Power distribution network multistage fault rapid positioning and self-healing control method and device and medium

Abstract

The invention belongs to the technical field of power system power distribution network automation, and relates to a power distribution network multistage fault rapid positioning and self-healing control method and device based on digital twinborn and hybrid enhanced intelligence, and a medium, and the method comprises the steps: constructing a digital twinborn body of a power distribution network, and carrying out the panoramic perception and simulation of the operation state of the power distribution network; multi-mode operation data are collected in real time through intelligent sensing devices deployed at all nodes of the power distribution network, and the multi-mode operation data comprise electrical quantity data, equipment state data and environment data; judging the fault occurrence possibility and the fault type, triggering dynamic virtual topology reconstruction, and generating a virtual network corresponding to the current operation mode in the digital twin; starting a multi-stage cooperative fault positioning process; generating a self-healing control strategy set through hybrid enhanced intelligent decision based on the accurate physical position of the fault point, the fault property and the real-time state of the power grid simulated by the digital twin; performing strategy security execution and closed loop correction; and carrying out self-evolution learning.

Description

Technical Field

[0001] This invention belongs to the field of power system distribution network automation technology, specifically relating to a method, device and medium for rapid location and self-healing control of multi-level faults in distribution networks based on digital twins and hybrid enhanced intelligence. Background Technology

[0002] With the continuous expansion of distribution network scale and the large-scale integration of distributed energy resources, the complexity and uncertainty of distribution networks have increased significantly. Traditional distribution network fault handling mainly suffers from the following technical shortcomings: 1. Insufficient fault location accuracy: Existing methods mostly rely on single technical means, such as simple overcurrent protection or impedance method, which are prone to misjudgment in complex multi-control power grids, resulting in limited location accuracy.

[0003] 2. Slow response speed: It usually takes several minutes or even longer from the occurrence of a fault to the completion of location, isolation and restoration, which cannot meet the high requirements of modern power distribution networks for power supply reliability.

[0004] 3. Limited self-healing strategies: Traditional self-healing control is mostly based on preset rules and lacks adaptability to the real-time state of the power grid. Especially in scenarios with a high proportion of distributed power sources, the effectiveness of the strategy is greatly reduced.

[0005] 4. Low level of system intelligence: The existing system lacks self-learning and self-evolution capabilities, cannot accumulate experience from historical failures, and it is difficult to continuously improve system performance.

[0006] Digital twin technology offers a new solution to these problems. By constructing a virtual mapping of the physical power distribution network, it is possible to achieve real-time perception of the network's operating status, accurate simulation of fault processes, and virtual verification of self-healing strategies. However, how to deeply integrate digital twins with artificial intelligence technology to build an intelligent fault handling system with continuous learning capabilities remains a challenge for current technology. Summary of the Invention

[0007] The purpose of this invention is to provide a method, device, and medium for rapid fault location and self-healing control of multi-level faults in distribution networks based on digital twins and hybrid enhanced intelligence, addressing the problems existing in the prior art, so as to achieve rapid and accurate fault location and intelligent adaptive recovery of distribution networks.

[0008] The technical solution of this invention is: This invention constructs an integrated intelligent system of "perception-cognition-decision-execution-learning". The core of the system is a virtual mapping of the power distribution network based on digital twins. Through multi-source data fusion, multi-level collaborative positioning, hybrid enhanced decision-making and closed-loop feedback correction, the entire process of fault handling is made intelligent.

[0009] The construction of a digital twin, which is the basic platform of this invention, includes: Physical model layer: Establish a physical model of the distribution network that includes detailed electrical parameters such as line parameters, transformer models, switching equipment, and distributed power sources.

[0010] Data-driven layer: Integrates real-time data from multiple sources, including SCADA systems, fault indicators, smart meters, and PMUs.

[0011] Simulation analysis layer: It has simulation capabilities such as power flow calculation, short circuit calculation, and stability analysis.

[0012] Visualization and Interaction Layer: Provides functions such as power grid status visualization, fault process reproduction, and strategy simulation display. The digital twin maintains synchronization with the physical distribution network status through real-time data injection and adaptive adjustment of model parameters.

[0013] Multi-level collaborative fault location Level 1: Wide-area perception and positioning, based on fault marker information such as overcurrent and grounding reported by power distribution terminals; combined with dynamic virtual topology, graph theory algorithm is used to search for fault areas; a hybrid algorithm of depth-first search (DFS) and breadth-first search (BFS) is used to determine the smallest suspected fault area within hundreds of milliseconds; the output is one or more smallest topological areas that may contain fault points.

