A method and system for evaluating the health of a tunnel drainage system

Abstract

The application discloses a kind of tunnel drainage system health degree evaluation method and system, to solve the problem that traditional centralized monitoring architecture is insufficient in real-time, lack of physical law embedding, cannot accurately simulate fault propagation and maintenance decision collaborative difficulty.Performance includes: based on physical topology deployment edge computing unit realizes distributed data sensing and state deduction;Build isomorphic virtual agent network, reach global state consensus through multi-round game optimization to form digital mirror image;In mirror image, simulate abnormal normal and reverse flow propagation path, quantify influence value and generate health situation spectrum;Based on spectrum analysis maintenance task, generate maintenance plan of time-space resource coordination through multi-agent negotiation.System corresponding includes distributed sensing, mirror image construction, situation assessment, collaborative decision-making and scheduling execution module.The application realizes closed-loop intelligent operation and maintenance from state sensing to decision execution, improves the accuracy of tunnel drainage system health evaluation and the overall benefit of maintenance decision.

Description

Technical Field

[0001] This invention relates to the field of tunnel drainage system operation and maintenance technology, specifically to a method and system for assessing the health of tunnel drainage systems, applicable to intelligent monitoring, health assessment, fault diagnosis, and maintenance decision optimization for complex tunnel drainage systems. Background Technology

[0002] As a critical infrastructure ensuring the structural safety and smooth operation of tunnels, tunnel drainage systems play a vital role in collecting and removing seepage water and operational wastewater from within the tunnel. Traditional tunnel drainage system maintenance relies primarily on periodic inspections, manual monitoring, and reactive repairs, which suffers from problems such as large monitoring blind spots, delayed fault response, and reliance on experience for maintenance decisions.

[0003] In existing technologies, the monitoring and evaluation of tunnel drainage systems mainly adopts a centralized data acquisition and analysis architecture. Sensors are installed at key nodes, and the collected data is transmitted to a central server for processing and analysis. This architecture has significant technical drawbacks: First, data transmission relies on network connections; when local communication is interrupted or delayed, the system cannot obtain real-time status information of key nodes, resulting in insufficient monitoring completeness. Second, in the centralized computing model, the system's anomaly identification for individual nodes mainly relies on preset thresholds or simple models, making it difficult to accurately distinguish between normal fluctuations and abnormal states, easily leading to false alarms or missed alarms. Third, traditional methods do not adequately consider the physical connections and influence propagation mechanisms between nodes within the drainage system, making it difficult to accurately assess the cascading effects of local anomalies on the overall system, resulting in a lack of a global perspective in maintenance decisions. Finally, maintenance plans are usually based on manual experience, making it difficult to achieve multi-task collaborative optimization under resource constraints, easily leading to resource waste and low maintenance efficiency.

[0004] Specifically, existing technologies face the following major technical shortcomings: 1) The centralized data processing architecture cannot adapt to the complex network topology and limited communication conditions within the tunnel, resulting in insufficient real-time performance and accuracy of state awareness; 2) There is a lack of deep embedding of the physical laws of the drainage system, and the anomaly detection model does not match the actual hydraulic behavior well; 3) The evaluation method is limited to single-point analysis and fails to effectively simulate the propagation effect of faults in the topological network; 4) The maintenance decision-making process lacks an intelligent collaborative mechanism, making it difficult to generate a globally optimal maintenance plan under multiple constraints. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method and system for assessing the health of a tunnel drainage system, so as to solve the technical defects of the prior art, such as insufficient real-time perception due to centralized monitoring architecture, lack of deep embedding of physical laws and topological correlations, inability to accurately simulate fault propagation effects, reliance on human experience, and difficulty in achieving global collaborative optimization of maintenance tasks under multiple constraints.

[0006] To achieve the above objectives, the present invention adopts the following technical solution;

[0007] A method for assessing the health of a tunnel drainage system includes the following steps:

[0008] S100. Based on the physical topology network of the tunnel drainage system, edge computing units are deployed at each node to synchronously collect and preprocess local multi-source sensor data, exchange status information with neighboring nodes, and generate node-level evolution feature vectors by running a local physical consistency model to fuse local observations, neighborhood collaboration and data credibility.

[0009] S200. Construct a virtual agent network isomorphic to the physical topology. Each agent takes the node-level evolutionary feature vector as input, and performs distributed multi-round game optimization with upstream and downstream agents based on local observation matching, upstream quality conservation and downstream demand satisfaction constraints. Adjust the model parameters until Nash equilibrium is reached, forming a distributed digital mirror with global state consensus.

[0010] S300. In the distributed digital mirror, it monitors the comprehensive performance indicators of each agent node in real time. When an anomaly is detected, it simulates the propagation of downstream impact and the source tracing of upstream anomaly starting from the agent node. Based on the performance degradation of nodes on the propagation path, the importance of topology and the vulnerability of equipment, it calculates the potential impact value and generates a health status map containing anomaly location, impact range, propagation path and risk heat information.

[0011] S400 analyzes the health status map into local intervention tasks and matches them with basic maintenance actions. Relevant agents generate local preference solutions, reach a maintenance consensus with the best global benefits through multiple rounds of negotiation, and generate and issue a distributed maintenance action plan that coordinates time, space, and resources.

[0012] As a further aspect of the present invention, the generation of node-level evolutionary feature vectors specifically includes:

[0013] Local sensor data is subjected to outlier filtering based on physical limits and noise filtering based on statistical characteristics. For each data point, a quantitative confidence weight is calculated to integrate the sensor’s own health, recent data stability and the degree of agreement with simple model predictions.

[0014] Based on the preset topological relationships, the observed feature vectors and the confidence weight information are exchanged with all directly adjacent nodes.

[0015] The preprocessed and weighted local observation data, the received upstream node state prediction information, and the downstream node state requirement expectation are jointly input into the local physical consistency model to deduce the key state prediction value of this node in a short period of time in the future.

[0016] The predicted state values, prediction bias, model confidence, and the sensitivity and importance of nodes in the topology are integrated and encapsulated to generate a structured node-level evolution feature vector.

[0017] As a further aspect of the present invention, the local physical consistency model is a lightweight mathematical model that describes the hydraulic characteristics of nodes or the operating rules of equipment, and is used to deduce node state changes based on physical boundary conditions.

[0018] As a further aspect of the present invention, the step of forming a distributed digital mirror specifically includes:

[0019] On cloud or edge server clusters, virtual proxies are created one-to-one with physical nodes based on the physical topology map, forming a virtual proxy network, and the corresponding local physical consistency model is loaded.

