Intelligent identification and recovery method and system for voltage sag

Abstract

The invention provides a voltage sag intelligent identification and recovery method and system. The method comprises the following steps: acquiring power grid data and establishing a digital twinborn model; generating simulated power grid voltage data and simulated load voltage data during operation of the power grid equipment, and identifying an abnormal operation state to obtain first abnormal data; power grid voltage data and load voltage data of the power grid equipment are collected and identified to obtain second abnormal data; comparing the first abnormal data with the second abnormal data to obtain an abnormal diagnosis result, and inputting the abnormal diagnosis result into a voltage abnormal reason analysis model to analyze a voltage sag reason; according to an analysis result, simulating different recovery schemes, and selecting an optimal scheme superior to a threshold value; and the scheme is sent to power grid equipment, so that multi-equipment cooperative control is realized, and combined control and recovery of voltage values are achieved. By adopting various technologies such as the Internet of Things and deep learning, dynamic monitoring and identification and cost optimization and recovery of voltage problems are realized, and the problem processing efficiency and success rate are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grids, and particularly to an intelligent voltage sag identification and recovery method and system. Background Art

[0002] Voltage sag is a common and significant power quality problem in distribution networks. With the increase of industrial sensitive loads, the losses caused by it are becoming more and more serious. In order to reduce such losses, it is crucial to timely identify the fault causes of voltage sags and perform voltage recovery, which will effectively guide the design of solutions. However, traditional technologies are often affected by subjective factors, resulting in problems such as inaccurate identification results of voltage sag causes and untimely recovery. Summary of the Invention

[0003] Based on the above problems, the present invention proposes an intelligent voltage sag identification and recovery method and system. By adopting various advanced technologies such as the Internet of Things and deep learning, it realizes the dynamic monitoring and identification of voltage problems and cost-optimized recovery, improving the problem-solving efficiency and success rate.

[0004] In view of this, one aspect of the present invention proposes an intelligent voltage sag identification and recovery method, including: Obtaining power grid data and establishing a power grid digital twin model of the power grid according to the power grid data; Synchronizing the power grid digital twin model with the real power grid in real time through Internet of Things technology; Based on the power grid digital twin model, using mathematical models and simulation algorithms to simulate and generate simulated power grid voltage data and simulated load voltage data when power grid equipment operates in a digital twin environment; Using artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data to determine the abnormal operation state data of power grid equipment and obtain first abnormal data; Collecting first power grid voltage data and first load voltage data of the operation state of power grid equipment, and identifying the first power grid voltage data and the first load voltage data to determine the abnormal operation state data of power grid equipment and obtain second abnormal data; Synchronously comparing and verifying the first abnormal data with the second abnormal data to obtain an abnormal diagnosis result; Inputting the abnormal diagnosis result into a voltage abnormal cause analysis model to analyze the voltage abnormal cause and obtain a voltage sag cause analysis result; According to the voltage sag cause analysis result, using a deep reinforcement learning platform to simulate different recovery schemes in a digital twin environment and select an optimal scheme with an effect and cost better than a preset threshold; Convert the optimal solution into control instructions and send them to the real power grid for multi-device collaborative control, jointly adjusting the power or switch state to achieve joint control and collaborative restoration of the voltage value.

[0005] Preferably, the step of acquiring power grid data and establishing a digital twin model of the power grid based on the power grid data includes: Extract the power grid topology structure data and historical monitoring data of each power grid device from the power grid management system; Extract the attribute data of various power grid devices and sensors from the equipment asset management system; Use Internet of Things technology to acquire the real-time operation status data of power grid devices; Through mathematical modeling and simulation technology, construct the digital twin model of the power grid according to the power grid topology structure data, the historical monitoring data, the attribute data and the real-time operation status data.

[0006] Preferably, the step of generating simulated power grid voltage data and simulated load voltage data during the operation of power grid devices in the digital twin environment based on the digital twin model of the power grid by using mathematical models and simulation algorithms includes: Establish a physical and mathematical model according to the device attributes and operation mechanism to describe the device shape and key components; According to the physical and mathematical model, use computer graphics technology to generate a 3D digital image, restore the device shape and internal structure, and establish an electrical model describing electrical properties; Based on the physical and mathematical model, use numerical simulation algorithms and electrical models to calculate the power grid voltage and load voltage during the operation of the device, and obtain simulated power grid voltage data and simulated load voltage data respectively; According to the physical changes under different operating states, dynamically adjust the digital image for restoration to obtain simulated power grid voltage data; Establish a device audio generation mechanism model according to sensor monitoring data and physical laws; Use acoustic model algorithms to perform numerical simulation calculations on complex acoustic wave interferences under different states, and generate multi-dimensional simulated load voltage data that restores the operating sound quality.

[0007] Preferably, the step of using artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data, determine the abnormal data of the operation state of power grid devices, and obtain the first abnormal data includes: Collect normal and abnormal power grid voltage data as the first training set, and collect normal and abnormal load voltage data as the second training set to construct a power grid voltage recognition model and a load voltage recognition model respectively; Preprocess the simulated grid voltage data and the simulated load voltage data respectively, and extract the grid voltage characteristics and the load voltage characteristics; Input the grid voltage characteristics and the load voltage characteristics into the trained grid voltage recognition model and the load voltage recognition model respectively; Take the abnormal data recognized by the grid voltage recognition model and the load voltage recognition model as the first abnormal data.

[0008] Preferably, the step of collecting the first grid voltage data and the first load voltage data of the operating state of the grid equipment, identifying the first grid voltage data and the first load voltage data, and determining the abnormal data of the operating state of the grid equipment to obtain the second abnormal data includes: Extract the first grid voltage characteristics from the first grid voltage data; Extract the first load voltage characteristics from the first load voltage data; Input the first grid voltage characteristics and the first load voltage characteristics into the grid voltage recognition model and the load voltage recognition model respectively to detect abnormalities, and obtain the second abnormal data.

