Circuit fault automatic recovery method and system

Abstract

The invention belongs to the technical field of power systems, and discloses a circuit fault automatic recovery method and system, and the system comprises a resource management module, a fault detection module, a strategy decision module and a task execution module. The intelligent decision-making system is constructed by fusing the generative AI and the decision-making tree algorithm, and automatic diagnosis and intelligent recovery of faults of the power optical transmission network are achieved; the system can analyze multi-source alarm data in real time, accurately identify a fault influence range, and automatically generate an optimal recovery strategy according to different service levels and protection types; compared with a traditional manual processing mode, the system has the advantages that the fault response speed is remarkably improved, continuous and stable transmission of key service signals is ensured, the problems of low recovery efficiency, high manual misoperation risk and the like in the prior art are effectively solved, and reliable guarantee is provided for safe operation of a smart power grid.

Description

Technical Field

[0001] This invention belongs to the field of power system technology, specifically a method and system for automatic circuit fault recovery. Background Technology

[0002] With the deepening of smart grid construction, the reliability requirements of power communication networks, as a crucial support for the safe and stable operation of the power grid, are increasingly stringent. Optical transmission networks, as the core carrier platform of power communication networks, undertake the transmission of critical business signals such as relay protection, stability control, and dispatch automation. However, existing fault recovery mechanisms for optical transmission networks still suffer from significant problems when dealing with complex fault scenarios, including slow response speed, low level of intelligence, and low resource utilization, making it difficult to meet the high reliability requirements of new power systems for communication networks.

[0003] Currently, fault recovery in power optical transmission networks mainly employs two technical solutions: manual intervention recovery and static protection switching. Manual intervention recovery relies on maintenance personnel monitoring alarm information through the network management system, manually analyzing the scope of fault impact, and manually performing fault location and service restoration operations. Static protection switching is based on pre-configured protection mechanisms such as SDH multiplex section protection (MSP) or subnet connection protection (SNCP).

[0004] Existing technical solutions share the following common problems: fault recovery efficiency is mismatched with the requirements of new power systems; they lack intelligent fault diagnosis and recovery decision-making capabilities; they cannot achieve dynamic optimization of network resource allocation; and they rely heavily on manual labor, resulting in high operation and maintenance costs. Therefore, improvements are needed. Summary of the Invention

[0005] The purpose of this invention is to provide an automatic circuit fault recovery method and system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an automatic circuit fault recovery method and system, comprising: a resource management module, a fault detection module, a strategy decision module, and a task execution module; The resource management module is used to collect and model the resource status of network elements, ports, cross-connections and protection subnets of the optical transmission network in real time, and synchronize the resource status data to the fault detection module and the strategy decision module. The fault detection module identifies circuit faults based on multi-source alarm data and performance monitoring data, generates a fault impact range report, and sends it to the strategy decision module. The strategy decision module integrates generative AI and decision tree algorithm, dynamically selects recovery strategy based on circuit level (VC12 / VC3 / VC4) and protection mode (no protection / partial protection / full protection), and sends strategy instructions to task execution module; The task execution module is used to perform "delete first, then create" or "delayed rerouting" operations, and supports adjustable delay recovery of 5-1000 seconds.

[0007] Preferably, the strategy decision module includes a routing algorithm library and a priority scheduling unit; The routing algorithm library stores the shortest path algorithm (Dijkstra), the least hops algorithm, and dynamic routing algorithms based on reinforcement learning; The priority scheduling unit allocates recovery priorities according to the service type (relay protection, stability, automation), with relay protection services having the highest priority.

[0008] Preferably, it also includes a human-machine collaboration interface and a topology visualization unit; The human-machine collaboration interface provides a floating window to display tasks to be processed in real time, and supports manual review, batch operation and path trajectory backtracking; The topology visualization unit distinguishes working routes (green), protection routes (blue), and self-healing routes (red) through color coding and supports interactive editing of topology elements.

[0009] Preferably, the resource management module further includes a dynamic resource pre-allocation mechanism and a status synchronization unit; The dynamic resource pre-allocation mechanism pre-calculates and locks available time slots and link resources before fault recovery to avoid concurrent conflicts; The state synchronization unit ensures the consistency of the network management system and physical device resource status through a heartbeat detection mechanism.

