Intelligent fault positioning system and method based on direct current system

Abstract

The invention provides an intelligent fault positioning system and method based on a DC system, and relates to the technical field of power fault positioning, and the system comprises a digital twin module, a sensor network module, a real-time data synchronization module, a graph neural network model module, an online reasoning module, an intelligent decision module and a decision self-healing module. Through deep fusion of the digital twinning technology and the graph neural network, complex electrical, thermal, mechanical and environmental parameters in the direct current system can be comprehensively analyzed. The digital twin module provides a high-precision virtual model of the system through multi-physics field integration, so that a graph neural network model can accurately capture tiny anomalies and complex fault modes in the system under the support of rich and high-quality data. Through combination of the graph convolutional network and the graph attention network, the graph neural network model can effectively identify relevance between key nodes and edges, weights of neighbor nodes are dynamically adjusted, and the fault detection and positioning precision is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power fault location, and in particular to an intelligent fault location system and method based on a direct current system. Background Art

[0002] According to Chinese patent publication number CN115864322A, a method and system for intelligent analysis of DC protection system faults include: step 1, establishing a DC protection system fault information model; step 2, initiating an intelligent analysis function based on characteristic quantities of the DC protection system fault; step 3, obtaining relevant alarm information and screening protection action information when the DC protection system fails; step 4, querying the DC protection system fault information model based on the screened protection action information to obtain the fault area and fault type corresponding to the protection action information; and step 5, generating a DC protection system fault analysis report.

[0003] Based on the collaborative architecture, device, and collaborative method for distribution protection including flexible interconnection devices disclosed in Chinese Patent Publication No. CN107332217A, within the collaborative control framework, the rapid state perception of the power electronic components of the AC / DC flexible interconnection device and the AC-side feeder protection control device are fully utilized to achieve rapid fault location and fault removal, thereby realizing rapid autonomous self-healing of regional faults in the entire AC / DC system. A GOOSE communication network based on regional high-speed information exchange is proposed to achieve global high-speed information communication among intelligent devices in the same AC / DC system, including the flexible DC interconnection device, and automatically achieve rapid fault location and isolation of protection devices according to preset conditions. Compared with traditional distribution system protection configuration methods, this method has faster fault handling speed and better collaborative fault handling capabilities.

[0004] The above patent documents and prior art have the following technical problems when used: Existing technologies and publicly available technical information lack comprehensive data analysis and fault information collection and analysis of DC systems, resulting in the need to further improve the accuracy and real-time performance of fault analysis and fault location. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention provides an intelligent fault location system based on a DC system, which solves the problem in the prior art that the fault analysis and fault location of the DC system are not ideal in terms of accuracy and real-time performance.

[0006] Another object of the present invention is to provide an intelligent fault location method based on a DC system.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: The intelligent fault location system based on DC system includes: The digital twin module uses simulation tools to generate a digital twin model of the DC system based on the DC system's physical model and historical operating data. Sensor network module, real-time collection of multi-dimensional operation data; The real-time data synchronization module pre-processes the real-time data collected by the sensor network module through the edge computing node and transmits it to the digital twin model for real-time data synchronization between the physical system and the digital twin; A graph neural network model module, which uses a system topology graph and feature vectors of nodes and edges to train a graph neural network model for fault detection and location. The nodes in the system topology graph represent devices, and the edges represent electrical connections. The online inference module deploys the trained graph neural network model on edge computing nodes and uses real-time data for fault detection and location. The intelligent decision-making module uses a reinforcement learning algorithm based on the fault detection results of the graph neural network model to generate a decision-making and self-healing strategy for fault response; The decision-making self-healing module executes control operations based on the decision-making self-healing strategy of the intelligent decision-making module.

[0008] Furthermore, the decision self-healing module includes: The self-healing execution module communicates with DC system equipment through industrial protocols to automatically perform control operations such as fault isolation, system switching, and load adjustment; Record and traceability module, which records fault detection and response data to the blockchain; The augmented reality-assisted operation and maintenance module displays fault location results and maintenance instructions through augmented reality devices.

[0009] Furthermore, the digital twin module includes physical field integration of electrical, thermal, mechanical and environmental parameters to comprehensively reflect the operating status of the DC system, and the physical field integration is achieved by coupling equations and models of different physical phenomena.

[0010] Furthermore, the sensor network module includes a DC voltage sensor, a DC current sensor, a temperature sensor and a vibration sensor, and the sensors have an adaptive sampling rate and intelligent data preprocessing. The adaptive sampling rate is dynamically adjusted according to system load and environmental changes. The intelligent data preprocessing includes data filtering, denoising and feature extraction; the multi-dimensional operating data includes voltage, current, temperature and vibration.