[0014] Level 2: Sectional directional collaborative verification. Within the suspected area, the alarm information from the fault indicator is comprehensively analyzed; the direction and amplitude of the fault current at both ends of the line are compared; a preliminary traveling wave arrival time scale is introduced as an auxiliary judgment; and a multi-source information fusion algorithm based on confidence weighting is adopted. Confidence score = w1 × I_dir + w2 × I_mag + w3 × FI + w4 × TW Wherein, I_dir is the directional consistency score, I_mag is the amplitude matching degree, FI is the fault indicator status score, TW is the traveling wave arrival time reasonableness score, and w1-w4 are adaptive weighting coefficients. When the comprehensive score exceeds the threshold, it is determined to be a faulty section.

[0015] Level 3: Node traveling wave impedance fusion localization. Within the determined fault section, high-precision dual-end / multi-end traveling wave ranging is initiated; simultaneously, full-frequency domain impedance analysis is used to calculate the fault distance; a data assimilation algorithm based on Kalman filtering is developed to fuse the traveling wave and impedance method results. X_final=(W_tw×X_tw+W_z×X_z) / (W_tw+W_z) Where X_tw and X_z are the calculation results of traveling wave and impedance method, respectively, and W_tw and W_z are weights dynamically adjusted based on measurement accuracy and environmental conditions. Combining the coordinates of the line towers, cable joint locations, and historical fault distributions, the precise physical location of the fault point is determined.

[0016] Hybrid augmented intelligent decision-making includes a rules engine module, which stores power grid operation procedures and expert experience rules to generate basic, conservative but absolutely safe fault isolation strategies, ensuring that there are feasible strategies available in any situation.

[0017] The deep reinforcement learning module has a state space that includes network topology, load distribution, distributed power output, and fault location; an action space that includes switching on and off, and DG control mode switching; a reward function that considers multiple objectives such as the amount of load restored, the number of switching operations, network loss, and voltage quality; and large-scale offline training and online fine-tuning are performed in a digital twin.

[0018] The multi-attribute utility decision-making module establishes an evaluation index system that includes multiple dimensions such as technicality, economy, and reliability; it uses the analytic hierarchy process (AHP) to determine the weight of each index, and uses the TOPSIS method to rank candidate strategies, outputting the comprehensive optimal self-healing control strategy.

[0019] The secure execution and closed-loop correction mechanism, its virtual verification process includes the following steps: 1. Visualize the entire strategy execution process in a digital twin; 2. Simulate changes in the power grid state after strategy execution; 3. Provide expected effect assessment and risk warnings; 4. Execute after virtual confirmation from the dispatcher.

[0020] The closed-loop correction mechanism specifically includes: 1. Real-time monitoring of the power grid status after the strategy is executed; 2. Comparing the actual status with the digital twin's predicted status; 3. If the deviation exceeds the threshold, initiating strategy correction; 4. Recording deviation information for system self-learning.

[0021] The self-evolutionary learning mechanism stores the complete process data for each fault, including fault characteristics, localization process, execution strategy, and actual effect, and establishes a multi-dimensional labeling system to support efficient retrieval.

[0022] The model update strategy includes: regularly updating the parameters of the deep learning model with new cases; accelerating model adaptation through transfer learning based on case similarity; and using an online learning mechanism to quickly optimize for frequently occurring failure modes.

[0023] The multi-level fault handling scheme for distribution networks based on digital twins and hybrid enhanced intelligence provided by this invention represents a leap from quantitative to qualitative change compared to existing technologies. Its technical effects are specifically reflected in the following five aspects: 1. Fault Location Capability: From "Area Judgment" to "Centimeter-Level Precise Location," traditional techniques typically can only narrow down the fault area to one or more line sections (potentially encompassing several kilometers of line and dozens of towers), with location accuracy ranging from "hundred-meter" to "kilometer." This invention, through data assimilation and fusion of traveling wave and impedance methods, and auxiliary correction using a historical case database, can pinpoint the fault point to a single tower or cable joint, achieving location accuracy at the "meter" or even "centimeter" level (depending on the traveling wave device's sampling rate). For example, on a 10kV line, the fault point can be pinpointed to "tower number 47, 325 meters from switch A," providing crucial information for subsequent rapid maintenance. Furthermore, traditional techniques rely on a single information source (such as overcurrent or impedance alone), making them prone to misjudgment or failure to operate when fault current is injected into complex branches, high-resistance grounding, or distributed power sources. This invention employs a multi-level collaborative mechanism with redundant information verification. The first level (area location) serves as a rapid screening process, providing a range of alternatives. The second level (segment verification) effectively identifies and eliminates erroneous or conflicting alarm information through direction and amplitude comparison and a multi-source information voting mechanism, greatly reducing the false alarm rate. Even if individual terminal information is abnormal, the system can still make a correct judgment based on confidence weighting. The third level (precise positioning) uses cross-validation; when the results of the traveling wave method and the impedance method deviate significantly, case reasoning is initiated for correction, avoiding systematic errors of a single method under specific operating conditions. This reduces the overall false alarm rate of fault location from 5%-10% of traditional methods to below 1%.