[0010] Preferably, the task execution module supports: The automatic return to the original route function determines whether to switch back to the original path after the fault is cleared by detecting optical power and bit error rate. The success rate statistics unit records quantitative indicators such as recovery success rate (≥99%) and time taken (≤20 seconds).

[0011] Preferably, it includes the following steps: A1. Real-time collection of alarm data, port status, and cross-connection information of the optical transmission network through the resource management module to construct a dynamic resource relationship model; A2. Identify circuit faults through the fault detection module, generate a fault impact range report, and send it to the strategy decision module; A3. The strategy decision module predicts the fault propagation path based on generative AI and dynamically selects the recovery strategy (delete and rebuild / delayed rerouting) by combining the decision tree algorithm. A4. The task execution module performs resource pre-allocation and path reconstruction according to policy instructions, and supports adjustable delay recovery of 5-1000 seconds; A5. After the fault is cleared, the optical power and bit error rate performance data are used to evaluate and decide whether to return to the original route.

[0012] Preferably, in A3: For unprotected services, the shortest path algorithm (Dijkstra) and resource pre-allocation mechanism are used first. For some protection services, the weights of the protection subnets are dynamically adjusted through reinforcement learning to optimize detour routes; For services under full protection, the preset routing table can be switched directly, and the recovery time is controlled within 10 seconds.

[0013] Preferably, A4 includes: Set an adjustable delay window of 5-1000 seconds; the delay window for high-priority services (relay protection) should be ≤30 seconds. Resources on faulty links are released through a dynamic resource pre-allocation mechanism, and concurrent conflicts are avoided by an optimistic locking mechanism.

[0014] Preferably, A5 further includes: Record the path and performance data of each recovery operation to form a knowledge base to optimize subsequent decisions; When the optical power fluctuation of the original route is ≤1dB and the bit error rate is ≤1E-9, the function of returning to the original route is automatically triggered.

[0015] Preferably, it also includes a manual intervention mechanism and a strategy backtracking function; The manual intervention mechanism provides "review-execute-ignore" operation options in the floating window, supporting manual overriding of automatic decision-making; The policy backtracking function uses timestamps to mark policy modification records, ensuring the traceability of operations.

[0016] The beneficial effects of this invention are as follows: 1. This invention constructs an intelligent decision-making system by integrating generative AI and decision tree algorithms, realizing automated diagnosis and intelligent recovery of faults in power optical transmission networks. The system can analyze multi-source alarm data in real time, accurately identify the scope of fault impact, and automatically generate the optimal recovery strategy according to different service levels and protection types. Compared with traditional manual processing methods, this system significantly improves fault response speed, ensures continuous and stable transmission of critical service signals, and effectively solves the problems of low recovery efficiency and high risk of human error in existing technologies, providing a reliable guarantee for the safe operation of smart grids.

[0017] 2. This invention innovatively adopts a dynamic resource pre-allocation mechanism and intelligent routing algorithm to achieve real-time optimized configuration of network resources. By establishing a network-wide resource status model, the system can quickly lock available resources when a fault occurs and intelligently allocate transmission paths according to service priorities. This design not only solves the problem of low resource utilization in traditional static protection schemes, but also flexibly adapts to different types of service requirements such as no protection, partial protection, and full protection, significantly improving the overall utilization efficiency of network resources and providing a new technical solution for the intelligent operation and maintenance of power communication networks.

