Power flexible adjustment method and system based on voltage out-of-limit risk prediction
By fusing multi-source heterogeneous data and digital twin simulation, an optimal power regulation strategy is generated, which solves the problem of insufficient integration of multi-source dynamic factors in the voltage control strategy in the existing technology, and improves the safety, stability and economy of power grid operation.
Patent Information
- Application Number
- CN202611313323.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-11-17
AI Technical Summary
Existing voltage control strategies lack comprehensive consideration and forward-looking integration of multiple dynamic factors such as meteorological environment, user behavior and equipment status during the generation process. This results in low matching degree between power regulation strategies and actual operating conditions, making it difficult to accurately respond to voltage over-limit risks and affecting the safety, stability and economic efficiency of power grid operation.
By fusing multi-source heterogeneous data and extracting voltage over-limit features, a graph neural network is constructed to predict voltage over-limit risks. Combined with digital twins for simulation, the optimal power regulation strategy is generated. The regulation strategy is then optimized using reinforcement learning algorithms to achieve multi-objective intelligent decision-making.
It improves the matching accuracy between power regulation strategies and actual operating conditions, enhances the timeliness and accuracy of responding to voltage over-limit risks, and takes into account voltage stability, regulation economy, user satisfaction and renewable energy consumption rate, thus ensuring the safe and stable operation of the power grid.
Smart Images

Figure CN123121665A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power regulation technology, and in particular to a power flexible regulation method and system based on voltage over-limit risk prediction. Background Technology
[0002] With the large-scale grid integration of distributed renewable energy generation and the increasing complexity of power load characteristics, voltage exceedance issues in grid operation have gradually become a key factor restricting power quality and power supply security. Voltage exceedance refers to the phenomenon where the voltage amplitude at grid nodes deviates from the allowable operating range. Frequent voltage exceedances may not only trigger protection devices, leading to power outages, but may also accelerate equipment insulation aging and increase grid losses, posing a significant threat to the stable operation of the power system.
[0003] Currently, existing voltage control methods mainly rely on offline simulation analysis or power regulation based on fixed rules. Offline simulation methods pre-determine control strategies by calculating power flow in historical typical scenarios, while fixed-rule regulation methods trigger corresponding control actions based on the comparison between measured voltage values and preset thresholds. These existing technologies have limitations in dealing with voltage exceedance issues, namely, the generation process of regulation strategies lacks comprehensive consideration and forward-looking optimization of multiple dynamic factors. Specifically, voltage fluctuations in actual power grid operation are affected by multiple factors, including changes in meteorological environment, randomness of user electricity consumption behavior, and real-time equipment status. Offline simulation and fixed-rule methods struggle to effectively integrate this real-time changing information when formulating strategies. Due to insufficient consideration of dynamic factors such as meteorological environment, user behavior, and equipment status, the generated power regulation strategies often have a low degree of matching with the actual operating conditions in the current or short-term future, further affecting the effectiveness and accuracy of regulation measures. In complex operating scenarios, under-regulation or over-regulation may occur, making it difficult to simultaneously ensure voltage stability while considering multiple objectives such as regulation costs and renewable energy consumption.
[0004] In summary, existing technologies suffer from a lack of comprehensive and forward-looking integration of multiple dynamic factors, such as meteorological environment, user behavior, and equipment status, during the voltage control strategy generation process. This results in a low degree of matching between the generated power regulation strategy and actual operating conditions, making it difficult to accurately address voltage over-limit risks and further affecting the safety and stability of power grid operation and the economic effectiveness of power regulation. Summary of the Invention
[0005] The purpose of this application is to provide a power flexible regulation method and system based on voltage over-limit risk prediction, in order to solve the technical problems in the prior art where the generation process of voltage control strategy does not take into account the comprehensive consideration and forward-looking integration of multiple dynamic factors such as meteorological environment, user behavior and equipment status, resulting in low matching degree between the generated power regulation strategy and actual operating conditions, making it difficult to accurately deal with voltage over-limit risks, and further affecting the safety and stability of power grid operation and the economic effectiveness of power regulation.
[0006] In view of the above problems, this application provides a power flexible regulation method and system based on voltage over-limit risk prediction.
[0007] In a first aspect, this application provides a power flexible regulation method based on voltage over-limit risk prediction, implemented through a power flexible regulation system based on voltage over-limit risk prediction, comprising: obtaining voltage over-limit features by fusing multi-source heterogeneous data and extracting voltage over-limit features; predicting voltage over-limit risks by combining the voltage over-limit features to obtain voltage over-limit risks; constructing a digital twin of the physical power grid, simulating the voltage over-limit risks by inputting the digital twin, and generating an optimal power regulation strategy; and distributing the optimal power regulation strategy to the physical power grid for execution.