[0024] 2. Response Speed: From "minutes" to "seconds" in terms of end-to-end self-healing. Firstly, regarding response speed, traditional technologies typically require several minutes to tens of minutes (including fault detection, manual analysis, and remote operation) from the occurrence of a fault to the completion of isolation and power restoration. This invention, through parallel processing and predictive inference, compresses the entire process to within 30 seconds. Fault location <5 seconds: The three-level location process is not entirely sequential; data processing at the first and second levels can be performed in parallel. Strategy generation and decision-making <10 seconds: The deep reinforcement learning model in the digital twin infers multiple strategies in parallel, avoiding the time-consuming problem of traditional linear computation. Execution and confirmation <15 seconds: The virtual confirmation process allows dispatchers to understand the strategy intent and consequences within seconds, significantly reducing decision-making hesitation time. This greatly shortens the average user outage time (CAIDI), providing a solid guarantee for continuous power supply to sensitive loads (such as semiconductor manufacturing and data centers). Rapid fault isolation helps suppress fault propagation and improves the transient stability of the system.

[0025] 3. Decision-Making Intelligence: From "Static Rules" to "Dynamic Adaptive Optimization" Traditional technologies rely on preset, fixed logical rules (such as "voltage-loss tripping, power-on reconnection"), resulting in singular strategies that struggle to handle complex operating conditions and typically focus solely on "restoring power supply." This invention, through hybrid augmented intelligent decision-making, generates multi-objective, dynamically optimal strategies. Objective Diversification: It simultaneously considers multiple objectives, including total load restoration, load importance level, number of switching operations, network losses, voltage quality, and distributed energy absorption rate. The reinforcement learning model can perceive the real-time state of the power grid (such as DG output and load level) and deduce the most suitable non-obvious optimal strategy for the current scenario in a digital twin (e.g., forming the optimal power transfer path through the coordinated operation of multiple interconnecting switches). Capability to Handle Complex Scenarios: In distribution networks with a high proportion of distributed power sources, this invention effectively handles seamless switching between islanded and grid-connected operation, proactively forming high-quality islands during faults to supply power to critical loads within those islands—a feat difficult to achieve with traditional rule-based systems.

[0026] 4. System security and reliability: Construct a "virtual sandbox" and "closed-loop immunity" mechanism. Traditional technology: self-healing strategies are directly issued to physical devices, which may lead to new failures due to inadequate strategy considerations.

[0027] This invention utilizes a "virtual confirmation" process to construct a secure "virtual sandbox" within the digital twin. Dispatchers can intuitively view the animated simulation and expected results of the entire strategy execution process, authorizing execution only after confirming there is no risk, achieving "what you think is what you get," fundamentally eliminating the possibility of escalating accidents due to misoperation. The system possesses real-time comparison and feedback correction capabilities. Once an unacceptable deviation occurs between the actual state of the power grid and the predictions of the digital twin after strategy execution, the system immediately activates a correction mechanism, adjusting the strategy or issuing an alarm, effectively giving the system an "immune system" and ensuring the robustness of the control process.

[0028] 5. The system's ability to continuously evolve: from "fixed programs" to "growth-oriented AI" Traditional technologies suffer from performance degradation after deployment, even deteriorating with grid changes. This invention, through a self-evolving learning module, transforms the system into a "growing AI." The case library is continuously enriched: each fault handling process, from phenomenon and location to decision-making and outcome, is stored as a structured case in the knowledge base. The model undergoes continuous iteration; deep learning models (such as fault screening models and reinforcement learning decision-making models) are periodically trained or fine-tuned using new case data, becoming increasingly adapted to the specific operating characteristics of the distribution network. With accumulated operating time, the system's location accuracy and speed further improve, and the optimization and effectiveness of its self-healing strategies continuously enhance, achieving a virtuous cycle of "getting smarter with use." This is the core long-term value that distinguishes this invention from all static systems. Attached Figure Description