[0018] 3. This invention designs an intelligent human-machine collaborative management mechanism, which provides a flexible manual intervention interface while maintaining the advantages of automated recovery. Through a visual operation interface and policy retrospective function, maintenance personnel can monitor the system's operating status in real time and make manual interventions or policy adjustments when necessary. This design fully leverages the automation advantages of intelligent systems while retaining necessary human decision-making power, effectively reducing operational risks and improving the system's reliability and maintainability, providing a complete solution for the intelligent operation and maintenance management of power communication networks. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the automatic circuit fault recovery method of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: This invention provides an automatic circuit fault recovery method and system, comprising: a resource management module, a fault detection module, a strategy decision module, and a task execution module; The resource management module is used to collect and model the resource status of network elements, ports, cross-connections and protection subnets of the optical transmission network in real time, and synchronize the resource status data to the fault detection module and the strategy decision module. The fault detection module identifies circuit faults based on multi-source alarm data and performance monitoring data, generates a fault impact range report, and sends it to the strategy decision module. The strategy decision module integrates generative AI and decision tree algorithm, dynamically selects recovery strategy based on circuit level (VC12 / VC3 / VC4) and protection mode (no protection / partial protection / full protection), and sends strategy instructions to task execution module; The task execution module is used to perform "delete first, then create" or "delayed rerouting" operations, and supports adjustable delay recovery of 5-1000 seconds.

[0022] In this embodiment, the generative AI integrated into the strategy decision-making module is a "power communication fault-specific generative model". Its core is a large model architecture based on industry knowledge enhancement, which is specifically adapted to optical transmission network fault recovery scenarios. The specific technical parameters and logic are as follows: The input data employs a "three-dimensional structured feature matrix" to ensure data integrity and relevance to decision-making, specifically including: Basic fault characteristics: circuit level (VC12 / VC3 / VC4), protection mode (no protection / partial protection / full protection), alarm type (link interruption / port failure / abnormal optical power / excessive bit error rate), fault occurrence time (whether it is during peak power consumption period: 9:00-11:00, 18:00-20:00 are marked as peak). Real-time operating characteristics: Current load rate of faulty network element (≤60% is low load, 60%-80% is medium load, >80% is high load), backup link bandwidth margin (unit: Gbps), real-time optical power (normal range: -8dBm to 2dBm), current bit error rate (unit: E-9). Historical experience characteristics: the success rate of recovery of similar faults (same circuit level + same protection mode) in the past 12 months, the resource consumption of the historical best recovery strategy, and records of manual intervention corrections (such as a certain type of fault that failed to recover due to "not pre-occupied time slots", which should be marked as a risk point).

[0023] The model structure adopts a three-layer architecture of "basic large model + retrieval enhancement (RAG) + power knowledge graph"; The base layer uses a 34-parameter general language model (such as Llama3), which is fine-tuned with power communication fault corpus to have the ability to understand the semantics of fault scenarios. The enhancement layer accesses the "network element-fault-policy" knowledge graph and uses GraphRAG technology to achieve multi-hop retrieval (e.g., "abnormal optical power" can be associated with the link "fiber attenuation → loose connector → switchover to backup fiber"). The output layer integrates a power operation rule engine and incorporates core clauses of the "Power Optical Transmission Network Fault Handling Procedures" (such as "Unprotected services are prohibited from occupying VC4 level link bandwidth"). Training objective: Maximize the success rate of recovery strategy execution (training achievement threshold ≥ 99.2%) and minimize recovery time (≤ 10 seconds for fully protected services, ≤ 30 seconds for partially protected services, and ≤ 60 seconds for unprotected services). Constraint objectives: Reduce resource conflict rate (≤0.5%) and reduce human intervention rate (≤5%). Reinforcement learning is used to give positive rewards (reward coefficient 1.2) to strategies that meet the requirements of the rule engine and have a history of successful execution, and to give negative penalties (penalty coefficient 0.3) to strategies that trigger risk clauses (such as affecting high-priority business).

[0024] Generative AI Inference Output and Policy Mapping Mechanism: The inference process consists of three steps: "candidate generation - scoring and filtering - instruction formatting," ensuring that the output can be directly connected to the task execution module. Based on input features, generate 3-5 differentiated recovery strategies. Each strategy includes "operation steps (e.g., 'disconnect the faulty link → lock the backup time slot → activate the detour route'), expected resource consumption (e.g., 'occupy 2 VC12 links and 1 backup port'), and risk level (high / medium / low, determined based on historical failure rate: >5% is high risk)". Select the optimal strategy through multi-dimensional weighted scoring (weight: compliance 40%, feasibility 30%, historical matching degree 30%). Check whether it conforms to the operation rules (e.g., "full protection service enables the pre-configured routing table" gets full marks, "unprotected service requests VC4 bandwidth" gets 0 marks). Determine based on real-time operation characteristics (e.g., backup link bandwidth margin ≥ required bandwidth gets full marks, insufficient bandwidth deducts points proportionally). Calculate the cosine similarity with high success rate (≥99%) cases of the same type in the past 6 months (≥0.8 gets full marks, <0.6 triggers secondary search). The resource management module is responsible for the real-time acquisition and modeling of resources across the entire network, the fault detection module realizes intelligent analysis of multi-source data, the strategy decision-making module uses AI algorithms to generate recovery strategies, and the task execution module completes specific recovery operations. This architecture achieves closed-loop management of fault handling through modular design, which significantly improves response speed compared to traditional manual handling methods. It can meet the stringent requirements of power communication networks for high reliability and real-time performance, and provides technical support for the safe operation of smart grids.

