Standardized centralized DTU multi-level fault diagnosis and self-healing method and system

Abstract

The invention discloses a standardized centralized DTU multi-level fault diagnosis and self-healing method and system, and relates to the technical field of intelligent power grid fault diagnosis, and the method comprises the steps: building a dynamic topological graph according to a unified coding rule, extracting spatial-temporal characteristics through a graph convolution network, and obtaining a spatial-temporal characteristic power grid topological graph; building a power grid digital twin environment based on a multi-level standardized fault diagnosis result, generating a plurality of sets of self-healing strategy candidate sets by adopting a reinforcement learning algorithm, screening an optimal self-healing strategy, and converting the optimal self-healing strategy into a control instruction conforming to a substation automation standard protocol; the control instruction is sent to the DTU and analyzed, and the intelligent power switch cabinet is controlled to execute self-healing operation. By automatically correcting the abnormal diagnosis result, false alarm and missing alarm are reduced, and the accuracy of fault diagnosis is improved. The intelligent level of power grid management is obviously improved, the maintenance cost is reduced, and the operation efficiency and the service quality of a power system are optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grid fault diagnosis, and in particular to a standardized centralized DTU multi-level fault diagnosis and self-healing method and system. Background Art

[0002] With the rapid development of smart grids, the role of distribution automation systems (DAS) in power system operations is becoming increasingly prominent. Distribution terminal units (DTUs), key equipment for real-time monitoring and fault handling of distribution networks, have a direct impact on grid reliability and operational efficiency. In recent years, fault diagnosis and self-healing control methods based on artificial intelligence and big data analysis have become a research hotspot, with significant progress particularly in building digital twin models of power grids and integrating spatiotemporal feature analysis. By introducing advanced algorithms such as graph neural networks, attention mechanisms, and reinforcement learning, effective identification of fault propagation paths and rapid response to them in complex power grids have been achieved.

[0003] Most existing solutions fail to achieve deep coordination between multi-level fault diagnosis and self-healing strategies, particularly in areas like standardized data processing and model generalization. For example, fault diagnosis based solely on a single physical model or local time series features is susceptible to noise interference and fails to uniformly characterize the global spatiotemporal characteristics of the power grid. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a standardized centralized DTU multi-level fault diagnosis and self-healing method to solve the problem of low efficiency of self-healing strategies.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, the present invention provides a standardized centralized DTU multi-level fault diagnosis and self-healing method, which includes collecting power grid equipment operating parameters and environmental monitoring data to form a standard data frame; establishing a dynamic topology map according to a unified coding rule, and extracting spatiotemporal features through a graph convolutional network to obtain a spatiotemporal feature power grid topology map; inputting the spatiotemporal feature power grid topology map into a spatiotemporal graph neural network, capturing faults through a graph attention mechanism, and generating multi-level fault labels; using a physical information neural network based on Kirchhoff's law to verify the multi-level fault labels of the spatiotemporal graph neural network to obtain multi-level standardized fault diagnosis results; building a power grid digital twin environment based on the multi-level standardized fault diagnosis results, using a reinforcement learning algorithm to generate multiple sets of self-healing strategy candidate sets, screening the optimal self-healing strategy and converting it into control instructions that comply with the substation automation standard protocol; sending the control instructions to the DTU and parsing them to control the intelligent power switch cabinet to perform self-healing operations.

[0007] As a preferred solution of the standardized centralized DTU multi-level fault diagnosis and self-healing method of the present invention, wherein: the standardized DTU collects power grid equipment operating parameters and environmental monitoring data through the substation automation standard protocol to form a standard data frame. The specific steps are as follows: Collect voltage, current and power data to obtain grid equipment operating parameters and environmental monitoring data; Through the substation automation standard protocol, the grid equipment operating parameters and environmental monitoring data are modeled, and the MMS in the substation automation standard protocol is used for encoding to generate standard data frames.

[0008] As a preferred solution of the standardized centralized DTU multi-level fault diagnosis and self-healing method described in the present invention, wherein: the dynamic topology map is established according to the unified coding rules, and the spatiotemporal features are extracted through the graph convolutional network to obtain the spatiotemporal feature power grid topology map. The specific steps are as follows: According to the unified coding rules, grid devices are assigned an identifier as a node. Edges are defined based on the electrical connection relationship between grid devices. The initial static topology of the grid is constructed based on the node and edge information. A time synchronization mechanism is introduced to form a dynamic topology. A feature matrix is ​​constructed based on the historical standard data frame, and an adjacency matrix is ​​constructed based on the historical connection relationship between power grid devices. The graph convolutional network is trained and the graph convolutional network parameters are adjusted to obtain the trained graph convolutional network. The real-time feature matrix and real-time adjacency matrix are input into the trained graph convolutional network to output a power grid topology map with rich spatiotemporal features.

