Gas leakage grading traceability system and method based on cooperation of fixed monitoring network and mobile robot

Abstract

The invention relates to a gas leakage grading traceability system and method based on cooperation of a fixed monitoring network and a mobile robot. The system comprises a fixed sensing assembly, a mobile execution assembly and a cooperative control assembly. The fixed sensing assembly is composed of a sensor array arranged in a limited space and is used for monitoring a concentration field in real time; the mobile execution assembly is composed of an autonomous mobile robot carrying multiple types of sensors, and has navigation, obstacle avoidance and active sniffing capabilities. The cooperative control assembly comprises a data processing unit, a source item inversion unit and a task scheduling unit, and is used for processing monitoring data, carrying out leakage source coarse positioning based on a differential evolution Markov chain algorithm, and scheduling the robot to rapidly maneuver to a target area. Compared with the prior art, through a grading cooperation mechanism of fixed network coarse positioning and robot fine traceability, response speed and positioning precision are considered, and the problems of low gas leakage traceability efficiency and poor precision in a large-scale limited space are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of public safety monitoring and industrial emergency rescue technology, and in particular to a gas leak classification and tracing system and method based on the collaboration of a fixed monitoring network and a mobile robot. Background Technology

[0002] With the deepening development of urban underground space, the construction scale of large-scale infrastructure such as underground integrated pipe corridors, large-scale petrochemical industrial parks, and long-distance natural gas pipeline networks is growing exponentially. In these complex, enclosed, and long-distance confined spaces, monitoring the leakage of flammable, explosive, or toxic gases is a core issue in ensuring the safe operation of urban lifelines. Currently, the industry mainly relies on two independent technical routes for locating gas leak sources: source term inversion based on fixed sensor networks and active sniffing based on mobile robots. However, when facing emergency response tasks in large-scale spaces, both of these single technical methods face insurmountable performance bottlenecks.

[0003] For fixed sensor networks, their advantage lies in their 24 / 7 real-time monitoring and alarm capabilities. To locate leak sources, existing technologies typically employ source term estimation (STE) algorithms based on Bayesian inference or regularization, using limited sensor data to infer leak source parameters, such as the "Improved Search Algorithm for Strong Back-Calculation of Gas Leak Sources" disclosed in Chinese Patent Application Publication No. CN111505205A. However, due to high construction and maintenance costs, the sensor deployment density at engineering sites is often low, leading to a high degree of "illness of determination" in the inversion problem mathematically. Furthermore, the airflow field within confined spaces is complex and variable, with significant deviations between theoretical diffusion models and actual flow fields. This results in large location errors in inversion results relying solely on fixed sensors, failing to meet the engineering requirements for precise point-to-point sealing during emergency repairs.

[0004] On the other hand, mobile robot active sniffing technology, due to its advantages of autonomous movement and close-range perception, can achieve centimeter-level high-precision positioning, such as the "A Gas Leakage Source Tracing Method Utilizing a Mobile Robot" disclosed in Chinese Patent Application Publication No. CN107402283A. However, this technology faces a severe contradiction between time and space efficiency in practical applications. In large-scale pipe gallery environments, if the robot starts a blind search across the entire area from the entrance, it needs to go through a lengthy plume detection phase before it can access effective concentration information. Experimental data shows that as the distance to the leak source increases, the robot's search time increases significantly and non-linearly. This inefficient search strategy not only causes the robot to consume a large amount of valuable electricity and computing resources in ineffective areas, but more seriously, for hazardous gases with rapid diffusion characteristics, a long search delay means missing the best window for initial response to an accident, which can easily lead to catastrophic consequences.

[0005] Currently, fixed monitoring systems and mobile inspection systems typically operate as two isolated units, lacking deep information fusion and collaborative control mechanisms. Existing technologies lack a hierarchical collaborative architecture capable of converting real-time global macroscopic distribution information acquired by fixed sensors into prior navigation knowledge for mobile robots, and driving the system to intelligently switch between "global coarse positioning" and "local fine tracing." Therefore, developing a hierarchical collaborative tracing system that combines the response speed of fixed networks with the positioning accuracy of mobile robots is an urgent need to solve the problem of emergency response to gas leaks in large-scale confined spaces. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a gas leak grading and tracing system and method based on the collaboration of a fixed monitoring network and a mobile robot. This effectively overcomes the shortcomings of large positioning errors of fixed sensors and long global search time of mobile robots, and realizes gas leak tracing with both high efficiency and high accuracy in large-scale complex environments.

[0007] The objective of this invention can be achieved through the following technical solutions: The first aspect of this invention provides a gas leak grading and tracing system based on the collaboration of a fixed monitoring network and a mobile robot, comprising: The fixed sensing component consists of a fixed gas sensor array arranged in a preset optimal topology along the longitudinal and vertical directions of the confined space, used for real-time monitoring of the ambient gas concentration field. The mobile execution component consists of several autonomous mobile robots equipped with lidar and gas sensors. These autonomous mobile robots are capable of autonomous navigation, local obstacle avoidance, and active sniffing for gas leaks. The collaborative control component includes a data processing unit, a source term inversion unit, and a task scheduling unit. The data processing unit is used to receive and process the concentration data monitored by the fixed sensing component to extract feature vectors. The source term inversion unit is used to perform a coarse inversion of the leakage source location based on the extracted feature vectors using the differential evolution Markov chain algorithm and output a coarse estimated coordinate. The task scheduling unit is used to generate scheduling instructions based on the coarse estimated coordinate to drive the mobile execution component to quickly maneuver to the target area.