[0020] Each virtual agent constructs a local optimization problem that includes local observation matching constraints, upstream quality conservation constraints, and downstream demand satisfaction constraints. The objective function of the local optimization problem is the weighted sum of the losses of the three constraints.

[0021] All agents execute multiple rounds of iterations in parallel. In each round of iteration, each agent optimizes and adjusts its own model parameters to minimize the objective function by assuming that the parameters of its neighbors remain unchanged. Then, it broadcasts the updated parameters and the latest state prediction calculated based on the new parameters to all upstream and downstream neighbor agents.

[0022] When the parameter changes of all agents in the network are lower than the preset threshold for several consecutive rounds, it is determined that a Nash equilibrium has been reached. The virtual agent network reaches a stable consensus on the global operating state of the system and forms the distributed digital mirror.

[0023] As a further aspect of the present invention, the comprehensive performance index of the agent node is obtained by weighted summation of three parts: model prediction residual, the degree of deviation of key operating parameters from the normal benchmark, and the inherent health score based on the service life of the equipment and historical maintenance records.

[0024] The potential impact value of the proxy node is obtained by multiplying the node's performance degradation, its normalized betweenness centrality weight in the entire drainage network topology, and its vulnerability score, which reflects the equipment's age, historical failure rate, and maintenance difficulty.

[0025] As a further aspect of the present invention, the downstream influence propagation is quantified by calculating the impact of abnormal flow at upstream nodes on water level changes at downstream nodes; the upstream anomaly tracing is determined by detecting whether abnormal pressure fluctuations or flow mutations at upstream nodes exceed a preset threshold to determine the anomaly propagation path.

[0026] As a further aspect of the present invention, the multi-round negotiation process specifically includes:

[0027] Each virtual agent is responsible for broadcasting its local preference maintenance plan. The maintenance plan should include at least the suggested actions, a list of required resources, an estimated time, a prediction of performance impact during construction, and a local utility score.

[0028] Each virtual agent automatically detects whether there are resource usage conflicts, overlapping schedules, or physical space interference after receiving other proposals;

[0029] If a conflict is detected, the conflict resolution plan will be automatically adjusted according to the preset conflict resolution priority rules or global optimization goals, and the process will proceed to the next round of negotiation.

[0030] Within the set decision-making time window, proposals, conflict detection, and adjustments are repeated until all conflicts are resolved and the overall maintenance benefit assessment reaches its optimal level. At this point, a maintenance consensus is considered to have been reached.

[0031] Another aspect of this application provides a health assessment system for a tunnel drainage system, the system comprising:

[0032] The distributed intelligent sensing and execution module consists of edge computing hardware units and sensor arrays deployed at each physical node of the tunnel drainage system, and is used to perform data acquisition, local preprocessing and neighborhood communication.

[0033] The digital mirror collaborative construction module is deployed on edge servers or in the cloud. It is used to instantiate virtual agents that correspond one-to-one with physical nodes, and to build and continuously update a distributed digital mirror that reflects the global state of the system through distributed game collaboration among the virtual agents.

[0034] The health status assessment and simulation module is used to perform real-time performance monitoring, anomaly detection, two-way impact propagation simulation, potential risk quantification calculation, and generation of a visualized health status map based on the distributed digital mirror.

[0035] The collaborative maintenance decision-making and scheduling module is used to parse the health status map into specific maintenance tasks and generate an optimized collaborative maintenance plan under constraints through a multi-agent negotiation mechanism.

[0036] The human-computer interaction and scheduling execution module provides a graphical user interface for system status monitoring, health status display, maintenance plan review and issuance, and tracking of plan execution status.

[0037] As a further embodiment of the present invention, the edge computing hardware unit in the distributed intelligent sensing execution module integrates a microprocessor, a memory, a wired or wireless communication module and a local power supply unit, and is connected to a sensor for monitoring one or more parameters among water level, flow rate, equipment vibration, temperature, electrical parameters or water quality.

[0038] As a further aspect of the present invention, each virtual agent in the digital mirror collaborative construction module has a built-in or associated local physical consistency model that matches the hydraulic characteristics or equipment operation rules of its corresponding physical node.

[0039] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are as follows:

[0040] The tunnel drainage system health assessment method and system provided by this invention effectively overcomes the shortcomings of existing technologies and produces significant technical effects through innovative distributed architecture and collaborative mechanism.

[0041] First, at the system perception and state assessment level, this invention achieves distributed collaborative perception by deploying edge computing units on each physical node and establishing a virtual agent network. The edge computing units run a local physical consistency model, integrating local observations, neighborhood collaboration, and data reliability to generate node-level evolutionary feature vectors. This distributed architecture avoids the single-point-of-failure risk of traditional centralized systems, enhancing system robustness. Each agent performs distributed game optimization based on local observation matching, upstream quality conservation, and downstream demand satisfaction constraints, forming a high-fidelity distributed digital mirror. This method overcomes the excessive reliance on communication stability in traditional methods, maintaining accurate perception of the system state even under network fluctuations, significantly improving the real-time performance and reliability of state assessment.

[0042] Secondly, at the level of anomaly detection and impact assessment, this invention achieves a multi-dimensional assessment of system health by simulating downstream impact propagation and upstream anomaly tracing in a distributed digital mirror. Based on the performance degradation of nodes along the propagation path, the importance of the topology, and the vulnerability of equipment, potential impact values ​​are calculated, generating a health status map that includes anomaly location, impact range, propagation path, and risk heatmap information. This method overcomes the shortcomings of traditional methods that insufficiently consider fault propagation mechanisms, accurately assessing the cascading impact of local anomalies on the overall system, and providing a comprehensive and quantitative basis for maintenance decisions. By embedding physical laws and network topology constraints, the accuracy of anomaly detection and the relevance of impact assessment are significantly improved.

[0043] Finally, at the level of maintenance decision-making and execution optimization, this invention analyzes the health status map into local intervention tasks, generates local preference solutions through relevant agents, and achieves a maintenance consensus with optimal global benefits through multiple rounds of negotiation. This generates and distributes a distributed maintenance action plan that coordinates time, space, and resources. This multi-agent negotiation-based decision-making mechanism overcomes the subjectivity and limitations of traditional manual decision-making, enabling globally optimized scheduling of maintenance tasks under complex constraints. By considering spatiotemporal conflicts and resource competition, it ensures efficient coordination of maintenance actions, avoids resource waste and operational conflicts, and significantly improves the overall efficiency and economy of maintenance operations.