[0025] Example 2: The strategy decision module includes a routing algorithm library and a priority scheduling unit; The routing algorithm library stores the shortest path algorithm (Dijkstra), the least hops algorithm, and dynamic routing algorithms based on reinforcement learning; The priority scheduling unit allocates recovery priorities according to the service type (relay protection, stability, automation), with relay protection services having the highest priority.

[0026] The module's built-in routing algorithm library contains a variety of intelligent algorithms that can automatically select the optimal recovery scheme based on different business needs; the priority scheduling unit ensures that critical services are processed first. This design enables the system to provide differentiated quality of service for different levels of business needs, especially achieving millisecond-level rapid response for critical services such as relay protection, while ensuring a high recovery success rate under various complex fault scenarios through algorithm diversity.

[0027] To further clarify the operating mechanism of the routing algorithm library, the following details the algorithm allocation mechanism, application scenarios of each algorithm, algorithm flow steps, and corresponding pseudocode and mathematical models; The algorithm allocation mechanism is the core logic of the strategy decision-making module. It is used to dynamically select the appropriate routing algorithm based on fault reports. This mechanism achieves automated selection by combining the output of a decision tree model with the output of generative AI. Specifically, the implementation logic of this mechanism is as follows: Input preparation phase: Receive the fault impact range report generated by the fault detection module. The report includes the fault type (such as link interruption or optical power abnormality), circuit level (VC12 / VC3 / VC4), protection mode (no protection / partial protection / full protection), and service type (relay protection / stability / automation). Generative AI (see Implementation 1) first predicts the fault propagation path and resource requirements (such as backup link bandwidth margin), and uses these prediction results as the input feature vector of the decision tree model.

[0028] Decision tree judgment phase: The C4.5 decision tree algorithm is adopted, where the root node branches based on business priority (relay protection has the highest priority, and a priority queue data structure implemented using heap sort is used to ensure that high-priority tasks are queued and processed first); subsequent branch nodes consider protection type and fault complexity (such as the distinction between single-point fault and multi-point propagation), and finally the leaf nodes are mapped to specific routing algorithms. The construction of the decision tree uses information gain as the splitting criterion to maximize the efficiency of the recovery strategy.

[0029] Selection criteria phase: Prioritize matching high-priority services, such as forcing low-latency algorithms (e.g., Dijkstra) for relay protection services; if the resource management module determines resource conflicts through heartbeat detection, switch to dynamic algorithms (e.g., dynamic routing algorithms based on reinforcement learning); allocation threshold is set as follows: if the backup link bandwidth margin is ≥50%, prioritize static algorithms (e.g., Dijkstra); otherwise, use reinforcement learning algorithms for adaptive adjustment; this criteria stem from the real-time requirements of power communication networks, ensuring recovery time is controlled within 10-60 seconds (≤10 seconds for full protection services).

[0030] Output and Execution Phase: After selecting an algorithm, policy instructions are issued to the task execution module, supporting concurrent processing (using an optimistic locking mechanism to avoid resource conflicts between multiple tasks); in addition, the decision tree weights are iteratively optimized through a historical knowledge base to achieve continuous learning of the system; this mechanism not only improves the intelligence level of decision-making, but also ensures compatibility with the high reliability requirements of the power system.