[0029] Figure 1 : Schematic diagram of the overall architecture of this invention; Figure 2 : Structure diagram of a digital twin; Figure 3 Multi-level collaborative fault location flowchart; Figure 4 : Schematic diagram of the principle of hybrid augmented intelligent decision-making; Figure 5 Schematic diagram of safe execution and closed-loop correction mechanism; Figure 6 : Flowchart of the self-evolutionary learning process. Detailed Implementation

[0030] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0031] Example 1 This embodiment provides a method for rapid fault location and self-healing control of multi-level faults in distribution networks based on digital twins and hybrid enhanced intelligence, including: S1: Construct a digital twin of the distribution network. This digital twin is dynamically updated based on the power grid physical model, real-time operating data, and historical data, and is used for panoramic perception and simulation of the distribution network's operating status. S2: Real-time collection of multimodal operation data through intelligent sensing devices deployed at various nodes of the distribution network, including electrical quantity data, equipment status data and environmental data; S3: Based on the multimodal operation data and digital twin, perform initial fault screening and type identification: Use a lightweight artificial intelligence model to analyze the data in real time, preliminarily determine the probability and type of fault occurrence, and trigger dynamic virtual topology reconstruction to generate a virtual network in the digital twin that corresponds to the current operation mode; S4: Initiate a multi-level collaborative fault location process: S41: Level 1: Wide-area perception and positioning: Based on the fault sign information reported by the power distribution terminal, combined with the dynamic virtual topology, one or more minimum suspected fault areas are quickly delineated in the digital twin using graph computing algorithms within hundreds of milliseconds. S42: Level 2: Section Direction Cooperative Verification: Within the minimum suspected fault area, retrieve and analyze the alarm information of the fault indicator, the fault current direction and amplitude comparison of the intelligent terminals at both ends of the line, and the initial arrival time of the traveling wave. Through multi-source information fusion and voting mechanism, accurately determine the specific fault section where the fault occurred. S43: Level 3: Node Traveling Wave Impedance Fusion Location: Within the specific fault section, a high-precision traveling wave ranging device is activated to obtain the precise time scale of the arrival of the double-ended or multi-ended traveling waves. At the same time, the full-frequency impedance is calculated using fault recording data. The traveling wave ranging results and the impedance calculation results are cross-validated and assimilated in a digital twin. Combined with line parameters, meteorological data and historical fault records, the precise physical location and nature of the fault point are determined. S5: Based on the precise physical location of the fault point, the nature of the fault, and the real-time state of the power grid simulated by the digital twin, a set of self-healing control strategies is generated through hybrid enhanced intelligent decision-making. S51: In the digital twin, an initial fault isolation scheme is automatically generated based on the fault location; S52: With multiple objectives of restoring the most critical load, maximizing power supply reliability, minimizing the number of switching operations and network losses, it uses a deep reinforcement learning model to simulate and deduce multiple power supply restoration paths in parallel in a digital twin. S53: Perform static security and transient stability checks on the derived feasible strategies, and evaluate the impact of the strategies on distributed energy consumption; S54: Select the optimal self-healing control strategy from the verified strategies using a multi-attribute utility decision algorithm; S6: Strategy Safety Execution and Closed-Loop Correction: The optimal self-healing control strategy is decomposed into a series of safe operation instructions, which are then sent to field equipment for execution after being confirmed or authorized by a human virtual agent; at the same time, the digital twin performs real-time comparison and feedback based on the new state of the power grid after execution, and if a deviation occurs, the strategy correction mechanism is activated. S7: After the fault is resolved, the entire process data, location results, execution strategies and effects of this fault will be automatically archived to the case library to drive the self-evolutionary learning of the digital twin and various AI models.

[0032] In step S43 of the method provided in this embodiment, the cross-validation and data assimilation specifically involve: when the distance difference between the traveling wave ranging result and the impedance method calculation result is within a preset threshold, the weighted average value is taken as the final positioning result; if the distance difference exceeds the threshold, reasoning based on historical similar cases is initiated, and typical positioning results of cases similar to those in the past in terms of meteorological conditions, line type, and fault nature are selected for correction.

[0033] In step S5 of the method provided in this embodiment, the hybrid enhanced intelligent decision-making refers to: generating a basic and safe set of policies by a rule engine, then optimizing and exploring the set by a deep reinforcement learning model to generate candidate policies with better performance, and finally making a final decision by multi-attribute utility decision-making, thus forming a human-machine collaborative hybrid enhanced decision-making mode.