[0011] Furthermore, the edge computing node is configured with a processor and supports real-time data processing, which includes data aggregation, preliminary analysis and anomaly detection. The processor includes a multi-core CPU and an acceleration unit, and the acceleration unit is a GPU or FPGA.

[0012] Furthermore, the graph neural network model module adopts an architecture that combines a graph convolutional network and a graph attention network, and the graph attention network is used to dynamically adjust the weights of neighbor nodes; The graph neural network model module is trained using historical fault data and simulation-generated datasets through a supervised learning method, and adopts a cross-entropy loss function and regularization to optimize classification performance. The supervised learning includes multi-category classification to distinguish different types of faults.

[0013] Furthermore, the online reasoning module uses TensorFlow Lite or ONNX Runtime to achieve real-time fault detection and location, and is deployed through edge computing nodes. The online reasoning module is used to process high-frequency data streams under low-latency conditions and respond in real time.

[0014] Furthermore, the intelligent decision-making module continuously optimizes the fault response strategy through simulation training and real-time feedback based on a deep reinforcement learning algorithm. The deep reinforcement learning algorithm includes a policy gradient method and a Q-learning method, which are used to adaptively adjust the control strategy in a dynamic environment.

[0015] Furthermore, the augmented reality-assisted operation and maintenance module includes real-time display of fault location results, system topology diagrams and maintenance instructions through AR glasses or mobile devices, which is used to support remote expert guidance and intelligent operations of on-site operation and maintenance personnel. The augmented reality-assisted operation and maintenance module provides three-dimensional visualization of fault point location and equipment status information through integration with the digital twin model.

[0016] The present invention is based on an intelligent fault location method for a DC system, comprising the following steps: Generate a digital twin model of the DC system using simulation tools based on the DC system's physical model and historical operating data. Real-time collection of multi-dimensional operating data including voltage, current, temperature and vibration; The collected real-time data is pre-processed by edge computing nodes and then transmitted to the digital twin model for real-time data synchronization between the physical system and the digital twin. Using the system topology and the feature vectors of nodes and edges, we train a graph neural network model for fault detection and location. This model is then deployed on edge computing nodes, where real-time data is used for fault detection and location. Based on the fault detection results of the graph neural network model, a reinforcement learning algorithm is used to generate a decision-making self-healing strategy for fault response; Perform control operations such as fault isolation, system switching, and load adjustment based on decision-making self-healing strategies.

[0017] The present invention has the following beneficial effects: 1. The present invention deeply integrates digital twin technology with graph neural networks to comprehensively analyze the complex electrical, thermal, mechanical, and environmental parameters in DC systems. The digital twin module provides a high-precision virtual model of the system through multi-physics field integration, enabling the graph neural network (GNN) model to accurately capture subtle anomalies and complex fault modes in the system with the support of rich and high-quality data. The combination of graph convolutional networks and graph attention networks enables the GNN model to effectively identify the correlation between key nodes and edges, dynamically adjust the weights of neighboring nodes, and further improve the accuracy of fault detection and location. Through supervised learning methods, using historical fault data and simulation-generated data sets for training, the GNN model has strong generalization and multi-category classification capabilities, can accurately distinguish different types of faults, significantly reduce false alarm and missed alarm rates, and improve the overall reliability and safety of the system.

[0018] 2. The present invention adopts edge computing nodes for real-time data preprocessing and GNN model reasoning, and utilizes high-performance multi-core CPUs and GPU or FPGA acceleration units to achieve low-latency fault detection and positioning. The online reasoning module is deployed on the edge computing node through TensorFlow Lite or ONNX Runtime, which can respond quickly under high-frequency data streams to ensure the immediacy of fault detection and response. Based on the deep reinforcement learning algorithm, the intelligent decision-making and self-healing control module can generate the optimal fault response strategy, continuously optimize the fault response strategy through simulation training and real-time feedback, and realize adaptive control in a dynamic environment. The self-healing execution module automatically performs fault isolation, system switching and load adjustment operations through industrial protocols such as Modbus and OPC UA, quickly restores the normal operation of the system, prevents the spread of faults, ensures the continuous and stable operation of the DC system, significantly shortens the system downtime, and improves power supply reliability.