[0031] The routing algorithm library contains three algorithms optimized for different power services and fault scenarios to achieve accurate matching. The shortest path algorithm (Dijkstra) is used for single-point link failure recovery in unprotected or low-priority services (such as automated scheduling); it is suitable for networks with sufficient resources and simple topology (such as backup time slots ≥ VC4 level). Least hops algorithm: Optimizes multi-hop detour routing for some protection services (such as stability control); suitable for scenarios with network congestion or limited backup links (such as peak load rate >80%). Dynamic routing algorithms based on reinforcement learning: complex faults (such as multi-source alarms or fault propagation) that protect or high-priority services (such as relay protection); suitable for dynamically changing network environments (such as peak electricity consumption periods 9:00-11:00). These scenarios are matched with the processing logic for different protection services, ensuring the flexible application of the system in actual power optical transmission networks.

[0032] The execution flow of each algorithm is integrated into the routing algorithm library, supporting modular invocation. The following describes the process steps in detail: Dijkstra's Shortest Path Algorithm: This algorithm is applicable to static topology optimization, and the steps are as follows: Initialization phase: Construct network graph G, where nodes represent network elements, edges represent links, and weights are latency and bandwidth cost; set the distance to the source node to 0 and the distance to other nodes to infinity; create a priority queue (using a min-heap structure). Node selection phase: Extract the unvisited node u with the smallest distance from the priority queue; Neighbor update phase: Iterate through the neighbor nodes v of node u. If the path distance through node u is less than the distance of the current node v, update the distance value of node v and record the predecessor node. At the same time, push node v into the priority queue. Repeated iteration phase: Repeat steps 2-3 until the priority queue is empty or the target node is reached; Path reconstruction phase: Backtrack from the target node to the predecessor node to form a recovery route; at the same time, check resource availability, and if there is a conflict, backtrack and trigger the backup algorithm.

[0033] Least Hop Count Algorithm (Based on a Variation of Breadth-First Search): This algorithm emphasizes minimizing transmission delay, and the steps are as follows: Initialization phase: Enqueue the source node and set the hop count to 0; mark all nodes as unvisited. Dequeue processing phase: Take out the current node u, traverse its neighbor nodes v; if node v has not been visited, set the hop count of node v to the hop count of node u plus 1, record the predecessor node, and enqueue node v. Repeated iteration phase: Repeat step 2 until the queue is empty or the target node is found; Path reconstruction phase: Reconstruction is performed based on the shortest hop path recorded by hop count; an integrated resource pre-allocation mechanism is used to lock in necessary time slots; Output phase: If no valid path is found, return a failure signal and trigger the backup algorithm.

[0034] The dynamic routing algorithm based on reinforcement learning (using the Q-learning framework) is as follows: This algorithm is suitable for dynamic environments, and the steps are as follows: State definition phase: Define state S, including network topology vector, fault location, resource utilization, and service priority; Action space phase: Define action A, including selecting a backup link, adjusting the weight of the protection subnet, or switching the preset routing table; Reward function phase: Define reward R, for example +1 for successful recovery, -0.5 for failure or delay exceeding the threshold, and +0.2 for resource utilization optimization; Training and inference phases: Initialize the Q-table; during the training phase, use an ε-greedy strategy (ε=0.1) for exploration; during the inference phase, select the action with the largest Q(s,a) value and update the Q-value. Path generation phase: Construct routes based on the optimized action sequence; if the recovery success rate is less than 99%, feedback is sent to the knowledge base for further optimization.