[0009] As a preferred solution of the standardized centralized DTU multi-level fault diagnosis and self-healing method described in the present invention, the spatiotemporal characteristic power grid topology map is input into the spatiotemporal graph neural network, faults are captured through the graph attention mechanism, and multi-level fault labels are generated. The specific steps are as follows: A multi-layer spatiotemporal graph convolution stacking structure is adopted, and a sliding window mechanism is used to capture the dynamic changes of faults in the time dimension. By stacking graph convolution layers in the spatial dimension, grid-related features of different ranges are extracted layer by layer, and preliminary fault diagnosis is obtained from local to global perspectives. Through the fully connected layer and classifier in the spatiotemporal graph neural network, the fault conditions of the nodes in the spatiotemporal characteristic power grid topology map are predicted and multi-level fault labels are generated.

[0010] As a preferred solution of the standardized centralized DTU multi-level fault diagnosis and self-healing method described in the present invention, the physical information neural network is used to verify the multi-level fault labels of the spatiotemporal graph neural network based on Kirchhoff's law to obtain a multi-level standardized fault diagnosis result. The specific steps are as follows: Match multi-level fault labels with the power grid topology and real-time operating parameters, extract electrical quantity data for corresponding nodes and branches, cleanse and format the electrical quantity data, and obtain a fault diagnosis verification data set; Based on Kirchhoff's current law and Kirchhoff's voltage law, a physical constraint model is constructed using a physical information neural network. The fault diagnosis verification data set is input to verify node current balance and branch power flow, and abnormal diagnosis results in multi-level fault labels are automatically corrected. The preliminary diagnosis results of the spatiotemporal graph neural network and the abnormal diagnosis results after verification of the physical constraint model are integrated, and the conflicting data are eliminated through a weighted fusion strategy to generate multi-level standardized diagnosis results.

[0011] As a preferred solution of the standardized centralized DTU multi-level fault diagnosis and self-healing method described in the present invention, the following specific steps are taken: Multi-level standardized diagnostic results are embedded as constraints in power flow calculation and transient stability analysis models. Long-term time-lapse prediction (LSTM) and convolutional neural networks (CNN) are then combined for load forecasting and feature extraction. Through standardized communication interfaces, these results are interactively verified with the actual power grid, forming a digital twin environment for the power grid. The state space is defined based on the grid topology state, and the action space is defined by the switch operation combination. The reinforcement learning model is trained using the grid digital twin environment to generate multiple sets of self-healing strategies to form a candidate set. Define evaluation criteria based on economic cost, safety, and reliability, assign corresponding weights, and conduct simulation tests using a power grid digital twin environment. By comparing the performance of various strategies, the optimal self-healing strategy is determined. According to the substation automation standard protocol, the optimal self-healing strategy is converted into a rule for specific control instructions, and the selected optimal self-healing strategy is converted into specific control instructions that comply with the substation automation standard protocol.

[0012] As a preferred solution of the standardized centralized DTU multi-level fault diagnosis and self-healing method of the present invention, wherein: the control instruction is sent to the DTU and parsed, and the intelligent power switch cabinet is controlled to perform the self-healing operation. The specific steps are as follows: The control instructions are sent from the master station to the DTU using the communication protocol. The DTU decodes the received data packets, extracts the specific control commands, and checks the integrity. The intelligent power switchgear performs corresponding actions according to the received control instructions and performs self-healing operations.

[0013] In a second aspect, the present invention provides a standardized centralized DTU multi-level fault diagnosis and self-healing system, including an acquisition module, a topology module, a labeling module, a fault diagnosis module, a strategy generation module, and an execution module; the acquisition module is used for the standardized DTU to collect power grid equipment operating parameters and environmental monitoring data through the substation automation standard protocol to form a standard data frame; the topology module is used to establish a dynamic topology map according to a unified coding rule, and extract spatiotemporal features through a graph convolutional network to obtain a spatiotemporal feature power grid topology map; the labeling module is used to input the spatiotemporal feature power grid topology map into a spatiotemporal graph neural network, capture faults through a graph attention mechanism, and generate multi-level fault labels; the fault diagnosis module is used to use a physical information neural network based on Kirchhoff's law to verify the multi-level fault labels of the spatiotemporal graph neural network to obtain multi-level standardized fault diagnosis results; the strategy generation module is used to build a power grid digital twin environment based on the multi-level standardized fault diagnosis results, use a reinforcement learning algorithm to generate multiple sets of self-healing strategy candidate sets, screen the optimal self-healing strategy and convert it into control instructions that comply with the substation automation standard protocol; the execution module is used to send the control instructions to the DTU and parse them, and control the intelligent power switch cabinet to perform self-healing operations.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the standardized centralized DTU multi-level fault diagnosis and self-healing method as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the standardized centralized DTU multi-level fault diagnosis and self-healing method as described in the first aspect of the present invention.

