power impact on its adjacent transformers or branches, and this dependency can be automatically modeled by graph convolution and identified and responded to in subsequent optimizations.

[0178] Finally, after multi-layer graph convolution calculations, each node will obtain a state representation that fuses its local load trend, the dynamic influence of adjacent nodes, and the structural context constraints. This state representation not only reflects the current operating environment of the node but also includes the coupling risks that may be caused by future load fluctuations, which is crucial for the rolling optimization module to construct control strategies.

[0179] In summary, this embodiment integrates the dynamic evolution of predicted load, the expression of model uncertainty, and the structural dependence of the power grid topology into a unified graph structure, which not only enhances the generalization ability and prediction interpretability of the model but also provides a high-dimensional and high-quality input basis for subsequent graph-based scheduling optimization.

[0180] Furthermore, the process of generating a scheduling control strategy through evolutionary policy search includes:

[0181] Based on the extracted node state features, construct a scheduling coding vector including the state of the control object, constraint conditions, a set of scheduling actions, and an initial policy score, which is used to initialize the population or the policy library;

[0182] In each rolling period, continuously optimize the scheduling action sequence through a crossover mutation and policy screening mechanism, and introduce a risk buffer distance and equipment protection margin as implicit penalty terms during the process of generating the policy to guide the search towards a robust operation path;

[0183] For each output joint operation sequence, append a scheduling area mask, a time period priority label, and a policy coverage range identifier to support subsequent regional distribution, priority sorting, and policy switching judgment by the scheduling execution system, and achieve distributed and hierarchical execution control.

[0184] In this embodiment, to achieve the joint optimal scheduling of various power distribution equipment and control resources, the system introduces an optimization method based on evolutionary policy search, which is used to dynamically generate a feasible and robust regulation policy sequence within the rolling prediction period. The basic idea is to encode the states, physical constraints, control action sets, and historical execution feedback of controllable objects in the power grid into a unified policy representation, and with the help of an evolutionary search algorithm, perform iterative optimization in a solution space containing multiple candidate policies.

[0185] Specifically, the system first constructs a scheduling encoding vector for each rolling period based on the node state features extracted by the graph neural network. This encoding vector not only includes the load state of the current node or line, the remaining adjustment capacity of the equipment, known physical boundary constraints, but also control constraint information such as equipment action frequency limits, interlock rules, or regulation order. The scheduling action set part covers the adjustment of transformer tap positions, the switching control of capacitor banks, the opening and closing selection of feeder tie switches, the charge and discharge strategies of energy storage units, and the response levels of adjustable loads. For each type of control resource, the system constructs a set of legal actions according to its current state and predicted reachable range. All these elements are unified and packaged into a structured encoding vector, which is used to initialize the population or policy library in the evolutionary algorithm as the starting point of policy search.

[0186] During the evolutionary search process, the system uses genetic algorithms, reinforcement learning, or other heuristic methods to perform crossover mutation and fitness evaluation on the policy set. In the crossover stage, fragment combinations are performed between different policy encodings to form new operation sequences; in the mutation stage, slight perturbations are introduced to discover potential optimal solutions. To make the generated policies have stronger robustness and engineering feasibility, the system introduces an "implicit penalty term" in the fitness evaluation process, which includes two indicators: the risk buffer distance and the equipment protection margin. The risk buffer distance is used to measure whether the system is close to the voltage or load operation limit after the policy is executed; the equipment protection margin reflects whether the policy operates within the acceptable action frequency, overload duration, or temperature rise range of the equipment. If a certain policy satisfies the objective function (such as minimum voltage deviation, balanced load distribution, etc.) and also has strong risk isolation ability and protection tolerance, then its fitness will be given a higher weight, so it will be preferentially retained and used for subsequent iterations.

[0187] Finally, the system will output a set of optimized joint operation sequences before the end of the rolling cycle. Each operation sequence not only specifies the operation type and execution parameters, but also comes with multiple dispatch-related attributes to support actual deployment and instruction distribution. These attributes include dispatch area masks, which indicate the feeders or substations within which the operation should take effect; time period priority labels, which are used to sort execution priorities according to urgency, scope of impact, or intensity of behavior drivers when multiple operations coexist; and policy coverage identifiers, which are used to mark high-risk nodes or potential fault areas targeted by the policy, so that the dispatch center can perform local intervention, distributed execution, or centralized coordination according to the risk level.