[0008] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: acquiring historical voltage data, historical load data, and historical renewable energy output data to obtain historical electrical data; collecting historical meteorological environment data, historical user behavior data, and historical equipment status data; constructing a graph neural network with the historical electrical data, historical meteorological environment data, historical user behavior data, and historical equipment status data as nodes and data association states as edges; and extracting voltage over-limit features from the graph neural network to obtain the voltage over-limit features.
[0009] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: monitoring the frequency anomalies of load peak keywords in a local information source to obtain load peak keywords; adding the load peak keywords to the user behavior data; extracting the historical electrical data and the historical meteorological data based on the user behavior data to identify the data association status and obtain the data association status; and adding the data association status to the graph neural network.
[0010] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: defining voltage over-limit feature labels; based on the voltage over-limit feature labels, traversing the graph neural network to perform voltage over-limit feature aggregation and outputting a voltage over-limit feature vector; performing importance analysis based on the voltage over-limit feature vector to obtain the voltage over-limit impact and adding it to the voltage over-limit feature.
[0011] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: using the historical electrical data, the historical meteorological environment data, the historical user behavior data, the historical equipment status data, and the voltage over-limit characteristics as inputs to construct a voltage over-limit risk prediction model; collecting real-time electrical data, real-time meteorological environment data, real-time user behavior data, and real-time equipment status data and inputting them into the voltage over-limit risk prediction model to output the voltage over-limit risk.
[0012] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: locating the risk state using the digital twin based on the voltage over-limit risk; constructing an action space according to the risk state; generating a state space by executing a power regulation strategy based on the action space; configuring a reward signal based on the state space; and performing multi-objective intelligent decision-making by combining the action space and the state space to generate the optimal power regulation strategy.
[0013] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: the power regulation strategy of the action space includes at least one of energy storage charging and discharging, flexible load response, and distributed power output regulation.
[0014] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: the state effect objectives of the state space include voltage stability, voltage economy, user satisfaction, and renewable energy consumption rate.
[0015] Preferably, the power flexible regulation method based on voltage over-limit risk prediction further includes: constructing an agent with the digital twin as the interaction environment, configuring the state space, the action space, and the reward signal; the agent selecting an adjustment action from the action space based on the current state and issuing it to the digital twin; the digital twin executing the adjustment action and feeding back a new state and the corresponding reward signal; the agent iteratively updating the parameters of the adjustment action using a reinforcement learning algorithm based on the reward signal; repeating the iterative process until the cumulative reward value reaches the convergence condition, and outputting the optimal power regulation strategy.
[0016] Secondly, this application also provides a power flexible regulation system based on voltage over-limit risk prediction, used to execute the power flexible regulation method based on voltage over-limit risk prediction as described in the first aspect, comprising: a voltage over-limit feature acquisition module, used to obtain voltage over-limit features by performing multi-source heterogeneous data fusion and voltage over-limit feature extraction; a voltage over-limit risk acquisition module, used to combine the voltage over-limit features to predict voltage over-limit risks and obtain voltage over-limit risks; an optimal power regulation strategy generation module, used to construct a digital twin of the physical power grid, input the voltage over-limit risks into the digital twin for simulation, and generate an optimal power regulation strategy; and a strategy execution module, used to distribute the optimal power regulation strategy to the physical power grid for execution.
[0017] The technical solution provided in this application has at least the following technical effects or advantages: by realizing the technical goal of predicting voltage over-limit risks and intelligently generating power regulation strategies based on deep fusion of multi-source heterogeneous data and forward-looking inference of digital twins, it achieves the technical effect of improving the matching accuracy of power regulation strategies with actual operating conditions, enhancing the timeliness and accuracy of responding to voltage over-limit risks, and simultaneously taking into account the multi-objective synergistic optimization of voltage stability, regulation economy, user satisfaction and renewable energy absorption rate, ultimately ensuring the safe and stable operation of the power grid and improving the comprehensive benefits of power regulation.
[0018] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0020] Figure 1 This is a flowchart illustrating the power flexible regulation method based on voltage over-limit risk prediction proposed in this application.
[0021] Figure 2 This is a schematic diagram of the power flexible regulation system based on voltage over-limit risk prediction in this application.