[0008] Furthermore, the sensor spacing of the fixed gas sensor array in the longitudinal direction of the confined space is set to 6 to 8 meters, and the spacing is determined by inversion positioning error sensitivity analysis.

[0009] Furthermore, the mobile execution component is configured to automatically switch to active sniffing mode after reaching the target area, use particle swarm optimization algorithm for collaborative search, and integrate grid map area restrictions and lidar obstacle avoidance mechanism during the search process to prevent collisions with facilities.

[0010] Furthermore, the collaborative control component is also configured with hierarchical tracing logic, which automatically switches between global monitoring mode, rapid response mode, and precise tracing mode according to the development stage of the leakage event.

[0011] Furthermore, when the collaborative control component is in global monitoring mode: it continuously processes the concentration data uploaded by the fixed sensing component, and when the monitored value continuously exceeds the preset alarm threshold, it automatically triggers the system to enter the rapid response mode; When the collaborative control component is in fast response mode: it extracts the concentration data vector of the fixed sensor array in the quasi-steady-state stage of gas diffusion, uses the differential evolution Markov chain algorithm to iteratively sample in the global parameter space, calculates the posterior probability distribution of the leak source location, and outputs the coarsely estimated coordinates. When the collaborative control component is in the precise source tracing mode: the scheduling mobile execution component switches to the active sniffing mode, uses the particle swarm optimization algorithm to perform a collaborative search around the coarsely estimated coordinates, and integrates grid map constraints and lidar obstacle avoidance mechanism. At the same time, the final leakage source location is confirmed by the global maximum value stability judgment logic of multi-robot monitoring data.

[0012] A second aspect of this invention provides a method for graded source tracing of gas leaks based on the collaboration of a fixed monitoring network and a mobile robot, comprising the following steps: S1. A fixed sensor array is used to monitor the entire monitoring area. When the monitored value exceeds the preset alarm threshold, the system is triggered to enter the fast response mode. S2. Based on the concentration data vector of the sensor array in the quasi-steady-state stage of gas diffusion, a likelihood function based on the forward Gaussian diffusion model is constructed. The differential evolution Markov chain algorithm is used to perform iterative sampling in the global parameter space to calculate the posterior probability distribution of the leakage source location and output the coarse estimated coordinates of the leakage source. S3. Map the estimated coordinates to the target navigation point on the global map, and dispatch multiple autonomous mobile robots to plan the optimal path and quickly travel to the vicinity of the target point. S4. After the autonomous mobile robot reaches the target area, it switches to active sniffing mode and uses particle swarm optimization algorithm to conduct a collaborative search around the roughly estimated coordinates. It integrates grid map area restrictions and lidar obstacle avoidance mechanism, and confirms the final location of the leakage source through the global maximum value stability judgment logic of multi-robot monitoring data.

[0013] Furthermore, in S1, the specific process includes: A fixed sensor array with a preset sampling frequency enables real-time monitoring of the entire monitoring area. It receives and processes concentration data acquired by a fixed sensor array in real time, extracts feature vectors, and automatically locks the abnormal moment when the concentration monitoring value of any sensor exceeds the preset alarm threshold for a continuous preset time during the monitoring process. It enters rapid response mode and captures the concentration vectors of all sensors within a preset time window from the moment of the anomaly, thus enabling a seamless switch from normal monitoring to emergency response.

[0014] Furthermore, in S2, the specific process includes: Based on the concentration data vector extracted from S1, a likelihood function based on the forward Gaussian diffusion model is constructed to quantify the degree of matching between the observed data and the model predictions. Initialize multiple Markov chains and set the initial number of chains; The proposed samples are generated using a differential evolution mechanism, and the positions of the new samples are calculated using a differential mutation operator, where the jump rate is set to a fixed value. The acceptance probability of the proposed sample is calculated based on the Metropolis criterion, and the chain state is updated by comparing the likelihood function values. The sampling process is executed iteratively, and the convergence status of the chain is monitored. Convergence is determined when the sample variance of all chains is lower than a preset threshold. The mean of the samples after statistical convergence is used as the rough coordinate of the leakage source.

[0015] Furthermore, in S3, the specific process includes: The coarse coordinates are mapped to target navigation points in the global raster map. The mapping process is based on a coordinate transformation algorithm to ensure that the point location is consistent with the actual space. An adaptive scheduling strategy is generated based on the robot's current position. If the estimated coordinates are in an area that is more than a preset distance from the robot, the optimal path is planned. The robot cruises along the path at maximum speed and uses LiDAR for dynamic obstacle avoidance. If the roughly estimated coordinates are located at a position smaller than the preset example position on the robot, the path planning stage is skipped directly, and the fine tracing mode is started immediately. The scheduling instructions are sent to the robot cluster via wireless network. After receiving the instructions, each robot enters a rapid maneuvering state, temporarily blocking gas sensor data to prioritize travel efficiency and quickly travels to the vicinity of the target point.