[0188] Through the above process, this embodiment not only realizes the multi-objective global optimization of the dispatching strategy, but also introduces the adaptive adjustment capability for the complex dynamic characteristics of the power grid through the evolutionary mechanism, which can effectively handle large-scale, diversified, and continuously changing control scenarios.

[0189] Step S105: Send the dispatch control strategy to the corresponding device control unit, and perform voltage control, load reconstruction, energy regulation or user load response before the prediction period according to the dispatch control strategy.

[0190] The core of step S105 is to convert the dispatching control strategy generated in the previous step into control instructions that can be accurately executed by on-site equipment, and send these instructions to the corresponding equipment control units, so as to complete the pre-regulation of the power grid before the load fluctuation actually occurs, and realize dynamic optimization control of voltage, power distribution and energy utilization.

[0191] In this step, we first need to clarify the target execution object of the dispatching control strategy. Usually, it includes various intelligent devices installed in the distribution network, such as transformer tap adjustment mechanism, capacitor switching controller, remote switching control unit of feeder tie switch, energy storage converter controller, and user-side energy consumption management system or intelligent terminal with two-way communication capability. These devices must have remote controllability and be connected to the main dispatching center or edge control node through a communication link. The communication mode can be wired (such as optical fiber, PLC) or wireless (such as 4G / 5G, LoRa, Wi-SUN, etc.), and are required to have appropriate real-time and anti-interference capabilities.

[0192] The process of issuing scheduling control strategies is uniformly coordinated and completed by the central scheduling system or the main station of distribution automation. This system will call the corresponding control logic module according to the control interface protocol of each type of device, and parse the abstract policy content into standardized control instructions that conform to the field device control protocol. For example, the policy of "transformer T1 adjusts the tap from +2 to +1 at 15:30" is converted into the Modbus TCP message format supported by the voltage regulator controller, with a checksum and timestamp attached to ensure the accurate execution and traceability of the instruction.

[0193] In terms of time control, all control instructions need to clearly define their execution time points and be strictly aligned with the prediction period. Therefore, the system introduces a time synchronization mechanism to ensure that all terminal devices operate based on a unified clock reference. The system can adopt the IEEE 1588 Precision Time Protocol (PTP) or GPS time synchronization method to ensure that scheduling actions are synchronized at the millisecond level. Especially for multi-point coordinated actions such as feeder reconfiguration or demand response, time consistency is particularly important for operation safety.

[0194] Voltage control is one of the most common strategies. The scheduling instruction can make the transformer automatically perform step-up or step-down operations, and through the feedback loop, it can monitor in real time whether the adjusted voltage value meets the expected target. If the adjustment fails, the system has an automatic fallback mechanism to restore to a safe gear position to avoid the expansion of abnormal voltage.

[0195] In terms of load reconfiguration, the issued instruction can control the closing or opening sequence of specific tie switches, thereby changing the distribution relationship of the load among each feeder. For example, when there is a predicted overload risk on a certain feeder, the system can drain some of the load to adjacent nodes through the standby feeder in advance. The instruction content needs to clearly define the switch number, action delay time, and reclosing protection settings.

[0196] For energy regulation operations, such as controlling the charging and discharging behavior of distributed energy storage systems, the scheduling system will issue specific charging and discharging power values, start and stop times, power curve control methods (constant power, tracking type, or prediction type), etc. to the energy storage controller based on the optimization results. The system will also collect and dynamically adjust key state variables such as SOC (state of charge), temperature, and voltage during the charging and discharging process to avoid overcharging or thermal runaway.

[0197] In terms of user load response, the scheduling system can send load adjustment signals to users or aggregators participating in demand response through a two-way communication platform. The methods can be automatic control (such as disconnecting non-critical loads, adjusting the cooling set temperature) or incentive-based guidance (such as sending temporary electricity price adjustment notices, electricity consumption reward prompts, etc.). The user terminal should have the ability to receive scheduling signals and automatically execute the response strategy, and the response results need to be feedback to the main system for evaluating the response effect and scheduling achievement rate.