[0009] Preferably, the step of synchronously comparing and verifying the first abnormal data and the second abnormal data to obtain the abnormal diagnosis result includes: Label the time and position tags of the first abnormal data and the second abnormal data, and store them in the distributed database; Extract the first abnormal sub-data and the second abnormal sub-data of the same device in the same time period according to the time and position tags; Compare the abnormal types, positions and coincidence degrees of the first abnormal sub-data and the second abnormal sub-data; If the types are the same and the position coincidence degree is high, it is determined that the detection results are consistent; Otherwise, analyze the difference points between the first abnormal sub-data and the second abnormal sub-data, and adjust the model with the worse performance; Take the consistent result directly as the final abnormal diagnosis result.

[0010] Preferably, the step of inputting the abnormal diagnosis result into the voltage abnormal cause analysis model to analyze the voltage abnormal cause and obtain the voltage sag cause analysis result includes: Collect historical voltage abnormal case data and determine the historical voltage sag cause; Construct labels for the historical voltage abnormal case data and the historical voltage sag cause and use them as the voltage sag training data; Use the neural network structure to establish a deep learning analysis model; Training the deep learning analysis model with the voltage sag training data to learn the anomaly-cause mapping relationship, and obtaining the voltage anomaly cause analysis model; Extracting features from the anomaly diagnosis results as input parameters and importing them into the voltage anomaly cause analysis model for voltage anomaly cause analysis and identification, so as to obtain the voltage sag cause analysis results.

[0011] Preferably, the step of simulating different recovery schemes in the digital twin environment using the deep reinforcement learning platform according to the voltage sag cause analysis results and selecting the optimal scheme with better effect and cost than the preset threshold includes: Generating several alternative voltage recovery schemes according to the voltage sag cause analysis results; Using the power grid digital twin model to simulate the execution of each voltage recovery scheme; Defining a reward function, and the recovery scheme in which the voltage is restored within the preset voltage threshold within the preset time range and the recovery cost is lower than the preset cost threshold will be rewarded; Obtaining historical voltage recovery schemes, using the historical voltage recovery schemes as actions, and establishing a deep reinforcement learning platform in combination with the reward function; Inputting the voltage recovery scheme into the deep reinforcement learning platform to obtain the scheme reward data of each voltage recovery scheme; Simulating each voltage recovery scheme in the digital twin environment according to the power grid digital twin model, the voltage recovery scheme and the scheme reward data, and identifying the scheme with the highest reward value as the optimal scheme.

[0012] Preferably, the step of converting the optimal scheme into a control instruction and sending it to the real power grid for multi-device collaborative control, jointly adjusting the power or switch state to achieve joint control and collaborative recovery of the voltage value includes: Refining the optimal scheme into specific control operations for each involved device; Serializing the specific control operations into standard control instructions; Sending the standard control instructions to each target control device through the Internet of Things platform; Each target control device performs joint operations of adjusting the power or switching the switch state according to the standard control instructions; The target control devices perform data synchronization through timed communication and adjust the operations to constrain the voltage value within the target range.

[0013] Another aspect of the present invention provides a voltage sag intelligent identification and recovery system, which is used to execute a voltage sag intelligent identification and recovery method, including: power grid equipment; cloud platform; Internet of Things platform; wherein, The cloud platform is used for: Obtain power grid data and establish a digital twin model of the power grid based on the power grid data; Synchronize the digital twin model of the power grid with the real power grid in real time through Internet of Things technology; Based on the digital twin model of the power grid, use mathematical models and simulation algorithms to simulate and generate simulated power grid voltage data and simulated load voltage data during the operation of power grid equipment in the digital twin environment; Use artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data, determine the abnormal operation state data of power grid equipment, and obtain the first abnormal data; Collect the first power grid voltage data and the first load voltage data of the operation state of power grid equipment, and identify the first power grid voltage data and the first load voltage data to determine the abnormal operation state data of power grid equipment, and obtain the second abnormal data; Synchronously compare and verify the first abnormal data with the second abnormal data to obtain an abnormal diagnosis result; Input the abnormal diagnosis result into a voltage abnormal cause analysis model for voltage abnormal cause analysis to obtain a voltage sag cause analysis result; According to the voltage sag cause analysis result, use a deep reinforcement learning platform to simulate different recovery schemes in the digital twin environment, and select the optimal scheme with better effect and cost than the preset threshold; Convert the optimal scheme into a control instruction and send it to the real power grid for multi-device collaborative control, jointly adjust the power or switch state, and achieve joint control and collaborative recovery of the voltage value.

[0014] Adopting the technical solution of the present invention, the intelligent voltage sag identification and recovery method includes: acquiring power grid data and establishing a digital twin model of the power grid based on the power grid data; synchronizing the digital twin model of the power grid with the real power grid in real time through Internet of Things technology; based on the digital twin model of the power grid, using mathematical models and simulation algorithms to simulate and generate simulated power grid voltage data and simulated load voltage data during the operation of power grid equipment in the digital twin environment; using artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data to determine the abnormal operation state data of power grid equipment and obtain the first abnormal data; collecting the first power grid voltage data and the first load voltage data of the operation state of power grid equipment, and identifying the first power grid voltage data and the first load voltage data to determine the abnormal operation state data of power grid equipment and obtain the second abnormal data; synchronously comparing and verifying the first abnormal data with the second abnormal data to obtain an abnormal diagnosis result; inputting the abnormal diagnosis result into a voltage abnormal cause analysis model to analyze the cause of voltage abnormality and obtain a voltage sag cause analysis result; according to the voltage sag cause analysis result, using a deep reinforcement learning platform to simulate different recovery schemes in the digital twin environment and select the optimal scheme whose effect and cost are better than the preset threshold; converting the optimal scheme into a control instruction and sending it to the real power grid for multi-device collaborative control, jointly adjusting the power or switch state to achieve joint control and collaborative recovery of the voltage value. By adopting a variety of advanced technologies such as the Internet of Things and deep learning, dynamic monitoring and identification of voltage problems and cost-optimized recovery are realized, and the problem handling efficiency and success rate are improved. Description of the Drawings

[0015] Figure 1 is a flowchart of the intelligent voltage sag identification and recovery method provided by an embodiment of the present invention; Figure 2 is a schematic block diagram of the intelligent voltage sag identification and recovery system provided by an embodiment of the present invention. Detailed Embodiments

[0016] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0017] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0018] The terms "first", "second", etc. in the description and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products, or devices.