[0016] Furthermore, in S4, the specific process includes: Once the robot swarm reaches the target area, it automatically switches to active sniffing mode, with each robot acting as a particle to execute the particle swarm optimization algorithm. The robot's speed update in the algorithm is based on inertia weight, individual learning factor, and group learning factor. Inertia weight controls the trend of maintaining the original speed, while learning factor adjusts the balance between individual experience and group cooperation. After the position is updated, the grid map index is used to check whether the target point is located in the drivable area. If the point is located outside the obstacle or the map, the velocity vector is forcibly adjusted to make the robot slide in the drivable direction. Simultaneously, it integrates a real-time obstacle avoidance mechanism with lidar. When an unmarked temporary obstacle is detected, the search is interrupted and a local replanning is performed. By using the global maximum value stability determination logic based on multi-robot monitoring data, if the rate of change of the maximum value is lower than the preset value within multiple consecutive iteration cycles, the location of the leak source is confirmed and the search is terminated.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1) By employing a hierarchical collaborative mechanism between a fixed monitoring network and mobile robots, the industry challenge of balancing response speed and positioning accuracy inherent in traditional single monitoring methods is fundamentally solved. A fixed sensor array enables rapid coarse positioning, followed by the dispatch of mobile robots to the target area for fine-grained searching. This architecture transforms global blind search into localized fine-grained search, significantly improving source tracing efficiency. Experiments show that this collaborative strategy can reduce the average positioning time by more than 30%. Simultaneously, the use of intelligent algorithms such as differential evolution Markov chains and particle swarm optimization ensures the accuracy of inversion and search in complex diffusion environments, achieving a balance between efficiency and accuracy.

[0018] 2) This system demonstrates excellent engineering practicality and robustness through deep integration of environmental perception and decision control. At the algorithm level, combining particle swarm optimization with grid map constraints and real-time obstacle avoidance via LiDAR ensures the safety and smoothness of the robot's autonomous search within confined spaces. At the system level, adaptive scheduling strategies and multi-mode switching logic enable it to intelligently respond to leak sources and sudden obstacles at different distances. Finally, through multi-robot data fusion and stability judgment logic, the system can reliably identify the leak source and automatically terminate the search, significantly reducing the need for manual intervention and safety risks. This provides an efficient and reliable automated solution for gas safety monitoring in chemical plants, pipeline corridors, and other similar scenarios. Attached Figure Description

[0019] Figure 1 : The overall system architecture diagram of this invention.

[0020] Figure 2 : Main flowchart of hierarchical collaborative traceability in this invention.

[0021] Figure 3 : A schematic diagram of the fixed sensor network topology and coarse inversion in this invention.

[0022] Figure 4 : Logic diagram of collaborative scheduling and mode switching in this invention.

[0023] Figure 5 : Schematic diagram of the robot precise tracing principle based on grid map constraints in this invention Figure 6 : Comparison chart of collaborative tracing and single-method tracing time in this invention.

[0024] Figure 7 The simulation verification trajectory diagram of the system in the utility tunnel scenario in this invention. Detailed Implementation

[0025] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, circuit structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0026] Example 1 The first aspect of the present invention provides a gas leak grading and tracing system based on the collaboration of a fixed monitoring network and a mobile robot, including a fixed sensing component, a mobile execution component, and a collaborative control component.

[0027] The fixed sensing component consists of a fixed gas sensor array arranged in a preset optimal topology along the longitudinal and vertical directions of the confined space, used for real-time monitoring of the ambient gas concentration field. The mobile execution component consists of several autonomous mobile robots equipped with lidar and gas sensors. These autonomous mobile robots are capable of autonomous navigation, local obstacle avoidance, and active sniffing for gas leaks. The collaborative control component includes a data processing unit, a source term inversion unit, and a task scheduling unit. The data processing unit is used to receive and process the concentration data monitored by the fixed sensing component to extract feature vectors. The source term inversion unit is used to perform a coarse inversion of the leakage source location based on the extracted feature vectors using the differential evolution Markov chain algorithm and output a coarse estimated coordinate. The task scheduling unit is used to generate scheduling instructions based on the coarse estimated coordinate to drive the mobile execution component to quickly maneuver to the target area.

[0028] The fixed gas sensor array has a sensor spacing of 6 to 8 meters in the longitudinal direction of the confined space, which is determined by inversion positioning error sensitivity analysis.

[0029] The mobile execution component is configured to automatically switch to active sniffing mode after reaching the target area, use particle swarm optimization algorithm for collaborative search, and integrate grid map area restrictions and lidar obstacle avoidance mechanism during the search process to prevent collision with facilities.

[0030] The collaborative control component is also equipped with hierarchical tracing logic, which automatically switches between global monitoring mode, rapid response mode, and precise tracing mode according to the development stage of the leakage event.

[0031] When the collaborative control component is in global monitoring mode: it continuously processes the concentration data uploaded by the fixed sensing component, and when the monitored value continuously exceeds the preset alarm threshold, it automatically triggers the system to enter the rapid response mode; When the collaborative control component is in fast response mode: it extracts the concentration data vector of the fixed sensor array in the quasi-steady-state stage of gas diffusion, uses the differential evolution Markov chain algorithm to iteratively sample in the global parameter space, calculates the posterior probability distribution of the leak source location, and outputs the coarsely estimated coordinates. When the collaborative control component is in the precise source tracing mode: the scheduling mobile execution component switches to the active sniffing mode, uses the particle swarm optimization algorithm to perform a collaborative search around the coarsely estimated coordinates, and integrates grid map constraints and lidar obstacle avoidance mechanism. At the same time, the final leakage source location is confirmed by the global maximum value stability judgment logic of multi-robot monitoring data.