[0044] In summary, this invention achieves full-chain intelligence in tunnel drainage systems, from condition monitoring and health assessment to maintenance decision-making, through end-to-end innovation in distributed collaborative sensing, physical law embedding, influence propagation simulation, and multi-agent intelligent decision-making. It effectively solves the technical deficiencies of traditional methods in terms of real-time performance, accuracy, globality, and collaboration, and provides reliable technical support for the safe, efficient, and economical operation of tunnel drainage systems. Attached Figure Description

[0045] Figure 1 A flowchart of a method for assessing the health of a tunnel drainage system;

[0046] Figure 2 This is a flowchart of a method S100 for assessing the health of a tunnel drainage system.

[0047] Figure 3 This is a flowchart of method S200 for assessing the health of a tunnel drainage system.

[0048] Figure 4 This is a flowchart of method S300 for assessing the health of a tunnel drainage system.

[0049] Figure 5 This is a flowchart of method S400 for assessing the health of a tunnel drainage system.

[0050] Figure 6 The convergence curve of the objective function in a distributed game;

[0051] Figure 7 The potential impact value of each node;

[0052] Figure 8 A heat map showing the health status;

[0053] Figure 9 Predicted performance recovery curves for critical nodes;

[0054] Figure 10 , Figure 11 , Figure 12 and Figure 13This is an interface diagram of the tunnel full drainage system health assessment system. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0056] 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.

[0057] The core logic of this method is as follows: establish a virtual agent network that is topologically isomorphic to the physical system, realize distributed data perception and local physical model inference through edge computing nodes, and form node-level features; based on these features, the virtual agents reach a global state consensus under the condition of satisfying physical constraints through game theory methods, and construct a high-fidelity digital mirror; on this mirror, simulate fault propagation, quantify the impact and generate a health status; finally, through a multi-agent negotiation mechanism, transform the maintenance task into a globally optimized collaborative action plan, forming a closed-loop intelligent operation and maintenance system from perception, assessment to decision execution.

[0058] Before implementing this method, the physical deployment and software initialization of the system must be completed. The specific steps are as follows:

[0059] Step 1. Physical Topology Construction and Smart Node Deployment:

[0060] First, based on the as-built drawings and on-site survey data of the tunnel drainage system, the entire drainage system is abstracted as a directed topological network. Nodes in the network represent specific physical facilities, including but not limited to: water inlets for each tunnel section, manholes distributed along the route, sedimentation tanks, various levels of booster pump stations, and the final outlet. Directed edges between nodes represent connecting pipes or ditches, their direction determined by the water flow direction. The sets of directly upstream and directly downstream nodes for each node are clearly defined.

[0061] At each key node in the network topology, an intelligent edge computing unit integrating a microprocessor, memory, local power supply, and wired / wireless communication modules is deployed. Simultaneously, a sensor array is configured for each node based on its function. For example, ultrasonic level gauges are installed at water level monitoring points, electromagnetic flow meters at flow monitoring points, and vibration sensors, temperature sensors, and electrical parameter acquisition modules are installed at pump station nodes. All sensor installations must comply with engineering specifications and undergo initial calibration.

[0062] Step 2. Data Acquisition Framework and Local Model Initialization:

[0063] All edge computing units synchronize their time using a unified clock synchronization protocol to ensure a consistent time base across the entire system. The system has a preset global data acquisition cycle, which can be flexibly configured according to actual operation and maintenance needs, typically ranging from 5 minutes to 1 hour. At the end of each acquisition cycle, all units synchronously trigger their connected sensors to acquire multi-dimensional raw data such as flow rate, pressure, water quality, and equipment status.

[0064] Upon startup, each edge computing unit loads a local physical consistency model matching its node type. The local physical consistency model is a set of mathematical relationships describing the hydraulic characteristics or operational patterns of the node. These models include: for pipeline nodes, the core model is based on the law of mass conservation, describing the relationship between internal water volume changes and inflow / outflow rates; for pump station nodes, the model is based on their performance curves, describing the relationship between flow rate, head, and power. In the initial stage of system operation, using collected historical data from normal operating conditions, an offline calibration algorithm is used to preliminarily calibrate the parameters of each local model, ensuring it accurately reflects the node's physical behavior under normal conditions. After calibration, the model transitions to online operation mode.

[0065] like Figure 1 As shown, this application illustrates an exemplary method for assessing the health of a tunnel drainage system, specifically including the following steps:

[0066] S100. Topology-aware distributed state evolution;

[0067] S200. Digital mirror consensus driven by local game theory;

[0068] S300. Emerging health trends affecting transmission simulation;

[0069] S400. Generation of distributed collaborative maintenance strategies.

[0070] Please refer to Figure 2 The diagram illustrates a flowchart of an exemplary energy storage rapid compensation method S100 based on the identification of voltage fluctuation characteristics of rail contact network. This step aims to enable each physical node not only to perceive its own state, but also to combine information from upstream and downstream neighbors to collaboratively deduce the evolution trend of its own state and form a comprehensive local state characteristic description.

[0071] To achieve the above technical objectives, the specific contents include:

[0072] S110. Local Data Acquisition and Feature Extraction:

[0073] At each preset acquisition time, the edge computing unit synchronously reads data from all connected sensors; after data reading, a standardized preprocessing pipeline is immediately executed:

[0074] First, filtering based on physical limits is performed to remove outliers that clearly exceed reasonable ranges; then, smoothing based on statistical characteristics is performed to filter out high-frequency noise.

[0075] Finally, a quantified confidence weight is calculated for each data point. The confidence weight takes into account the sensor’s own health status, the recent stability of the data, and the degree of agreement with the predictions of the local simple physical model.

[0076] Among them, the quantified credibility weight The calculation formula is: ; For sensors At the node Real-time health factors; For sensors At the present moment The observed original physical quantity values; This is the average of historical observations; Standard deviation of historical observations;

[0077] After completing preprocessing steps such as data cleaning, filtering, and confidence weighting, a set of key feature values ​​that can characterize the current operating state and instantaneous change trend of a node are extracted to form a node. The observed feature vector;

[0078] Key characteristic values ​​include, but are not limited to: verified water level measurements, water level change rate calculated based on time series, average flow rate over the period, and vibration energy of equipment in specific frequency bands extracted through spectrum analysis.

