A virtual power plant optimization scheduling system and method

Abstract

The present application relates to the technical field of virtual power plant, and discloses a virtual power plant optimization scheduling system and method, wherein the optimization scheduling system comprises a data acquisition module, an edge computing module, a prediction module, a scheduling controller, a topology reconstruction module and an intelligent terminal device cluster; the present application realizes the rapid generation and issuance of the scheduling scheme by localizing the processing and prediction of the sensing data through the edge computing module, avoids the response delay problem caused by network transmission and centralized calculation in the traditional centralized architecture, thereby supporting the millisecond-level scheduling feedback, significantly improving the real-time response capability under the condition of sudden load fluctuation or fault, enhancing the identification and trend prediction capability of the system to the meteorological disturbance, equipment aging and operation anomaly by constructing the multi-dimensional sensing and prediction mechanism based on the LSTM neural network prediction model, improving the intelligent level of the virtual power plant system, and the system can dynamically generate the optimal scheduling strategy to ensure the stable operation of the virtual power plant under various working conditions.

Description

Technical Field

[0001] This invention relates to the field of virtual power plant technology, specifically a virtual power plant optimized scheduling system and method. Background Technology

[0002] A Virtual Power Plant (VPP) is an intelligent system that integrates various distributed energy resources (such as photovoltaic power generation, wind power generation, energy storage systems, and adjustable loads) and achieves optimized scheduling and operation through unified coordination and control.

[0003] In existing technologies, virtual power plants primarily rely on traditional data acquisition and control architectures, consisting of a perception layer, a communication layer, and a central decision-making layer. Operational data from various energy nodes are first aggregated to a central server via a communication network. A centralized platform then handles data storage, visualization, analysis, processing, and scheduling decisions. Scheduling strategies are mostly based on static rules, and prediction and regulation depend heavily on human experience or simplified models, lacking dynamic perception and rapid response mechanisms for real-time operational status. Currently, some systems have introduced preliminary load forecasting models and SCADA (Supervisory Control and Controller Area Network) systems. While the DataAcquisition (DAO) system employs an autoregressive integrated moving average forecasting model for monitoring, its overall architecture remains primarily centralized. Consequently, existing virtual power plant systems suffer from numerous structural defects during actual operation. First, these systems heavily rely on a central server for centralized data processing and scheduling, resulting in long data links, heavy processing loads, and significant delays when dealing with high-frequency data changes, making it difficult to achieve millisecond-level response and real-time closed-loop control. Second, due to architectural limitations, the system has weak perception capabilities for environmental disturbances such as sudden weather changes, load fluctuations, and equipment malfunctions, hindering precise scheduling and making it difficult to ensure stable operation of the virtual power plant under various operating conditions. Summary of the Invention

[0004] The purpose of this invention is to provide a virtual power plant optimization scheduling system and method to solve the problems of slow response speed and low scheduling accuracy in existing virtual power plant optimization scheduling systems.

[0005] The technical solution of this invention is: A virtual power plant optimized dispatching system includes a data acquisition module, an edge computing module, a prediction module, a dispatch controller, a topology reconfiguration module, and a cluster of intelligent terminal devices. The data acquisition module acquires real-time and historical data on meteorological conditions, grid load, equipment status, and external interference. The edge computing module is equipped with a data fusion unit for fusing and analyzing the acquired real-time data to form a real-time operating status view of the power grid. The prediction module uses an LSTM neural network-based prediction model to predict the power grid operating status information for a future set time period based on the acquired historical data. This power grid operating status information includes grid load change trends and fault risks. The dispatch controller receives the real-time operating status view and the predicted operating status of the power grid. The system receives information and generates scheduling instructions. The scheduling controller has a built-in multi-objective optimization algorithm framework, which performs global optimization calculations on the scheduling nodes in the virtual power plant based on genetic algorithms or particle swarm optimization algorithms, and generates a preliminary scheduling strategy. The topology reconstruction module is used to dynamically generate multiple backup network topology schemes based on the operating status of the virtual power plant nodes and the predicted grid operating load, and selects the optimal topology scheme through a built-in decision tree model. The scheduling controller receives the optimal topology scheme and integrates it with the preliminary scheduling strategy to generate a node-level scheduling instruction for the globally optimal scheduling scheme. The intelligent terminal device cluster includes circuit breaker drivers, automatic reclosing controllers, and energy storage system regulators deployed in the grid nodes, which are used to receive and execute the node-level scheduling instructions issued by the scheduling controller.

[0006] Preferably, as a further improvement of the present invention, the scheduling controller constructs a power grid topology model based on graph theory, models each node device and load as vertices of the graph, models the lines as edges of the graph, and dynamically adjusts network weights and connection relationships by obtaining data in real time.

[0007] Preferably, as a further improvement of the present invention, during the adjustment and optimization process, the scheduling controller combines real-time collected line load rate, equipment fault risk score, and regional load difference data, incorporates the objectives of minimizing system loss, achieving load balance, and reducing equipment risk into the weight calculation logic, and dynamically adjusts the weight values ​​of the edges by weighting these parameters through a genetic algorithm or particle swarm algorithm.

[0008] Preferably, as a further improvement of the present invention, the multi-objective optimization algorithm framework introduces a fitness function to comprehensively evaluate energy efficiency loss, equipment status score and regional load balance coefficient index, and considers equipment health status, maintenance cycle and remaining life factors in the optimization calculation process.