[0198] The entire process of dispatching policy issuance has a confirmation and receipt mechanism. After each instruction is issued, it is necessary to receive the feedback on the execution status sent back by the device, including results such as success, failure, partial execution, or requiring manual confirmation. The system can adjust the policy or initiate standby instructions based on this. At the same time, the system supports archiving the issuance records, including timestamps, device numbers, execution parameters, and response results, which is convenient for post-event auditing, operation and maintenance analysis, and policy improvement.

[0199] All in all, step S105 is not only a process of instruction transfer, but also a key link that deeply integrates the prediction-driven optimization strategy with the actual operation of the distribution network.

[0200] The second embodiment of the application provides an electronic device, and the electronic device includes:

[0201] a processor;

[0202] a memory for storing a program, and when the program is read and executed by the processor, it executes an intelligent distribution load prediction and adaptive scheduling method provided in the first embodiment of the present application.

[0203] The third embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it executes an intelligent distribution load prediction and adaptive scheduling method provided in the first embodiment of the present application.

[0204] Although the present application is disclosed above with preferred embodiments, it is not used to limit the present application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application should be subject to the scope defined by the claims of the present application.

Claims

1. An intelligent distribution load forecasting and adaptive scheduling method, characterized in that, Including: Obtain historical load data, real-time meteorological information and user electricity consumption behavior data in the target distribution area, and establish a comprehensive load forecasting input data set; Based on the comprehensive load forecasting input data set, construct a short-term load forecasting model to predict the load levels of multiple feeders and key nodes in a preset future time period, and obtain the predicted load time series at the node level and feeder level; According to the predicted load time series, judge the possible voltage deviation, load imbalance or equipment overload risks of each node and feeder during the prediction period; Based on the power grid topology structure, electrical parameter constraints and the judgment results of the load forecasting time series, adopt a rolling horizon optimization method to generate a dispatching control strategy, and the dispatching control strategy includes the advance adjustment of the transformer tap position, the pre-switching plan of the capacitor bank, the reclosing sequence and load distribution logic of the feeder tie switch, the charge and discharge time and power setting of the distributed energy storage device, or the demand response execution instruction for the user side; Send the dispatching control strategy to the corresponding device control unit, and perform voltage control, load reconstruction, energy regulation or user load response before the prediction period according to the dispatching control strategy; Among them, the judging the possible voltage deviation, load imbalance or equipment overload risks of each node and feeder during the prediction period according to the predicted load time series includes: Map the predicted load time series to the power distribution network topology structure, and combine the node voltage regulation ability, electrical connection relationship and load conduction path to deduce the predicted voltage response trend; Based on the prediction results of the three-phase load distribution, calculate the inter-phase imbalance rate of the feeder, and combine the historical asymmetric operation records to identify the continuous imbalance evolution trend; According to the rated capacity of the equipment and the load evolution rate, perform time series analysis on the load rates of each node and feeder in the prediction interval to judge whether they cross the equipment operation safety threshold; Cross-compare the judgment results with the load surge characteristics that characterize high-power load usage and demand response characteristics in the behavior labels formed by clustering analysis based on user electricity consumption behavior data, screen out the highly sensitive nodes that are significantly affected by behavior and have overload risks, and construct a node-level risk list.

2. The intelligent power distribution load forecasting and adaptive scheduling method according to claim 1, wherein The obtaining the historical load data, real-time meteorological information and user electricity consumption behavior data in the target distribution area, and establishing a comprehensive load forecasting input data set includes: Obtain the historical active and reactive load data of multiple monitoring points within the time window, classify them according to the feeder and node structure, and perform missing value filling, anomaly elimination and trend stability analysis on them to form a basic load sample set with complete time series; Receive and fuse real-time meteorological information from multiple sources, including temperature, humidity, wind speed, sunshine intensity and weather type, map the meteorological data to the coverage area of each feeder through a spatial interpolation algorithm, and construct a meteorological feature set with geographical annotation; Collect the electricity consumption behavior information on the user side, including periodic electricity consumption patterns, peak-valley electricity consumption response habits, electricity price sensitivity and equipment operation logs, and perform clustering modeling based on user categories to generate a behavior label feature matrix; Based on the load sample set, meteorological feature set, and behavior label feature matrix, a unified multi-dimensional input data structure is constructed, and the data set is dynamically updated according to the rolling prediction period to drive the subsequent short-term load prediction model.