[0019] Reference to "embodiment" herein means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of this application. The phrase may not necessarily refer to the same embodiment when it appears in various places in the specification, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0020] The following will be described with reference to Figures 1 to 2 a voltage sag intelligent identification and recovery method and system provided according to some embodiments of the present invention.

[0021] As Figure 1 shown, an embodiment of the present invention provides a voltage sag intelligent identification and recovery method, including: Obtain power grid data (including the topological structure data of the power grid, historical operation and maintenance data, and the attribute data, structure data, function data, etc. of each power grid device, each sensor / monitoring terminal, etc.), and establish a power grid digital twin model of the power grid according to the power grid data; Synchronize the power grid digital twin model with the real power grid in real time through Internet of Things technology; In this step, IoT devices are deployed on real power grid devices and control systems to collect various types of monitoring data in real time; low-power communication technologies such as LoRa and NB-IoT are used to upload the monitoring data to the Internet of Things platform; the Internet of Things platform performs primary processing on the data through methods such as edge computing and fog computing to screen important data; the processed data is synchronously sent to the digital twin model system in real time; the digital twin model system updates the parameter settings of the model in real time according to the received data; so that the power grid digital twin model can immediately reflect the changes in the operating state of the real power grid; on the basis of ensuring synchronization, the digital twin model system can also send the prediction results to the control system; through closed-loop synchronization management, "real" and "virtual" are seamlessly coordinated; the monitoring level of power grid operation is improved, which is conducive to predicting and responding to faults in advance, reducing the risk of power outages, and is also more conducive to fault optimization and new device evaluation tests in the digital space, and can effectively realize the near-real-time combination of the digital twin model and the real scenario.

[0022] Based on the power grid digital twin model, using mathematical models and simulation algorithms, simulate and generate simulated power grid voltage data and simulated load voltage data during the operation of power grid equipment in the digital twin environment; Use artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data, determine the abnormal data of the operation state of power grid equipment, and obtain the first abnormal data; Collect the first power grid voltage data and the first load voltage data of the operation state of power grid equipment, and identify the first power grid voltage data and the first load voltage data to determine the abnormal data of the operation state of power grid equipment, and obtain the second abnormal data; Synchronously compare and verify the first abnormal data with the second abnormal data to obtain an abnormal diagnosis result; Input the abnormal diagnosis result into the voltage abnormal cause analysis model for voltage abnormal cause analysis to obtain the voltage sag cause analysis result; According to the voltage sag cause analysis result, use the deep reinforcement learning platform to simulate different recovery schemes in the digital twin environment, and select the optimal scheme with better effect and cost than the preset threshold; Convert the optimal scheme into a control instruction and send it to the real power grid for multi-device collaborative control, jointly adjust the power or switch state, and achieve joint control and collaborative recovery of the voltage value.

[0023] Adopting the technical solution of this embodiment, through the use of various advanced technologies such as the Internet of Things and deep learning, realize the dynamic monitoring and identification of voltage problems and cost-optimized recovery, and improve the problem handling efficiency and success rate.

[0024] In some possible implementation manners of the present invention, the step of obtaining power grid data and establishing a power grid digital twin model of the power grid according to the power grid data includes: Extract the power grid topology structure data (including connection relationship data of lines, substations, switches, etc.) and historical monitoring data of each power grid equipment (such as historical operation and maintenance data such as operation parameters, fault records, etc.) from the power grid management system; Extract the attribute data of various power grid equipment and sensors (such as structure data such as type, specification, performance parameters, 3D images, etc.) from the equipment asset management system; Use the Internet of Things technology to obtain the real-time operation state data of power grid equipment; Through mathematical modeling and simulation technology, construct the power grid digital twin model according to the power grid topology structure data, the historical monitoring data, the attribute data, and the real-time operation state data.

[0025] In this embodiment, a network model is established based on the power grid topology data, and each node and connection relationship are defined; digital models of various devices are established according to the device attribute data, and parameters and operating principles are defined; using physical laws such as power electronics and magnetic fields, model equations for describing the operation of the devices are established; the numerical model parameters of each device are trained and verified according to historical monitoring / operation data; numerical analysis and solution of the network are carried out using methods such as finite element; the values of physical quantities in the model are updated and iteratively calculated online according to the real-time operation state data; real-time tracking simulation calculation of physical quantities such as current, voltage, and electric power in the virtual power grid is realized; digital reproduction and analysis of abnormal events can be carried out to test the effects of various operation strategies; operation data is collected for a long time to optimize and enrich the model and improve the simulation authenticity; the digital twin model can quantitatively restore the real power grid operation state, provide digital support for analysis and optimization, and improve the operation management level.

[0026] In this embodiment, the power grid digital twin model can be refreshed in real time to restore the detailed operation state of the real power grid; faults can be simulated in the model environment and solutions can be evaluated to reduce the number of on-site commissioning times; it can also be used for the training and testing of artificial intelligence algorithms, which is beneficial to improving the prediction and diagnosis capabilities; the "real-virtual" closed-loop management of the power grid is realized, the operation monitoring level and solution efficiency are improved, and it is beneficial to improving the intelligent level of power grid operation management.