[0019] 3. The present invention integrates an augmented reality-assisted operation and maintenance module and a blockchain recording module, which significantly improves the work efficiency of operation and maintenance personnel and the data security of the system. The augmented reality-assisted operation and maintenance module uses AR glasses or mobile devices to display fault location results, system topology maps and maintenance instructions in real time, supports remote expert guidance and intelligent operations of on-site operation and maintenance personnel, reduces misoperations caused by incomplete information, and improves the accuracy and efficiency of fault handling. The record tracing module automatically records all fault events and response operations to the Hyperledger Fabric blockchain through smart contracts, ensuring the immutability and traceability of data, and enhancing the data security and credibility of the system. The application of blockchain technology not only ensures the integrity of fault records, but also provides reliable data support for subsequent fault analysis and system optimization. At the same time, the intelligent decision-making module continuously optimizes the fault response strategy through continuous feedback and learning mechanisms, improves the system's adaptability and intelligence level, and comprehensively improves the operational efficiency and maintenance management level of the DC system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a system structure diagram of the present invention; Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] Example 1 like Figure 1 As shown in the figure, the intelligent fault location system based on DC system includes the following parts: The digital twin module uses simulation tools to generate a digital twin of the DC system based on the system's physical model and historical operating data. A digital twin is a highly accurate virtual replica of the physical DC system, fully reflecting the system's operating status through real-time data synchronization and multi-physics integration. Building a digital twin requires modeling based on the DC system's physical model and historical operating data using simulation tools such as MATLAB / Simulink and COMSOL Multiphysics.

[0023] The digital twin model integrates electrical, thermal, mechanical, and environmental parameters to fully reflect the operating status of the DC system. By coupling equations and models of different physical phenomena, it enables collaborative simulation of multiple physical fields. The electrical part uses Kirchhoff's voltage law (KVL) and current law (KCL) to describe circuit behavior: ; The thermal part describes the temperature change of the device through the heat conduction equation: ; The mechanical part uses Newton's second law to describe vibrations and stresses: ; ; Environmental parameters are coupled with the above physical fields through empirical models or physical equations to ensure that the model can accurately simulate the operating status of the system under different environmental conditions.

[0024] Real-time data synchronization is achieved through edge computing nodes. Multi-dimensional data, such as voltage, current, temperature, and vibration, collected by the sensor network is first preprocessed at the edge computing nodes, including data filtering, denoising, and feature extraction. This preprocessed data is then transmitted to the digital twin model via high-bandwidth, low-latency communication protocols such as MQTT, achieving real-time synchronization with the physical system.

[0025] Bidirectional data flow ensures that the digital twin model dynamically reflects the latest state of the physical system and supports model-based prediction and optimization. For example, when a sensor detects an abnormal increase in current, the real-time data synchronization module will update the current distribution in the digital twin model in real time, triggering further fault analysis and response strategies.

[0026] The sensor network module installs sensors at key nodes of the DC system to collect multi-dimensional operating data such as voltage, current, temperature, and vibration in real time.

[0027] The sensor network module includes a DC voltage sensor, a DC current sensor, a temperature sensor, and a vibration sensor. These sensors feature adaptive sampling rates and intelligent data preprocessing to adapt to system load and environmental changes. The adaptive sampling rate dynamically adjusts based on these changes. For example, when the system load is high or the ambient temperature fluctuates dramatically, the sensor increases the sampling rate to capture more detailed data.

[0028] Intelligent data preprocessing includes data filtering, denoising, and feature extraction. It uses digital Kalman filters to remove sensor noise and extract key features such as voltage spikes, temperature gradients, and vibration frequencies for subsequent analysis: ; in, is the filtered state estimate, is the Kalman gain, is the observed value, is the observation matrix.

[0029] The real-time data synchronization module pre-processes the real-time data collected by the sensor network module through the edge computing node and transmits it to the digital twin model for real-time data synchronization between the physical system and the digital twin. The edge computing node is equipped with a high-performance processor that supports real-time data processing and model inference. Real-time data processing includes data aggregation, preliminary analysis, and anomaly detection. The high-performance processor includes a multi-core CPU and an acceleration unit, and the acceleration unit is a GPU or FPGA. The edge computing node first aggregates and performs preliminary analysis on the collected data to identify potential abnormal data points: ; in, is the observation data, is the system matrix, is the system status, Anomaly detection uses statistical anomaly detection or neural network machine learning algorithms to perform preliminary anomaly detection, filter out non-fault anomalies, and improve the detection accuracy of subsequent graph neural network (GNN) models.

[0030] The graph neural network (GNN) model module uses the system topology graph and the feature vectors of nodes and edges to train the GNN model for fault detection and location.

[0031] The DC system's electrical topology is generated into a graphical model using graph processing tools such as NetworkX. Nodes in the system topology represent equipment such as converters, circuit breakers, and cables, while edges represent electrical connections. Each node and edge contains a feature vector reflecting its operating status and connection characteristics. The GNN model utilizes an architecture that combines a graph convolutional network (GCN) and a graph attention network (GAT) to enhance focus on key nodes and edges, improving the accuracy of fault detection and localization.

[0032] GCN extracts local features of nodes through convolution operations: ; in, , To add the self-connected adjacency matrix, is the degree matrix, For the Layer node features, is the weight matrix, is the activation function.