[0032] This embodiment presents a gas leak classification and tracing method based on the collaboration of a fixed monitoring network and a mobile robot. See also the overall method described below. Figure 1 See the main flowchart for hierarchical collaborative traceability. Figure 2 The overall method includes the following steps: (1) Global monitoring and anomaly triggering: The monitoring area is fully covered by a fixed sensor array. When the monitored value exceeds the preset alarm threshold, the system is triggered to enter the fast response mode. (2) Coarse inversion of source terms based on DE-MC: The concentration data vector of the sensor array in the quasi-steady-state stage of gas diffusion is extracted, and a likelihood function based on the forward Gaussian diffusion model is constructed. The differential evolution Markov chain (DE-MC) algorithm is used to iteratively sample in the global parameter space to calculate the posterior probability distribution of the leakage source location and output the coarse estimated coordinates of the leakage source. x coarse , y coarse ); (3) Coordinated scheduling and rapid maneuver: The central control center will roughly estimate the coordinates ( x coarse , y coarse The target navigation point is mapped to the global map, and multiple mobile robots are dispatched from the standby area to plan the optimal path to quickly travel to the vicinity of the target point; (4) Local fine-grained tracing based on PSO: After the robot reaches the target area, it switches to active tracing mode and uses the Particle Swarm Optimization (PSO) algorithm to treat each robot as a particle and perform a collaborative search within a local area around the coarsely estimated coordinates. During the search process, the grid map area restriction and the lidar obstacle avoidance mechanism are integrated to prevent collisions with the utility tunnel facilities; (5) Terminal confirmation and result output: The final leak source location is confirmed by using the global maximum value stability determination logic of multi-robot monitoring data. x fine , y fine It will also bring the positioning accuracy within the preset error range.

[0033] In step 2 above, the specific implementation of the DE-MC algorithm includes: initialization N A Markov chain, using the difference mutation operator X new = X i + γ ( X a - X b ) + ζ Generate suggested samples, where γ For jump rate, ζ For random disturbance terms, X i This represents the current state of the Markov chain or the position vector of the current sample. X a ,X b This represents the state vector of two different samples or chains randomly selected from the current population; it determines whether to accept the suggested sample based on the Metropolis criterion; it calculates the mean of the samples after all chains converge as a coarse coordinate; this step does not require precise positioning and aims to quickly narrow down the search range.

[0034] In step 4, the control strategy for local fine-tuning is as follows: the robot's speed and position updates follow the formula: (1) In the formula, The inertial weight represents the robot's performance in maintaining its original speed; and The learning factor represents the speed that the robot learns from its own motion and the motion of the robot swarm. and Representing randomness, these values ​​range from 0 to 1 and are updated in each iteration to improve the robot's ability to explore more space. and These are the robot's maximum and minimum speeds, respectively. P g This represents the location of the highest concentration detected by all robots at the current moment. Vi(t) and Vi(t- Δt) These represent the velocity vectors of the i-th robot at the current time t and the previous time t-Δt, respectively. Pi(t) This represents the spatial coordinates of the i-th robot at the current time t.

[0035] After calculating the target location, the robot first checks whether the location is within the drivable area using the grid map index. If it is located outside of an obstacle or the map, the robot is forced to maintain its current position and wait for the next iteration.

[0036] The collaborative scheduling mechanism in step 3 also includes: if the fixed sensor inversion results show that the leak source is located in an area far from the robot's current location, the fixed sensor estimation results are used to guide the robot to perform long-distance navigation; if the inversion results show that the leak source is located near the robot, the long-distance navigation is skipped and PSO fine-tuning is started immediately.

[0037] In step 1, the fixed sensor array is deployed to meet the following requirements: in the longitudinal direction of the pipe gallery, the sensor spacing is set to 6~8m. This spacing is a critical threshold determined by the sensitivity analysis of the inversion positioning error, in order to balance the inversion accuracy and hardware cost.

[0038] The specific implementation process includes the following steps: first step: A fixed sensor array with a preset sampling frequency enables real-time monitoring of the entire monitoring area. It receives and processes concentration data acquired by a fixed sensor array in real time, extracts feature vectors, and automatically locks the abnormal moment when the concentration monitoring value of any sensor exceeds the preset alarm threshold for a continuous preset time during the monitoring process. It enters rapid response mode and captures the concentration vectors of all sensors within a preset time window from the moment of the anomaly, thus enabling a seamless switch from normal monitoring to emergency response.

[0039] Step Two: Based on the concentration data vector extracted from S1, a likelihood function based on the forward Gaussian diffusion model is constructed to quantify the degree of matching between the observed data and the model predictions. Initialize multiple Markov chains and set the initial number of chains; The proposed samples are generated using a differential evolution mechanism, and the positions of the new samples are calculated using a differential mutation operator, where the jump rate is set to a fixed value. The acceptance probability of the proposed sample is calculated based on the Metropolis criterion, and the chain state is updated by comparing the likelihood function values. The sampling process is executed iteratively, and the convergence status of the chain is monitored. Convergence is determined when the sample variance of all chains is lower than a preset threshold. The mean of the samples after statistical convergence is used as the rough coordinate of the leakage source.