[0035] The above process can be implemented using Python's NetworkX library. To facilitate programming implementation, pseudocode examples and mathematical models are provided below. The pseudocode assumes that the network graph is a dictionary structure. Dijkstra pseudocode: import heapq # Priority queue module def dijkstra(graph, start, end): # graph: {node: {neighbor: weight}} distances = {node: float('inf') for node in graph} # Initialize the distance dictionary distances[start] = 0 # The distance to the source node is 0 previous = {node: None for node in graph} # Dictionary of predecessor nodes pq = [(0, start)] # Priority queue: (distance, node) while pq: current_distance, current_node = heapq.heappop(pq) # Extract the node with the minimum distance if current_distance > distances[current_node]: continue # Skip optimized nodes for neighbor, weight in graph[current_node].items(): distance = current_distance + weight # Calculate the new distance, weight = latency + bandwidth cost if distance < distances[neighbor]: distances[neighbor] = distance # Update distance previous[neighbor] = current_node # Update the predecessor heapq.heappush(pq, (distance, neighbor)) # Push into the queue # Rebuild Path path = [] current = end while current is not None: path.append(current) current = previous[current] path = path[::-1] return path if path and path[0] == start else None # Returns the path or None; The computational complexity of this pseudocode is . Where O is the highest-order complexity of the running time as the input size increases, V is the number of network elements, and E is the number of links, which is applicable to the actual calculation of power grid topology; Pseudocode for the least hops algorithm (BFS implementation): from collections import deque # Double-ended queue module def min_hops(graph, start, end): # graph: {node: [neighbors]} visited = set() # Collection of visited nodes queue = deque([(start, 0, [])]) # Queue: (node, hop count, path) visited.add(start) while queue: node, hops, path = queue.popleft() # Dequeue path = path + [node] # Update path if node == end: return path, hops # Returns the path and the number of hops for neighbor in graph[node]: If the neighbor is not visited: visited.add(neighbor) queue.append((neighbor, hops + 1, path)) # Join the queue return None, float('inf') # Returns no path; Mathematical model and pseudocode of dynamic routing algorithm based on reinforcement learning: Mathematical model: Q-learning update formula: , in, This represents the current state, such as the topology matrix or fault vector. For actions, such as selecting a link; As a reward, a successful recovery will be worth +1. The learning rate is 0.1. This is a discount factor with a value of 0.9; For the next state, For the next action; Pseudocode (simplified Q-learning implementation): import numpy as np # Numerical computation module def q_learning(states, actions, rewards, alpha=0.1, gamma=0.9,episodes=1000): Q = np.zeros((len(states), len(actions))) # Initialize the Q table for _ in range(episodes): s = np.random.choice(states) # Randomly select the initial state while not is_terminal(s): # Loop until the terminal state is restored successfully. a = epsilon_greedy(Q, s) # Use ε-greedy to select actions s_next, r = execute_action(s, a) # Execute the action to obtain the new state and reward. Q[s, a] += alpha * (r + gamma * np.max(Q[s_next]) - Q[s,a]) # Update Q value s = s_next # Update status return Q # Return the optimized Q table def epsilon_greedy(Q, s, epsilon=0.1): if np.random.rand() < epsilon: return np.random.choice(actions) # Random selection return np.argmax(Q[s]) # Select the action that maximizes Q (utilizes); In practice, states can be defined as network state vectors, and actions are the routing selection set; historical fault data is used for iterative optimization during the training phase.

[0036] Example 3: It also includes a human-machine collaboration interface and a topology visualization unit; The human-machine collaboration interface provides a floating window to display tasks to be processed in real time, and supports manual review, batch operation and path trajectory backtracking; The topology visualization unit distinguishes working routes (green), protection routes (blue), and self-healing routes (red) through color coding and supports interactive editing of topology elements.

[0037] The floating window displays task status in real time, supporting manual review and intervention; the topology visualization interface intuitively presents network status. This human-machine collaborative design retains the efficiency of automated systems while providing maintenance personnel with necessary control, effectively reducing the risks that purely automated systems may bring. At the same time, visualization significantly improves the work efficiency of maintenance personnel and their ability to control the system status.

[0038] Example 4: The resource management module further includes a dynamic resource pre-allocation mechanism and a status synchronization unit; The dynamic resource pre-allocation mechanism pre-calculates and locks available time slots and link resources before fault recovery to avoid concurrent conflicts; The state synchronization unit ensures the consistency of the network management system and physical device resource status through a heartbeat detection mechanism.

[0039] A dynamic resource pre-allocation mechanism ensures resource availability during the recovery process, while a state synchronization unit guarantees data consistency. This design solves the resource conflict problem commonly found in traditional systems. Pre-computation and locking mechanisms avoid resource contention during multi-task concurrency, while real-time synchronization ensures precise control of actual network resources, laying a resource foundation for rapid and accurate fault recovery.