[0009] Preferably, as a further improvement of the present invention, the data acquisition module includes a temperature sensor, a humidity sensor, an anemometer, a light intensity meter, a voltage transformer, a current transformer, a partial discharge detector, and a device status detector. A local sensor network is constructed through an RS485 or LoRa communication module, and a clock synchronization mechanism is used to ensure the time consistency of various data acquisitions.

[0010] Preferably, as a further improvement of the present invention, the preprocessing process includes filtering and denoising, anomaly removal, time alignment, and data standardization.

[0011] Preferably, as a further improvement of the present invention, the data fusion unit adopts a multi-source fusion algorithm based on Kalman filtering and wavelet transform to construct a unified data fusion framework for different types of information. After fusing and analyzing meteorological data, power grid load, and equipment status, it performs correlation modeling to generate a visualized real-time operation status view.

[0012] Preferably, as a further improvement of the present invention, the power grid load trend prediction model is constructed based on a supervised learning algorithm. The training data includes power grid load and meteorological data from the past 12 months, and an LSTM time series network is used to predict the load trend for the next hour. The equipment failure probability model is constructed using unsupervised K-means clustering and isolated forest algorithms. The input data for the equipment failure probability model includes vibration amplitude, current anomalies, and thermal imaging data in the equipment status. The input data is processed by the unsupervised K-means clustering algorithm to obtain normal clusters in the new equipment data. Then, based on the normal clusters of the new equipment data, the isolated forest algorithm is used to detect anomalies, determine whether the new equipment data deviates from the feature distribution corresponding to the normal clusters, output the equipment anomaly score, and identify the faulty equipment nodes.

[0013] Preferably, as a further improvement of the present invention, it also includes a closed-loop feedback control module and a fault self-healing control module. The closed-loop feedback control module is used to continuously monitor the equipment operating parameters and dynamically compare them with the predicted values. When a deviation exceeding a preset threshold is detected, feedback correction is performed. The fault self-healing control module is used to perform backup power switching and fault isolation emergency operations when a fault occurs or the load fluctuates abnormally. The fault self-healing control module includes: a fault diagnosis unit, which identifies fault nodes based on real-time collected equipment status data, including voltage surges, current imbalances, and frequency anomalies, using a support vector machine algorithm; a self-healing path calculation unit, which calculates the optimal control path from the current node state to the fault recovery node based on a graph search algorithm; and an automatic execution unit, which controls the disconnecting switch and starts the bypass power supply equipment.

[0014] This invention also discloses a virtual power plant optimization scheduling method, implemented based on the aforementioned virtual power plant optimization scheduling system, comprising the following steps: Acquire real-time and historical data on meteorological conditions, power grid load, equipment status, and external interference. The acquired real-time data is fused and analyzed through edge node computing to generate a real-time power grid operation status view. Based on historical data, a power grid load trend prediction model is used to predict the power grid load change trend in a future set period, and an equipment failure probability model is used to predict the equipment failure risk in a future set period. Based on the real-time operation status view of the power grid, the predicted power grid load change trend, and the equipment failure risk, a preliminary scheduling strategy is generated by global optimization of the virtual power plant scheduling node based on the genetic algorithm or particle swarm optimization algorithm. During the optimization process, multiple backup network topology schemes are dynamically generated based on the operation status of the virtual power plant node and the predicted power grid operation load. The optimal topology scheme is selected by the decision tree model. The optimal topology scheme is integrated with the preliminary scheduling strategy to generate the node-level scheduling instruction of the global optimal scheduling scheme. Clusters of intelligent terminal devices deployed in power grid nodes receive and automatically execute generated node-level scheduling instructions.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 0. By using the edge computing module to perform localized processing and prediction of the sensing data, the scheduling scheme can be generated and distributed quickly, avoiding the response delay problem caused by network transmission and centralized computing in the traditional centralized architecture. This enables millisecond-level scheduling feedback and significantly improves the real-time response capability in the event of sudden load fluctuations or failures.

[0016] 1. By introducing multi-type sensor fusion for data acquisition and constructing a multi-dimensional perception and prediction mechanism through power grid load trend prediction model and equipment failure probability model, the system's ability to identify and predict trends of meteorological disturbances, equipment aging and operational anomalies is enhanced. This breaks through the bottleneck of traditional systems with limited perception dimensions and lack of prediction capabilities, resulting in low dispatch accuracy. It comprehensively improves the intelligence level, dispatch reliability and operation efficiency of the virtual power plant system.

[0017] 2. The coordination between the dispatch controller and the topology reconfiguration module enables the system to dynamically generate the optimal dispatch strategy, ensuring the stable operation of the virtual power plant under various operating conditions. Attached Figure Description

[0018] Figure 1 This is a flowchart of a virtual power plant optimization scheduling method according to the present invention. Detailed Implementation

[0019] The following is combined with Figure 1The specific embodiments of the present invention will be described in detail below. In the description of the invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0020] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature; in the description of the invention, unless otherwise stated, "a plurality of" means two or more.

[0021] Example 1 This invention provides a virtual power plant optimized scheduling system. The system constructs a multi-layered scheduling architecture, including a perception layer, a scheduling layer, and a response layer. It uses a multi-dimensional sensor network, edge computing nodes, and intelligent control mechanisms to perform real-time perception, optimized decision-making, and dynamic response to the power grid.