[0040] More specifically, the above process includes the following specific steps: Based on the concentration data vector extracted in the first step, a mathematical tool for quantifying the model's matching degree is established. First, a physical model suitable for gas diffusion in confined spaces—the forward Gaussian diffusion model—is selected as the forward predictor. This model can predict the theoretical concentration values ​​that should occur at each fixed sensor location based on hypothetical leak source location and intensity parameters. Next, a likelihood function is constructed, whose core function is to calculate the matching probability between the concentration data vector obtained from actual monitoring and the model's predicted values. Typically, it is assumed that the observation error follows a normal distribution; the probability density of the residuals is calculated to assess the likelihood of actual observed data occurring under specific leak source parameters. A higher matching degree results in a larger likelihood function value, indicating that the set of parameters is more likely to be the actual leak source conditions.

[0041] To improve the comprehensiveness of the search and avoid getting trapped in local optima, the algorithm employs a strategy of parallel sampling across multiple Markov chains. During initialization, a certain number of initial states for each chain are randomly generated within a predefined range of leakage source parameters (e.g., location coordinate range, leakage intensity range). Each chain represents an independent search trajectory. The number of chains is a critical parameter, requiring a balance between computational efficiency and global exploration capability. Too few chains may fail to adequately explore the multi-peak characteristics of complex probability distributions, while too many chains will increase unnecessary computational overhead.

[0042] For each Markov chain in the current iteration, instead of simply randomly perturbing its vicinity to generate new samples, a differential evolution mechanism is introduced to intelligently generate proposed samples using information from other chains in the population. Specifically, for the current chain's state, two distinct individuals are randomly selected from the population, and their vector difference is calculated. This difference is multiplied by a fixed jump rate (scaling factor) and added to the current chain's state, possibly with a small random perturbation, thus generating a new candidate sample (proposed sample).

[0043] After generating a proposed sample, the current state is not directly replaced. Instead, a decision on whether to accept it is made with a certain probability based on the Metropolis criterion. First, the likelihood function values ​​corresponding to the new proposed sample and the current chain state are calculated. The acceptance probability is proportional to the ratio of the likelihood function values ​​of the new sample and the current sample. If the likelihood value of the new sample is higher, it is always accepted; if it is lower, it is accepted according to probability. This decision-making process is implemented by generating a random number and comparing it with the calculated acceptance probability.

[0044] The process of generating suggested samples and accepting them probabilistically involves multiple iterations (such as the thousands of iterations mentioned in the documentation) for each chain. During these iterations, the overall behavior of all chains is continuously monitored. Convergence is determined by observing the sample statistics (such as variance) of multiple chains. After a sufficient number of iterations, if the sample variance of all chains stabilizes below a certain threshold, it indicates that these chains are no longer wandering aimlessly but have converged to one or more stable regions (i.e., high-probability regions) in the parameter space. At this point, the sampling process can be considered to have reached a stable state.

[0045] Step 3: The coarse coordinates are mapped to target navigation points in the global raster map. The mapping process is based on a coordinate transformation algorithm to ensure that the point location is consistent with the actual space. An adaptive scheduling strategy is generated based on the robot's current position. If the estimated coordinates are in an area that is more than a preset distance from the robot, the optimal path is planned. The robot cruises along the path at maximum speed and uses LiDAR for dynamic obstacle avoidance. If the roughly estimated coordinates are located at a position smaller than the preset example position on the robot, the path planning stage is skipped directly, and the fine tracing mode is started immediately. The scheduling instructions are sent to the robot cluster via wireless network. After receiving the instructions, each robot enters a rapid maneuvering state, temporarily blocking gas sensor data to prioritize travel efficiency and quickly travels to the vicinity of the target point.

[0046] More specifically, the above process includes the following specific steps: The output coarsely estimated coordinates are precisely mapped to the target navigation point in the global grid map using a coordinate transformation algorithm, ensuring that the point is consistent with the actual physical location in the confined space, providing an accurate spatial reference for subsequent navigation. Next, an adaptive scheduling strategy is generated based on the robot's current position. By comparing the distance between the coarsely estimated coordinates and the robot with a preset threshold, a dynamic decision branch is made: if the distance is greater than the threshold, a long-distance navigation mode is activated, using the path planner to calculate the optimal path. The robot cruises along the path at maximum speed in a straight line and relies on LiDAR for real-time dynamic obstacle avoidance to ensure safe travel; if the distance is less than the threshold, the path planning stage is skipped directly, and the system immediately switches to a fine-tracking mode to save response time. Finally, the scheduling command is sent to the robot cluster via a wireless network. After receiving the command, each robot enters a rapid maneuver state, temporarily disabling gas sensor data to prioritize travel efficiency and quickly travel to the vicinity of the target point.