[0016] The present invention achieves the following beneficial effects: by innovatively combining dynamic topology modeling with fault diagnosis and verification based on a physical information neural network, it enables real-time monitoring of power grid operating status, fault prediction, and automatic recovery. By using a physical information neural network to verify preliminary fault diagnosis results based on Kirchhoff's laws and automatically correcting abnormal diagnostic results, it reduces false positives and missed positives, thereby improving the accuracy of fault diagnosis. This significantly enhances the intelligent level of power grid management, reduces maintenance costs, and optimizes the operating efficiency and service quality of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flowchart of the standardized centralized DTU multi-level fault diagnosis and self-healing method.

[0019] Figure 2 Schematic diagram of a standardized centralized DTU multi-level fault diagnosis and self-healing system.

[0020] Figure 3 Flowchart for dynamic topology map construction and spatiotemporal feature extraction.

[0021] Figure 4 Flowchart for generating and executing a flow chart for a self-healing strategy. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a standardized centralized DTU multi-level fault diagnosis and self-healing method, including the following steps: S1. Collect power grid equipment operating parameters and environmental monitoring data to form a standard data frame.

[0026] Collect voltage, current and power data to obtain grid equipment operating parameters and environmental monitoring data.

[0027] Specifically, according to substation automation standard protocols (such as IEC 61850), the DTU's communication parameters, including data transmission rate, data frame format, and encoding method, are set. The data type and format of the connected device are correctly identified. The DTU then establishes a communication connection with the grid equipment based on the set parameters, collecting voltage, current, and power data from the equipment in real time. The DTU not only obtains basic operating parameters of the grid equipment, such as voltage, current intensity, active power, and reactive power, but also collects environmental monitoring data, such as temperature and humidity, for comprehensive assessment of the grid equipment's operating status. This provides both grid equipment operating parameters and environmental monitoring data.

[0028] Through the substation automation standard protocol, the grid equipment operating parameters and environmental monitoring data are modeled, and the MMS in the substation automation standard protocol is used for encoding to generate standard data frames.

[0029] Specifically, the collected power grid equipment operating parameters (such as voltage, current, and power) and environmental monitoring data (such as temperature and humidity) are modeled and described in a standardized manner according to the object model defined in the protocol (such as logical node LN, data class Data Class, and data attribute DA); they are mapped to the MMS (Manufacturing Message Specification) service model, and the power grid equipment operating parameters are serialized using the MMS encoding rules (such as ASN.1 Basic Encoding Rules BER); the encoded power grid equipment operating parameters are encapsulated into standard data frames that comply with the protocol specifications, including fields such as protocol header, service type, data identifier, and payload, to form standard data frames.

[0030] S2. Establish a dynamic topology map based on the unified coding rules, and extract spatiotemporal features through a graph convolutional network to obtain a spatiotemporal feature power grid topology map.

[0031] A dynamic topology model of power grid equipment is constructed based on standard data frames and unified coding rules. Power grid equipment is defined as nodes, and the electrical connection relationship between devices is defined as edges. The node characteristics and edge weights are updated through a time synchronization mechanism.

[0032] Specifically, standardized data frames are parsed according to substation automation standard protocols such as IEC 61850 to extract real-time operating parameters of power grid equipment, including voltage, current, power factor, and environmental monitoring data such as temperature and humidity. Fields such as the device ID, timestamp, and parameter type in the standardized data frames are then parsed in a standardized manner. The data format is unified, and each power grid device is assigned a unique device code according to the substation automation standard protocol. For example, a three-segment code structure of "device type + geographic location + serial number" is used, with the device code serving as the unique node identifier. This allows various types of power grid equipment to be defined as nodes in a dynamic topology graph. Based on the actual connections in the power grid, the electrical connection information between devices in the data frames is parsed to define the electrical connections between devices as edges in the graph. Each edge is assigned an initial weight, which can be calculated based on electrical distance, topological distance, or real-time operating status (such as line load factor). Furthermore, a time synchronization mechanism is established to dynamically update node characteristics and edge weights by subscribing to the timestamp field in the standardized data frames. Specifically, new data frames are received at fixed intervals and parsed for device status change information.

[0033] The dynamic topology map is input into the graph convolutional network, and the node neighborhood features are aggregated through multiple layers of graph convolutional layers to extract spatial correlation. The temporal convolutional network is used to capture the dynamic evolution of fault characteristics in the time dimension, and the grid topology map with spatiotemporal characteristics is obtained.