[0040] Example 5: The task execution module supports: The automatic return to the original route function determines whether to switch back to the original path after the fault is cleared by detecting optical power and bit error rate. The success rate statistics unit records quantitative indicators such as recovery success rate (≥99%) and time taken (≤20 seconds).

[0041] The automatic return to the original route function enables intelligent route switching based on performance detection, and the success rate statistics unit provides quantitative evaluation. These functions enable the system not only to complete basic recovery operations, but also to automatically optimize route configuration after the fault is eliminated. At the same time, data statistics provide a basis for continuous system optimization, forming a complete "recovery-optimization-evaluation" closed-loop management system.

[0042] Example 6: Includes the following steps: A1. Real-time collection of alarm data, port status, and cross-connection information of the optical transmission network through the resource management module to construct a dynamic resource relationship model; A2. Identify circuit faults through the fault detection module, generate a fault impact range report, and send it to the strategy decision module; A3. The strategy decision module predicts the fault propagation path based on generative AI and dynamically selects the recovery strategy (delete and rebuild / delayed rerouting) by combining the decision tree algorithm. A4. The task execution module performs resource pre-allocation and path reconstruction according to policy instructions, and supports adjustable delay recovery of 5-1000 seconds; A5. After the fault is cleared, the optical power and bit error rate performance data are used to evaluate and decide whether to return to the original route.

[0043] From resource acquisition and fault detection to strategy decision-making and recovery, a standardized processing chain is formed. This method breaks down the complex fault recovery process into clear steps, ensuring optimal handling at each stage. This guarantees processing efficiency while reducing system complexity through process design, making the entire recovery process more controllable and reliable.

[0044] Example 7: In A3: For unprotected services, the shortest path algorithm (Dijkstra) and resource pre-allocation mechanism are used first. For some protection services, the weights of the protection subnets are dynamically adjusted through reinforcement learning to optimize detour routes; For services under full protection, the preset routing table can be switched directly, and the recovery time is controlled within 10 seconds.

[0045] Differentiated algorithms are employed for services with different protection levels to achieve precise recovery. This tiered strategy selection mechanism ensures the rational allocation of system resources, guaranteeing both rapid recovery of critical services and efficient processing of ordinary services, enabling the system to provide the most suitable recovery solution based on the importance and characteristics of different services.

[0046] Example 8: A4 includes: Set an adjustable delay window of 5-1000 seconds; the delay window for high-priority services (relay protection) should be ≤30 seconds. Resources on faulty links are released through a dynamic resource pre-allocation mechanism, and concurrent conflicts are avoided by an optimistic locking mechanism.

[0047] Adjustable latency and optimistic locking are used to ensure the coordination of resource operations. These mechanisms effectively solve the resource contention problem when multiple tasks are running concurrently, ensuring processing efficiency while avoiding operational conflicts. In particular, stricter latency requirements are set for high-priority tasks to ensure they receive priority processing rights.

[0048] Example 9 The A5 further includes: Record the path and performance data of each recovery operation to form a knowledge base to optimize subsequent decisions; When the optical power fluctuation of the original route is ≤1dB and the bit error rate is ≤1E-9, the function of returning to the original route is automatically triggered.

[0049] By recording recovery process data, the decision-making algorithm is continuously improved. This design enables the system to learn from historical experience, continuously optimize recovery strategies, and gradually improve processing efficiency and success rate. It realizes the transformation from "fixed strategy" to "adaptive optimization," giving the system the ability to continuously evolve.

[0050] Example 10: It also includes a manual intervention mechanism and a strategy backtracking function; The manual intervention mechanism provides "review-execute-ignore" operation options in the floating window, supporting manual overriding of automatic decision-making; The policy backtracking function uses timestamps to mark policy modification records, ensuring the traceability of operations.

[0051] It provides a manual intervention interface and operation traceability function to ensure controllability. These functions fully leverage the advantages of automation while retaining necessary human supervision methods, which not only meets the stringent safety requirements of the power system, but also provides a basis for post-event analysis and accountability through complete operation records.