[0047] Step 4: Once the robot swarm reaches the target area, it automatically switches to active sniffing mode, with each robot acting as a particle to execute the particle swarm optimization algorithm. The robot's speed update in the algorithm is based on inertia weight, individual learning factor, and group learning factor. Inertia weight controls the trend of maintaining the original speed, while learning factor adjusts the balance between individual experience and group cooperation. After the position is updated, the grid map index is used to check whether the target point is located in the drivable area. If the point is located outside the obstacle or the map, the velocity vector is forcibly adjusted to make the robot slide in the drivable direction. Simultaneously, it integrates a real-time obstacle avoidance mechanism with lidar. When an unmarked temporary obstacle is detected, the search is interrupted and a local replanning is performed. By using the global maximum value stability determination logic based on multi-robot monitoring data, if the rate of change of the maximum value is lower than the preset value within multiple consecutive iteration cycles, the location of the leak source is confirmed and the search is terminated.

[0048] More specifically, the above process includes the following specific steps: Upon arriving at the target area, the robot swarm automatically switches to active detection mode, mapping each robot as an intelligent particle in a particle swarm optimization algorithm, initiating collaborative search. Driven by the algorithm, the robot's speed update mechanism comprehensively considers inertia weights, individual learning factors, and group learning factors. Inertia weights dominate the robot's inertia in maintaining its original motion trend, while individual and group learning factors respectively adjust the intensity of its learning towards its own historical best discovery position and the group's shared best position, thus balancing individual exploration experience with collective collaborative wisdom. Subsequently, after updating its theoretical position according to the algorithm, the system immediately verifies whether the target point is located within a practically passable area using a grid map index. If the verification finds a new point outside of obstacles or map boundaries, its velocity vector is forcibly corrected, guiding the robot to slide along a feasible direction allowed by the grid map, ensuring safe movement. Simultaneously, the entire process deeply integrates the real-time obstacle avoidance function of the LiDAR. Once a temporary obstacle not marked in the pre-map is detected, the current search behavior is immediately interrupted, and a local path replanning procedure is initiated. Finally, based on the monitoring data uploaded by multiple robots, the system executes the global maximum value stability judgment logic. If the rate of change of the global maximum value of the monitored gas concentration is lower than the preset stability threshold in multiple consecutive algorithm iteration cycles, it is determined that the location of the leak source has been successfully locked, and the entire search task is terminated.

[0049] Application Example 1 The overall hardware architecture and deployment of the system.

[0050] This application example demonstrates the construction of a gas leak grading and tracing system. The system's physical architecture consists of three parts: Fixed sensing components: The sensor employs 16 industrial-grade infrared methane sensors (model S-M4T-CH4, range 0-100% Vol), equipped with an RS485 communication interface. Based on the topology optimization criteria proposed in this invention, sensors are positioned every 6 m along the longitudinal direction of the pipe gallery. D opt A monitoring node is deployed at a height of 6 m. Vertically, taking advantage of the fact that methane is less dense than air, the sensor is installed at the top of the pipe rack, 0.3 m above the ceiling. All sensors are connected to the area controller (PLC) via a Modbus-RTU bus, and then the data is transmitted to the central control server via a fiber optic ring network.

[0051] Mobile Execution Components: The robotic platform is equipped with three Turtlebot3 Waffle Pi mobile robots as particles for collaborative search. Each robot is equipped with an RPlidar A3 lidar (for SLAM localization and obstacle avoidance), a high-precision portable methane detector (sampling frequency 10 Hz), and a Raspberry Pi 4B main control board. The robots establish a TCP / IP connection with the central server via the WiFi network (or 5G private network) within the tunnel, subscribing to task instructions and publishing their own status based on the ROS (Robot Operating System) Topic mechanism.

[0052] Collaborative control component: Deploy a high-performance workstation to run inversion algorithms and schedulers at all levels.

[0053] Application Example 2 Phase 1—Comprehensive Monitoring and Coarse Source Term Inversion.

[0054] The system is normally in global monitoring mode with a sampling frequency of 1 Hz.

[0055] Abnormal trigger: When any sensor in the fixed network... S i If the reading exceeds the preset alarm threshold for 3 consecutive seconds, the system will automatically lock the current time. T 0 and extract [ T 0 , T 0 Concentration vectors of all sensors within the +60 s time window O vector .

[0056] DE-MC inversion calculation: The central server calls the built-in "Confined Space Gaussian Plume Model" as the forward prediction engine, initializes 10 Markov chains, and randomly scatters points in the coordinate space of the entire pipe gallery (0-100 m, 0-3 m). The calculation is then performed using the formula... X new = X i + 0.6·( X a - X b ) + ζ Proposed locations are generated, and differential evolution iterations are performed. After approximately 1000 iterations (computation time < 5 seconds), multiple chains converge to the same region. The centroid of the sampled point cloud is calculated, and the coarsely estimated coordinates of the leakage source are output. P coarse = (25.0 m, 1.5 m), with confidence radius. R conf= 1.5 m. See the schematic diagram of the fixed sensor network topology and coarse inversion. Figure 3 , Figure 3 The layout topology of the fixed sensor array and the DE-MC inversion process are displayed intuitively.

[0057] Application Example 3 Phase Two - Multi-machine Cooperative Scheduling and Rapid Maneuver.

[0058] The server generates a scheduling instruction {"Action": "Dispatch", "Target_X": 25.0, "Target_Y": 1.5} and sends it to the robot cluster.

[0059] Path planning: After receiving instructions, the robot uses the Global Planner in the move_base package to plan the optimal path from the current standby point to the target point on a static grid map. During this stage, the robot ignores fluctuations in gas concentration along the way and moves at its maximum linear velocity (…). V max =0.5 m / s) to perform straight-line cruise, using only lidar for dynamic obstacle avoidance, thereby minimizing travel time. Figure 4 It demonstrates the logic from coarse coordinate mapping to robot scheduling, including mode switching (such as from rapid response mode to precise source tracing mode).