[0022] Figure labeling: Module 1 for obtaining voltage over-limit characteristics, Module 2 for obtaining voltage over-limit risk, Module 3 for generating optimal power regulation strategy, and Module 4 for strategy execution. Detailed Implementation
[0023] This application provides a power flexible regulation method and system based on voltage limit risk prediction. It addresses the technical problem in existing technologies where the generation of voltage control strategies suffers from insufficient comprehensive consideration and forward-looking integration of multiple dynamic factors such as meteorological environment, user behavior, and equipment status. This results in low matching degree between the generated power regulation strategy and actual operating conditions, making it difficult to accurately respond to voltage limit risks and further affecting the safety and stability of power grid operation and the economic effectiveness of power regulation. The application achieves the technical goal of predicting voltage limit risks and intelligently generating power regulation strategies based on deep fusion of multi-source heterogeneous data and forward-looking digital twin simulation. This improves the matching accuracy between power regulation strategies and actual operating conditions, enhances the timeliness and accuracy of voltage limit risk response, and simultaneously considers multiple objectives of voltage stability, regulation economy, user satisfaction, and renewable energy absorption rate through synergistic optimization. Ultimately, this ensures the safe and stable operation of the power grid and improves the overall benefits of power regulation.
[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.
[0025] Example 1, please refer to the appendix. Figure 1 This application provides a power flexible regulation method based on voltage over-limit risk prediction, which is applied to a power flexible regulation system based on voltage over-limit risk prediction, and specifically includes the following steps: Voltage over-limit features are obtained by fusing multi-source heterogeneous data and extracting voltage over-limit features.
[0026] Furthermore, this application also includes: acquiring historical voltage data, historical load data, and historical renewable energy output data over historical periods to obtain historical electrical data; collecting historical meteorological environment data, historical user behavior data, and historical equipment status data over historical periods; constructing a graph neural network using the historical electrical data, the historical meteorological environment data, the historical user behavior data, and the historical equipment status data as nodes and the data association status as edges; and extracting voltage over-limit features from the graph neural network to obtain the voltage over-limit features.
[0027] Furthermore, this application also includes: monitoring the frequency anomalies of load peak keywords in a local information source to obtain load peak keywords; adding the load peak keywords to the user behavior data; extracting the historical electrical data and the historical meteorological data based on the user behavior data to identify the data association status and obtain the data association status; and adding the data association status to the graph neural network.
[0028] Furthermore, this application also includes: defining voltage over-limit feature labels; based on the voltage over-limit feature labels, traversing the graph neural network to perform voltage over-limit feature aggregation and outputting a voltage over-limit feature vector; performing importance analysis based on the voltage over-limit feature vector to obtain the voltage over-limit impact and adding it to the voltage over-limit feature.
[0029] Specifically, historical voltage data, historical load data, and historical renewable energy output data are obtained to obtain historical electrical data. Historical voltage data refers to the historical record value of node voltage in the power grid, historical load data refers to the historical power value of electrical load, and historical renewable energy output data refers to the historical output power value of renewable energy power generation equipment such as wind power and photovoltaic power generation. Historical electrical data is a collective term for the above data, which constitutes the basic electrical quantity information for voltage over-limit analysis.
[0030] Next, historical meteorological environmental data, historical user behavior data, and historical equipment status data are collected. Historical meteorological environmental data includes environmental parameters such as temperature, humidity, wind speed, and light intensity. Historical user behavior data reflects users' electricity consumption patterns and habits. Historical equipment status data describes the operating status and health level of power grid equipment such as transformers and circuit breakers. These data serve as non-electrical quantity information to assist in the analysis of external influencing factors of voltage exceeding limits.