[0079] S120. Neighborhood status information exchange:

[0080] Each edge computing unit identifies all its direct upstream and downstream node sets based on a predefined topology. The edge computing unit encapsulates its processed observation feature vector and its confidence weight vector into a standard format data packet. The data packet is then sent to all neighboring nodes via an industrial Ethernet or wireless mesh network. Simultaneously, the edge computing unit continuously listens for and receives similar data packets from all its neighboring nodes. Through this bidirectional communication, each node obtains real-time status information of its local neighborhood, including upstream water inflow forecasts and downstream drainage needs.

[0081] The pre-defined topology is stored in the edge computing unit in the form of a structured node connection configuration table;

[0082] The configuration table uses node IDs as an index and explicitly records the set of direct upstream node IDs, the set of direct downstream node IDs, and connection attribute parameters for each node. This information is predetermined based on the as-built drawings of the tunnel drainage system and on-site survey data, and is fixed to the local memory of each edge computing unit during system deployment.

[0083] Specifically:

[0084] Node Identifier: Each physical facility is assigned a unique ID.

[0085] Flow direction relationship: Based on the actual water flow direction, clarify the upstream source and downstream destination of each node to form a directed connection relationship.

[0086] Connection parameters: Preset key hydraulic parameters for each connecting pipe or channel for subsequent state deduction and impact simulation.

[0087] S130. State deduction based on local model:

[0088] Each edge computing unit is configured and runs a local physical consistency model corresponding to its physical node type. The local physical consistency model is a lightweight mathematical or data-driven model embedded in the edge computing unit to describe the hydraulic behavior and evolution law that the node should follow under the current known physical boundary conditions.

[0089] Specifically, the local physical consistency model receives the following three parts of input information:

[0090] Preprocessed local high-confidence observation data: including local sensor measurements such as water level, pressure, and flow rate that have been cleaned, filtered, and weighted for confidence.

[0091] Received upstream node state prediction information: future state predictions derived from its direct upstream neighbor nodes based on its own model;

[0092] The expected state of the water output from the downstream node is the hydraulic condition required for its normal operation, which comes from its direct downstream neighbor nodes.

[0093] The model's derivation process is as follows: taking the state of the node at the previous acquisition time as the initial state, and using the three parts of input information as boundary conditions and disturbance information, the model derives the key state prediction value of the node in a short period after the current time by solving the simplified hydraulic equation or state transition relationship within a time step. Typical prediction outputs include: predicted water level, predicted outflow, or predicted pressure.

[0094] S140. Generate node-level evolutionary feature vectors:

[0095] Finally, the edge computing unit integrates and encapsulates quantitative information from multiple dimensions, including the state prediction output by the local physical consistency model, the deviation between the prediction and the actual observation, the model confidence, the sensitivity of the node to upstream changes, and the importance of the node's state to meeting downstream needs, into a structured node-level evolutionary feature vector. The node-level evolutionary feature vector is a comprehensive and quantitative snapshot of the node's future state and its local environment, and it serves as the basic data unit for subsequently constructing a global digital mirror.

[0096] Please refer to Figure 3 The diagram illustrates a flowchart of an exemplary energy storage rapid compensation method S200 based on the identification of voltage fluctuation characteristics of the rail contact network. This step aims to construct a virtual agent network at the computing layer that is completely isomorphic to the topology of the physical system. Through distributed game and collaborative optimization among agents, all agents can reach a stable consensus on the global state of the system, forming a high-fidelity distributed digital mirror.

[0097] To achieve the above technical objectives, the specific contents include:

[0098] S210. Virtual Proxy Network Initialization:

[0099] On cloud or edge server clusters, virtual agents are created one-to-one according to the physical topology map to form a virtual agent network; each agent program establishes an association with its corresponding physical node and edge computing unit; the agent loads the local physical consistency model corresponding to the node and uses the node-level evolution feature vector generated in step S140 as its initial input data.

[0100] S220. Constraint Construction and Prediction Coordination:

[0101] Each virtual agent constructs a local optimization problem with multiple local constraints based on the role of the physical node it represents;

[0102] The constraints mainly include:

[0103] Local observation matching constraints mean that the model predictions should be as close as possible to the high-confidence local actual observations.

[0104] The upstream quality conservation constraint requires that the predicted inflow should be consistent with the sum of the predicted outflows of all upstream agents to satisfy the quality conservation.

[0105] If downstream demand meets the constraints, the predicted output state should be able to meet the minimum operating requirements of all downstream agent nodes.

[0106] Optimization objective function of the agent Defined as: ; Match constraint weight coefficients to local observations; For upstream quality conservation constraints; The upstream quality conservation constraint weighting coefficient; For upstream quality conservation constraints; The constraint weighting coefficient is used to satisfy downstream demand. To meet constraints on downstream demand;

[0107] The agent's goal is to adjust its own model parameters. The goal is to minimize the overall bias of the prediction while satisfying these constraints.

[0108] S230. Distributed Game Theory and Parameter Iteration:

[0109] All agents execute the following iterative process in parallel and in multiple rounds:

[0110] In each round, each agent optimizes and adjusts its own model parameters only, assuming that the parameters of its neighboring agents remain unchanged, in order to better meet local constraints.

[0111] Parameter optimization is performed using gradient descent. ; For the first After round of iteration, the agent Model parameters; For the first After round of iteration, the agent Model parameters; The learning rate controls the step size for parameter updates; To optimize the objective function with respect to parameters The gradient;

[0112] After completing its own parameter optimization, the agent immediately broadcasts its new parameter values ​​and the latest state prediction values ​​calculated based on the new parameters to all its upstream and downstream neighbor agents; all agents synchronously perform the local optimization-information exchange steps.

[0113] S240. Nash Equilibrium and Consensus Reached:

[0114] The parameter iteration process continues, monitoring the changes in parameters of all agents across the network in real time. When the parameter changes are consistently below a preset small threshold for several rounds, the parameters and state predictions of all agents no longer change significantly, and the distributed game reaches a Nash equilibrium. At this point, the entire virtual agent network reaches a stable consensus on the global operating state of the system. The stable consensus set constitutes a high-fidelity distributed digital mirror of the current moment, which is a consistent estimate of the most likely true state of the system under the constraints of all local physical laws and global topological connections.

[0115] Please refer to Figure 4The document illustrates a flowchart of an exemplary energy storage rapid compensation method S300 based on the identification of voltage fluctuation characteristics of the rail contact network. This step aims to simulate the propagation path of the abnormal impact based on the mirror image when a node in the digital image is detected to have an abnormal performance, quantify potential risks, and automatically generate an intuitive global health status map.