[0022] In the constructed perception layer: Deploy multiple types of sensors as data acquisition modules to collect meteorological data, power grid load data, equipment status data, and external interference information.

[0023] It should be noted that, in order to improve the coverage and data quality of the sensing layer regarding the power grid's operating status, the system deployment described in this embodiment includes temperature sensors, humidity sensors, anemometers, light intensity meters, voltage transformers, current transformers, partial discharge detectors, and equipment status detectors. A clock synchronization mechanism (such as NTP or GPS) is used to ensure the time consistency of various data acquisitions and avoid misjudgments due to data offset. External interference information includes lightning strikes, wind disasters, etc.

[0024] An edge computing module is deployed to preprocess the multidimensional raw data collected by the data acquisition module at the edge computing node. Timestamp alignment and Kalman filtering algorithms are used for multi-source data synchronization and cleaning, including filtering and denoising, anomaly removal, time alignment, and data standardization. The edge computing module is equipped with a data fusion unit, which uses a multi-source fusion algorithm based on Kalman filtering and wavelet transform to achieve unified modeling of different types of information, forming a real-time operational status view of the power grid. Specifically, wavelet transform is used to extract the time-frequency features of multi-source sensor data, and Kalman filtering is combined to construct a dynamic state estimation model, forming a cross-domain correlation framework of meteorological, power grid, and equipment data. This is achieved by fusing and analyzing real-time meteorological data including wind speed, temperature, humidity, and light intensity; power grid load data including real-time total load, regional load rate, and real-time voltage or current values ​​of key nodes; equipment status data including equipment operating temperature, vibration amplitude, fault risk score, number and identification of abnormal equipment; line status data including real-time line load rate and on / off status; and overall system operation status data including normal / early warning / fault status. This establishes a correlation model between equipment health and power grid load, generating a comprehensive visualization interface for the power grid operation status.

[0025] During the analysis phase, edge computing nodes call the power grid load prediction model and equipment failure probability model to predict and assess the load trend changes and equipment status risks in the next hour after cleaning the data. The prediction results are uploaded to the dispatch layer in real time. At the same time, the empirical pattern recognition model is combined to extract features and label the sudden external interference (such as lightning strikes and wind disasters), thereby providing the dispatch layer with accurate and dynamic power grid operation status information input, improving the system's ability to anticipate abnormal states and the level of operational safety assurance.

[0026] Among them, the power grid load forecasting model is built based on a supervised learning algorithm. The training data includes power grid load and meteorological data from the past 12 months. The LSTM time series network is used to predict the load trend for the next hour. The equipment fault detection model is built upon unsupervised K-means clustering and the Isolation Forest algorithm. These two algorithms are a collaborative combination. The collaborative mechanism leverages the intra-cluster partitioning capability of K-means clustering to provide a more accurate normal sample training set for the Isolation Forest algorithm. This addresses the issue of the Isolation Forest algorithm being susceptible to variations in normal data distribution and exhibiting low anomaly detection accuracy when directly processing the full dataset, achieving a complementary optimization of "clustering before detection." The collaborative architecture employs a sequential process, with the K-means clustering module serving as the pre-processing data unit and the Isolation Forest algorithm as the core unit for subsequent anomaly detection. Sample transfer between the two is achieved through a data interface. The processing procedure is as follows: First, the system receives device data (vibration amplitude, current anomaly, thermal imaging data, etc.) collected by the sensing layer (environmental monitoring sensors, current and voltage sensors, etc.) and cleaned by timestamp alignment and Kalman filtering algorithm. Then, the K-means clustering algorithm clusters these unlabeled data, aggregating the data under normal device operation into multiple normal clusters (removing initial outliers with extremely high dispersion). Subsequently, samples are extracted from each normal cluster as the training set for the Isolation Forest algorithm. After training, the Isolation Forest algorithm performs anomaly detection on new real-time input device data, judging whether the data deviates from the feature distribution corresponding to the normal cluster, and finally outputs the device anomaly score, realizing the identification of potentially faulty device nodes.

[0027] The aforementioned power grid load forecasting model and equipment fault detection model are deployed on edge nodes, and the model weights are updated periodically by the central server. The power grid load forecasting model supports a rolling update mechanism, dynamically adjusting the training set through a sliding window to enhance the model's adaptability to new load patterns. The equipment health assessment model supports comprehensive scoring of multi-dimensional anomaly indicators and can automatically remove outliers to improve the accuracy of fault identification. The central server can remotely schedule parameter synchronization of edge models and record the model version evolution trajectory, facilitating backtracking analysis and long-term stable operation.

[0028] It should be noted that, in constructing the perception layer, the system described in this embodiment flexibly configures the density and type of sensors according to the on-site deployment requirements, and uniformly identifies various data sources through node coding and timestamps to ensure data traceability and consistency; the edge computing module presets a data tolerance range and automatically triggers denoising and compensation algorithms when receiving abnormal signals to ensure the integrity and accuracy of the perceived data; the input parameters of the prediction model not only include the current value, but also consider the rate of change, gradient trend and periodic factors to improve the stability of the prediction model under load fluctuations and sudden events.