[0027] In some possible implementation manners of the present invention, the step of simulating and generating simulated power grid voltage data and simulated load voltage data during the operation of power grid devices in the digital twin environment based on the power grid digital twin model includes: Establish a physical and mathematical model according to the device attributes and operating mechanism to describe the device appearance and key components; In this step, the device attributes mainly refer to the basic specification parameters of the device, such as type, size, structural material, etc.; the operating mechanism refers to the device working principle, such as the electromagnetic conversion mechanism of the transformer, the opening and closing electromechanics of the switch, etc.; these attributes and mechanisms are abstracted into physical quantities, such as current, voltage, current density, etc.; according to physical laws such as electromagnetism and thermodynamics, mathematical model equations describing the relationships between these physical quantities are established, such as establishing an electromagnetic field equation set for the transformer to describe the magnetic field distribution and establishing a heat conduction equation for the switch to describe the temperature distribution; these equations describe and quantify the external shape structure and key structural components of the device in a mathematical form, such as the coil structure of the transformer is described by a coordinate equation, and the geometric structure of each part of the switch is described by a finite element topological relationship; by establishing a physical and mathematical model based on the device attributes and operating mechanism, the external shape structure characteristics of the device can be quantitatively and digitally described. This is very important for subsequent digital twin simulation.

[0028] According to the physical and mathematical model, use computer graphics technology to generate 3D digital images, restore the external shape and internal structure of the device, and establish an electrical model describing electrical properties; In this step, the physical and mathematical model quantitatively describes the relationships of physical quantities such as the shape and structure of each component of the device in the form of coordinates / equations, etc.; using computer graphics technology, such as 3D modeling software, digitize and restore these quantity relationships into three-dimensional coordinate information; represent each surface of the device's external shape with three-dimensional objects to form a shell structure; construct the internal key components three-dimensionally with similar block or curved surface objects according to the model equations. For example, the coil structure of the transformer automatically generates a curved surface structure according to the coordinate equation, and each part of the switch automatically generates three-dimensional parts that fit the real shape according to the finite element relationship; different components are distinguished by different colors or materials to obtain a three-dimensional digital image model of the real device that can be observed from various perspectives. Driven by the physical and mathematical model, use computer graphics technology to generate a three-dimensional digital image that restores the real external and internal structure of the device.

[0029] Based on the physical and mathematical model, use numerical simulation algorithms and the electrical model to calculate the grid voltage and load voltage during the operation of the device, and obtain simulated grid voltage data and simulated load voltage data respectively; In this step, perform simulations according to physical changes / electrical property changes (such as voltage changes, current changes, etc.) under different operating states to obtain simulated grid voltage data and simulated load voltage data; In this embodiment, the simulated data matches the real data, which can be used for AI training and the development and testing of recognition algorithms, reduces the difficulty of on-site testing, is beneficial to early research and development and equipment optimization; can efficiently use digital twin technology for equipment research and development analysis, and improve the intelligent monitoring level.

[0030] In some possible implementation manners of the present invention, the step of using artificial intelligence technology to analyze and process the generated simulated grid voltage data and simulated load voltage data to determine the abnormal data of the operating state of the grid equipment and obtain the first abnormal data includes: Collect normal and abnormal grid voltage data as the first training set, and collect normal and abnormal load voltage data as the second training set to construct a grid voltage recognition model and a load voltage recognition model respectively (the model will judge whether the data belongs to the normal class seen in the training set or which specific type of abnormality it is); In this step, a large amount of grid voltage data under normal and different types of abnormal states of real devices is collected as the first training set; real load voltage data corresponding to normal and abnormal operations is collected as the second training set; signal processing methods are used to extract features from these voltage data; the feature data is input into the network to train the model parameters and learn the classification mapping relationship; for new data, repeat the above feature extraction and input the model for classification. The model will determine whether the data features are in the learned normal classes, otherwise, it will determine the type of abnormality; unsupervised learning methods such as clustering can also be used to divide abnormal data into different new classes; as the data continues to increase, iteratively train and optimize the model to improve the recognition accuracy. This method can achieve efficient and automatic identification of device operating states using deep learning.

[0031] Preprocess the simulated grid voltage data and the simulated load voltage data respectively to extract grid voltage features and load voltage features. In this step, standardize and cluster the features to reduce the dimension. Extract the effective information in the voltage data to help subsequent algorithms better learn the mapping relationship and improve the recognition effect; feature engineering has a great impact on the quality of model training.

[0032] Input the grid voltage features and the load voltage features into the trained grid voltage recognition model and load voltage recognition model respectively. In this step, the grid voltage features and load voltage features produced during the feature extraction process are digitally represented; the trained grid voltage recognition model and load voltage recognition model have already learned the mapping relationship between features and classes internally; each model outputs a determination result by comparing the input features with the internal learned mapping: if it is a known class, it outputs the class, otherwise, it outputs the abnormality and the type of abnormality, etc. This step is a process of automatically classifying and identifying the extracted digital features by the trained model.

[0033] Take the abnormal data identified by the grid voltage recognition model and the load voltage recognition model as the first abnormal data.

[0034] In this embodiment, a threshold can be set, and data with a severe degree is judged as the first abnormal data; unsupervised learning can also be used to directly cluster abnormal data into different types; the abnormal recognition process is in the digital space, and can be repeatedly tested with high accuracy and low error; the recognition results are helpful for optimizing device design and operation management; the recognition ability is continuously enhanced after supplementing real data; after identifying the abnormality, the cause of the fault can be further analyzed and located; using AI technology, sensitive and reliable device anomaly monitoring can be carried out in the digital twin simulation environment.

[0035] In some possible embodiments of the present invention, the steps of collecting first grid voltage data and first load voltage data of the operating state of grid equipment, identifying the first grid voltage data and the first load voltage data, determining abnormal data of the operating state of grid equipment, and obtaining second abnormal data include: Extracting first grid voltage features from the first grid voltage data; Extracting first load voltage features from the first load voltage data; Inputting the first grid voltage features and the first load voltage features into a grid voltage identification model and a load voltage identification model respectively to detect abnormalities, and obtaining second abnormal data.