[0031] Then, a graph neural network is constructed using historical electrical data, historical meteorological data, historical user behavior data, and historical equipment status data as nodes and data association states as edges. Nodes represent each data entity or feature, and edges represent the inherent connections or interactions between nodes. By combining nodes and edges, a graph neural network structure capable of capturing complex topological relationships between data is constructed. Specifically, frequency anomaly monitoring of load peak keywords is performed on local information sources to obtain load peak keywords. Local information sources refer to data collection sources such as social media platforms, news websites, or internal power grid communication systems deployed within the power grid area; load peak keywords refer to text terms closely related to peak electricity load events, such as expressions like air conditioner startup or increased industrial electricity consumption; frequency anomaly monitoring refers to real-time monitoring of the frequency of specific keywords within a unit of time and identifying anomalies exceeding normal fluctuation ranges through statistical methods; after frequency anomaly monitoring processing, keywords characterizing load peak events are extracted from massive text data, i.e., load peak keywords. Next, load peak keywords are added to user behavior data, and historical electrical data and the aforementioned historical meteorological data are extracted based on user behavior data to identify data association states, thus obtaining data association states. User behavior data is a comprehensive dataset reflecting users' electricity consumption habits, response patterns, and event-triggered behaviors. Historical electrical data includes historical voltage amplitude, load power, and renewable energy power generation, among other electrical measurements. Historical meteorological data covers environmental parameters such as temperature, humidity, wind speed, and solar irradiance. By embedding peak load keywords into user behavior data, the data's ability to represent peak load events is enhanced. Then, guided by the enhanced user behavior data, the intrinsic relationships between historical electrical and meteorological data are mined, such as the correlation between peak load and high temperatures, and the correlation between renewable energy output fluctuations and user responses. This process employs association rule mining, correlation analysis, or causal inference methods to identify the association states between multi-source data. The data association states are specifically reflected in the connection strength or dependency between data nodes. Furthermore, the data association states are added to the graph neural network. Graph Neural Networks (GNNs) are deep learning models that use graph structures for information transmission and feature aggregation. Their core consists of nodes representing data entities and edges representing relationships between entities. By incorporating data association states as edge attributes or connection weights into GNNs, the original graph structure, which uses historical electrical data, historical meteorological data, historical user behavior data, and historical equipment status data as nodes, gains richer topological information. After adding data association states, GNNs can more accurately simulate the complex interactions between multi-source heterogeneous data, providing a more refined graph structure foundation for subsequent voltage over-limit feature extraction.
[0032] Furthermore, voltage limit violation features are extracted from the graph neural network. This process utilizes the aggregation and propagation mechanism of the graph neural network to mine deep patterns related to voltage limit violations from node features and edge relationships. The extracted voltage limit violation features are used for subsequent risk prediction and control decisions. Voltage limit violation feature labels are defined. Voltage limit violation feature labels refer to pre-defined identification information based on historical voltage limit violation event records or expert experience rules, used to mark node attributes or sample categories closely related to voltage limit violations in the graph neural network, serving as a benchmark for subsequent supervised learning or feature extraction.
[0033] Subsequently, based on the voltage over-limit feature labels, the graph neural network is traversed to aggregate voltage over-limit features, outputting a voltage over-limit feature vector. A graph neural network is a deep learning model based on a graph structure composed of nodes and edges. Nodes represent multi-source data entities, and edges represent the relationships between data. Traversing the graph neural network refers to performing information transmission and update operations layer by layer according to the graph structure. Voltage over-limit feature aggregation refers to using the message passing mechanism of the graph neural network, guided by the voltage over-limit feature labels, to weightedly fuse the feature information of adjacent nodes, thereby extracting a deep representation reflecting the essence of voltage over-limit risk. The voltage over-limit feature vector output after the aggregation process is a multi-dimensional numerical combination that comprehensively characterizes the features related to voltage over-limit.
[0034] Furthermore, importance analysis is performed based on the voltage exceedance feature vector to obtain the voltage exceedance impact, which is then added to the voltage exceedance feature. Importance analysis refers to quantifying the contribution of each dimension of the voltage exceedance feature vector to the voltage exceedance event using attention mechanisms, feature weight calculation, or interpretability analysis methods. The voltage exceedance impact refers to the quantitative index obtained through importance analysis, reflecting the magnitude of the effect of different original factors on the voltage exceedance risk. The voltage exceedance impact is integrated as supplementary information into the voltage exceedance feature, so that the voltage exceedance feature not only includes the aggregated feature vector but also covers the influence weights of each factor, thereby enhancing the expressive power and interpretability of the feature.
[0035] By combining the voltage over-limit characteristics, voltage over-limit risk is predicted, and the voltage over-limit risk is obtained.
[0036] Furthermore, this application also includes: using the historical electrical data, the historical meteorological environment data, the historical user behavior data, the historical equipment status data, and the voltage over-limit characteristics as inputs to construct a voltage over-limit risk prediction model; collecting real-time electrical data, real-time meteorological environment data, real-time user behavior data, and real-time equipment status data and inputting them into the voltage over-limit risk prediction model to output the voltage over-limit risk.
[0037] Specifically, historical electrical data, historical meteorological data, historical user behavior data, historical equipment status data, and voltage over-limit characteristics are used as inputs, i.e., as training samples, into a machine learning framework or deep learning architecture. The model is trained and the parameters are optimized through supervised or unsupervised learning methods to construct a voltage over-limit risk prediction model that can map the complex relationship between input data and voltage over-limit risk.