[0029] The constructed scheduling layer is used to receive the real-time status view of the power grid and the predicted load change trend and equipment failure risk information output from the perception layer. Based on the load optimization algorithm (including genetic algorithm or particle swarm optimization algorithm) and the power grid topology adjustment strategy, a global scheduling scheme is generated. The scheduling process takes into account equipment health status, maintenance cycle, remaining life and regional load balance factors.

[0030] In one embodiment of the present invention, the scheduling layer includes: a scheduling controller with a built-in multi-objective optimization module, which uses a genetic algorithm or particle swarm optimization algorithm to perform global optimization of node scheduling under a preset cost function; and a topology reconstruction module, which dynamically generates multiple backup network topology schemes based on the operating status of virtual power plant nodes and the predicted power grid operating load, and selects the optimal topology scheme through a built-in decision tree model. The scheduling controller receives the optimal topology scheme and integrates it with the preliminary scheduling strategy to generate a node-level scheduling instruction for the globally optimal scheduling scheme. The generated scheduling instruction is sent to the response layer control interface in the form of an XML structured file for automatic execution and feedback correction.

[0031] It should be noted that, in order to achieve precise management and efficient control of complex power grid systems by the scheduling layer, the system first constructs a power grid topology based on graph theory modeling. Each node device, line, and load is modeled as a vertex and edge of the graph. Specifically, the vertex is defined as the node device (such as distributed energy equipment, transformer, switching equipment, user-side terminal, etc.) and load (user-side power load node) in the power grid. Its parameters include the rated power of the device, current voltage / current, load size, fault risk score, and operating status (normal / abnormal). The edge is defined as the line (transmission line, distribution line) connecting the vertex. Its parameters include line impedance, transmission capacity, real-time load rate, on / off status, etc. The network weight (such as load rate) and connection relationship (such as on / off status) are dynamically updated through real-time data, and the network weight and connection relationship are dynamically adjusted through real-time updated sensing data. During the scheduling optimization process, a multi-objective optimization algorithm framework is embedded within the scheduling controller. Based on genetic algorithms or particle swarm optimization algorithms, the scheduling solution space is searched, and a fitness function is introduced to comprehensively evaluate key indicators such as energy efficiency loss, equipment status score, and regional load balancing coefficient. Different objective functions are weighted and combined through weight adjustments (e.g., minimizing system losses, load balancing, and equipment risk control). Furthermore, the scheduling process also references equipment lifecycle models and maintenance plans to avoid high-risk or near-life nodes in the generated scheduling schemes, improving the overall feasibility and grid stability. During topology reconfiguration, a deep learning model analyzes topology switching strategies in historical emergency scenarios to form a rapid reconfiguration candidate set for different fault conditions, thereby obtaining multiple scheduling candidate schemes. The optimal topology scheme is selected through a decision tree model. The decision tree model's judgment conditions mainly revolve around node operating status (e.g., equipment fault risk score, remaining life, current operating parameters), predicted load (future load trends and fluctuation range), line load rate, system energy efficiency loss (e.g., line transmission loss), and regional load. The system uses core indicators such as load balancing coefficient (load difference between different zones). The selection logic is as follows: the decision tree model constructs hierarchical decision rules based on historical scheduling data and emergency scenario cases. It prioritizes "minimizing system loss," "maximum load balancing," and "lowest equipment risk" as the top-level judgment nodes, and filters backup topology schemes layer by layer. For example, it first eliminates overload risk schemes based on line load rate, then narrows down the candidate range based on load balancing coefficient, and finally determines the optimal scheme by combining equipment remaining life and fault warning score. Ultimately, it selects the topology scheme that simultaneously satisfies safety (low fault risk), economy (low loss), and stability (load balancing) under the current node state and predicted load. To improve real-time scheduling efficiency, the system is configured with a parallel solution module to perform simulation calculations and robustness evaluations on multiple scheduling topology candidate schemes. Finally, through multiple rounds of iteration, it selects the optimal scheduling scheme that meets the requirements of real-time performance, safety, and economy. The system generates scheduling instructions in XML format and transmits them to the response layer for precise execution. The XML scheduling instructions include version control, execution timestamp, and device tag fields to ensure that the response layer can quickly identify and execute them accurately.