[0034] Specifically, the dynamic topology graph is input into the GCN, where node neighborhood features are aggregated through multiple graph convolutional layers to extract spatial correlations. The TCN uses causal and dilated convolutional layers to capture fault evolution patterns in the temporal dimension, forming a spatiotemporal grid topology map. This includes the input layer, graph convolutional layer, activation function layer, and fully connected layer in the GCN, and the causal convolutional layer, dilated convolutional layer, and residual connection in the TCN. The training process prepares historical power grid data as a sample set, defines a loss function such as the cross-entropy loss function, and uses optimization algorithms such as Adam to calculate predictions through forward propagation. Backpropagation is then used to update weights. This process is iterated until convergence, resulting in a spatiotemporal graph neural network. Multiple layers of graph convolutional layers aggregate node neighborhood features layer by layer. Each layer of graph convolution is based on a predefined adjacency matrix and feature transformation matrix. A message passing mechanism is used to calculate the feature interactions between each node and its directly connected nodes, gradually expanding the receptive field to capture local to global spatial correlations. Learnable weight matrices and activation functions (such as ReLU) are introduced at each layer to enhance feature representation. The multi-layer node feature sequence output by the graph convolutional network is input into a temporal convolutional network. A combination of causal convolution and dilated convolution is used to capture the dynamic evolution of fault features at different time scales along the temporal dimension. Long-term dependencies are gradually extracted by stacking multiple temporal convolutional layers, and residual connections are used to prevent gradient vanishing. The output of the spatiotemporal feature fusion is mapped to a preset fault classification space through a fully connected layer, resulting in a power grid topology map containing spatiotemporal features, including information such as device status, fault type, and impact range.

[0035] S3. Input the spatiotemporal feature power grid topology map into the spatiotemporal graph neural network, capture faults through the graph attention mechanism, and generate multi-level fault labels.

[0036] A multi-layer spatiotemporal graph convolution stacking structure is adopted, and a sliding window mechanism is used to capture the dynamic change law of faults in the time dimension. By stacking graph convolution layers in the spatial dimension, grid-related features of different ranges are extracted layer by layer to obtain preliminary fault diagnosis from local to global.

[0037] Specifically, a sliding time window mechanism is constructed to divide the input dynamic topology graph sequence into multiple continuous time segments at fixed time intervals (such as 5 seconds), and each segment contains topology graph data of multiple time steps; within each time window, node neighborhood features are aggregated layer by layer through multi-layer graph convolution layers. The first layer of graph convolution focuses on the status of directly connected devices, and subsequent layers gradually capture a wider range of power grid correlation features through a recursive message passing mechanism, such as the communication relationship between substations and the coupling characteristics of regional power grids; in the time dimension, one-dimensional convolution layers or time-series graph convolutions are used to extract features from the graph sequence within the sliding window to capture the dynamic evolution of faults in time, such as fault propagation speed and transient process characteristics. By stacking multiple spatiotemporal graph convolution blocks, the local correlation features of the spatial dimension are deeply integrated with the dynamic change features of the time dimension to form a multi-scale fault representation from the device level to the system level. By introducing a graph attention mechanism, the fault type, location and impact range are output, and complete fault diagnosis from local anomaly detection to global pattern recognition is achieved.

[0038] Through the fully connected layer and classifier in the spatiotemporal graph neural network, the fault conditions of the nodes in the spatiotemporal characteristic power grid topology map are predicted and multi-level fault labels are generated.

[0039] Specifically, the node dimensions in the spatiotemporal characteristic power grid topology are expanded to obtain the feature vector representation of each node; the node feature vector is input into the fully connected layer for nonlinear transformation. The number of neurons in the fully connected layer is designed according to the hierarchical structure of the fault label system (such as three layers: device level, feeder level, and substation level); a classifier (such as the Softmax classifier) ​​is used to perform multi-classification prediction on the feature vector of each node to generate a device-level fault label (such as transformer overheating and circuit breaker abnormality).

[0040] S4. Based on Kirchhoff's law, the physical information neural network is used to verify the multi-level fault labels of the spatiotemporal graph neural network and obtain multi-level standardized fault diagnosis results.

[0041] The multi-level fault labels are matched with the spatiotemporal characteristic power grid topology map and standard data frames, and the electrical quantity data of the corresponding nodes and branches are extracted. The electrical quantity data are cleaned and format converted to obtain the fault diagnosis verification data set.

[0042] Specifically, according to the device ID and branch identifier in the multi-level fault label, the corresponding nodes and branches are located from the dynamic spatiotemporal characteristic power grid topology map, and a mapping relationship between the multi-level fault label and the dynamic spatiotemporal characteristic power grid topology map is established; based on this mapping relationship, the voltage, current, power and other electrical quantity data of the corresponding nodes and the operating parameters such as the flow and impedance of the branch are extracted from the standard data frame; the extracted electrical quantity data are cleaned, including processing missing values ​​(using interpolation or mean filling), outlier detection (eliminating outliers based on the 3σ principle or box plot method), deduplication of duplicate data, and unifying the timestamp format (converting to UTC time) and unit system (such as unifying voltage to kV and power to MW); the cleaned operating data is formatted according to the fault label hierarchical structure, and a standardized data set containing time series features, topological association features and electrical quantity features is constructed, and partitioned storage is performed according to dimensions such as fault type and occurrence time to form a fault diagnosis verification data set.