[0036] In this step, the first grid voltage features and the first load voltage features are the feature data obtained after the foregoing preprocessing; these feature data are input into the already trained grid voltage identification model and load voltage identification model respectively; through learning existing normal and abnormal samples inside these two models, the classification criteria are mastered; the models make judgments: if the input features match well with the known normal samples (exceeding the preset matching value), it is judged as normal; if the matching degree is low or it matches well with the abnormal samples, it indicates abnormal data.

[0037] In this embodiment, through the identification of the grid voltage identification model and the load voltage identification model, obvious abnormalities are output as second abnormal data and an alarm is sent; monitoring multiple information sources improves the accuracy and reduces false alarms; realizing real-time on-site monitoring is beneficial to event response and on-site diagnosis; continuously optimizing the AI model as the data increases further improves the intelligent monitoring level; multi-sensor fusion monitoring can improve the reliability and provide support for fault diagnosis.

[0038] In some possible embodiments of the present invention, the steps of synchronously comparing and verifying the first abnormal data and the second abnormal data to obtain an abnormal diagnosis result include: Labeling time and position tags for the first abnormal data and the second abnormal data, and storing them in a distributed database; In this step, define the standard data structure for abnormal data, including fields such as grid voltage data / load voltage data itself, timestamp, and location coordinates; parse the first abnormal data and the second abnormal data into the temporary database in the standard format, and mark the collection time and location; design a distributed database architecture (such as a master-slave replication architecture based on MySQL or a MongoDB cluster); develop a data synchronization tool to regularly extract newly marked data from the temporary database and synchronize it to the master node of the distributed database; synchronize the master node data to multiple slave nodes for cluster backup; develop an API to perform read and write operations on the distributed database: the query interface supports filtering abnormal data by time and location; the write interface supports writing real-time collected data into the database; use a message middleware (such as Kafka) to transmit the collected data to ensure data consistency; dynamically expand the slave nodes to achieve high-availability storage of massive data; implement the annotation storage and efficient query of abnormal data to provide support for subsequent analysis.

[0039] Extract the first abnormal sub-data and the second abnormal sub-data of the same device within the same time period according to the time and location tags; In this step, define the query parameters, including device ID, time range conditions, etc.; call the distributed database API interface to execute the filtered query according to the conditions; the database query engine matches the tag fields with the conditions to filter out the matching data; group the data according to the device ID, and then sort the data within each group according to the timestamp; judge whether the collection time of each piece of data is within the query time range; then, according to the data source (the first abnormal data or the second abnormal data), save the matching data respectively; the data can be directly saved as a list / dataset, etc. in memory; perform flexible processing on the filtered sub-dataset: count the number of records, serialize and transmit it to other server-side analysis, perform online analysis and calculation based on spark / flink, etc. This can specifically extract the multi-round abnormal sub-datasets of a specific device within a period of time, providing data support for subsequent research.

[0040] Compare the abnormal types, locations, and degrees of conformity of the first abnormal sub-data and the second abnormal sub-data; In this step, compare whether the abnormal categories identified for the first abnormal sub-data and the second abnormal sub-data are consistent; compare whether the locations where the abnormalities occur match, such as whether they are in the same area / device; if the types are the same and the locations are the same, the degree of conformity is high; if the types are different but the locations are close, the degree of conformity is medium; if the types and locations are completely inconsistent, the degree of conformity is low. The specific operation is as follows: extract the abnormal type and location tag field values of each sub-dataset; count the number of records or proportions of the same type, the same location, and different locations; judge the consistency degree of the two-round detection results according to the statistical results; the judgment results are helpful to evaluate the stability performance of the recognition model. By comparing and analyzing the detection results, evaluate the consistency of the abnormal detection process and the model recognition ability.

[0041] If the types are the same and the degree of position overlap is high, it is determined that the detection results are the same; Otherwise, analyze the difference points between the first abnormal sub-data and the second abnormal sub-data, and adjust the model of the poorer side; In this step, when the degree of conformity between the first abnormal sub-data and the second abnormal sub-data is low (lower than the preset value), further compare the difference points: check the specific differences in the type and position tag values; analyze the possible reasons for the differences, such as feature extraction methods, classification standard settings, etc.; take a certain number of samples for the difference points, manually check and label the true types; feedback the true type results to the original model training process: if the grid voltage recognition result is poor, add the difference samples to retrain the grid voltage recognition model; if the load voltage recognition result is poor, add it to the load voltage recognition model for continuous training; it may be necessary to adjust the model structure, optimize the algorithm and other means to improve the accuracy, re-run the recognition process, and evaluate whether the adjusted model results are improved. Through iterative improvement, the consistency of the recognition results at the difference points is continuously improved to enhance the overall accurate performance of abnormal detection and recognition.

[0042] Directly use the consistent results as the final abnormal diagnosis results (this diagnosis result includes location information at the same time, which is beneficial for on-site maintenance).

[0043] The solution of this embodiment can improve the monitoring accuracy rate, avoid misjudgment or missed judgment; refine the two detection processes through programming for the differences, and collect feedback for a long time to continuously optimize the model; achieve multi-dimensional and comprehensive monitoring, and provide a reliable reference for the intelligent diagnosis system; so synchronous comparison can improve the credibility of the results and bring support for fault detection.

[0044] In some possible implementation manners of the present invention, the step of inputting the abnormal diagnosis result into a voltage abnormal cause analysis model to perform voltage abnormal cause analysis and obtain a voltage sag cause analysis result includes: Collect historical voltage abnormal case data and determine the historical voltage sag causes; Construct tags for the historical voltage abnormal case data and the historical voltage sag causes and use them as voltage sag training data; In this step, the historical voltage anomaly case data is labeled to indicate the true anomaly causes of each case (such as voltage sag cases caused by gas lightning, line short circuits, etc.), and a set of labeled tags for different cause types is compiled; according to the labeling standard, each anomaly case data is matched with the tag corresponding to the cause; in this way, a paired training data set is formed: the input is the data of historical different voltage anomaly cases, and the output is the tag indicating the voltage anomaly cause for each case. These training acquisition data with correct answers can be used to build a model for identifying the causes of voltage anomalies; increasing the accuracy of training samples through manual labeling is beneficial to subsequent modeling work.