[0052] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0053] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An automatic circuit fault recovery system, characterized in that, It includes: a resource management module, a fault detection module, a strategy decision-making module, and a task execution module; The resource management module is used to collect and model the resource status of network elements, ports, cross-connections and protection subnets of the optical transmission network in real time, and synchronize the resource status data to the fault detection module and the strategy decision module. The fault detection module identifies circuit faults based on multi-source alarm data and performance monitoring data, generates a fault impact range report, and sends it to the strategy decision module. The strategy decision module integrates generative AI and decision tree algorithm, dynamically selects recovery strategy according to circuit level and protection mode, and sends strategy instructions to task execution module; The task execution module is used to perform delete-then-create or delayed rerouting operations, and supports adjustable delay recovery of 5-1000 seconds.

2. The automatic circuit fault recovery system according to claim 1, characterized in that: The strategy decision module includes a routing algorithm library and a priority scheduling unit; The routing algorithm library stores shortest path algorithms, least hop number algorithms, and dynamic routing algorithms based on reinforcement learning; The priority scheduling unit assigns recovery priorities based on service type, with relay protection services having the highest priority.

3. The automatic circuit fault recovery system according to claim 1, characterized in that: It also includes a human-machine collaboration interface and a topology visualization unit; The human-machine collaboration interface provides a floating window to display tasks to be processed in real time, and supports manual review, batch operation and path trajectory backtracking; The topology visualization unit distinguishes working routes (green), protection routes (blue), and self-healing routes (red) using color coding, and supports interactive editing of topology elements.

4. The automatic circuit fault recovery system according to claim 1, characterized in that: The resource management module further includes a dynamic resource pre-allocation mechanism and a status synchronization unit; The dynamic resource pre-allocation mechanism pre-calculates and locks available time slots and link resources before fault recovery to avoid concurrent conflicts; The state synchronization unit ensures the consistency of the network management system and physical device resource status through a heartbeat detection mechanism.

5. The automatic circuit fault recovery system according to claim 1, characterized in that: The task execution module supports: The automatic return to the original route function determines whether to switch back to the original path after the fault is cleared by detecting optical power and bit error rate. The success rate statistics unit records quantitative indicators such as recovery success rate and time taken.

6. The automatic circuit fault recovery method for the system according to any one of claims 1-5, characterized in that, Includes the following steps: A1. Real-time collection of alarm data, port status, and cross-connection information of the optical transmission network through the resource management module to construct a dynamic resource relationship model; A2. Identify circuit faults through the fault detection module, generate a fault impact range report, and send it to the strategy decision module; A3. The strategy decision-making module predicts the fault propagation path based on generative AI and dynamically selects the recovery strategy by combining the decision tree algorithm. A4. The task execution module performs resource pre-allocation and path reconstruction according to policy instructions, and supports adjustable delay recovery of 5-1000 seconds; A5. After the fault is cleared, the optical power and bit error rate performance data are used to evaluate and decide whether to return to the original route.

7. The automatic circuit fault recovery method according to claim 6, characterized in that: In A3: For unprotected services, the shortest path algorithm and resource pre-allocation mechanism will be used first. For some protection services, the weights of the protection subnets are dynamically adjusted through reinforcement learning to optimize detour routes; For services under full protection, the preset routing table can be switched directly, and the recovery time is controlled within 10 seconds.

8. The automatic circuit fault recovery method according to claim 6, characterized in that: A4 includes: Set an adjustable delay window of 5-1000 seconds; the delay window for high-priority services should be ≤30 seconds. Resources on faulty links are released through a dynamic resource pre-allocation mechanism, and concurrent conflicts are avoided by an optimistic locking mechanism.

9. The automatic circuit fault recovery method according to claim 6, characterized in that: The A5 further includes: Record the path and performance data of each recovery operation to form a knowledge base to optimize subsequent decisions; When the optical power fluctuation of the original route is ≤1dB and the bit error rate is ≤1E-9, the function of returning to the original route is automatically triggered.

10. The automatic circuit fault recovery method according to claim 6, characterized in that: It also includes a manual intervention mechanism and a strategy backtracking function; The manual intervention mechanism provides "review-execute-ignore" operation options in the floating window, supporting manual overriding of automatic decision-making; The policy backtracking function uses timestamps to mark policy modification records, ensuring the traceability of operations.

Images