[0043] Based on Kirchhoff's current law and Kirchhoff's voltage law, a physical constraint model is constructed using a physical information neural network. The fault diagnosis verification data set is input to verify the node current balance and branch power flow, and the abnormal diagnosis results in the multi-level fault labels are automatically corrected.

[0044] Specifically, the node current, branch power and other electrical quantity data in the fault diagnosis verification dataset are matched with the node-branch association relationship in the dynamic topology diagram to construct a spatiotemporal feature matrix that conforms to the PINNs input; according to the physical constraint model - the current sub-network constrains the sum of the node inflow current to zero based on the KCL constraint, and the voltage sub-network constrains the voltage difference between the beginning and end of the branch to be equal to the line voltage drop based on the KVL constraint, and joint training is achieved by sharing the underlying topological feature extraction layer; the fault diagnosis verification dataset is input, and the forward propagation is used to calculate the current balance error of each node and the branch power flow deviation. During the back propagation, physical constraint terms (such as the L1 norm of the current imbalance and the sum of the squares of the voltage deviation) are added to the loss function, and the network parameters are optimized by gradient descent; the abnormal diagnosis results are automatically adjusted according to the output correction coefficient: the fault state of the node with current imbalance exceeds is re-determined, the associated fault label of the branch with power reverse is corrected, and the corrected multi-level fault label is confidence-weighted fused with the original diagnosis result to output the abnormal diagnosis result.

[0045] The preliminary diagnosis results of the spatiotemporal graph neural network and the abnormal diagnosis results after verification of the physical constraint model are integrated, and the conflicting data are eliminated through a weighted fusion strategy to generate multi-level standardized diagnosis results.

[0046] Specifically, feature alignment is performed on the abnormal diagnosis results and the fault diagnosis verification data set, and the node-level fault probability matrix (dimension N×L, N is the number of nodes, L is the number of fault types) output by the spatiotemporal graph neural network is spatiotemporally matched with the branch power flow data (dimension M×3, M is the number of branches, and 3 is the voltage / current / power parameters) corrected by the physical constraint model. A corresponding relationship is established through device ID and topological association. A weighted fusion strategy is used to set dynamic weight coefficients. For situations where there are contradictory diagnoses for the same node, fault types that have passed the physical model verification are retained first (for example, if the spatiotemporal graph diagnoses a circuit breaker fault but the branch current is normal, it is corrected to a false alarm). The spatiotemporal graph diagnosis results are forcibly overwritten for physical abnormalities such as branch power backflow. Finally, the diagnosis results are standardized at multiple levels.

[0047] S5. Build a digital twin environment for the power grid based on multi-level standardized fault diagnosis results, use reinforcement learning algorithms to generate multiple sets of self-healing strategy candidates, screen the optimal self-healing strategy, and convert it into control instructions that comply with the substation automation standard protocol.

[0048] The multi-level standardized diagnostic results are embedded as constraints in the power flow calculation and transient stability analysis models, and combined with LSTM and CNN for load forecasting and feature extraction. Through standardized communication interfaces, interactive verification with the actual power grid is obtained to form a digital twin environment for the power grid.

[0049] Specifically, the Newton-Raphson method is used to iteratively solve the node voltage and branch power, and a transient stability analysis model is added. The diagnosed fault location and type are used as boundary conditions, and the stability is evaluated through time domain simulation. The LSTM network is used to model the time series of historical load data to extract long-term dependency features, and CNN is combined to extract the spatial correlation features of the grid topology and operating parameters, and the model parameters are optimized through joint training. The multi-level standardized diagnosis results (including equipment status, fault location and impact range) are embedded as hard constraints in the power flow calculation model (such as the Newton-Raphson method) and the transient stability analysis model (such as time domain simulation), and the node injection power and branch admittance matrix are modified to reflect the grid operation status after the fault. The LSTM network processes time-series load data (such as historical load curves and meteorological data) and extracts long-term dependency features. At the same time, CNN is used to extract the spatial features of load distribution (such as regional load density and topological correlation characteristics), and a joint prediction model is constructed to obtain load forecast results for future time periods. The predicted load is combined with the diagnostic results to dynamically update the node load parameters and branch flow distribution in the digital twin model. The simulation results of the digital twin model are compared and verified with the actual power grid operation data in real time through a standardized communication interface. When the simulation error exceeds the threshold, the model parameter self-correction mechanism is triggered, and the interaction log is recorded to form a closed-loop feedback system, resulting in a power grid digital twin environment that includes a physical model, an AI prediction module, and a real-time verification mechanism.