[0045] Establish a deep learning analysis model using a neural network structure; Use the voltage sag training data to train the deep learning analysis model to learn the anomaly-cause mapping relationship, and obtain the voltage anomaly cause analysis model; In this step, first, preprocess the voltage sag training data (such as regularizing and standardizing the data format), convert the input features and output labels into a format that can be recognized by the neural network (such as digital vectors); select an appropriate deep learning architecture (such as convolutional neural network or recurrent network, etc.); initialize the network parameters, define the optimizer and loss function; stream the training data batches into the network, continuously perform forward propagation to calculate the output; perform backpropagation to calculate the parameter gradients, and use the optimizer to continuously adjust the network parameters; iterate the training until the loss value converges or reaches the maximum number of epochs; construct a test set, evaluate the model performance on the test set, and perform hyperparameter tuning if necessary; save the trained analysis model for subsequent use; receive new voltage anomaly data as input, and the model outputs the predicted cause tag; adopt the end-to-end method of deep learning to automatically learn the mapping relationship between anomalies and causes.

[0046] Extract features from the anomaly diagnosis results as input parameters (such as equipment error categories, operating parameters, locations, etc.) and import them into the voltage anomaly cause analysis model for voltage anomaly cause analysis and identification, and obtain the voltage sag cause analysis result.

[0047] In this embodiment, the confidence level of each possible cause can be given, and the one with the highest confidence is selected; the prediction result considers multiple factors and is more comprehensive than expert experience; as the data increases, the model's learning ability continuously improves; cause identification helps to quickly locate the problem and take remedial measures promptly; realizing intelligent power grid maintenance management; therefore, this method can analyze the causes of voltage problems with high accuracy and efficiency, providing a basis for disease diagnosis.

[0048] In some possible implementation manners of the present invention, the step of simulating different recovery schemes in the digital twin environment using a deep reinforcement learning platform according to the voltage sag cause analysis result and selecting the optimal scheme with an effect and cost better than a preset threshold includes: Generate several alternative voltage restoration plans according to the analysis results of the voltage sag causes; Use the digital twin model of the power grid to simulate and execute each of the voltage restoration plans; Define a reward function. A restoration plan that restores the voltage within a preset time range to within a preset voltage threshold and has a restoration cost lower than a preset cost threshold (the cost includes equipment operation cost, user inconvenience cost, etc.) will receive a reward; Obtain historical voltage restoration plans, use the historical voltage restoration plans as actions, and establish a deep reinforcement learning platform in combination with the reward function; In this step, Deep Reinforcement Learning is a technology that uses deep learning methods to solve reinforcement learning problems; an Agent is the entity that makes decisions and learns in the environment, and the Environment refers to the state space composed of all possible state situations that may occur during the decision-making process / problem-solving process; at each moment, the Agent will take corresponding actions according to the environmental state; the Environment will generate the next state based on the action and provide a reward signal to the Agent; the Agent learns by continuous trial and error to find a decision-making strategy that can maximize the cumulative reward (i.e., the sequence of plans with the maximum utility); in the reinforcement learning framework, the Agent uses actions to interact with the Environment to maximize the cumulative reward to the greatest extent. That is, using deep reinforcement learning technology, a series of plan actions are optimally selected in the problem environment described as the state space to solve this task. The specific implementation method is: define the environmental state space (the state set composed of conditions such as the voltage levels and loads of each node in the power grid); define the action space (various voltage restoration plans, such as increasing the number of generators in operation, expanding the capacity of transformers, etc.); set the Agent (deep reinforcement learning model (such as DQN), which selects actions based on the input state); define the reward function (evaluate the reward of the measure action according to the power grid target indicators (meeting the load, voltage level, etc.); specifically as in the previous step); simulate the interaction between the Agent and the Environment (the Agent selects an action, and the Environment updates the state and gives a reward); the Agent uses a deep neural network to fit the Q function and optimizes the parameters using a reinforcement learning algorithm; after sufficient training, the Agent can quickly select a high-utility action plan according to any state; combine actions with the ultimate goal of restoring the voltage; continue to learn and optimize operations in new situations; regularly iterate to achieve automatic optimal action selection and establish a dynamically balanced deep reinforcement learning scheduling platform.

[0049] Input the voltage restoration plans into the deep reinforcement learning platform to obtain the plan reward data for each of the voltage restoration plans; Simulate each of the voltage recovery schemes in the digital twin environment based on the power grid digital twin model, the voltage recovery scheme, and the scheme reward data, and identify the scheme with the highest reward value as the optimal scheme.

[0050] In this embodiment, through iterative training, the deep reinforcement learning scheduling platform can learn a more efficient control policy; in the digital twin space, it can select the one with the highest action score in real time; ensure that the scheme with high effect and low cost is used as the optimal scheme; feedback it to the real power grid for execution to complete actual control; improve the reward function to continuously optimize the training effect of the platform; and use digital evaluation to quickly screen practical and feasible optimization schemes in the virtual space while taking into account multi-objective requirements.

[0051] In some possible implementation manners of the present invention, the step of converting the optimal scheme into a control instruction and sending it to the real power grid for multi-device collaborative control, jointly adjusting the power or switch state, and realizing the joint control and collaborative recovery of the voltage value includes: Refine the optimal scheme into specific control operations for each involved device (such as time points, adjustment amounts, etc.); Serialize the specific control operations into standard control instructions; Through the Internet of Things platform, send the standard control instructions to each target control device; Each of the target control devices performs a joint operation of adjusting the power or switching the switch state according to the standard control instructions; Data synchronization is performed between the target control devices through timed communication, and the operations are adjusted to constrain the voltage value within the target range.