[0034] In step S6 of the method provided in this embodiment, the manual virtual confirmation refers to: before the formal execution of the strategy, in the virtual environment of the digital twin, visually displaying the entire process of strategy execution animation and expected results to the scheduler. After the scheduler confirms that the virtual confirmation is correct, he authorizes the system to perform physical operations.

[0035] Example 2 This embodiment provides a multi-level fault rapid location and self-healing control device for a distribution network to implement the method described in Embodiment 1. The overall system structure is as follows: Figure 1 As shown, it includes: The digital twin construction and maintenance module is used to create and update digital twins of the distribution network; The multimodal data fusion sensing module is used to access and process multi-source heterogeneous data from different smart sensing devices; The fault screening and intelligent triggering module is used to run a lightweight AI model for fault warning and type identification, and to trigger dynamic topology reconstruction. The multi-level collaborative fault location module includes: a wide-area perception location unit, used to delineate the smallest suspected fault area based on graph calculation; a direction collaborative verification unit, used to determine the fault section through multi-source information fusion; and a traveling wave impedance fusion location unit, used to determine the precise location of the fault point through data assimilation and cross-validation. The hybrid augmented intelligent decision-making module includes: a rule engine unit for generating basic isolation strategies; a reinforcement learning inference unit for simulating and optimizing recovery strategies in a digital twin; and a security verification and utility decision-making unit for evaluating and selecting the optimal strategy. The safe execution and closed-loop control module is used for safe command issuance, virtual confirmation, and execution feedback correction. The self-evolutionary learning module is used to manage the case library and drive model updates.

[0036] Example 3 This embodiment provides an electronic device, including: at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described in Embodiment 1.

[0037] Example 4 This embodiment provides a computer-readable storage medium storing a computer program thereon, characterized in that, when the computer program is executed by a processor, it implements the method described in Embodiment 1. Example 5 This embodiment uses a medium-sized city power distribution network as an example to illustrate the method and apparatus provided by the present invention.

[0038] The power distribution network system of a medium-sized city is configured as follows: a 10kV distribution network, including 8 feeders, 65 sectionalizing switches, 8 tie switches, and connected to a 12MW distributed photovoltaic system. 125 intelligent distribution units (FTUs) are deployed at key nodes, 85 fault indicators with wireless communication capabilities are installed on the lines, and traveling wave ranging devices are installed at the exits of 4 key substations and at the 8 tie switches. Data acquisition and communication layer: An industrial Ethernet dual-ring network architecture is adopted, with the core switch supporting the IEEE 1588 precision clock protocol to ensure time synchronization accuracy better than 1μs. A data acquisition server is deployed, communicating with terminal devices via the IEC 61850 protocol, with a data acquisition cycle of 100ms, and fault data supports burst upload. Core layer: Two application servers (CPU: 2*Intel Xeon Silver 4314, memory: 256GB) and one GPU server (equipped with an NVIDIA A100) are deployed, forming a digital twin platform. The software platform is developed based on a Java Spring Cloud microservice architecture, integrating a digital twin engine, a multi-level collaborative fault location module, a hybrid enhanced intelligent decision-making module, a safe execution and closed-loop control module, and a self-evolving learning module. A human-machine interface for dispatchers has been developed, supporting panoramic visualization of the power grid, display of fault location results, animated demonstrations of strategy deduction, and virtual confirmation operations.

[0039] Reference Figure 2 The digital twin structure diagram shown below illustrates the construction process as follows: Basic data source construction: Import the CIM / E format power grid model from the power distribution GIS system, containing detailed parameters of 1258 nodes and 156 lines. Configure real-time data interfaces to connect to the SCADA system (acquisition cycle 2s), PMU devices (acquisition cycle 20ms), and the fault information system. Establish a historical case database to store fault records for the past 5 years; access the meteorological bureau's API to obtain real-time environmental data.

[0040] Core engine development: Dynamic virtual topology reconfiguration service: Receives switch change information in real time and dynamically updates the power grid connection relationships based on the Neo4j graph database. When a switch state change is detected, topology reconfiguration is completed within 100ms.

[0041] Multi-scale simulation engine: integrates OpenDSS for power flow calculation and ATLAB / Simulink for electromagnetic transient simulation, supporting multi-scale analysis from second to microsecond.

[0042] Functional service layer implementation: Develop four service interfaces: Status perception and panoramic mapping service (provides real-time power grid status), fault process tracing and reproduction service (supports fault inversion), control strategy deduction and simulation service (previews the effect of strategy execution), and data assimilation and model calibration service (calibrates model parameters based on actual measurement data).