[0032] Among them, the genetic algorithm is an improved genetic algorithm. The improvement principle revolves around the multi-constraint and multi-objective requirements of virtual power plant optimization scheduling. Addressing the shortcomings of traditional genetic algorithms in adapting to dynamic grid topology adjustments, equipment state constraints (health status, maintenance cycle, remaining lifespan), and regional load balancing, a constraint processing mechanism is introduced to filter invalid solutions that do not meet equipment capacity and load supply-demand balance. Simultaneously, a multi-objective fitness evaluation system is constructed in conjunction with a preset cost function, comprehensively considering energy efficiency loss, equipment state scores, and regional load balance coefficients to avoid scheduling imbalances caused by single-objective optimization. Furthermore, an elite retention mechanism is added to prevent the loss of high-quality scheduling solutions during iteration, improving the search efficiency of the global optimal solution. Its architecture is based on the traditional genetic algorithm's basic process of "encoding-initialization-fitness evaluation-selection-crossover-mutation-termination condition," and adds functional modules specifically for the virtual power plant scenario, including a constraint processing layer (quantitatively verifying constraints such as equipment status and maintenance cycle), a multi-objective fusion module (integrating multi-dimensional evaluation indicators based on a preset cost function), and an optimal solution screening module (consolidation layer). The system predicts load change trends and fault risks, and selects solutions that adapt to the real-time grid status from the candidate solutions generated iteratively. During data processing, it first receives multi-source data (meteorological data, grid load, equipment status) cleaned by the perception layer (Kalman filter algorithm). Distributed energy output, load allocation ratio, equipment access / exit status, and other scheduling-related parameters are encoded into real chromosomes (to adapt to the requirements of continuous scheduling variables). Then, a fitness function is constructed with a preset cost function as the core. The scheduling schemes corresponding to each chromosome are comprehensively scored. Then, high-fitness chromosomes are retained through selection operations, scheme gene recombination is achieved through crossover operations, and new search directions are introduced through mutation operations (crossover and mutation parameters are dynamically adjusted according to population diversity to avoid premature convergence). After multiple rounds of iteration, the termination conditions are met (such as the number of iterations reaching the target or the fitness value stabilizing). The optimal chromosome is decoded to generate a global scheduling scheme that includes node power allocation and grid topology adjustment (line on / off, node access / exit), ensuring that the output meets the requirements of real-time operation of the virtual power plant and subsequent resource utilization.

[0033] The constructed response layer is used to automatically execute node-level scheduling instructions based on the scheme generated by the scheduling layer, and corrects them in real time through an adaptive feedback mechanism to ensure stable system operation. In the event of sudden failures or load fluctuations, it can quickly perform emergency operations such as load redistribution, backup power switching, and fault isolation.

[0034] It should be noted that, in order to achieve rapid execution and dynamic correction of scheduling commands by the response layer as described in this embodiment, the system deploys intelligent terminal devices with local control functions at each key node, such as circuit breaker drivers, automatic reclosing controllers, and energy storage system regulators. These devices achieve millisecond-level information interaction with the scheduling controller through smart grid communication modules that support GOOSE and MMS protocols. After the scheduling plan is issued, the response layer uses a closed-loop feedback control module to continuously monitor key operating parameters such as voltage, current, and frequency, and compares them with predicted values. When a deviation exceeding the threshold is detected, the control logic is immediately triggered to start the dynamic correction process, including automatically adjusting the power output of the energy storage system, opening or cutting off some branch loads, and starting backup power to restore balance. In the event of a sudden fault, the system-integrated fault self-healing control module will quickly identify the fault area and generate an isolation path to restore power supply in the shortest time based on a preset path optimization algorithm (such as the Dijkstr algorithm or the A* algorithm). It will automatically drive the switching equipment to complete the fault area isolation and bypass power supply switching, thereby effectively preventing the fault from spreading and ensuring the stable and continuous operation of the entire power grid system.

[0035] It should be understood that the triggering conditions are mainly based on the deviation between real-time monitoring parameters and preset thresholds, the degree of deviation between predicted and actual values, and abnormal equipment status indicators. Specifically, these include: 1. Load fluctuation triggering conditions: The deviation between the actual grid load and the load trend predicted by the LSTM model for the next hour exceeds the preset threshold (e.g., ±10%), or the regional load rate exceeds the normal range (e.g., the load rate of a single region suddenly rises to over 90%); 2. Sudden fault triggering conditions: Voltage mutation (e.g., a sudden rise / fall in voltage at a node exceeds ±15% of the rated value), current imbalance (the difference between three-phase currents exceeds 20%), frequency abnormality (deviation from 50Hz ±0.5Hz), or the equipment status detector detects vibration amplitude exceeding the standard (e.g., exceeding 1.5 times the rated vibration value of the equipment), and thermal imaging data shows abnormal temperature (e.g., the transformer temperature exceeds 85℃). Example: When the actual load in a certain area increases from 500kW to 700kW within 10 minutes, exceeding the 600kW upper limit predicted by LSTM (deviation of 16.7%), load redistribution is triggered. The main load regulation unit automatically transfers 20% of the load in that area to a nearby low-load area. When the current in a certain line suddenly increases from 300A to 500A (far exceeding the rated 400A) and the three-phase current difference reaches 25%, the fault diagnosis unit identifies it as a line short-circuit fault through the SVM algorithm, immediately triggers fault isolation, automatically disconnects the isolating switches at both ends of the line, and starts the bypass power supply equipment to restore power supply.

[0036] In one embodiment of the present invention, the fault self-healing control module includes a fault diagnosis unit, a self-healing path calculation unit, and an automatic execution unit; the fault diagnosis unit identifies fault nodes based on real-time collected equipment status data, including voltage surges, current imbalances, and frequency anomalies, using a support vector machine algorithm; the self-healing path calculation unit calculates the optimal control path from the current node status to the fault recovery node based on a graph search algorithm; and the automatic execution unit controls the disconnecting switch and starts the bypass power supply equipment.

[0037] It should be noted that the fault diagnosis model described in this embodiment is deployed in edge nodes to achieve localized rapid judgment and support event source tracing analysis; the search time is greatly reduced based on the reducible graph optimization algorithm, improving the efficiency of emergency recovery decision-making; the fault isolation process supports fully automatic switching or human-machine collaborative control, and has a redundant switching scheme to ensure the safety and stability of the control process.