[0052] In this embodiment, it is also possible to simultaneously monitor whether the control cost is better than the setting. For example, if there is an over-adjustment part, it can be corrected according to new data, and a closed-loop joint control is formed after correction. This embodiment can effectively rectify power grid faults, ensure power supply quality and safety; and realize the self-healing of intelligent power grid faults through real-time collaborative multi-device control, improving power supply capacity and efficiency.

[0053] Please refer to Figure 2 , another embodiment of the present invention provides a voltage sag intelligent identification and recovery system, which is used to execute the voltage sag intelligent identification and recovery method as described above, including: power grid equipment; a cloud platform; an Internet of Things platform; wherein, The cloud platform is used for: Obtain power grid data, and establish a power grid digital twin model of the power grid according to the power grid data; Synchronize the power grid digital twin model with the real power grid in real time through Internet of Things technology; Based on the power grid digital twin model, using mathematical models and simulation algorithms, simulate and generate simulated power grid voltage data and simulated load voltage data during the operation of power grid equipment in the digital twin environment; Use artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data, determine the abnormal data of the operation state of power grid equipment, and obtain the first abnormal data; Collect the first power grid voltage data and the first load voltage data of the operation state of power grid equipment, and identify the first power grid voltage data and the first load voltage data to determine the abnormal data of the operation state of power grid equipment, and obtain the second abnormal data; Synchronously compare and verify the first abnormal data and the second abnormal data to obtain an abnormal diagnosis result; Input the abnormal diagnosis result into a voltage abnormal cause analysis model to analyze the cause of voltage abnormality and obtain a voltage sag cause analysis result; According to the voltage sag cause analysis result, use a deep reinforcement learning platform to simulate different recovery schemes in the digital twin environment, and select the optimal scheme whose effect and cost are better than the preset threshold; Convert the optimal scheme into a control instruction and send it to the real power grid for multi-device collaborative control, jointly adjust the power or switch state, and achieve joint control and collaborative recovery of the voltage value.

[0054] It should be known that Figure 2 The block diagram of the voltage sag intelligent identification and recovery system shown is only for illustration, and the number of each module shown does not limit the protection scope of the present invention. The voltage sag intelligent identification and recovery system provided in this embodiment can be used to execute the various embodiment schemes of the corresponding voltage sag intelligent identification and recovery method. For the specific implementation process, please refer to the description of each method embodiment, which will not be elaborated here.

[0055] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0056] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0057] In several embodiments provided in this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the above division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical or other forms.

[0058] The units described above as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0059] In addition, in each embodiment of this application, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0060] If the above integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the above methods in each embodiment of this application. And the aforementioned memory includes: USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical disks and other media that can store program codes.

[0061] Those of ordinary skill in the art can understand that all or part of the steps in the above various methods can be completed by instructing relevant hardware through a program. This program can be stored in a computer-readable memory. The memory can include: flash drives, read-only memories (abbreviation: ROM), random access memories (abbreviation: RAM), magnetic disks, or optical disks, etc.

[0062] The above has introduced the embodiments of the present application in detail. Specific examples are used herein to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

[0063] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions without departing from the spirit and scope of the present invention, and various modifications and alterations can be made, including combinations of the above different functions and implementation steps, including software and hardware implementation manners, all within the protection scope of the present invention.

Claims

1. An intelligent voltage sag identification and recovery method, characterized in that Including: Obtain power grid data and establish a digital twin model of the power grid based on the power grid data; Synchronize the digital twin model of the power grid with the real power grid in real time through Internet of Things technology; Based on the digital twin model of the power grid, use mathematical models and simulation algorithms to simulate and generate simulated power grid voltage data and simulated load voltage data when the power grid equipment is operating in the digital twin environment; Use artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data to determine the abnormal operation state data of the power grid equipment and obtain the first abnormal data; Collect the first power grid voltage data and the first load voltage data of the operation state of the power grid equipment, and identify the first power grid voltage data and the first load voltage data to determine the abnormal operation state data of the power grid equipment and obtain the second abnormal data; Synchronously compare and verify the first abnormal data with the second abnormal data to obtain an abnormal diagnosis result; Input the abnormal diagnosis result into a voltage abnormal cause analysis model to analyze the cause of voltage abnormality and obtain a voltage sag cause analysis result; According to the voltage sag cause analysis result, use a deep reinforcement learning platform to simulate different recovery schemes in the digital twin environment and select the optimal scheme with an effect and cost better than a preset threshold; Convert the optimal scheme into a control instruction and send it to the real power grid for multi-device collaborative control, jointly adjust the power or switch state, and achieve joint control and collaborative recovery of the voltage value.

2. The voltage sag intelligent recognition and recovery method according to claim 1, characterized in that The steps of obtaining power grid data and establishing a digital twin model of the power grid based on the power grid data include: Extract the power grid topology structure data and the historical monitoring data of each power grid device from the power grid management system; Extract the attribute data of various power grid devices and sensors from the equipment asset management system; Use Internet of Things technology to obtain the real-time operation state data of the power grid equipment; Through mathematical modeling and simulation technology, construct the digital twin model of the power grid according to the power grid topology structure data, the historical monitoring data, the attribute data, and the real-time operation state data.

3. The voltage sag intelligent identification and recovery method according to claim 2, characterized in that The steps of, based on the digital twin model of the power grid, using mathematical models and simulation algorithms to simulate and generate simulated power grid voltage data and simulated load voltage data when the power grid equipment is operating in the digital twin environment include: Establish a physical mathematical model according to the equipment attributes and operation mechanism to describe the equipment appearance and key components; According to the physical mathematical model, use computer graphics technology to generate a 3D digital image, restore the equipment appearance and internal structure, and establish an electrical model describing electrical properties; Based on the physical mathematical model, use numerical simulation algorithms and electrical models to calculate the power grid voltage and load voltage when the equipment is operating, and respectively obtain simulated power grid voltage data and simulated load voltage data.