[0043] Model calibration: In the initial stage of system commissioning, by comparing the power flow results calculated by the digital twin with the SCADA measurements, a parameter identification algorithm based on sensitivity analysis was used to fine-tune the line resistance and reactance parameters. After calibration, the calculation error of the critical node voltage stabilized within ±0.5%.

[0044] Reference Figure 3 The multi-level collaborative fault location flowchart shown below illustrates the location process using a permanent A / B phase-to-phase short circuit in feeder F25 as an example: Fault screening and triggering: After the fault occurred, the FTU along the line detected phase-to-phase overcurrent. The lightweight 1D-CNN model analyzed the current waveform and identified it as "AB phase-to-phase short circuit" with a confidence level of 92%.

[0045] The system immediately triggers the location process and calls the dynamic virtual topology reconfiguration service to obtain the current power grid topology.

[0046] Level 1 Positioning – Wide-Area Aware Positioning: The positioning module receives overcurrent signals from FTU-2503, FTU-2505, and FTU-2507. An improved breadth-first search algorithm is run in the digital twin, starting with these FTUs and searching in conjunction with the current topology. The algorithm identifies the common upstream device as CB2501 and the common downstream boundary as SW2505, and outputs the smallest suspected fault area as the line segment between "CB2501→SW2505" within 80ms.

[0047] Second-level positioning – Segmental direction collaborative verification: Collecting multi-source information within suspicious areas: FI-2504 alarm, direction is downstream; FTU-2505 fault current direction is positive; FTU-2507 fault current direction is negative.

[0048] A confidence-weighted fusion algorithm is adopted: confidence weights: direction 0.6, FI alarm 0.3, amplitude 0.1. Orientation consistency score: 1.0 when the orientation of FTU-2505 contradicts that of FTU-2507, and 0.0 when they are consistent (weight 0.6). FI Status Score: 1.0 when FI-2504 alarms, 0.0 when no alarms (weight 0.3). Amplitude matching degree: The score range is set to 0.0-1.0 (weight 0.1).

[0049] Comprehensive scoring formula: Overall score = 0.6 * Directional consistency score + 0.3 * FI state score + 0.1 * Amplitude matching degree The overall score of 0.95 is greater than the threshold of 0.75. The faulty section is determined to be "SW2505→SW2507", taking 150ms.

[0050] Third-level positioning – nodal traveling wave impedance fusion positioning: The traveling wave ranging devices at CB2501 and SW2507 were activated. Using the wavelet transform wavefront identification algorithm, the traveling wave ranging result was calculated to be 2.85 km from CB2501. Using FTU waveform data, the fault distance was calculated using the full-frequency impedance method to be 12.78 km from CB2501.

[0051] Based on the tower coordinate database and historical cases, the fault point was determined to be cable joint No. 3, located 5821 meters from SW250. The entire three-level positioning process was completed within 2.2 seconds.

[0052] Reference Figure 4 The diagram shown illustrates the principle of hybrid augmented intelligent decision-making, and the decision-making process is as follows: The rule engine generates a basic policy: receive the fault location "Cable Connector No. 3". Trigger preset rules: 1. Isolate the nearest switch to the fault point → generate instructions: trip (SW2505), trip (SW2507); 2. Identify the power loss area as the downstream load of SW2507.

[0053] Deep reinforcement learning model derivation: State space: [Topology encoding, load 0.85 pu, photovoltaic 0.2 pu, fault location]; Action space: {Close L08, adjust capacitor, disconnect load, form island}; Reward function: R = 10 * load recovery - 2 * switching operation - 1 * network loss + 5 * critical load recovery - 100 * voltage limit violation; Three candidate strategies are extrapolated in parallel within a digital twin, and their respective reward values ​​are calculated.

[0054] Multi-attribute evaluation and decision-making: Multi-attribute evaluation of strategies that pass the initial screening:

[0055] The TOPSIS method was used for the final decision, and action A (closing the contact switch L08) was selected as the optimal strategy.

[0056] According to Figure 5 The diagram shown illustrates the safe execution and closed-loop correction mechanism. Virtual confirmation: The optimal strategy is sent to the dispatcher's workstation, and the digital twin displays a strategy simulation animation: the faulty section turns red for isolation, the tie switch L08 is highlighted and closed, the power flow direction changes dynamically, and the load rate of F28 increases from 45% to 78%. The dispatcher completes the review within 8 seconds and clicks the "Virtual Confirmation" button to authorize execution.