[0038] Subsequently, real-time electrical data, real-time meteorological data, real-time user behavior data, and real-time equipment status data are collected and input into the voltage limit exceedance risk prediction model, which outputs the voltage limit exceedance risk. Real-time electrical data refers to the electrical quantities such as voltage, load, and renewable energy output obtained in real time through the power grid measurement system; real-time meteorological data refers to environmental parameters such as temperature, humidity, wind speed, and light intensity collected in real time from meteorological monitoring stations; real-time user behavior data refers to the current user electricity consumption patterns and response behaviors obtained in real time through smart meters or user interaction platforms; real-time equipment status data refers to the operating status and health indicators of power grid equipment fed back in real time by online monitoring devices. The multi-source data collected in real time are synchronously input into the pre-built voltage limit exceedance risk prediction model. The voltage limit exceedance risk prediction model calculates and infers based on the mapping relationship learned in the training phase, and finally outputs a voltage limit exceedance risk value that represents the possibility of voltage exceedance at the current moment or in the future.
[0039] A digital twin of the physical power grid is constructed, the voltage over-limit risk is input into the digital twin for simulation, and an optimal power regulation strategy is generated.
[0040] Furthermore, this application also includes: locating the risk state using the digital twin based on the voltage over-limit risk; constructing an action space based on the risk state; generating a state space by executing a power regulation strategy based on the action space; configuring a reward signal based on the state space; and performing multi-objective intelligent decision-making by combining the action space and the state space to generate the optimal power regulation strategy.
[0041] Furthermore, this application also includes: the power regulation strategy of the action space includes at least one of energy storage charging and discharging, flexible load response, and distributed power output regulation.
[0042] Furthermore, this application also includes: the state effect objectives of the state space include voltage stability, voltage economy, user satisfaction, and renewable energy consumption rate.
[0043] Furthermore, this application also includes: constructing an agent with the digital twin as the interaction environment, configuring the state space, the action space, and the reward signal; the agent selecting an adjustment action from the action space based on the current state and issuing it to the digital twin; the digital twin executing the adjustment action and feeding back a new state and the corresponding reward signal; the agent iteratively updating the parameters of the adjustment action using a reinforcement learning algorithm based on the reward signal; repeating the iterative process until the cumulative reward value reaches the convergence condition, and outputting the optimal power adjustment strategy.
[0044] Specifically, a digital twin of the physical power grid is constructed, and the risk status is located through the digital twin based on voltage exceedance risk. The physical power grid refers to the actual operating power network entity and its equipment; the digital twin is a high-fidelity digital mirror of the physical power grid in virtual space, simulating the operating characteristics and dynamic behavior of the physical power grid through real-time data mapping and simulation models; voltage exceedance risk is a quantitative indicator representing the probability of voltage exceedance obtained based on the aforementioned steps; the risk status refers to the specific operating conditions related to voltage exceedance risk simulated in the digital twin, including key attributes such as the location node where the risk occurs, the risk level, and the scope of the risk's impact; by inputting the voltage exceedance risk into the digital twin, the simulation computing capabilities of the digital twin are used to extrapolate the power grid's operating status, thereby accurately locating the risk status.
[0045] Based on the risk status, an action space is constructed. The action space is the set of all possible control actions that can be executed during power regulation. This set is designed and limited according to the specific characteristics of the risk status to ensure that the action space covers effective control measures that can mitigate or eliminate voltage over-limit risks. The power regulation strategies in the action space include at least one of energy storage charging and discharging, flexible load response, and distributed power generation output regulation. Energy storage charging and discharging refers to absorbing or releasing active power by controlling the charging or discharging process of the energy storage system. Flexible load response refers to the regulation operation of power reduction or transfer of adjustable loads. Distributed power generation output regulation refers to adjusting the active or reactive power output of distributed photovoltaic, wind power, and other power sources. The power regulation strategies together constitute the basic elements of the action space, providing optional control measures for subsequent strategy generation.
[0046] A state space is generated based on the execution of power regulation strategies within the action space. The state space refers to the set of power grid operating states presented after executing different power regulation strategies in the digital twin. Each state corresponds to the implementation effect of a specific power regulation strategy. The state effect objectives of the state space include voltage stability, voltage economy, user satisfaction, and renewable energy absorption rate. Voltage stability reflects the degree to which the voltage at power grid nodes is maintained within the allowable range. Voltage economy reflects the control costs and economic benefits during the power regulation process. User satisfaction characterizes the acceptance level and comfort impact of power regulation on the load side. Renewable energy absorption rate measures the proportion of distributed power generation that is effectively utilized. By traversing the regulation strategies in the action space and simulating their execution in the digital twin, the corresponding effect indicators after the execution of each strategy are collected, forming a state space representing different regulation results.