[0033] GAT dynamically adjusts the weights of neighboring nodes through the attention mechanism to improve the model's sensitivity to important connections; ; : represents the attention weight between node i and node j, indicating the importance of node j to node i; : exponential function used to normalize the attention score to a probability distribution; is the activation function; Used to introduce nonlinearity and avoid complete neuronal inactivation; The weight vector of the attention mechanism (transposed), used to calculate the attention score of the relationship between nodes; A learnable weight matrix for linearly transforming node features; represent the feature vectors of node i and node i respectively; Indicates that and Splice into a vector; The neighbor set of node i, that is, the nodes directly connected to i; The neighbor set of node i, that is, the nodes directly connected to i.

[0034] The GNN model is trained using supervised learning methods, utilizing historical fault data and simulation-generated datasets. A cross-entropy loss function and regularization techniques are used to optimize classification performance. Supervised learning includes multi-class classification to distinguish different types of faults, improving the model's adaptability and generalization capabilities.

[0035] ; is the true label, is the predicted probability, is the regularization parameter, is the model weight.

[0036] The online inference module deploys the trained GNN model on the edge computing node and uses real-time data for fault detection and location. The online inference module uses TensorFlowLite or ONNXRuntime to achieve real-time fault detection and location and is deployed through the edge computing node. The online inference module is used to process high-frequency data streams under low-latency conditions and respond in real time.

[0037] The intelligent decision-making module uses a reinforcement learning algorithm based on the fault detection results of the GNN model to generate a decision-making self-healing strategy for fault response. The intelligent decision-making module is based on a deep reinforcement learning algorithm. Through simulation training and real-time feedback, it continuously optimizes the fault response strategy and realizes intelligent decision support. The deep reinforcement learning algorithm includes the policy gradient method and the Q-learning method, which are used to adaptively adjust the control strategy in a dynamic environment.

[0038] ; in, is the state-action value function, is the learning rate, For instant rewards, is the discount factor, For the next state, For the next action.

[0039] The decision-making self-healing module includes a self-healing execution module, a record tracing module, and an augmented reality-assisted operation and maintenance module. The self-healing execution module communicates with DC system equipment through industrial protocols such as Modbus (OPC UA) to automatically perform control operations such as fault isolation, system switching, and load adjustment. The specific operation process is as follows: Fault isolation: Upon detecting a fault, the system automatically identifies the faulty area and isolates the faulty portion by disconnecting the relevant circuit breakers or switches to prevent the fault from spreading. For example, if a short circuit is detected on a power transmission line, the system uses the GNN model to locate the faulty node, sends a signal to the edge computing node, and automatically controls the circuit breaker to disconnect the line.

[0040] System switching: After fault isolation, the system automatically switches to a backup power supply path or device to ensure power continuity. For example, if the main converter fails, the system automatically starts the backup converter and reconfigures the electrical topology to achieve seamless switchover.

[0041] Load adjustment dynamically adjusts load distribution based on system status and fault conditions, optimizing energy utilization and preventing system overload. A real-time load management algorithm is used to distribute or reduce load in specific areas based on current power supply capacity and load demand.

[0042] The record and traceability module records fault detection and response data to the blockchain to ensure the data is tamper-proof and traceable. The specific implementation is as follows: Data is uploaded to the blockchain, and through smart contracts, key events such as fault detection time, location, and response measures are automatically recorded on the blockchain to ensure data transparency and security.

[0043] Data query and audit, using the distributed ledger characteristics of blockchain, supports querying and auditing historical records of failure events, improving the credibility and traceability of the system.

[0044] The augmented reality (AR) assisted operation and maintenance module displays fault location results and maintenance instructions through augmented reality devices. The AR assisted operation and maintenance module includes real-time display of fault location results, system topology diagrams, and maintenance instructions through AR glasses or mobile devices to support remote expert guidance and intelligent operation of on-site operation and maintenance personnel. The AR assisted operation and maintenance module provides three-dimensional visualization of fault point location and equipment status information through integration with digital twin models. Specific functions include: Real-time display uses AR devices to display the fault location, equipment status and system topology in the digital twin model, helping operation and maintenance personnel to intuitively understand the fault situation.

[0045] Maintenance instructions provide maintenance steps and operation instructions based on fault types, and combine with digital twin models to guide operation and maintenance personnel to perform precise maintenance operations.

[0046] Remote collaboration allows remote experts to provide real-time guidance to on-site operation and maintenance personnel through AR devices, improving the efficiency and accuracy of troubleshooting.