[0060] Application Example 4 The third stage—detailed source tracing within a confined space Figure 5 This demonstrates the principle by which a robot integrates grid map constraints and obstacle avoidance mechanisms during local search.

[0061] When the robot cluster reaches a preset range near the target point, the system automatically switches to active sniffing mode.

[0062] PSO Collaborative Search: Three robots disperse around the target point, forming a triangular encirclement. Each robot treats itself as a particle, sharing position and concentration information, and calculates the fine-tuning direction for the next moment based on the particle swarm velocity update formula. Before each move, the algorithm maps the target coordinates to a grid map index. If the target point falls inside a wall or outside a pipe gallery, the normal component of the velocity vector is forcibly set to zero, causing the robot to slide along the wall instead of crashing into it.

[0063] Dynamic obstacle avoidance: If the lidar detects a temporary obstacle not marked on the map at a distance of 0.5m ahead, the robot immediately interrupts the PSO logic and executes a local obstacle avoidance action of "backward-rotate-replan".

[0064] Application Example 5 Phase Four - Source Identification and Termination.

[0065] Stability assessment: The system monitors the maximum concentration values ​​uploaded by the three robots in real time. C max_global Introduce a counter, Count: If C max_global If no significant increase (change rate <5%) occurs within 5 consecutive iterations (approximately 10 seconds), and the current concentration value is already at a high level, then the source is determined to have been identified.

[0066] Result reporting: The system automatically stops the movement of all robots and sets the current optimal position for the group. G best Marked as the final leak point P fine In the test simulating L3 condition (leakage source located at 7.5 m), the final location result using the cooperative method of this embodiment was (7.51 m, 1.51 m), with a location error of only 0.014 m, and the entire process took only 52.05 s. Compared with single fixed sensor inversion (error 1 m) and single robot search (time 85 s), this system achieves optimal efficiency and accuracy. Figure 6 The comparison curves demonstrate the advantages of collaborative tracing over a single method in terms of time consumption. Here, a represents a near-distance leakage scenario (7.5m), b represents a medium-distance leakage scenario (50m), and c represents a long-distance leakage scenario (80m). Figure 7 The simulation results show the robot's trajectory and motion in the utility tunnel environment.

[0067] Application Example 6 Special operating conditions handling (escape mechanism).

[0068] If, during the detailed source tracing phase, the robot is affected by local airflow vortices and repeatedly circles around a high-concentration point (not at the source) for more than 30 seconds, the system will trigger an escape mechanism, temporarily blocking the population optimum. G best The attraction of the gradient. Give the robot a random impulse, forcing it to leave its current region and search for a new direction of gradient ascent over a larger area.

[0069] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A gas leak grading and tracing system based on the collaboration of a fixed monitoring network and a mobile robot, characterized in that, include: The fixed sensing component consists of a fixed gas sensor array arranged in a preset optimal topology along the longitudinal and vertical directions of the confined space, used for real-time monitoring of the ambient gas concentration field. The mobile execution component consists of several autonomous mobile robots equipped with lidar and gas sensors. These autonomous mobile robots are capable of autonomous navigation, local obstacle avoidance, and active sniffing for gas leaks. The collaborative control component includes a data processing unit, a source term inversion unit, and a task scheduling unit. The data processing unit is used to receive and process the concentration data monitored by the fixed sensing component to extract feature vectors. The source term inversion unit is used to perform coarse inversion of the leakage source location based on the extracted feature vectors using the differential evolution Markov chain algorithm and output coarsely estimated coordinates. The task scheduling unit is used to generate scheduling instructions based on the coarsely estimated coordinates to drive the mobile execution component to quickly maneuver to the target area.

2. The gas leak grading and tracing system based on the collaboration of a fixed monitoring network and a mobile robot as described in claim 1, characterized in that, The fixed gas sensor array has a sensor spacing of 6 to 8 meters in the longitudinal direction of the confined space, which is determined by inversion positioning error sensitivity analysis.

3. The gas leak grading and tracing system based on the collaboration of a fixed monitoring network and a mobile robot as described in claim 1, characterized in that, The mobile execution component is configured to automatically switch to active sniffing mode after reaching the target area, use particle swarm optimization algorithm for collaborative search, and integrate grid map area restrictions and lidar obstacle avoidance mechanism during the search process to prevent collision with facilities.

4. A gas leak grading and tracing system based on the collaboration of a fixed monitoring network and a mobile robot, as described in claim 1, is characterized in that... The collaborative control component is also equipped with hierarchical tracing logic, which automatically switches between global monitoring mode, rapid response mode, and precise tracing mode according to the development stage of the leakage event.