[0047] A smart agent is constructed using a digital twin as its interactive environment, configuring a state space, action space, and reward signal. The digital twin is a high-fidelity digital mirror of the physical power grid in virtual space, capable of mapping the operating characteristics and dynamic behavior of the physical power grid in real time. The smart agent is a computational entity that makes autonomous decisions based on reinforcement learning algorithms, learning optimal control strategies through interaction with the digital twin. The state space encompasses a set of power grid operating states, including key parameters such as voltage levels, load distribution, and renewable energy output. The action space contains all executable power regulation actions, such as energy storage charging and discharging, flexible load response, and distributed power generation output regulation. The reward signal is a quantified feedback value configured according to the effect target in the state space, used to evaluate the quality of the regulation actions. By configuring these core elements, the foundation for the smart agent's learning and decision-making within the digital twin is laid.
[0048] Subsequently, the agent selects an adjustment action from the action space based on the current state and issues it to the digital twin. The current state refers to the real-time power grid operating conditions simulated by the digital twin, including the specific risk state after the voltage over-limit risk is located. The agent searches for and selects an adjustment action in the action space based on the current state. This adjustment action is issued to the digital twin in the form of a control command to trigger power regulation operations in the virtual environment, thus realizing the mapping from state to action.
[0049] Next, after executing the adjustment action, the digital twin feeds back the new state and the corresponding reward signal. After receiving the adjustment action from the agent, the digital twin simulates the changes in power grid operation caused by the action in the virtual environment, generating a new state space vector. At the same time, it calculates the reward signal value corresponding to the new state according to the preset reward function, and feeds back the new state and reward signal to the agent, providing experience data for the agent's learning and forming a closed-loop interaction.
[0050] Furthermore, based on the reward signal, the agent iteratively updates the parameters of the adjustment action using reinforcement learning algorithms. After receiving the reward signal, the agent uses reinforcement learning algorithms such as Q-learning, deep deterministic policy gradient, or proximal policy optimization to adjust the internal parameters of the action selection policy based on the positive and negative feedback of the reward signal. This makes the agent more inclined to choose the adjustment action that can obtain higher cumulative rewards in subsequent decisions, thereby achieving continuous optimization of the policy and parameter updates.
[0051] Finally, the iterative process is repeated until the cumulative reward value reaches the convergence condition, and the optimal power regulation strategy is output. The agent and digital twin continuously interact cyclically, selecting actions, receiving environmental feedback, and updating parameters. The cumulative reward value gradually increases and stabilizes. When the preset convergence condition is met, such as reward value fluctuations being less than a threshold or the number of iterations reaching the upper limit, the iteration terminates. The strategy learned by the agent at this point is the optimal power regulation strategy, which effectively addresses the risk of voltage exceeding limits while also considering multi-objective optimization.
[0052] The optimal power regulation strategy is then sent to the physical power grid for execution.
[0053] Specifically, the optimal power regulation strategy is distributed to the physical power grid for execution. The optimal power regulation strategy refers to the power control scheme obtained through the interaction and optimization between the digital twin and the reinforcement learning agent in the aforementioned steps. This scheme includes specific control parameters such as energy storage charging and discharging commands, flexible load response plans, and distributed power generation output adjustment values. Distribution refers to transmitting the optimal power regulation strategy to the field execution unit of the physical power grid through the power grid dispatch communication network or the interface of the automated control system. The physical power grid refers to the actual power system entity composed of transmission lines, substations, distribution equipment, distributed power sources, and various loads. Execution refers to the field execution unit receiving the strategy and driving the energy storage system, controllable loads, and distributed power sources to perform actual adjustment operations according to the power values set by the strategy, thereby controlling the risk of voltage exceeding limits within the allowable range and achieving the safe and stable operation of the power grid.
[0054] In summary, the power flexible regulation method based on voltage over-limit risk prediction provided in this application has the following technical effects: by achieving the technical goal of predicting voltage over-limit risks and intelligently generating power regulation strategies based on deep fusion of multi-source heterogeneous data and forward-looking digital twin simulation, it achieves the technical effects of improving the matching accuracy between power regulation strategies and actual operating conditions, enhancing the timeliness and accuracy of responding to voltage over-limit risks, and simultaneously taking into account the multi-objective synergistic optimization of voltage stability, regulation economy, user satisfaction and renewable energy absorption rate, ultimately ensuring the safe and stable operation of the power grid and improving the comprehensive benefits of power regulation.