[0050] The state space is defined according to the grid topology state, the action space is defined by the switch operation combination, and the reinforcement learning model is trained using the grid digital twin environment to generate multiple sets of self-healing strategies to form a candidate set.

[0051] Specifically, a state space is constructed based on the real-time topological state of the power grid (node ​​voltage, branch power, switch position, etc.), and the switch operation combination (such as the sequence of opening and closing instructions) is defined as a discrete action space, and action constraints (such as N-1 safety criteria and equipment operation locking conditions) are set; the power grid operation state is initialized in the digital twin environment, and a variety of fault scenarios (such as single-phase grounding, line short circuit, etc.) are generated through Monte Carlo simulation as training samples for the reinforcement learning model; the proximal policy optimization algorithm is used to train the intelligent agent, and when defining the reward function, the fault isolation speed, load loss minimization, operational safety and number of switch operations are comprehensively considered, and the policy network is optimized through policy gradient updating; during the training process, the simulation feedback of the digital twin environment of the power grid is monitored in real time. When the safety constraints are violated, it is marked as an invalid strategy and the exploration strategy is adjusted; multiple sets of self-healing strategy candidate sets are generated, each strategy contains a complete switch operation sequence, expected recovery time, load loss assessment and risk level labeling.

[0052] Evaluation criteria are defined based on economic cost, safety, and reliability, and corresponding weights are assigned. Simulation tests are conducted using the digital twin environment of the power grid, and the optimal self-healing strategy is obtained by comparing the performance of various strategies.

[0053] Specifically, a multi-dimensional evaluation standard system is defined. Economic indicators include load loss cost and switching operation loss, safety indicators cover voltage stability margin and equipment overload risk, and reliability indicators evaluate fault recovery time and islanding risk. A simulation scenario consistent with the actual power grid is built in the digital twin environment, historical fault data is injected, and extreme operating conditions (such as multiple fault superposition) are set. Simulation tests are performed on each set of candidate strategies, and indicator data is recorded. A dynamic weighted scoring is used to calculate the comprehensive score, and the optimal solution set that balances safety and economy is screened out through Pareto frontier analysis. Safety hard constraints (such as voltage drop ≤ 8%) are set to filter out dangerous strategies. The robustness of the strategy is verified through Monte Carlo simulation, and the strategy with the highest success rate and smallest standard deviation in 1,000 random perturbation tests is selected as the optimal self-healing strategy.

[0054] According to the substation automation standard protocol, rules for converting self-healing strategies into specific control instructions are formulated, and the optimal self-healing strategy is converted into self-healing control instructions.

[0055] Specifically, the switch operation sequence in the optimal self-healing strategy is parsed to extract parameters such as device ID, operation type (open / close), priority, and execution time; the device type is mapped to the corresponding logical node, and the control block is defined); a GOOSE message data set is constructed, including mandatory fields such as SwitchID, OperateType, and Priority, ctlModel is set to direct control mode, and the stVal status value is determined; then, the GOOSE message is encoded and encapsulated according to IEC 61850-8-1, and a timestamp and security check code are added to the GOOSE message; standardized control instructions are generated in binary / text format, including complete device identification, operation instructions, and security authentication information, forming self-healing control instructions that meet the requirements of substation automation.

[0056] S6. Send the control command to the DTU for parsing, and control the smart power switch cabinet to perform self-healing operations.

[0057] The control instructions are sent from the master station to the DTU using the communication protocol. The DTU checks the integrity of the received data packets and decodes them to extract the device ID and operation type. The DTU sends control commands to the intelligent switch cabinet through the serial port, executes corresponding actions according to the received control instructions, and performs self-healing operations.

[0058] Specifically, it is sent to the target DTU via the TCP network. After receiving the data packet, the DTU checks whether the synchronization identifier and length field format of the message header are correct, calculates the CRC check code and compares it with the check value carried in the message to verify data integrity. The DTU parses the message to obtain the device ID and operation type (opening / closing), sends a control command to the corresponding intelligent switch cabinet via the RS485 serial port, and converts the operation instruction into binary code using the Modbus RTU protocol. The control module of the intelligent switch cabinet parses the operation type and drives the actuator to operate. When opening, the trip coil is triggered to cut off the circuit. When closing, the electric mechanism is driven to close the contacts to complete the self-healing operation.