[0047] Example 2 The sensor network module includes a DC voltage sensor, a DC current sensor, a temperature sensor, and a vibration sensor. The specific configuration is as follows: DC voltage sensor: uses a high-precision, low-drift fluxgate effect sensor to monitor system voltage in real time.

[0048] DC current sensor: uses Hall effect sensor to accurately measure the magnitude and direction of current.

[0049] Temperature sensor: Use high-precision RTD or thermocouple to monitor key equipment and ambient temperature.

[0050] Vibration sensor: uses an accelerometer to detect the vibration status of the equipment and warn of mechanical failures.

[0051] Edge computing nodes are equipped with high-performance processors and software modules that support real-time data processing. Under specific configurations, High-performance processor: including multi-core CPU (ARM Cortex-A series) and acceleration unit (GPU or FPGA), supporting efficient data processing and GNN model inference.

[0052] Software module: runs real-time data processing, GNN model inference, and control operations, using containerization technology (Docker) for modular deployment and management.

[0053] The communication network uses a high-bandwidth, low-latency protocol. Under specific configuration, Communication protocol: MQTT protocol is used for data transmission, supporting efficient and reliable message delivery.

[0054] Industrial protocols: Modbus and OPC UA protocols are used to communicate with DC system equipment to achieve remote control and data exchange.

[0055] The blockchain node is deployed on the edge computing node. Under specific configuration, Blockchain platform: Using Hyperledger Fabric or Ethereum 2.0 to support high throughput and low latency data on-chain requirements.

[0056] Smart Contract: Develop dedicated smart contracts to achieve automatic recording of failure events, data access control, and permission management.

[0057] Augmented reality devices include AR glasses and mobile devices. Under specific configurations, AR glasses: Microsoft HoloLens, integrated display modules and sensors, support real-time data display and interactive operations.

[0058] Mobile devices: Smartphones or tablets with dedicated AR applications installed to support mobile operations and remote collaboration for on-site operation and maintenance personnel.

[0059] Example 3 A method for locating a fault based on an intelligent fault location system of a DC system includes the following steps: Generate a digital twin model of the DC system using simulation tools based on the DC system's physical model and historical operating data. Real-time collection of multi-dimensional operating data including voltage, current, temperature and vibration; After filtering, denoising, and feature extraction of the collected real-time data through edge computing nodes, the data is transmitted to the digital twin model via the MQTT protocol for real-time data synchronization between the physical system and the digital twin. Using the system topology and the feature vectors of nodes and edges, we train a GNN model for fault detection and location. The trained GNN model is deployed on edge computing nodes, and real-time data is used for fault detection and location. Based on the fault detection results of the GNN model, a reinforcement learning algorithm is used to generate a decision-making self-healing strategy for fault response; Based on the decision-making self-healing strategy, it communicates with DC system equipment through Modbus or OPC UA protocols to perform control operations such as fault isolation, system switching, and load adjustment to ensure stable system operation.

[0060] Fault events and response data are recorded through blockchain to ensure the data’s immutability and traceability, supporting subsequent auditing and analysis.

[0061] The augmented reality-assisted operation and maintenance module uses AR devices to display fault location results and maintenance instructions, supports remote expert guidance and intelligent operations of on-site operation and maintenance personnel, and improves fault handling efficiency and accuracy.

[0062] The operations and feedback from the operation and maintenance personnel are transmitted back to the GNN model training module through the feedback mechanism, continuously optimizing the model performance and decision-making strategy, and improving the system's adaptability and intelligence level.

[0063] During system operation, the sensor network continuously collects multi-dimensional operational data, and edge computing nodes pre-process and synchronize the data to the digital twin model in real time. The GNN model uses the system topology map and feature vectors to perform real-time fault detection and location. When a fault is detected, the intelligent decision-making module generates the optimal self-healing control strategy based on the reinforcement learning algorithm, and automatically performs fault isolation, system switching, and load adjustment operations through industrial protocols. At the same time, the fault event is recorded in the blockchain to ensure data security and traceability. Operations and maintenance personnel obtain real-time fault location information and maintenance instructions through augmented reality devices, supporting intelligent operation and remote collaboration. The entire system continuously improves the accuracy and response speed of fault location through continuous learning and optimization mechanisms, ensuring the efficient, stable, and intelligent operation of the DC system.

[0064] Application Test Example 1 A cross-provincial high-voltage direct current (HVDC) transmission line is responsible for transmitting electricity generated by a large-scale wind farm in Inner Mongolia to major load centers in and around Beijing. The line, spanning 1,500 kilometers, traverses a complex geographical landscape. Traditional fault detection and location methods struggle to meet the requirements for efficiency and accuracy, often resulting in delayed fault response and inaccurate location, impacting power supply stability and economic efficiency.