[0038] In one embodiment of the present invention, the virtual power plant optimized scheduling system includes machine learning models for load forecasting and equipment health assessment. The load forecasting model is constructed based on a supervised learning algorithm, with training data including electricity load and weather conditions from the past 12 months. An LSTM time series network is used to predict load trends for the next hour. The equipment fault detection model employs unsupervised K-means clustering and isolated forest algorithms, with input variables including vibration amplitude, current anomalies, and thermal imaging data. The model is used to identify potentially faulty device nodes. It is deployed on edge nodes and its weights are updated periodically by the central server.

[0039] It should be noted that the load forecasting model described in this embodiment supports a rolling update mechanism, which dynamically adjusts the training set through a sliding window to enhance the model's adaptability to new load patterns; the equipment health assessment model supports comprehensive scoring of multi-dimensional abnormal indicators and can automatically remove outliers to improve the accuracy of fault identification; the central server can remotely schedule the parameter synchronization of edge models and record the model version evolution trajectory, which is convenient for backtracking analysis and long-term stable operation.

[0040] In one embodiment of the present invention, the virtual power plant optimized dispatch system is configured with a multi-level load adjustment module, including a main load adjustment unit and a user-side terminal integrated communication module; the main load adjustment unit automatically adjusts the power allocation of different areas or terminals based on load priority settings; the user-side terminal integrated communication module receives demand response signals through the MQTT (Message Queuing Telemetry Transport) message queue telemetry transmission protocol to realize user-side load peak shaving or peak shifting control; all load adjustment data are fed back to the dispatch controller in real time to update the system load balance parameters.

[0041] It should be noted that the multi-level load adjustment strategy described in this embodiment introduces hierarchical scheduling logic, distinguishes between basic load, adjustable load and emergency load, and matches different levels of control rules respectively; the MQTT protocol supports QoS (Quality of Service) to ensure stable transmission of instructions, and is compatible with user terminals such as home energy gateways and commercial building management systems; the system continuously adjusts the load curve prediction model based on real-time feedback to make the load distribution more in line with the actual operating status.

[0042] In one embodiment of the present invention, the virtual power plant optimization scheduling system includes a renewable energy scheduling module, which collects meteorological data and inputs it into a neural network regression model to predict the output of wind power and photovoltaic power generation in the next 15 minutes, dynamically adjusts the renewable energy access ratio according to the difference between the predicted output and the scheduling load, and starts the backup power system when necessary. The module supports unified access and power regulation of multiple new energy power plants.

[0043] It should be noted that the scheduling module described in this embodiment establishes an equipment parameter library based on the power generation characteristics of each new energy power station. Dynamic factors such as wind shear and cloud cover change rate are introduced into the regression model to improve the model's sensitivity to output fluctuations. The system sets upper and lower limits for the access ratio and automatically starts backup power or energy storage system to buffer anomalies based on the load tolerance band of the scheduling layer. The module adopts a unified data access standard and can access multiple stations in parallel to achieve cross-site power complementarity and unified control.

[0044] In one embodiment of the present invention, the virtual power plant optimized scheduling system is deployed on a big data cloud platform supporting Hadoop or Spark, and has the following functions: The centralized data management module uploads the status data collected by sensors and edge nodes in each area to the HDFS storage system; The optimization analysis module uses MapReduce or Spark Streaming technology for real-time computation and outputs cross-regional scheduling optimization strategies. The generated optimization results are published to the scheduling nodes in each region through the REST API interface, realizing multi-region load coordination and unified control.

[0045] It should be noted that the cloud platform described in this embodiment adopts a data lake architecture to uniformly manage structured and unstructured data, and supports automatic stratification of cold / hot data; Spark Streaming technology enables second-level data analysis and multi-scenario concurrent scheduling simulation; the standardized design of the REST interface enables the platform to have good scalability and can be connected to third-party energy management systems or local power dispatch controllers.

[0046] In one embodiment of the present invention, the virtual power plant optimization scheduling system includes high-performance computing nodes deployed on a cloud platform or local cluster. Based on historical load data and real-time monitoring data, it calls a short-term load forecasting model and generates a load trend forecasting curve for the next 30 minutes to 2 hours through multi-threaded parallel computing. Based on the forecasting results, it calls the equipment operation database and uses a parallel health status assessment algorithm to perform life assessment and risk scoring on key node equipment. The predicted load and equipment scoring results are input into the scheduling optimization module, which generates a multi-scheme scheduling candidate set through a parallel solution strategy and selects the optimal scheduling strategy to be sent to the response layer.

[0047] It should be noted that the system described in this embodiment uses a distributed parallel computing framework (such as MPI or OpenMP) to enable concurrent operation of the prediction model and the evaluation model; the lifetime assessment introduces fault tree analysis (FTA) and accelerated aging modeling technology to improve the scoring accuracy; the scheduling candidate set is clustered and prioritized to reduce response latency, and finally the optimal scheduling scheme is selected by an intelligent selection algorithm (such as TOPSIS or AHP) and transmitted to the execution module through a standardized interface.