[0059] This embodiment also provides a standardized centralized DTU multi-level fault diagnosis and self-healing system, including: a collection module, a topology module, a label module, a fault diagnosis module, a policy generation module, and an execution module; The acquisition module is used to collect power grid equipment operating parameters and environmental monitoring data to form a standard data frame; the topology map module is used to establish a dynamic topology map according to a unified coding rule, and extract spatiotemporal features through a graph convolutional network to obtain a spatiotemporal feature power grid topology map; the label module is used to input the spatiotemporal feature power grid topology map into the spatiotemporal graph neural network, capture faults through the graph attention mechanism, and generate multi-level fault labels; the fault diagnosis module is used to use the physical information neural network based on Kirchhoff's law to verify the multi-level fault labels of the spatiotemporal graph neural network to obtain multi-level standardized fault diagnosis results; the strategy generation module is used to build a power grid digital twin environment based on the multi-level standardized fault diagnosis results, use a reinforcement learning algorithm to generate multiple sets of self-healing strategy candidate sets, screen the optimal self-healing strategy and convert it into control instructions that comply with the substation automation standard protocol; the execution module is used to send the control instructions to the DTU and parse them to control the intelligent power switch cabinet to perform self-healing operations.

[0060] This embodiment further provides a computer device applicable to the standardized centralized DTU multi-level fault diagnosis and self-healing method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the standardized centralized DTU multi-level fault diagnosis and self-healing method proposed in the above embodiment.

[0061] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0062] This embodiment further provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the program implements the standardized centralized DTU multi-level fault diagnosis and self-healing method proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0063] In summary, this invention achieves real-time monitoring of power grid operating status, fault prediction, and automatic recovery through the innovative combination of dynamic topology modeling and fault diagnosis verification based on a physical information neural network. Preliminary fault diagnosis results are verified using Kirchhoff's laws using a physical information neural network. Automatic correction of abnormal diagnostic results reduces false positives and missed positives, improving the accuracy of fault diagnosis. This significantly enhances the intelligent level of power grid management, reduces maintenance costs, and optimizes the operating efficiency and service quality of the power system.

[0064] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. Standardized centralized DTU multi-level fault diagnosis and self-healing method, characterized by: include, Collect power grid equipment operating parameters and environmental monitoring data to form standard data frames; A dynamic topology map is established according to unified coding rules, and spatiotemporal features are extracted through a graph convolutional network to obtain a spatiotemporal feature power grid topology map. The spatiotemporal grid topology graph is input into the spatiotemporal graph neural network, and faults are captured through the graph attention mechanism to generate multi-level fault labels. The physical information neural network is used to verify the multi-level fault labels of the spatiotemporal graph neural network based on Kirchhoff's law, and multi-level standardized fault diagnosis results are obtained; A digital twin environment for the power grid is built based on multi-level standardized fault diagnosis results. A reinforcement learning algorithm is used to generate multiple candidate sets of self-healing strategies. The optimal self-healing strategy is selected and converted into control instructions that comply with standard substation automation protocols. The control instructions are sent to the DTU for parsing, and the smart power switch cabinet is controlled to perform self-healing operations.

2. The standardized centralized DTU multi-level fault diagnosis and self-healing method according to claim 1, characterized in that: The specific steps of collecting the operating parameters of the power grid equipment and the environmental monitoring data to form a standard data frame are as follows: Collect voltage, current and power data to obtain grid equipment operating parameters and environmental monitoring data; Through the substation automation standard protocol, the grid equipment operating parameters and environmental monitoring data are modeled, and the MMS in the substation automation standard protocol is used for encoding to generate standard data frames.

3. The standardized centralized DTU multi-level fault diagnosis and self-healing method according to claim 2, characterized in that: The dynamic topology map is established according to the unified coding rules, and the spatiotemporal features are extracted through the graph convolutional network to obtain the spatiotemporal feature power grid topology map. The specific steps are as follows: According to the unified coding rules, grid devices are assigned an identifier as a node. Edges are defined based on the electrical connection relationship between grid devices. The initial static topology of the grid is constructed based on the node and edge information. A time synchronization mechanism is added to form a dynamic topology. A feature matrix is ​​constructed based on the historical standard data frame, and an adjacency matrix is ​​constructed based on the historical connection relationship between power grid devices. The graph convolutional network is trained and the graph convolutional network parameters are adjusted to obtain the trained graph convolutional network. The real-time feature matrix and real-time adjacency matrix are input into the trained graph convolutional network to obtain the spatiotemporal feature power grid topology map.

4. The standardized centralized DTU multi-level fault diagnosis and self-healing method according to claim 3, characterized in that: The spatiotemporal feature power grid topology graph is input into the spatiotemporal graph neural network, faults are captured through the graph attention mechanism, and multi-level fault labels are generated. The specific steps are as follows: A multi-layer spatiotemporal graph convolution stacking structure is adopted, and a sliding window mechanism is used to capture the dynamic changes of faults in the time dimension. By stacking graph convolution layers in the spatial dimension, grid-related features of different ranges are extracted layer by layer, obtaining preliminary diagnostic results from local to global perspectives. Through the fully connected layer and classifier in the spatiotemporal graph neural network, the fault conditions of the nodes in the spatiotemporal characteristic power grid topology map are predicted and multi-level fault labels are generated.