[0065] Research and Evaluation: State Grid Corporation of China, in collaboration with technology providers, conducted a detailed investigation of the HVDC system's operating status, key nodes, and common fault types, identifying specific requirements and implementation objectives for the intelligent fault location system.

[0066] Solution design: A design scheme for an intelligent fault location system based on digital twins and graph neural networks (GNN) was developed, and the required hardware equipment, software modules, and communication architecture were clarified.

[0067] High-precision modeling: Use MATLAB / Simulink to build an electrical model of the HVDC line, covering the dynamic characteristics of key equipment such as converter stations, transmission cables, and circuit breakers.

[0068] Multi-physics integration: Integrate electrical, thermal, mechanical, and environmental parameters into the digital twin model through COMSOL Multiphysics to simulate system operation under different environmental conditions, such as high temperature, high humidity, and wind and sand.

[0069] Sensor Installation: DC voltage, DC current, temperature, and vibration sensors are installed at key nodes along the HVDC line, including converter stations, circuit breakers, and transmission cable sections crossing mountains and rivers. High-precision, low-drift fluxgate and Hall-effect sensors are used to ensure accurate data collection.

[0070] Adaptive sampling rate configuration: The sensor is configured to dynamically adjust its sampling rate based on system load and environmental changes. When wind power output fluctuates significantly or ambient temperature changes dramatically, the sensor automatically increases the sampling frequency to capture more detailed data.

[0071] Device deployment: NVIDIA Jetson AGX Xavier edge computing devices are deployed at each key node, equipped with a multi-core ARM Cortex-A72 CPU and a 256-core Volta GPU to accelerate data processing and GNN model inference.

[0072] Software Installation: Edge nodes run Docker-based containerized software modules, including real-time data processing, data filtering, denoising, and feature extraction functions to ensure efficient data preprocessing.

[0073] Data Collection: Continuously collect HVDC system operating data, including normal operation data and historical fault data, through a sensor network. Use the digital twin model to generate simulated fault datasets to expand the amount of training data.

[0074] Model Design and Training: We designed a GNN architecture combining a graph convolutional network (GCN) and a graph attention network (GAT). We used NetworkX to generate a system topology diagram, including the electrical connections between nodes. We used supervised learning to train the GNN model on historical fault data and simulation data, optimizing the cross-entropy loss function and L2 regularization.

[0075] Model deployment: Export the trained GNN model to TensorFlow Lite format, deploy it to edge computing nodes, and use TensorFlow Lite Runtime to achieve efficient online inference.

[0076] Reinforcement learning integration: Deep reinforcement learning (DRL) algorithms (such as Deep Q-Network, DQN) are integrated into the intelligent decision-making module to optimize fault response strategies through simulation training and real-time feedback.

[0077] Automatic Control: When the GNN model detects a fault, the DRL algorithm generates an optimal fault response strategy, including automatically disconnecting the faulty line, activating backup converters, and adjusting load distribution. The self-healing execution module communicates with the HVDC equipment via Modbus protocol to automatically execute these control actions, ensuring continuous and stable system operation.

[0078] Data on-chain: All failure events and response operations are automatically recorded on the Hyperledger Fabric blockchain through smart contracts, ensuring the immutability and traceability of data.

[0079] Audit and Query: Operation managers and auditors can query historical fault records through blockchain to analyze fault causes and optimize the system.

[0080] AR device usage: Operations and maintenance personnel wear Microsoft HoloLens AR glasses to view fault location results, system topology maps, and maintenance instructions in real time.

[0081] Remote collaboration: On-site operation and maintenance personnel can share real-time video and data with remote experts through AR devices, receive immediate guidance and support, and improve the efficiency and accuracy of troubleshooting.

[0082] Improved fault location accuracy: The GNN model combines system topology and multi-dimensional sensor data to achieve high-precision fault location and reduce positioning errors. Faster response: The real-time reasoning capabilities of edge computing nodes shorten fault detection and response times, significantly reducing system downtime. Improved operation and maintenance efficiency: The augmented reality-assisted operation and maintenance module improves the efficiency and accuracy of operation and maintenance personnel and reduces errors caused by incomplete information. Enhanced data security: Through blockchain technology, all fault records are tamper-proof and traceable, improving the overall security and credibility of the system. Improved system stability: Intelligent decision-making and self-healing control mechanisms effectively prevent the spread of faults and ensure the continued stable operation of the HVDC system.

[0083] Application Test Example 2 A city operates an electric vehicle (EV) charging network, jointly built by the municipal government and private enterprises. This network encompasses over 5,000 charging stations located in public parking lots, commercial areas, and residential areas. Due to the large number and widespread distribution of charging stations, traditional fault detection and location methods struggle to detect and locate faults in a timely manner, resulting in a poor charging experience for users and hindering both EV adoption and user satisfaction.