[0116] To achieve the above technical objectives, the specific contents include:

[0117] S310. Anomaly Detection and Assessment Trigger:

[0118] Define a comprehensive performance metric for each proxy node in the digital mirror; the comprehensive performance metric is composed of three weighted parts:

[0119] First, the magnitude of the model prediction residuals is calculated by comparing the model predictions with the actual observed values.

[0120] Second, the degree to which its key operating parameters deviate from the normal baseline is obtained by calculating the weighted deviation of multiple key parameters;

[0121] Third, the inherent health score is based on the equipment's history and is calculated by weighting factors such as the equipment's service life, historical failure frequency, and maintenance records.

[0122] Overall performance indicators Represented as: ; Predict residuals for the model; The degree to which key operating parameters deviate from the normal baseline; Assess inherent health score; Weighting coefficients for model prediction residuals; Weighting coefficients for the degree to which key operating parameters deviate from the normal baseline; Weighting factor for inherent health score;

[0123] A dynamic performance threshold is preset for each type of node. When the monitoring system finds that the performance index of a certain node is consistently lower than its threshold, it automatically triggers the system-level health impact assessment process.

[0124] S320. Simulation of Two-Way Influence Propagation:

[0125] Starting with the triggered abnormal node, simulations are performed on a distributed digital mirror that has reached a consensus. The simulations are divided into two directions:

[0126] First, the simulation of downstream impact propagation is conducted, analyzing the cascading effects of a node's performance degradation on all downstream nodes along the flow direction, such as rising water levels and insufficient flow. During the simulation, for each downstream node, the changes in hydraulic state caused by anomalies at upstream nodes are calculated. ;in For downstream nodes The change in water level, upstream node Abnormal traffic volume upstream node to downstream nodes The length of the pipe, For downstream nodes The water flow area, It is the acceleration due to gravity. This is the hydraulic loss coefficient for the pipeline.

[0127] Second, the simulation of anomalies in the reverse flow direction analyzes whether anomalies at upstream nodes are the root cause of problems at this node. The simulation process detects abnormal characteristic parameters of upstream nodes, such as abnormal pressure fluctuations. Or sudden changes in flow To determine the abnormal propagation path: or ;in A preset threshold for abnormal pressure fluctuations; A preset abnormal threshold for traffic surges;

[0128] The simulation process records all visited nodes, forming a complete chain of influence propagation.

[0129] S330. Calculation of Potential Impact Value:

[0130] For each node in the propagation chain, calculate the potential performance degradation it may suffer during this anomaly.

[0131] Potential performance degradation that may occur during abnormal events The calculation formula is: ; For nodes Baseline performance values ​​under normal operating conditions; For nodes At the present moment The actual performance index values; The function ensures that the degradation amount is non-negative, meaning that degradation is only included when the current performance is below the baseline;

[0132] Then, overall performance degradation The structural importance weight of nodes in the entire topology network And the vulnerability score of the node device itself. Calculate the potential impact value of this node: The structural importance weight of a node in the entire network topology. The weight value represents the importance of a node in the entire network topology and is obtained by calculating the network centrality index of the node. For example, the weight value is higher for key pump stations, main pipelines, or network hub nodes, and lower for edge nodes or secondary branch nodes. Normalized betweenness centrality can be used in the specific calculation.

[0133] Vulnerability score of the node device itself The results are obtained by weighting factors such as equipment age, historical failure rate, and maintenance difficulty.

[0134] Finally, the potential impact values ​​of all nodes in the propagation chain are summed to obtain the total potential impact value of this abnormal event.

[0135] S340. Generation of Health Status Map:

[0136] The system integrates and visualizes the above analysis results to generate a multi-layered health status map. The health status map uses a physical topology map as its base map and overlays the following information layer by layer:

[0137] Anomaly nodes are highlighted; arrows of different colors and thicknesses are used to dynamically display the propagation paths of the impact along and against the current; a risk heatmap of the entire network is rendered using color gradients based on the potential impact value of each node; and the most critical propagation paths with the most severe impact are specially marked.

[0138] The graph is accompanied by a structured report that details all anomalous nodes, their scope of impact, propagation paths, and quantified impact values.

[0139] Please refer to Figure 5 The document illustrates a flowchart of an exemplary energy storage rapid compensation method S400 based on the identification of voltage fluctuation characteristics of the rail contact network. This step aims to transform the health status map into a specific, executable, and globally optimized collaborative maintenance action plan, resolve resource conflicts through multi-agent negotiation, and maximize maintenance benefits.

[0140] To achieve the above technical objectives, the specific contents include:

[0141] S410. Task Decomposition and Action Mapping:

[0142] The system analyzes the health status map, transforming each identified anomalous node-impact value pair into a specific local intervention task. Each task includes the target node, the performance indicators to be repaired, the expected target value, and the task urgency level calculated based on the total potential impact value and node importance. The system maintains a knowledge base containing various standard maintenance operations. Based on the anomaly type of the task and the target equipment model, the system automatically matches and recommends one or more candidate basic maintenance actions from the knowledge base.

[0143] S420. Local Preference Scheme Generation:

[0144] Each maintenance task is assigned to the virtual agent corresponding to the physical node it involves, which acts as the responsible agent for that task. Based on the local context information it possesses, the responsible agent selects a specific solution from the candidate actions and generates a detailed local preference maintenance plan. The local preference maintenance plan includes: the specific action to be executed, a list of required resources, estimated time consumption, a prediction of the impact on node performance during construction, and a local utility score that integrates effectiveness, cost, and impact.

[0145] S430. Multiple rounds of negotiation and consensus reached:

[0146] All agents that generated maintenance tasks enter a multi-round negotiation process. In each round, each agent broadcasts its current preferred solution. Upon receiving other solutions, each agent checks for resource, time, or spatial conflicts. Once a conflict is detected, the agents automatically adjust their solutions according to preset conflict resolution rules, such as modifying construction time or adjusting resource allocation. Through multiple rounds of proposals, conflict detection, and adjustments, all agents ultimately reach a consensus within a given decision-making time window on a final solution set that maximizes overall maintenance efficiency and eliminates all conflicts.

[0147] S440. Generation and Distribution of Collaborative Maintenance Plans:

[0148] After the consultation, the system integrates all the finalized maintenance plans to generate a detailed distributed collaborative maintenance action plan. The distributed collaborative maintenance action plan is a complete schedule that clarifies the execution actions, start and end times, resource allocation, responsible team, expected performance recovery curve, and dependencies between tasks for each task. The distributed collaborative maintenance action plan ensures that the system-level performance impact can be minimized to the greatest extent under limited resource constraints, while avoiding mutual interference between maintenance actions. Finally, the plan is distributed to the relevant edge computing units and the central scheduling system to guide the orderly execution of on-site maintenance operations.