3. The intelligent power distribution load prediction and adaptive scheduling method according to claim 2, characterized in that The short-term load prediction model includes a topology encoding sub-module, a feature fusion sub-module, a time series prediction sub-module, and an output reconstruction sub-module; Among them, the topology encoding sub-module takes the power grid topology structure of the target distribution area as input, and uses a graph neural network to construct a node vector embedding representation with electrical attributes, encoding the connection relationship, impedance parameters, equipment constraints, and physical location relationships of each feeder and node in the topology into low-dimensional continuous vectors, as the explicit expression of the impact of the topology structure on load prediction; The feature fusion sub-module takes the topology encoding result, historical load time series, meteorological features after spatial interpolation, and behavior label matrix as joint input, uses a multi-layer perception fusion mechanism, extracts key features based on the attention mechanism and feature gating mechanism, and eliminates the scale differences and semantic conflicts between data, and outputs a multi-dimensional dynamic feature vector sequence of unified length as the comprehensive driving factor for load changes; The time series prediction sub-module takes the fused multi-dimensional dynamic feature vector sequence as input, and uses a bidirectional gated recurrent unit network enhanced by time series attention to predict the node-level and feeder-level loads step by step in time. By introducing an adaptive time window mechanism and an abnormal state dynamic correction structure, the model has the ability to identify load mutation events in the short term and actively suppress the prediction error; The output reconstruction sub-module maps the predicted values back to the physical numbers of each node or feeder according to the actual configuration relationship between the time series prediction results and the power grid nodes, and attaches confidence indicators, upper and lower bounds of prediction errors, and typical electricity consumption scenario labels to provide multi-level prediction information support for subsequent dispatching strategy optimization and execution control.

4. The intelligent power distribution load prediction and adaptive scheduling method according to claim 3, characterized in that The abnormal state dynamic correction structure in the time series prediction sub-module dynamically adjusts the initial predicted value at the node level through the following formula 1: ； Among them, represents the predicted load value of the -th node after correction at time ; is the initial predicted value of the -th node at time ; is the individual correction coefficient of node , which is obtained by training based on the historical load variance and prediction confidence of this node; represents the overall load variability index at the current moment, which is dynamically calculated based on the standard deviation of the whole - network predicted load sequence; is the abnormal - load sensitivity factor of the node, which is obtained by statistical regression according to the load deviation frequency and amplitude of this node during the past meteorological disturbance periods; is the behavior disturbance index of the node at the current moment, which is constructed based on the fluctuation - type feature dimensions in the behavior label matrix and reflects the driving degree of user behavior on the prediction deviation; is 's topological embedding vector, which is generated by the topological coding sub - module; is the dynamic weight vector of behavior features, which is calculated by the feature fusion module at the current time step and is used to measure the local contribution weight of behavior labels to load prediction.

5. The intelligent power distribution load forecasting and adaptive scheduling method according to claim 3, wherein, The feature fusion sub-module includes a feature construction structure based on the behavior-meteorology joint decoupling attention mechanism, and generates the fusion input representation of each node through the following formula 2: ； Among them, represents the node at the moment of the fused input vector, which is used as the input of the time series prediction sub-module; represents the node extracted from the behavior label feature matrix at the moment of the behavior-driven vector; represents the local environmental factor vector of the meteorological feature after spatial interpolation at the node ; are respectively the trainable weight matrices for decoupling transformation of behavior and meteorological features; is the individual behavior weight coefficient of the node and is calculated using the following formula 3: ； in, Representation Node The historical behavior dominance index is obtained by statistically learning the explanation intensity of user behavior on load changes in the past scheduling cycle; Representation Node The meteorological sensitivity parameters are estimated based on the partial derivatives of node load changes during historical meteorological upheavals.