[0048] In summary, the virtual power plant optimization scheduling system of this invention constructs a multi-layered architecture of "perception layer—scheduling layer—response layer," combining key technologies such as edge computing, machine learning, optimization algorithms, and cloud platform computing to form a closed-loop system covering the entire process from data acquisition, status analysis, intelligent optimization to scheduling execution. The perception layer deploys multi-source sensors for environmental monitoring, power acquisition, and equipment status, and uses protocols such as RS485 or LoRa to build a local area communication network. Real-time perception data is transmitted to the edge computing module. Edge nodes integrate Kalman filtering and LSTM prediction models, performing filtering, fusion, and time series modeling to generate real-time status views and future load and fault trend predictions, achieving high-precision prediction of load trends and fault probabilities. The scheduling layer receives status and prediction information transmitted from the perception layer. Based on genetic algorithms or particle swarm optimization algorithms, it calculates the optimal scheduling scheme that balances equipment health, maintenance plans, and load balancing within the power grid structure constructed using a graph model. Standardized instructions are then generated using an XML structure and sent to the response layer. Simultaneously, this layer integrates a high-performance computing platform and machine learning modules, supporting short-term parallel load forecasting and equipment health assessment to ensure real-time, dynamic, and sustainable updates to the scheduling strategy. Upon receiving scheduling instructions, the response layer rapidly executes node-level control operations through control terminals supporting MMS and GOOSE protocols. A closed-loop feedback system is constructed to monitor the execution status in real time and automatically correct operational deviations. Through a self-healing terminal execution module, the response layer identifies fault sources based on support vector machines and generates optimal recovery paths using graph search algorithms when faults occur or loads fluctuate abnormally. In emergencies, it can quickly implement load redistribution, backup power switching, and fault isolation, effectively ensuring stable system operation and adaptive adjustment capabilities. Furthermore, the entire system architecture supports deployment on Hadoop or Spark platforms, integrating edge computing and cloud computing to achieve cross-regional data management and collaborative scheduling.

[0049] This technical solution addresses the issues of high data latency, slow response speed, and inability to achieve millisecond-level scheduling feedback in existing virtual power plant systems. By offloading real-time computing tasks to edge computing nodes, it significantly improves the system's dynamic response efficiency. Simultaneously, the predictive mechanism integrating multi-source sensing and machine learning models enhances the ability to identify and judge trends in environmental disturbances, meteorological changes, and equipment health status. This overcomes the technical shortcomings of existing systems, such as insufficient sensing dimensions, weak predictive capabilities, and inaccurate scheduling, thereby achieving more intelligent, efficient, and reliable power grid operation and scheduling management. The coordination between the scheduling controller and the topology reconfiguration module enables the system to dynamically generate optimal scheduling strategies, effectively ensuring the stable operation of the virtual power plant under various operating conditions. The construction of a closed-loop feedback and fault self-healing control module, in conjunction with the scheduling controller and topology reconfiguration module, allows the system to dynamically generate optimal scheduling strategies and automatically correct operational deviations, effectively ensuring stable operation and adaptive adjustment capabilities, guaranteeing sub-second response and stable power grid operation in emergencies. Through precise execution of node-level commands and dynamic correction, the stable operation of the virtual power plant under various operating conditions is ensured.

[0050] Example 2 This embodiment, based on Embodiment 1, discloses a virtual power plant optimized scheduling method, such as... Figure 1 As shown, the implementation based on the virtual power plant optimized scheduling system in Example 1 includes the following steps: Acquire real-time and historical data on meteorological conditions, power grid load, equipment status, and external interference. The acquired real-time data is fused and analyzed through edge node computing to generate a real-time power grid operation status view. Based on historical data, a power grid load trend prediction model is used to predict the power grid load change trend in a future set period, and an equipment failure probability model is used to predict the equipment failure risk in a future set period. Based on the real-time operation status view of the power grid, the predicted power grid load change trend, and the equipment failure risk, a preliminary scheduling strategy is generated by global optimization of the virtual power plant scheduling node based on the genetic algorithm or particle swarm optimization algorithm. During the optimization process, multiple backup network topology schemes are dynamically generated based on the operation status of the virtual power plant node and the predicted power grid operation load. The optimal topology scheme is selected by the decision tree model. The optimal topology scheme is integrated with the preliminary scheduling strategy to generate the node-level scheduling instruction of the global optimal scheduling scheme. After receiving node-level scheduling instructions from the globally optimal scheduling scheme, the cluster of intelligent terminal devices deployed in the power grid nodes automatically executes them. During the execution process, it monitors and corrects abnormal equipment operating parameters through loop feedback control. In the event of a fault or abnormal load fluctuation, it quickly performs emergency operations such as load redistribution, backup power switching, and fault isolation.