5. The standardized centralized DTU multi-level fault diagnosis and self-healing method according to claim 4, characterized in that: The physical information neural network is used to verify the multi-level fault labels of the spatiotemporal graph neural network based on Kirchhoff's law to obtain multi-level standardized fault diagnosis results. The specific steps are as follows: Match multi-level fault labels with the spatiotemporal grid topology and standard data frames, extract electrical quantity data for corresponding nodes and branches, clean and format-convert the electrical quantity data, and obtain a fault diagnosis verification data set. Based on Kirchhoff's current law and Kirchhoff's voltage law, a physical constraint model is constructed using a physical information neural network. The fault diagnosis verification data set is input to verify node current balance and branch power flow, and abnormal diagnosis results in multi-level fault labels are automatically corrected. The preliminary diagnosis results of the spatiotemporal graph neural network and the abnormal diagnosis results after verification of the physical constraint model are integrated, and the conflicting data are eliminated through a weighted fusion strategy to generate multi-level standardized diagnosis results.

6. The standardized centralized DTU multi-level fault diagnosis and self-healing method according to claim 5, characterized in that: The digital twin environment of the power grid is built based on the multi-level standardized fault diagnosis results, and a reinforcement learning algorithm is used to generate multiple sets of self-healing strategy candidates. The optimal self-healing strategy is selected and converted into control instructions that comply with the substation automation standard protocol. The specific steps are as follows: Multi-level standardized diagnostic results are embedded as constraints in power flow calculation and transient stability analysis models. Long-term time-lapse prediction (LSTM) and convolutional neural networks (CNN) are then combined for load forecasting and feature extraction. Through standardized communication interfaces, these results are interactively verified with the actual power grid, forming a digital twin environment for the power grid. The state space is defined based on the grid topology state, and the action space is defined by the switch operation combination. The reinforcement learning model is trained using the grid digital twin environment to generate multiple sets of self-healing strategies to form a candidate set. Define evaluation criteria based on economic cost, safety, and reliability, assign corresponding weights, and conduct simulation tests using a power grid digital twin environment. By comparing the performance of various strategies, the optimal self-healing strategy is determined. According to the substation automation standard protocol, the self-healing strategy is converted into the rules of specific control instructions, and the optimal self-healing strategy is converted into self-healing control instructions.

7. The standardized centralized DTU multi-level fault diagnosis and self-healing method according to claim 6, characterized in that: The control instructions are sent to the DTU for parsing, and the smart power switch cabinet is controlled to perform self-healing operations. The specific steps are as follows: The control instructions are sent from the master station to the DTU using the communication protocol. The DTU checks the integrity of the received data packets and decodes them to extract the device ID and operation type. The DTU sends control commands to the intelligent switch cabinet through the serial port, and performs corresponding actions to perform self-healing operations according to the received control instructions.

8. A standardized centralized DTU multi-level fault diagnosis and self-healing system, based on the standardized centralized DTU multi-level fault diagnosis and self-healing method according to any one of claims 1 to 7, characterized in that: Including, acquisition module, topology module, label module, fault diagnosis module, strategy generation module and execution module; The acquisition module is used to collect power grid equipment operating parameters and environmental monitoring data through the standard substation automation protocol using the standardized DTU to form a standard data frame; The topology map module is used to establish a dynamic topology map according to a unified coding rule, and extract spatiotemporal features through a graph convolutional network to obtain a spatiotemporal feature power grid topology map; The labeling module is used to input the spatiotemporal characteristic power grid topology map into the spatiotemporal graph neural network, capture faults through the graph attention mechanism, and generate multi-level fault labels; The fault diagnosis module is used to verify the multi-level fault labels of the spatiotemporal graph neural network based on Kirchhoff's law using the physical information neural network to obtain multi-level standardized fault diagnosis results; The strategy generation module is used to build a digital twin environment for the power grid based on multi-level standardized fault diagnosis results, use a reinforcement learning algorithm to generate multiple sets of self-healing strategy candidates, screen the optimal self-healing strategy, and convert it into control instructions that comply with the substation automation standard protocol; The execution module is used to send the control instruction to the DTU and parse it, and control the intelligent power switch cabinet to perform self-healing operations.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the standardized centralized DTU multi-level fault diagnosis and self-healing method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the standardized centralized DTU multi-level fault diagnosis and self-healing method according to any one of claims 1 to 7 are implemented.

Images