[0149] Example 2

[0150] I. Case Background and Physical Topology

[0151] A highway tunnel in a mountainous area is 3.2 km long and designed to drain 5.0 m³ / s. Based on the as-built drawings and site survey data, the drainage system is abstracted as a directed topological network, containing five key nodes: inlet (N1), inspection well (N2), sedimentation tank (N3), booster pump station (N4), and outlet (N5). The directed edges between nodes are determined by the direction of water flow, and the lengths of the connecting pipes and hydraulic parameters are shown in Table 1.

[0152] Table 1 Parameters of Tunnel Drainage System Nodes and Pipelines

[0153] N1 Inlet — N2 200 2.5 9.8 0.025 N2 Inspection well N1 N3 350 3 9.8 0.03 N3 Sedimentation tank N2 N4 150 8 9.8 0.02 N4 pumping station N3 N5 80 2.0 (Exit pipe) 9.8 0.035 N5 water outlet N4 — — — — —

[0154] Each node is equipped with an edge computing unit and corresponding sensors: N1 is equipped with an ultrasonic level gauge and an electromagnetic flow meter; N2 is equipped with a level gauge; N3 is equipped with a level gauge and a turbidity meter; N4 is equipped with a vibration sensor, a temperature sensor, an electrical parameter acquisition module, and a pump outlet pressure gauge; N5 is equipped with a level gauge and a flow meter. The data acquisition cycle is set to 15 minutes, and the clock synchronization accuracy is ±1ms.

[0155] II. Distributed State Evolution of S100 Topology Sensing

[0156] The system synchronously triggers sensors at each node. The raw data is then filtered by physical limits, smoothed, and weighted to obtain the observed feature vector. Taking the normal time before the anomaly occurs as an example, the raw data and confidence weights of node N4 are shown in Table 2.

[0157] Table 2 N4 Node Sensor Data and Reliability Weight Calculation

[0158] Outflow rate (m³ / s) 4.52 4.5 0.2 0.98 0.95 Pump outlet pressure (bar) 3.21 3.18 0.15 0.96 0.93 Vibrational energy (mm / s) 1.85 1.2 0.3 0.85 0.79 Motor current (A) 342 335 12 0.97 0.9

[0159] The extracted key feature vectors are: [water level = 1.82m, water level change rate = +0.03m / min, average flow rate = 4.52m³ / s, vibration characteristic frequency band energy = 1.85mm / s].

[0160] Each node broadcasts its feature vector to its neighboring nodes. N4 receives the predicted flow rate of 4.60 m³ / s from upstream N3 and the demand water level of 1.50 m from downstream N5. N4's local physical consistency model, based on the pump performance curve and incorporating boundary conditions, extrapolates the predicted water level of 1.85 m, predicted flow rate of 4.53 m³ / s, and predicted pressure of 3.20 bar for the next 15 minutes.

[0161] N4 integration forms a node-level evolutionary feature vector: {Predicted water level deviation: +0.03m, Model confidence: 0.92, Upstream sensitivity: 0.74, Downstream importance: 0.88}. The feature vectors of all nodes constitute a local state snapshot, serving as input to the digital mirror.

[0162] III. Digital Mirror Consensus Driven by S200 Local Game Theory

[0163] Five virtual agents (A1~A5) were created on the edge server, and the models of each node were loaded. Each agent constructed a constrained optimization problem involving local observation matching, upstream quality conservation, and downstream demand satisfaction. Gradient descent was used for 20 iterations with a learning rate η=0.1.

[0164] Table 3. Changes in parameters (pump station efficiency coefficient) of each agent model during the game iteration process.

[0165] 0 0.52 0.48 0.75 0.85 0.62 3.21 5 0.531 0.491 0.748 0.832 0.634 1.85 10 0.535 0.495 0.745 0.812 0.641 0.94 15 0.536 0.496 0.743 0.796 0.645 0.38 20 0.536 0.496 0.743 0.791 0.646 0.15

[0166] After 20 iterations, the parameter change amplitude was below 1e-4 for three consecutive iterations, reaching Nash equilibrium. Figure 6 The convergence curve of the global objective function value with each iteration is shown. It can be seen that the system value drops rapidly after the 10th iteration and stabilizes at around 0.15 by the 20th iteration, indicating that all agents have reached a high-precision consensus on the global state of the system and formed a high-fidelity digital mirror.

[0167] IV. Emerging Health Trends in S300 Impact Propagation Simulation

[0168] The comprehensive performance index of N4, Φ = 0.75·residual + 0.20·parameter deviation + 0.05·inherent health, is calculated as follows: model prediction residual 0.28 (normal ≤ 0.10), parameter deviation 0.55 (normal ≤ 0.20), and inherent health score 0.70. The calculated Φ = 0.75 × 0.28 + 0.20 × 0.55 + 0.05 × 0.70 = 0.355, which is below the dynamic threshold of 0.65, triggering an impact assessment.

[0169] Simulation along the flow starting from N4: The water level change at downstream N5 is 0.18m, exceeding the allowable value of 0.10m. Tracing back to the source: The abnormal pressure fluctuation at upstream N3, ΔP=0.12bar, exceeds the threshold of 0.08bar, indicating that N3 is also affected by the anomaly. The impact propagation chain is N3←N4→N5.

[0170] S330 Potential Impact Value Calculation

[0171] Table 4 shows the performance degradation of each node, its structural importance, vulnerability score, and the final potential impact value.

[0172] Table 4 Calculation of Potential Impact Values ​​for Each Node

[0173] N3 0.92 0.78 0.14 0.45 0.62 0.039 N4 0.95 0.55 0.4 0.85 0.88 0.299 N5 0.88 0.73 0.15 0.3 0.45 0.02

[0174] The total potential impact value is 0.358. Figure 7 The bar chart showing the potential impact values ​​of each node clearly shows that pump station N4 has the highest potential impact value at 0.299, making it the main source of risk; sedimentation tank N3 is next at 0.039; and outlet N5 has the least impact.

[0175] The system generates a multi-layered health status map: the abnormal node N4 is highlighted; the downstream impact path of N4→N5 is shown with a thick red arrow, and the upstream source tracing path of N3←N4 is shown with a blue arrow; the risk heat map is rendered with color gradient (N4 area is dark red, N3 is light red, and N5 is slightly yellow). Figure 8A heatmap showing the distribution of the normalized risk index is presented. The risk index for node N4 is as high as 0.85, N3 is 0.21, N1 and N2 are close to 0, and N5 is 0.15, consistent with the calculation results in Table 4. The map includes a structured report: Anomaly node N4 affects areas N3 and N5, and the key propagation path is N4→N5.