[0051] The above-disclosed embodiments are merely preferred embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A virtual power plant optimized dispatching system, characterized in that, include: The data acquisition module is used to acquire real-time and historical data on meteorological conditions, power grid load, equipment status, and external interference. The edge computing module is equipped with a data fusion unit, which is used to fuse and analyze the acquired real-time data to form a real-time operating status view of the power grid; The prediction module is used to predict the trend of power grid load changes within a future set period using a power grid load trend prediction model and to predict the risk of equipment failure within a future set period using an equipment failure probability model, based on acquired historical data. The power grid load trend prediction model is constructed based on a supervised learning algorithm, and the training data includes power grid load and meteorological data from the past 12 months. An LSTM time series network is used to predict the load trend within the next hour. The equipment failure probability model is constructed using unsupervised K-means clustering and an isolated forest algorithm. The input data for the equipment failure probability model includes vibration amplitude, current anomalies, and thermal imaging data in the equipment status. The input data is processed by the unsupervised K-means clustering algorithm to obtain normal clusters in the new equipment data. Then, based on the normal clusters of the new equipment data, the isolated forest algorithm is used to detect anomalies, determine whether the new equipment data deviates from the feature distribution corresponding to the normal clusters, output an equipment anomaly score, and identify faulty equipment nodes. The dispatch controller receives real-time operating status views of the power grid, predicted power grid load change trends, and equipment failure risks, and generates dispatch instructions. The dispatch controller has a built-in multi-objective optimization algorithm framework, which performs global optimization calculations on dispatch nodes in the virtual power plant based on genetic algorithms or particle swarm optimization algorithms, and generates preliminary dispatch strategies. The dispatch controller constructs a power grid topology model based on graph theory, models each node device and load as vertices of the graph, models lines as edges of the graph, and dynamically adjusts network weights and connection relationships based on real-time data. The topology reconfiguration module is used to dynamically generate multiple backup network topology schemes based on the operating status of virtual power plant nodes and the predicted power grid operating load, and select the optimal topology scheme through the built-in decision tree model. The scheduling controller receives the optimal topology scheme and integrates it with the preliminary scheduling strategy to generate a node-level scheduling instruction for the global optimal scheduling scheme. The intelligent terminal device cluster, including circuit breaker drivers, automatic reclosing controllers, and energy storage system regulators deployed in grid nodes, is used to receive and execute node-level scheduling instructions issued by the scheduling controller.

2. The virtual power plant optimized dispatching system according to claim 1, characterized in that, During the adjustment and optimization process, the scheduling controller combines real-time collected data on line load rate, equipment fault risk score, and regional load difference to incorporate the goals of minimizing system loss, achieving load balance, and reducing equipment risk into the weight calculation logic. It then uses a genetic algorithm or particle swarm optimization algorithm to weight these parameters and dynamically adjust the edge weight values.

3. The virtual power plant optimized scheduling system according to claim 2, characterized in that, The multi-objective optimization algorithm framework introduces a fitness function to comprehensively evaluate energy efficiency loss, equipment status score and regional load balance coefficient index, and considers equipment health status, maintenance cycle and remaining life factors in the optimization calculation process.

4. The virtual power plant optimized dispatching system according to claim 1, characterized in that, The data acquisition module includes a temperature sensor, a humidity sensor, an anemometer, a light intensity meter, a voltage transformer, a current transformer, a partial discharge detector, and an equipment status detector. A local sensor network is constructed through an RS485 or LoRa communication module, and a clock synchronization mechanism ensures the time consistency of various data acquisitions.

5. The virtual power plant optimized dispatching system according to claim 1, characterized in that, The edge computing module has a built-in preprocessing unit, which is used to filter and denoise the acquired real-time data, remove anomalies, align the time, and standardize the data.

6. The virtual power plant optimized dispatching system according to claim 1, characterized in that, The data fusion unit uses a multi-source fusion algorithm based on Kalman filtering and wavelet transform to construct a unified data fusion framework for different types of information. After fusing and analyzing meteorological data, power grid load, and equipment status, it performs correlation modeling to generate a visualized real-time operation status view.

7. The virtual power plant optimized dispatching system according to claim 1, characterized in that, It also includes a closed-loop feedback control module and a fault self-healing control module. The closed-loop feedback control module is used to continuously monitor the equipment operating parameters and dynamically compare them with the predicted values. When a deviation exceeding a preset threshold is detected, feedback correction is performed. The fault self-healing control module is used to perform emergency operations such as backup power switching and fault isolation when a fault occurs or the load fluctuates abnormally. The fault self-healing control module includes: a fault diagnosis unit, which identifies fault nodes based on real-time collected equipment status data, including voltage surges, current imbalances and frequency anomalies, using a support vector machine algorithm. The self-healing path calculation unit calculates the optimal control path from the current node state to the fault recovery node based on a graph search algorithm; the automatic execution unit is used to control the disconnecting switch and start the bypass power supply equipment.

8. A virtual power plant optimized scheduling method, implemented based on the scheduling system described in any one of claims 1 to 7, characterized in that, Includes the following steps: Acquire real-time and historical data on meteorological conditions, power grid load, equipment status, and external interference. The acquired real-time data is fused and analyzed through edge node computing to generate a real-time power grid operation status view. Based on historical data, a power grid load trend prediction model is used to predict the power grid load change trend in a future set period, and an equipment failure probability model is used to predict the equipment failure risk in a future set period. Based on the real-time operation status view of the power grid, the predicted power grid load change trend, and the equipment failure risk, a preliminary scheduling strategy is generated by global optimization of the virtual power plant scheduling node based on the genetic algorithm or particle swarm optimization algorithm. During the optimization process, multiple backup network topology schemes are dynamically generated based on the operation status of the virtual power plant node and the predicted power grid operation load. The optimal topology scheme is selected by the decision tree model. The optimal topology scheme is integrated with the preliminary scheduling strategy to generate the node-level scheduling instruction of the global optimal scheduling scheme. Clusters of intelligent terminal devices deployed in power grid nodes receive and automatically execute generated node-level scheduling instructions.

Images