[0084] Research and evaluation: Urban transportation management departments work with charging station operators to evaluate the operating status of the existing charging network and identify common fault types such as critical charging stations and charging station downtime, low charging efficiency, and power failure.

[0085] Solution design: Develop a design plan for an intelligent fault location system based on digital twins and graph neural networks (GNNs), and clarify the required hardware equipment, software modules, and communication architecture.

[0086] High-precision modeling: Use MATLAB / Simulink to build an electrical model of the charging network, covering the dynamic characteristics of key components such as charging piles, power supply equipment, and communication equipment.

[0087] Multi-physics integration: Thermal and mechanical parameters are integrated through COMSOL Multiphysics to simulate the temperature changes and mechanical stress of the charging pile under different operating conditions, ensuring that the digital twin model can accurately reflect the actual operating status.

[0088] Sensor Installation: DC voltage sensors, DC current sensors, temperature sensors, and vibration sensors are installed at each charging station. High-precision resistance temperature detectors (RTDs) and accelerometers are used to ensure accurate and reliable data collection.

[0089] Adaptive sampling rate configuration: The sensor dynamically adjusts the sampling rate based on charging load and environmental conditions. For example, under high charging load or when the ambient temperature fluctuates drastically, the sampling frequency is increased to capture more data details.

[0090] Device deployment: NVIDIA Jetson Nano edge computing devices are deployed in the charging station area. They are equipped with a multi-core ARM Cortex-A57 CPU and a 128-core Maxwell GPU to accelerate data processing and GNN model inference.

[0091] Software Installation: Edge nodes run Docker-based containerized software modules, including real-time data processing, data filtering, denoising, and feature extraction functions to ensure efficient data preprocessing.

[0092] Data Collection: Continuously collect charging pile operating data, including normal operation data and historical fault data, through a sensor network. Use the digital twin model to generate simulated fault datasets to expand the amount of training data.

[0093] Model Design and Training: We designed a GNN architecture combining a graph convolutional network (GCN) and a graph attention network (GAT). We used NetworkX to generate a charging network topology diagram, including the electrical connections between nodes. We used supervised learning to train the GNN model on historical fault data and simulation data, optimizing the cross-entropy loss function and L2 regularization.

[0094] Model deployment: Export the trained GNN model to TensorFlow Lite format, deploy it to edge computing nodes, and use TensorFlow Lite Runtime to achieve efficient online inference.

[0095] Reinforcement learning integration: Integrate deep reinforcement learning (DRL) algorithms (such as DQN) into the intelligent decision-making module to optimize fault response strategies through simulation training and real-time feedback.

[0096] Automatic Control: When the GNN model detects a charging station fault, the DRL algorithm generates an optimal fault response strategy, including automatically restarting the charging station, switching to a backup power source, and adjusting load distribution. The self-healing execution module communicates with the charging station device via the Modbus protocol to automatically execute these control operations, ensuring the continuous and stable operation of the charging network.

[0097] Data on-chain: All failure events and response operations are automatically recorded on the Hyperledger Fabric blockchain through smart contracts, ensuring the immutability and traceability of data.

[0098] Audit and Query: Operation managers and auditors can query historical fault records through blockchain to analyze fault causes and optimize the system.

[0099] AR device usage: Operations and maintenance personnel wear Microsoft HoloLens AR glasses to view fault location results, system topology maps, and maintenance instructions in real time.

[0100] Remote collaboration: On-site operation and maintenance personnel can share real-time video and data with remote experts through AR devices, receive immediate guidance and support, and improve the efficiency and accuracy of troubleshooting.

[0101] Improved fault location accuracy: The GNN model combines the charging network topology and multi-dimensional sensor data to achieve high-precision positioning of the fault point and reduce positioning errors. Faster response speed: The real-time reasoning capability of the edge computing node shortens fault detection and response time, significantly reducing charging pile downtime. Improved operation and maintenance efficiency: The augmented reality-assisted operation and maintenance module improves the work efficiency and accuracy of operation and maintenance personnel, reduces misoperations caused by incomplete information, and improves overall operation and maintenance efficiency. Improved user satisfaction: The timeliness of fault location and response is improved, reducing the time users wait for charging, improving the overall charging experience and user satisfaction. Enhanced data security: Through blockchain technology, all fault records are tamper-proof and traceable, improving the overall security and credibility of the system. Improved system reliability: Intelligent decision-making and self-healing control mechanisms effectively prevent the spread of faults in the charging network and ensure the continuous and stable operation of the charging system.