[0055] Example 2: Based on the same inventive concept as the power flexible regulation method based on voltage over-limit risk prediction in the foregoing examples, this application also provides a power flexible regulation system based on voltage over-limit risk prediction. Please refer to the appendix. Figure 2 The system includes: a voltage over-limit feature acquisition module 1, used to obtain voltage over-limit features by performing multi-source heterogeneous data fusion and voltage over-limit feature extraction; a voltage over-limit risk acquisition module 2, used to combine the voltage over-limit features to predict voltage over-limit risk and obtain voltage over-limit risk; an optimal power regulation strategy generation module 3, used to construct a digital twin of the physical power grid, input the voltage over-limit risk into the digital twin for simulation, and generate an optimal power regulation strategy; and a strategy execution module 4, used to distribute the optimal power regulation strategy to the physical power grid for execution.
[0056] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used to: acquire historical voltage data, historical load data, and historical renewable energy output data to obtain historical electrical data; collect historical meteorological environment data, historical user behavior data, and historical equipment status data; construct a graph neural network using the historical electrical data, historical meteorological environment data, historical user behavior data, and historical equipment status data as nodes and data association states as edges; and extract voltage over-limit features from the graph neural network to obtain the voltage over-limit features.
[0057] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used for: monitoring the frequency anomalies of load peak keywords in the local information source to obtain load peak keywords; adding the load peak keywords to the user behavior data; extracting the historical electrical data and the historical meteorological data based on the user behavior data to identify the data association status and obtain the data association status; and adding the data association status to the graph neural network.
[0058] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used for: defining voltage over-limit feature labels; based on the voltage over-limit feature labels, traversing the graph neural network to perform voltage over-limit feature aggregation and outputting a voltage over-limit feature vector; performing importance analysis based on the voltage over-limit feature vector to obtain the voltage over-limit impact and adding it to the voltage over-limit feature.
[0059] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used to: construct a voltage over-limit risk prediction model by taking the historical electrical data, the historical meteorological environment data, the historical user behavior data, the historical equipment status data, and the voltage over-limit characteristics as inputs; collect real-time electrical data, real-time meteorological environment data, real-time user behavior data, and real-time equipment status data as inputs to the voltage over-limit risk prediction model, and output the voltage over-limit risk.
[0060] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used for: locating the risk state through the digital twin based on the voltage over-limit risk; constructing an action space according to the risk state; generating a state space by executing a power regulation strategy based on the action space; configuring a reward signal based on the state space; and performing multi-objective intelligent decision-making by combining the action space and the state space to generate the optimal power regulation strategy.
[0061] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used for: the power regulation strategy of the action space includes at least one of energy storage charging and discharging, flexible load response, and distributed power output regulation.
[0062] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used for: the state effect objectives of the state space include voltage stability, voltage economy, user satisfaction, and renewable energy consumption rate.
[0063] Furthermore, the power flexible regulation system based on voltage over-limit risk prediction is also used for: constructing an agent with the digital twin as the interaction environment, configuring the state space, the action space, and the reward signal; the agent selecting an adjustment action from the action space based on the current state and issuing it to the digital twin; the digital twin executing the adjustment action and feeding back a new state and the corresponding reward signal; the agent iteratively updating the parameters of the adjustment action using a reinforcement learning algorithm based on the reward signal; repeating the iterative process until the cumulative reward value reaches the convergence condition, and outputting the optimal power regulation strategy.
[0064] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The power flexible regulation method and specific examples based on voltage over-limit risk prediction in the aforementioned Embodiment 1 are also applicable to the power flexible regulation system based on voltage over-limit risk prediction in this embodiment. Through the foregoing detailed description of the power flexible regulation method based on voltage over-limit risk prediction, those skilled in the art can clearly understand the power flexible regulation system based on voltage over-limit risk prediction in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.
[0065] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
[0066] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.
Claims
1. A method for power flexibility regulation based on voltage excursion risk prediction, characterized in that, include: By fusing multi-source heterogeneous data and extracting voltage over-limit features, voltage over-limit features are obtained; By combining the voltage over-limit characteristics, voltage over-limit risk is predicted, and the voltage over-limit risk is obtained. A digital twin of the physical power grid is constructed, the voltage over-limit risk is input into the digital twin for simulation, and an optimal power regulation strategy is generated; The optimal power regulation strategy is then sent to the physical power grid for execution.