6. The intelligent power distribution load prediction and adaptive scheduling method according to claim 1, characterized in that Based on the judgment results of the power grid topology structure, electrical parameter constraints, and load prediction time series, a rolling horizon optimization method is used to generate a dispatching control strategy, including: Using a graph structure modeling method to map the power grid topology of the target distribution area into a graph data structure, where nodes represent feeder connection points, transformer positions, or load access points, and edges represent line connection relationships, and considering electrical attributes such as line impedance, voltage level, and switch status, a graph topology representation that fuses topology and parameters is constructed; Embed the node-level load prediction time series obtained from the load prediction model into the above graph structure as the dynamic additional attribute of the time evolution feature, and extract the load response feature distribution under the global network structure dependence through the graph neural network to capture the coupling effects between system levels; Based on the extracted state characteristics of the nodes in the entire network, a rolling time domain optimization framework is introduced to construct a multi-objective nonlinear optimization problem in each rolling cycle. Minimization of voltage deviation, suppression of feeder imbalance, reduction of equipment overload risk and minimization of load adjustment cost are used as joint objective functions, and the operation boundaries, action frequency constraints and chain action logic of various electrical equipment are integrated as constraint conditions. Through evolutionary strategy search, a joint operation sequence covering different control resources is dynamically generated in each rolling cycle, including transformer tap adjustment, capacitor bank switching, tie switch opening and closing, energy storage unit power allocation and user load priority sorting. Each strategy is attached with its execution window, predicted risk area and priority level information to ensure that the dispatch strategy has regional awareness, adaptability and high timeliness.

7. The intelligent distribution load prediction and adaptive scheduling method according to claim 6, wherein The graph structure modeling method includes: The initial topology diagram is generated based on the distribution network GIS data and the equipment master data in the SCADA system. The node type is determined by the equipment attribute field analysis, and the electrical parameters such as line impedance and voltage level are supplemented by real-time monitoring data and standard parameter library. The topology connectivity verification algorithm is used to identify the breakpoints, island nodes or loop structures in the graph, and the graph structure is closed in combination with the electrical operation rules to ensure the integrity and decoupling of the topology graph. Structured parameter vectors are embedded in the nodes and edges of the topology graph respectively. The node parameters include access capacity, node type and control capability identification, and the edge parameters include impedance value, maximum current capacity and current switch state, so as to realize the fusion expression of topological structure and operating characteristics.

8. The intelligent power distribution load prediction and adaptive scheduling method according to claim 6, wherein The process of embedding the node-level load forecast time series into the graph structure includes: The load value of each node at multiple prediction moments is constructed as a dynamic feature vector, and the load confidence, prediction upper and lower bounds, and abnormal fluctuation signs are added to form a time series feature set with uncertainty. The time series feature set is mapped one-to-one with the topology graph nodes through the node and time dual index mechanism, and the prediction sequence is segmented in a rolling manner using the time window mechanism to ensure that the graph neural network input has time consistency within each rolling cycle. A multi-layer graph convolution structure is used to fuse static topological attributes and dynamic load characteristics, so that the node state representation not only reflects the evolution trend of its local load, but also reflects the coupling influence with neighboring nodes in structure and timing, providing context-related input for rolling optimization.

9. The intelligent distribution load prediction and adaptive scheduling method according to claim 6, wherein The process of generating a scheduling control strategy through evolutionary strategy search includes: Based on the extracted node state features, a scheduling encoding vector including the state of the control object, constraints, scheduling action set and initial strategy score is constructed to initialize the population or strategy library; In each rolling cycle, the scheduling action sequence is continuously optimized through cross-mutation and strategy screening mechanisms. Risk buffer distance and equipment protection margin are introduced as implicit penalty items in the strategy generation process to guide the search towards a robust operation path. For each output joint operation sequence, a scheduling area mask, time period priority label and policy coverage range identifier are attached to support the subsequent scheduling execution system to perform regional distribution, priority sorting and policy switching judgment, and realize distributed and hierarchical execution control.

Images