[0102] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further restrictions, an element defined by the statement "comprising a reference structure" does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0103] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. Intelligent fault location system based on DC system, characterized by: include: The digital twin module uses simulation tools to generate a digital twin model of the DC system based on the DC system's physical model and historical operating data. Sensor network module, real-time collection of multi-dimensional operation data; The real-time data synchronization module pre-processes the real-time data collected by the sensor network module through the edge computing node and transmits it to the digital twin model for real-time data synchronization between the physical system and the digital twin; A graph neural network model module, which uses a system topology graph and feature vectors of nodes and edges to train a graph neural network model for fault detection and location. The nodes in the system topology graph represent devices, and the edges represent electrical connections. The online inference module deploys the trained graph neural network model on edge computing nodes and uses real-time data for fault detection and location. The intelligent decision-making module uses a reinforcement learning algorithm based on the fault detection results of the graph neural network model to generate a decision-making and self-healing strategy for fault response; The decision-making self-healing module executes control operations based on the decision-making self-healing strategy of the intelligent decision-making module.

2. The intelligent fault location system based on DC system according to claim 1, characterized in that: The decision-making self-healing module includes: The self-healing execution module communicates with DC system equipment through industrial protocols to automatically perform control operations such as fault isolation, system switching, and load adjustment; Record and traceability module, which records fault detection and response data to the blockchain; The augmented reality-assisted operation and maintenance module displays fault location results and maintenance instructions through augmented reality devices.

3. The intelligent fault location system based on DC system according to claim 1, characterized in that: The digital twin module includes a physical field integration of electrical, thermal, mechanical and environmental parameters to comprehensively reflect the operating status of the DC system. The physical field integration couples equations and models of different physical phenomena.

4. The intelligent fault location system based on DC system according to claim 1, characterized in that: The sensor network module includes a DC voltage sensor, a DC current sensor, a temperature sensor and a vibration sensor, and the sensors are equipped with an adaptive sampling rate and intelligent data preprocessing. The adaptive sampling rate is dynamically adjusted according to system load and environmental changes. The intelligent data preprocessing includes data filtering, denoising and feature extraction; the multi-dimensional operating data includes voltage, current, temperature and vibration.

5. The intelligent fault location system based on DC system according to claim 1, characterized in that: The edge computing node is configured with a processor and supports real-time data processing, which includes data aggregation, preliminary analysis and anomaly detection. The processor includes a multi-core CPU and an acceleration unit, and the acceleration unit is a GPU or FPGA.

6. The intelligent fault location system based on DC system according to claim 1, characterized in that: The graph neural network model module adopts an architecture that combines a graph convolutional network and a graph attention network. The graph attention network is used to dynamically adjust the weights of neighbor nodes. The graph neural network model module uses a supervised learning method to train historical fault data and simulation-generated data sets, and adopts a cross-entropy loss function and regularization to optimize classification performance. The supervised learning includes multi-category classification to distinguish different types of faults.

7. The intelligent fault location system based on DC system according to claim 1, characterized in that: The online reasoning module uses TensorFlow Lite or ONNX Runtime to achieve real-time fault detection and location, and is deployed through edge computing nodes. The online reasoning module is used to process high-frequency data streams under low-latency conditions and respond in real time.

8. The intelligent fault location system based on DC system according to claim 1, characterized in that: The intelligent decision-making module is based on a deep reinforcement learning algorithm that continuously optimizes fault response strategies through simulation training and real-time feedback. The deep reinforcement learning algorithm includes a policy gradient method and a Q-learning method for adaptively adjusting control strategies in dynamic environments.

9. The intelligent fault location system based on DC system according to claim 2, characterized in that: The augmented reality-assisted operation and maintenance module includes real-time display of fault location results, system topology diagrams and maintenance instructions through AR glasses or mobile devices, which is used to support remote expert guidance and intelligent operations of on-site operation and maintenance personnel. The augmented reality-assisted operation and maintenance module provides three-dimensional visualization of fault point location and equipment status information through integration with the digital twin model.

10. The method for locating a fault based on an intelligent fault location system of a DC system according to any one of claims 1 to 9, characterized in that: The steps include: Generate a digital twin model of the DC system using simulation tools based on the DC system's physical model and historical operating data. Real-time collection of multi-dimensional operating data including voltage, current, temperature and vibration; The collected real-time data is pre-processed by edge computing nodes and then transmitted to the digital twin model for real-time data synchronization between the physical system and the digital twin. Using the system topology and the feature vectors of nodes and edges, we train a graph neural network model for fault detection and location. This model is then deployed on edge computing nodes, where real-time data is used for fault detection and location. Based on the fault detection results of the graph neural network model, a reinforcement learning algorithm is used to generate a decision-making self-healing strategy for fault response; Perform control operations such as fault isolation, system switching, and load adjustment based on decision-making self-healing strategies.

Images