5. A gas leak grading and tracing system based on the collaboration of a fixed monitoring network and a mobile robot, as described in claim 4, is characterized in that... When the collaborative control component is in global monitoring mode: it continuously processes the concentration data uploaded by the fixed sensing component, and when the monitored value continuously exceeds the preset alarm threshold, it automatically triggers the system to enter the rapid response mode; When the collaborative control component is in fast response mode: it extracts the concentration data vector of the fixed sensor array in the quasi-steady-state stage of gas diffusion, uses the differential evolution Markov chain algorithm to iteratively sample in the global parameter space, calculates the posterior probability distribution of the leak source location, and outputs the coarsely estimated coordinates. When the collaborative control component is in the precise source tracing mode: the scheduling mobile execution component switches to the active sniffing mode, uses the particle swarm optimization algorithm to perform a collaborative search around the coarsely estimated coordinates, and integrates grid map constraints and lidar obstacle avoidance mechanism. At the same time, the final leakage source location is confirmed by the global maximum value stability judgment logic of multi-robot monitoring data.

6. A method for graded source tracing of gas leaks based on the collaboration of a fixed monitoring network and a mobile robot, characterized in that, Includes the following steps: S1. A fixed sensor array is used to monitor the entire monitoring area. When the monitored value exceeds the preset alarm threshold, the system is triggered to enter the fast response mode. S2. Based on the concentration data vector of the sensor array in the quasi-steady-state stage of gas diffusion, a likelihood function based on the forward Gaussian diffusion model is constructed. The differential evolution Markov chain algorithm is used to perform iterative sampling in the global parameter space to calculate the posterior probability distribution of the leakage source location and output the coarse estimated coordinates of the leakage source. S3. Map the estimated coordinates to the target navigation point on the global map, and dispatch multiple autonomous mobile robots to plan the optimal path and quickly travel to the vicinity of the target point. S4. After the autonomous mobile robot reaches the target area, it switches to active sniffing mode and uses particle swarm optimization algorithm to conduct a collaborative search around the roughly estimated coordinates. It integrates grid map area restrictions and lidar obstacle avoidance mechanism, and confirms the final location of the leakage source through the global maximum value stability judgment logic of multi-robot monitoring data.

7. The gas leak classification and tracing method based on the collaboration of a fixed monitoring network and a mobile robot as described in claim 6, characterized in that, In S1, the specific process includes: A fixed sensor array with a preset sampling frequency enables real-time monitoring of the entire monitoring area. It receives and processes concentration data acquired by a fixed sensor array in real time, extracts feature vectors, and automatically locks the abnormal moment when the concentration monitoring value of any sensor exceeds the preset alarm threshold for a continuous preset time during the monitoring process. It enters rapid response mode and captures the concentration vectors of all sensors within a preset time window from the moment of the anomaly, thus enabling a seamless switch from normal monitoring to emergency response.

8. A gas leak classification and source tracing method based on the collaboration of a fixed monitoring network and a mobile robot, as described in claim 6, is characterized in that... In S2, the specific process includes: Based on the concentration data vector extracted from S1, a likelihood function based on the forward Gaussian diffusion model is constructed to quantify the degree of matching between the observed data and the model predictions. Initialize multiple Markov chains and set the initial number of chains; The proposed samples are generated using the differential evolution mechanism, and the positions of the new samples are calculated using the differential mutation operator, where the jump rate is set to a fixed value. The probability of accepting the proposed sample is calculated based on the Metropolis criterion, and the chain state is updated by comparing the likelihood function values. The sampling process is executed iteratively, and the convergence status of the chain is monitored. Convergence is determined when the sample variance of all chains is lower than a preset threshold. The mean of the samples after statistical convergence is used as a rough coordinate for estimating the leakage source.

9. A gas leak classification and source tracing method based on the collaboration of a fixed monitoring network and a mobile robot, as described in claim 6, is characterized in that... In S3, the specific process includes: The coarse coordinates are mapped to target navigation points in the global raster map. The mapping process is based on a coordinate transformation algorithm to ensure that the point location is consistent with the actual space. An adaptive scheduling strategy is generated based on the robot's current position. If the estimated coordinates are in an area that is more than a preset distance from the robot, the optimal path is planned. The robot cruises along the path at maximum speed and uses LiDAR for dynamic obstacle avoidance. If the roughly estimated coordinates are located at a position smaller than the preset example position on the robot, the path planning stage is skipped directly, and the fine tracing mode is started immediately. The scheduling instructions are sent to the robot cluster via wireless network. After receiving the instructions, each robot enters a rapid maneuvering state, temporarily blocking gas sensor data to prioritize travel efficiency and quickly travels to the vicinity of the target point.

10. A gas leak classification and source tracing method based on the collaboration of a fixed monitoring network and a mobile robot, as described in claim 6, is characterized in that... In S4, the specific process includes: Once the robot swarm reaches the target area, it automatically switches to active sniffing mode, with each robot acting as a particle to execute the particle swarm optimization algorithm. The robot's speed update in the algorithm is based on inertia weight, individual learning factor, and group learning factor. Inertia weight controls the trend of maintaining the original speed, while learning factor adjusts the balance between individual experience and group cooperation. After the position is updated, the grid map index is used to check whether the target point is located in the drivable area. If the point is located outside the obstacle or the map, the velocity vector is forcibly adjusted to make the robot slide in the drivable direction. Simultaneously, it integrates a real-time obstacle avoidance mechanism with lidar. When an unmarked temporary obstacle is detected, the search is interrupted and a local replanning is performed. By using the global maximum value stability determination logic based on multi-robot monitoring data, if the rate of change of the maximum value is lower than the preset value within multiple consecutive iteration cycles, the location of the leak source is confirmed and the search is terminated.

Images