[0176] V. Generation of S400 Distributed Collaborative Maintenance Strategy

[0177] Based on the anomaly type, a local intervention task is generated for N4: replace the pump bearing + impeller dynamic balancing, with an urgency level of A+. The knowledge base matches two candidate actions: (a) on-site repair, taking 4 hours and costing 8k; (b) complete replacement of the backup unit, taking 1.5 hours and costing 25k. The N4 agent calculates the local utility score and selects option (b). Task N3 involves cleaning the sedimentation tank sludge, requiring 2 hours and costing 3k; task N5 involves temporarily lowering the outlet threshold, requiring 0.5 hours and costing 0.5k.

[0178] The three agents entered into two rounds of negotiations: In the first round, it was discovered that both the N3 cleanup and the N4 replacement required the same mini excavator; the N4 agent proposed adjusting the construction time from 14:00-15:30 to 22:00-23:30, and the N3 agent agreed to adjust it to 1:00-3:00 AM. There were no conflicts with the N5 task. In the second round, no new conflicts arose, and all agents reached a consensus.

[0179] The final distributed collaborative maintenance action plan is shown in Table 5.

[0180] Table 5 Collaborative Maintenance Action Plan

[0181] T1 N4 Replace the backup pump 22:00-23:30 One crane and three electricians Electromechanical Class Recovered to 95% within 30 minutes T2 N3 Dredging 01:00-03:00 One excavator and two general workers Maintenance Team Recovery 2 hours after dredging T3 N5 Threshold adjustment 22:30-23:00 One automation engineer. Self-control class Effective immediately

[0182] The plan was distributed to the edge computing unit and the central dispatch system, and on-site operations were carried out in sequence. Figure 9 The performance recovery prediction curves for key node pump station N4 and sedimentation tank N3 after maintenance are shown: Within 30 minutes of replacing the backup pump, the performance index of N4 jumped from 0.55 to 0.95 and remained stable; N3 gradually recovered from 0.78 to 0.96 within 2 hours after the dredging operation was completed. The system's drainage capacity was fully restored by 04:00 the following day. Compared to traditional immediate repair solutions, this method avoids resource contention, reduces the total downtime impact from 7 hours to 5 hours, and reduces potential losses by approximately 35%.

[0183] VI. Conclusion

[0184] This embodiment, based on real tunnel drainage system topology and simulated fault data, fully verifies the effectiveness of the proposed method in distributed sensing, digital mirror consensus, impact propagation simulation, and collaborative maintenance decision-making. Experimental data shows that high-precision convergence is achieved after 20 rounds of game iteration. Figure 1 The anomaly detection sensitivity reached 0.355, which is lower than the threshold of 0.65, affecting the accurate location of key risk sources in the propagation simulation. Figure 7 , Figure 8 The maintenance plan generated through multi-agent negotiation reduces total downtime by approximately 29% compared to traditional methods. Figure 9 This method significantly improves the level of intelligent operation and maintenance of tunnel drainage systems.

[0185] Example 3

[0186] A health assessment and maintenance decision optimization system for a tunnel full drainage system, the system interface has four interfaces, such as... Figure 10 , Figure 11 , Figure 12 and Figure 13 As shown, the system includes: a topology evolution interface, a mirror consensus interface, an impact propagation interface, and a collaborative maintenance interface. The system comprises:

[0187] The distributed intelligent sensing execution layer consists of edge computing units deployed at each physical node of the tunnel drainage system. Each unit integrates a microprocessor, memory, communication module and local power supply, and is connected to one or more of the following: water level sensor, flow meter, water quality sensor, vibration sensor and equipment status acquisition module, for real-time acquisition, preprocessing and temporary storage of multi-source monitoring data of the node.

[0188] The digital mirror collaborative construction layer is deployed on edge servers or in the cloud to build a virtual agent network that is isomorphic to the physical topology. Each virtual agent is bound to a physical node and its edge computing unit and loads the corresponding local physical consistency model. By receiving node-level evolution feature vectors and based on local observation matching constraints, upstream quality conservation constraints, and downstream demand satisfaction constraints, the model parameters are iteratively optimized in a distributed game manner until a Nash equilibrium is reached, forming a high-fidelity distributed digital mirror of the entire system.

[0189] The health status assessment and inference layer, based on the distributed digital mirror, performs real-time performance index calculation and threshold comparison for each agent node. When an anomaly is detected, it automatically triggers a two-way impact propagation simulation, calculates the chain effect in the forward direction and traces the source of the anomaly in the reverse direction. Combining the node performance degradation, topology importance and equipment vulnerability, it quantifies the potential impact value and generates a multi-layer health status map that integrates abnormal nodes, impact paths and risk heat maps.

[0190] The collaborative maintenance decision and scheduling layer receives the health status map, parses it into multiple local intervention tasks, matches basic maintenance actions through the task-action mapping knowledge base, and generates preferred solutions based on the local context by relevant agents. Through a negotiation protocol of multiple rounds of proposal-conflict detection-adjustment, a maintenance consensus with the best global benefits is reached within a given time, and finally outputs a distributed maintenance action plan that coordinates time, space and resources.

[0191] The human-computer interaction and scheduling execution layer provides a graphical monitoring interface to display the real-time status of the system topology, health status map and maintenance plan. It supports operation and maintenance personnel to confirm, fine-tune and issue the plan, and decompose the plan into specific instructions to issue to the corresponding edge computing units and field maintenance terminals, so as to realize the plan execution tracking and feedback closed loop.

[0192] The system exchanges data and transmits instructions between its various layers through standard data interfaces and communication protocols, forming a full-link intelligent operation and maintenance system from state awareness, image construction, health assessment, decision generation to plan execution.