4. The voltage sag intelligent identification and restoration method according to claim 3, characterized in that The steps of using artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data to determine the abnormal operation state data of the power grid equipment and obtain the first abnormal data include: Collect normal and abnormal grid voltage data as the first training set, and collect normal and abnormal load voltage data as the second training set to construct a grid voltage recognition model and a load voltage recognition model respectively; Preprocess the simulated grid voltage data and the simulated load voltage data respectively, and extract grid voltage features and load voltage features; Input the grid voltage features and the load voltage features into the trained grid voltage recognition model and load voltage recognition model respectively; Take the abnormal data identified by the grid voltage recognition model and the load voltage recognition model as the first abnormal data.

5. The voltage sag intelligent identification and restoration method according to claim 4, wherein The steps of collecting the first grid voltage data and the first load voltage data of the operating state of the grid equipment, identifying the first grid voltage data and the first load voltage data, and determining the abnormal data of the operating state of the grid equipment to obtain the second abnormal data include: Extract the first grid voltage features from the first grid voltage data; Extract the first load voltage features from the first load voltage data; Input the first grid voltage features and the first load voltage features into the grid voltage recognition model and the load voltage recognition model respectively to detect abnormalities and obtain the second abnormal data.

6. The voltage sag intelligent identification and recovery method according to claim 5, characterized in that The steps of synchronously comparing and verifying the first abnormal data and the second abnormal data to obtain an abnormal diagnosis result include: Label the time and position tags of the first abnormal data and the second abnormal data, and store them in a distributed database; Extract the first abnormal sub-data and the second abnormal sub-data of the same device in the same time period according to the time and position tags; Compare the abnormal types, positions and degrees of coincidence of the first abnormal sub-data and the second abnormal sub-data; If the types are the same and the position coincidence degree is high, it is determined that the detection results are consistent; Otherwise, analyze the difference points between the first abnormal sub-data and the second abnormal sub-data, and adjust the model with the worse performance; Take the consistent result directly as the final abnormal diagnosis result.

7. The voltage sag intelligent identification and restoration method according to claim 6, characterized in that, The steps of inputting the abnormal diagnosis result into a voltage abnormal cause analysis model to analyze the cause of voltage abnormality and obtain a voltage sag cause analysis result include: Collect historical voltage abnormal case data and determine the historical voltage sag cause; Construct labels for the historical voltage abnormal case data and the historical voltage sag cause and use them as voltage sag training data; Establish a deep learning analysis model using a neural network structure; Use the voltage sag training data to train the deep learning analysis model to learn the abnormal-cause mapping relationship to obtain the voltage abnormal cause analysis model; Extract features from the abnormal diagnosis result as input parameters and import them into the voltage abnormal cause analysis model to analyze and identify the cause of voltage abnormality to obtain a voltage sag cause analysis result.

8. The voltage sag intelligent identification and restoration method according to claim 7, characterized in that, The steps of simulating different recovery schemes in a digital twin environment using a deep reinforcement learning platform according to the voltage sag cause analysis result and selecting the optimal scheme with an effect and cost better than a preset threshold include: Generate several alternative voltage recovery schemes according to the voltage sag cause analysis result; Use the grid digital twin model to simulate the execution of each voltage recovery scheme; Define a reward function, and a recovery plan that restores the voltage within a preset voltage threshold within a preset time range and has a recovery cost lower than a preset cost threshold will receive a reward; Obtain historical voltage recovery plans, use the historical voltage recovery plans as actions, and establish a deep reinforcement learning platform in combination with the reward function; Input the voltage recovery plan into the deep reinforcement learning platform to obtain the plan reward data for each voltage recovery plan; Simulate each voltage recovery plan in the digital twin environment according to the power grid digital twin model, the voltage recovery plan, and the plan reward data, and identify the plan with the highest reward value as the optimal plan.

9. The voltage sag intelligent identification and recovery method according to claim 8, wherein The step of converting the optimal plan into a control instruction and sending it to the real power grid for multi-device collaborative control, jointly adjusting the power or switch state, and realizing the joint control and collaborative recovery of the voltage value includes: Refine the optimal plan into specific control operations for each involved device; Serialize the specific control operations into standard control instructions; Through the Internet of Things platform, send the standard control instructions to each target control device; Each target control device performs a joint operation of adjusting the power or switching the switch state according to the standard control instructions; The target control devices perform data synchronization through timed communication and adjust the operations to constrain the voltage value within the target range.

10. An intelligent voltage sag identification and recovery system, characterized in that, The system is used to execute the voltage sag intelligent identification and recovery method according to any one of claims 1 to 9, including: power grid equipment; cloud platform; Internet of Things platform; wherein, The cloud platform is used for: Obtain power grid data and establish a power grid digital twin model of the power grid according to the power grid data; Synchronize the power grid digital twin model with the real power grid in real time through Internet of Things technology; Based on the power grid digital twin model, use mathematical models and simulation algorithms to simulate and generate simulated power grid voltage data and simulated load voltage data when the power grid equipment is running in the digital twin environment; Use artificial intelligence technology to analyze and process the generated simulated power grid voltage data and simulated load voltage data to determine the abnormal operation state data of the power grid equipment and obtain the first abnormal data; Collect the first power grid voltage data and the first load voltage data of the power grid equipment operation state, and identify the first power grid voltage data and the first load voltage data to determine the abnormal operation state data of the power grid equipment and obtain the second abnormal data; Synchronously compare and verify the first abnormal data with the second abnormal data to obtain an abnormal diagnosis result; Input the abnormal diagnosis result into a voltage abnormal cause analysis model for voltage abnormal cause analysis to obtain a voltage sag cause analysis result; According to the voltage sag cause analysis result, use the deep reinforcement learning platform to simulate different recovery plans in the digital twin environment, and select the optimal plan with better effect and cost than the preset threshold; Convert the optimal plan into a control instruction and send it to the real power grid for multi-device collaborative control, jointly adjusting the power or switch state, and realizing the joint control and collaborative recovery of the voltage value.

Images