[0057] Execution process: The control command sequence of opening (SW2505) → opening (SW2507) → closing (L08) is sent to the field equipment after safety verification. The system monitors the switch position changes in real time to confirm that the strategy is executed correctly.

[0058] Monitoring and Correction Phase: After the strategy was executed, the downstream voltage of SW2507 was monitored to be 9.8kV, while the digital twin predicted 10.2kV. The deviation of 0.4kV exceeded the threshold of 0.15kV, triggering the correction mechanism. The system analysis indicated that the load on line F28 was higher than expected, generating an auxiliary suggestion to "adjust the tap changer on line F28" for the dispatcher's reference in decision-making.

[0059] Reference Figure 6 The diagram below illustrates the self-evolutionary learning process: First, data acquisition and archiving: All data from the entire fault handling process (waveforms, location process, decision-making basis, and execution results) was automatically collected. After structured processing, the data was tagged with labels such as [`phase-to-phase short circuit`, `cable joint`, `interconnection and transfer`, `voltage deviation`]. The complete case was stored in the InfluxDB and MySQL databases of the historical case repository.

[0060] Next comes analysis and mining: the self-evolutionary learning module extracts case features: fault type, location accuracy, strategy effectiveness, and reasons for voltage deviation. Through correlation analysis, a pattern was discovered: "Voltage prediction is prone to deviation in high-load power transfer scenarios."

[0061] Model retraining and optimization: Incremental training tasks are initiated weekly, adding new cases to the training set. The policy value network of the deep reinforcement learning model is fine-tuned, with a focus on optimizing voltage prediction accuracy. Validation shows that the model reduces voltage prediction error by 35% in similar scenarios.

[0062] Finally, there's model updating and deployment: the optimized model is updated to the production environment using a blue-green deployment approach. The system continuously monitors the performance of the new model to ensure a stable and reliable learning process.

[0063] After a 6-month trial run, the system handled 27 faults, and its performance indicators are as follows: Mean time to locate fault: 2.8 seconds; Average self-healing time: 24.5 seconds; Fault location accuracy: 96.3%; Success rate of self-healing strategy: 94.7%; Power supply reliability improved: 0.99985 → 0.99992; Compared with traditional methods, the fault handling time is reduced by 68%, and the power supply reliability is significantly improved, verifying the effectiveness and advancement of the present invention.

[0064] This invention achieves a transformation from "experience-driven" to "intelligent-driven" fault handling in distribution networks through the deep integration of digital twins and hybrid augmented intelligence. It has technical advantages such as accurate positioning, rapid response, strong adaptability, safety and reliability, and continuous evolution, and can effectively improve the power supply reliability and operating efficiency of distribution networks.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A method for rapid fault location and self-healing control in a distribution network at multiple levels, characterized in that, Includes the following steps: S1: Construct a digital twin of the distribution network. This digital twin is dynamically updated based on the power grid physical model, real-time operating data, and historical data, and is used for panoramic perception and simulation of the distribution network's operating status. S2: Real-time collection of multimodal operation data through intelligent sensing devices deployed at various nodes of the distribution network, including electrical quantity data, equipment status data and environmental data; S3: Based on the multimodal operation data and digital twin, perform initial fault screening and type identification: Use a lightweight artificial intelligence model to analyze the data in real time, preliminarily determine the probability and type of fault occurrence, and trigger dynamic virtual topology reconstruction to generate a virtual network in the digital twin that corresponds to the current operation mode; S4: Initiate a multi-level collaborative fault location process: S41: Level 1: Wide-area perception and positioning: Based on the fault sign information reported by the power distribution terminal, combined with the dynamic virtual topology, one or more minimum suspected fault areas are quickly delineated in the digital twin using graph computing algorithms within hundreds of milliseconds. S42： Level 2: Section Direction Collaborative Verification: Within the smallest suspected fault area, retrieve and analyze the alarm information of the fault indicator, the fault current direction and amplitude comparison of the intelligent terminals at both ends of the line, and the initial arrival time of the traveling wave. Through multi-source information fusion and voting mechanism, accurately determine the specific fault section where the fault occurred. S43: Level 3: Node Traveling Wave Impedance Fusion Location: Within the specific fault section, a high-precision traveling wave ranging device is activated to obtain the precise time scale of the arrival of the double-ended or multi-ended traveling waves. At the same time, the full-frequency impedance is calculated using fault recording data. The traveling wave ranging results and the impedance calculation results are cross-validated and assimilated in a digital twin. Combined with line parameters, meteorological data and historical fault records, the precise physical location and nature of the fault point are determined. S5: Based on the precise physical location of the fault point, the nature of the fault, and the real-time state of the power grid simulated by the digital twin, a set of self-healing control strategies is generated through hybrid enhanced intelligent decision-making. S51: In the digital twin, an initial fault isolation scheme is automatically generated based on the fault location; S52: With multiple objectives of restoring the most critical load, maximizing power supply reliability, minimizing the number of switching operations and network losses, it uses a deep reinforcement learning model to simulate and deduce multiple power supply restoration paths in parallel in a digital twin. S53: Perform static security and transient stability checks on the derived feasible strategies, and evaluate the impact of the strategies on distributed energy consumption; S54: Select the optimal self-healing control strategy from the verified strategies using a multi-attribute utility decision algorithm; S6: Strategy Safety Execution and Closed-Loop Correction: The optimal self-healing control strategy is decomposed into a series of safe operation instructions, which are then sent to field equipment for execution after being confirmed or authorized by a human virtual agent; at the same time, the digital twin performs real-time comparison and feedback based on the new state of the power grid after execution, and if a deviation occurs, the strategy correction mechanism is activated. S7: After the fault is resolved, the entire process data, location results, execution strategies and effects of this fault will be automatically archived to the case library to drive the self-evolutionary learning of the digital twin and various AI models.