[0193] The sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0194] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for assessing the health of a tunnel drainage system, characterized in that, include: Edge computing units are deployed at each node based on the physical topology network of the tunnel drainage system. Local multi-source sensor data are collected and preprocessed synchronously, and status information is exchanged with neighboring nodes. By running a local physical consistency model, local observations, neighborhood collaboration and data credibility are fused to generate node-level evolution feature vectors. A virtual agent network is constructed that is isomorphic to the physical topology. Each agent takes the node-level evolutionary feature vector as input, and performs distributed multi-round game optimization with upstream and downstream agents based on local observation matching, upstream quality conservation and downstream demand satisfaction constraints. The model parameters are adjusted until Nash equilibrium is reached, forming a distributed digital mirror with global state consensus. In the distributed digital mirror, the comprehensive performance indicators of each agent node are monitored in real time. When an anomaly is detected, the propagation of the downstream impact and the source tracing of the upstream anomaly are simulated starting from the agent node. The potential impact value is calculated based on the performance degradation of the nodes on the propagation path, the importance of the topology, and the vulnerability of the equipment. A health status map containing anomaly location, impact range, propagation path, and risk heat information is generated. The health status map is analyzed into local intervention tasks and matched with basic maintenance actions. Relevant agents generate local preference solutions, and through multiple rounds of negotiation, a maintenance consensus with the best global benefits is reached. A distributed maintenance action plan with time, space, and resource coordination is generated and issued.

2. The method for assessing the health of a tunnel drainage system according to claim 1, characterized in that, The generation of node-level evolutionary feature vectors specifically includes: Local sensor data is subjected to outlier filtering based on physical limits and noise filtering based on statistical characteristics. For each data point, a quantitative confidence weight is calculated to integrate the sensor’s own health, recent data stability and the degree of agreement with simple model predictions. Based on the preset topological relationships, the observed feature vectors and the confidence weight information are exchanged with all directly adjacent nodes. The preprocessed and weighted local observation data, the received upstream node state prediction information, and the downstream node state requirement expectation are jointly input into the local physical consistency model to deduce the key state prediction value of this node in a short period of time in the future. The predicted state values, prediction bias, model confidence, and the sensitivity and importance of nodes in the topology are integrated and encapsulated to generate a structured node-level evolution feature vector.

3. A method for assessing the health of a tunnel drainage system according to any one of claims 1 or 2, characterized in that, The local physical consistency model is a lightweight mathematical model that describes the hydraulic characteristics of nodes or the operating rules of equipment, and is used to deduce node state changes based on physical boundary conditions.

4. The method for assessing the health of a tunnel drainage system according to claim 1, characterized in that, The steps for forming a distributed digital mirror specifically include: On cloud or edge server clusters, virtual proxies are created one-to-one with physical nodes based on the physical topology map, forming a virtual proxy network, and the corresponding local physical consistency model is loaded. Each virtual agent constructs a local optimization problem that includes local observation matching constraints, upstream quality conservation constraints, and downstream demand satisfaction constraints. The objective function of the local optimization problem is the weighted sum of the losses of the three constraints. All agents execute multiple rounds of iterations in parallel. In each round of iteration, each agent optimizes and adjusts its own model parameters to minimize the objective function by assuming that the parameters of its neighbors remain unchanged. Then, it broadcasts the updated parameters and the latest state prediction calculated based on the new parameters to all upstream and downstream neighbor agents. When the parameter changes of all agents in the network are lower than the preset threshold for several consecutive rounds, it is determined that a Nash equilibrium has been reached. The virtual agent network reaches a stable consensus on the global operating state of the system and forms the distributed digital mirror.

5. The method for assessing the health of a tunnel drainage system according to claim 1, characterized in that, The comprehensive performance index of the agent node is obtained by weighted summation of three parts: model prediction residual, the degree of deviation of key operating parameters from the normal benchmark, and the inherent health score based on the service life of the equipment and historical maintenance records. The potential impact value of the proxy node is obtained by multiplying the node's performance degradation, its normalized betweenness centrality weight in the entire drainage network topology, and its vulnerability score, which reflects the equipment's age, historical failure rate, and maintenance difficulty.

6. The method for assessing the health of a tunnel drainage system according to claim 1, characterized in that, The downstream impact propagation is quantified by calculating the effect of abnormal flow at upstream nodes on water level changes at downstream nodes; the upstream anomaly tracing is determined by detecting whether abnormal pressure fluctuations or flow mutations at upstream nodes exceed preset thresholds to determine the anomaly propagation path.

7. The method for assessing the health of a tunnel drainage system according to claim 1, characterized in that, The multi-round negotiation process specifically includes: Each virtual agent is responsible for broadcasting its local preference maintenance plan. The maintenance plan should include at least the suggested actions, a list of required resources, an estimated time, a prediction of performance impact during construction, and a local utility score. Each virtual agent automatically detects whether there are resource usage conflicts, overlapping schedules, or physical space interference after receiving other proposals; If a conflict is detected, the conflict resolution plan will be automatically adjusted according to the preset conflict resolution priority rules or global optimization goals, and the process will proceed to the next round of negotiation. Within the set decision-making time window, proposals, conflict detection, and adjustments are repeated until all conflicts are resolved and the overall maintenance benefit assessment reaches its optimal level. At this point, a maintenance consensus is considered to have been reached.

8. A health assessment system for a tunnel drainage system, characterized in that, The system for implementing the method of any one of claims 1 to 7 comprises: The distributed intelligent sensing and execution module consists of edge computing hardware units and sensor arrays deployed at each physical node of the tunnel drainage system, and is used to perform data acquisition, local preprocessing and neighborhood communication. The digital mirror collaborative construction module is deployed on edge servers or in the cloud. It is used to instantiate virtual agents that correspond one-to-one with physical nodes, and to build and continuously update a distributed digital mirror that reflects the global state of the system through distributed game collaboration among the virtual agents. The health status assessment and simulation module is used to perform real-time performance monitoring, anomaly detection, two-way impact propagation simulation, potential risk quantification calculation, and generation of a visualized health status map based on the distributed digital mirror. The collaborative maintenance decision-making and scheduling module is used to parse the health status map into specific maintenance tasks and generate an optimized collaborative maintenance plan under constraints through a multi-agent negotiation mechanism. The human-computer interaction and scheduling execution module provides a graphical user interface for system status monitoring, health status display, maintenance plan review and issuance, and tracking of plan execution status.

9. The tunnel drainage system health assessment system according to claim 8, characterized in that, The edge computing hardware unit in the distributed intelligent sensing execution module integrates a microprocessor, memory, wired or wireless communication module and local power supply unit, and is connected to a sensor for monitoring one or more parameters such as water level, flow rate, equipment vibration, temperature, electrical parameters or water quality.

10. The tunnel drainage system health assessment system according to claim 8, characterized in that, Each virtual agent in the digital mirror collaborative construction module has a built-in or associated local physical consistency model that matches the hydraulic characteristics or equipment operation rules of its corresponding physical node.

Images