2. The method for power flexibility adjustment based on voltage excursion risk prediction as claimed in claim 1, wherein, By fusing multi-source heterogeneous data and extracting voltage over-limit features, voltage over-limit features are obtained, including: Historical voltage data, historical load data, and historical renewable energy output data are obtained from historical time periods to obtain historical electrical data. Collect historical meteorological and environmental data, historical user behavior data, and historical device status data for different time periods; A graph neural network is constructed using the historical electrical data, historical meteorological environment data, historical user behavior data, and historical equipment status data as nodes and the data association status as edges; Voltage over-limit feature extraction is performed on the graph neural network to obtain the voltage over-limit feature.
3. The method for power flexibility adjustment based on voltage excursion risk prediction as claimed in claim 2, wherein, Constructing a graph neural network includes: The frequency anomalies of keywords during peak load periods are monitored using local information sources to identify keywords during peak load periods. The peak load keywords are added to the user behavior data. Based on the user behavior data, the historical electrical data and historical meteorological data are extracted to identify the data association status, and the data association status is obtained. The data association state is added to the graph neural network.
4. The method for power flexibility adjustment based on voltage excursion risk prediction as claimed in claim 2, wherein, Voltage over-limit feature extraction is performed on the graph neural network to obtain the voltage over-limit features, including: Define voltage over-limit feature labels; Based on the voltage over-limit feature labels, the graph neural network is traversed to perform voltage over-limit feature aggregation and output a voltage over-limit feature vector. An importance analysis is performed based on the voltage over-limit feature vector to obtain the voltage over-limit impact, which is then added to the voltage over-limit feature.
5. The method for power flexibility adjustment based on voltage excursion risk prediction as claimed in claim 2, wherein, By combining the voltage over-limit characteristics, voltage over-limit risk prediction is performed, resulting in voltage over-limit risk, including: Using the historical electrical data, historical meteorological data, historical user behavior data, historical equipment status data, and voltage over-limit characteristics as inputs, a voltage over-limit risk prediction model is constructed. The voltage over-limit risk prediction model is input by collecting real-time electrical data, real-time meteorological environmental data, real-time user behavior data, and real-time equipment status data, and outputs the voltage over-limit risk.
6. The method for power flexibility adjustment based on voltage excursion risk prediction as claimed in claim 1, wherein, The voltage over-limit risk is input into the digital twin for simulation, and an optimal power regulation strategy is generated, including: Based on the aforementioned voltage over-limit risk, the risk status is located using the digital twin; Based on the described risk status, construct the action space; Based on the action space, a power adjustment strategy is executed to generate a state space; Based on the state space configuration reward signal, multi-objective intelligent decision-making is performed by combining the action space and the state space to generate the optimal power adjustment strategy.
7. The power flexible regulation method based on voltage over-limit risk prediction as described in claim 6, characterized in that, The power regulation strategy of the action space includes at least one of energy storage charging and discharging, flexible load response, and distributed power output regulation.
8. The power flexible regulation method based on voltage over-limit risk prediction as described in claim 6, characterized in that, The state-space state performance objectives include voltage stability, voltage economy, user satisfaction, and renewable energy absorption rate.
9. The power flexible regulation method based on voltage over-limit risk prediction as described in claim 6, characterized in that, Based on the state space, a reward signal is configured, and multi-objective intelligent decision-making is performed by combining the action space and the state space to generate the optimal power adjustment strategy, including: Construct an intelligent agent with the digital twin as the interaction environment, and configure the state space, the action space and the reward signal; The intelligent agent selects an adjustment action from the action space based on the current state and sends it to the digital twin; After the digital twin performs the adjustment action, it provides feedback on the new state and the corresponding reward signal; The agent iteratively updates the parameters of the adjustment action based on the reward signal using a reinforcement learning algorithm; Repeat the iterative process until the cumulative reward value reaches the convergence condition, and then output the optimal power adjustment strategy.
10. A power flexible regulation system based on voltage over-limit risk prediction, characterized in that, The steps for implementing the power flexible regulation method based on voltage over-limit risk prediction as described in any one of claims 1 to 9 include: The voltage over-limit feature acquisition module is used to obtain voltage over-limit features by performing multi-source heterogeneous data fusion and voltage over-limit feature extraction; The voltage over-limit risk acquisition module is used to combine the voltage over-limit characteristics to predict the voltage over-limit risk and obtain the voltage over-limit risk. The optimal power regulation strategy generation module is used to construct a digital twin of the physical power grid, input the voltage over-limit risk into the digital twin for simulation, and generate the optimal power regulation strategy. The strategy execution module is used to send the optimal power regulation strategy to the physical power grid for execution.