2. The multi-level fault rapid location and self-healing control method according to claim 1, characterized in that, In step S3, the dynamic virtual topology reconstruction specifically involves: automatically updating the electrical connection relationship of the power grid in the digital twin based on the real-time switching status, forming a virtual network model that corresponds one-to-one with the physical power grid and includes distributed power source access points, providing an accurate network foundation for subsequent fault analysis and power flow calculation.

3. The multi-level fault rapid location and self-healing control method according to claim 1, characterized in that, In step S42, the multi-source information fusion and voting mechanism specifically includes: setting confidence weights for fault indicator alarm, current direction judgment, and preliminary traveling wave information; directly outputting the judgment result when all information is consistent; and adopting a confidence-weighted voting decision or triggering a higher-level precise positioning process when information conflicts exist.

4. The multi-level fault rapid location and self-healing control method according to claim 1, characterized in that, In step S43, the cross-validation and data assimilation specifically involve: when the distance difference between the traveling wave ranging result and the impedance method calculation result is within a preset threshold, the weighted average value is taken as the final positioning result; if the distance difference exceeds the threshold, reasoning based on historical similar cases is initiated, and typical positioning results of cases similar to those in the past in terms of meteorological conditions, line type, and fault nature are selected for correction.

5. The multi-level fault rapid location and self-healing control method according to claim 1, characterized in that, In step S5, the hybrid augmented intelligent decision-making refers to: generating a basic and secure set of policies by a rule engine, then optimizing and exploring the set by a deep reinforcement learning model to generate candidate policies with better performance, and finally making a final decision by a multi-attribute utility decision-making system, thus forming a human-machine collaborative hybrid augmented decision-making mode.

6. The multi-level fault rapid location and self-healing control method according to claim 1, characterized in that, In step S6, the manual virtual confirmation refers to: before the strategy is officially executed, in the virtual environment of the digital twin, visually displaying the entire process of strategy execution and the expected results to the scheduler. After the scheduler confirms that the virtual confirmation is correct, he authorizes the system to perform physical operations.

7. A multi-level fault rapid location and self-healing control device for a distribution network for implementing the method according to any one of claims 1-6, characterized in that, include: The digital twin construction and maintenance module is used to create and update digital twins of the distribution network; The multimodal data fusion sensing module is used to access and process multi-source heterogeneous data from different smart sensing devices; The fault screening and intelligent triggering module is used to run a lightweight AI model for fault warning and type identification, and to trigger dynamic topology reconstruction. The multi-level collaborative fault location module includes: a wide-area perception location unit, used to delineate the smallest suspected fault area based on graph calculation; a direction collaborative verification unit, used to determine the fault section through multi-source information fusion; and a traveling wave impedance fusion location unit, used to determine the precise location of the fault point through data assimilation and cross-validation. The hybrid augmented intelligent decision-making module includes: a rule engine unit for generating basic isolation strategies; a reinforcement learning inference unit for simulating and optimizing recovery strategies in a digital twin; and a security verification and utility decision-making unit for evaluating and selecting the optimal strategy. The safe execution and closed-loop control module is used for safe command issuance, virtual confirmation, and execution feedback correction. The self-evolutionary learning module is used to manage the case library and drive model updates.

8. An electronic device, comprising: At least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images