Coal mine gas inspection cooperative decision system and method based on multi-agent reinforcement learning

Abstract

The present application relates to a kind of coal mine gas inspection collaborative decision system and method based on multi-agent reinforcement learning, belong to coal mine safety informatization field.The system uses inspection robot to collect gas concentration in roadway and roadway structure parameters in real time by sensor;Edge computing layer uses the improved MADDPG algorithm framework to carry out data processing and analysis, realizes dynamic task allocation;Digital twin system uses grey correlation analysis module to evaluate environmental risk coefficient, carries out dynamic path planning based on roadway three-dimensional digital twin model, constructs three-dimensional dynamic gas diffusion prediction model based on LSTM network;Decision execution and feedback: distribute decision result to inspection robot execution, and correct robot pose and environmental parameters in real time by Kalman filtering.The present application can autonomously complete gas monitoring under complex geological conditions, has good adaptability and prediction ability, improves coal mine gas inspection efficiency and accuracy.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety information technology, and relates to a collaborative decision-making system and method for coal mine gas inspection based on multi-agent reinforcement learning. Background Technology

[0002] With the continuous development of coal resources, the coal mining environment is becoming increasingly complex, and gas hazards remain one of the important factors affecting coal mine safety. Many factors need to be considered in the underground coal mine environment, such as gas concentration, wind speed, pressure, air temperature, and dust concentration, all of which need to be measured and updated regularly. Traditional coal mine gas monitoring methods mainly rely on manual inspections, which suffer from high intensity, low efficiency, poor accuracy, and high safety risks. Although some intelligent inspection technologies have been developed, these technologies still struggle to effectively control and monitor the distribution and dynamic changes of gas in coal seams under complex geological conditions, and are unable to promptly detect and warn of potential hazards.

[0003] In recent years, with the development of intelligent technologies, coal mining has gradually achieved mechanization, automation, and intelligence. However, in complex mining environments, such as sudden gas eruptions, fires, or localized collapses, existing intelligent systems often suffer from response delays or misjudgments due to a lack of sufficient training data or models that rely too heavily on normal operating conditions. Furthermore, existing gas prediction methods are mainly divided into two categories: statistical methods and machine learning-based prediction models. These methods mostly operate at a single-granularity level, ignoring the heterogeneity and impact of multi-time-granularity data, resulting in low gas prediction accuracy.

[0004] Several invention patents have been granted to address the issues of insufficient efficiency and accuracy in coal mine gas inspections. For example:

[0005] CN116307637A discloses a method and apparatus for generating and distributing coal mine gas inspection tasks. The method includes acquiring the roadways requiring inspection and a first inspection item set from a mine map; identifying available gas inspection robots associated with the roadways requiring inspection; generating a set of coal mine gas inspection tasks based on the location information of the available gas inspection robots and the first inspection item set; and distributing the coal mine gas inspection tasks. This method can flexibly allocate roadway inspection tasks. However, this patent application still suffers from insufficient signal processing capabilities, requiring further improvement in the efficiency and accuracy of gas inspection tasks.

[0006] CN114183198A discloses a cluster system based on multiple gas inspection robots. This system includes multiple gas inspection robot systems, a cloud server, and a host control platform system. The gas inspection robot systems are installed at gas checkpoints, each connected to the cloud server via wired or wireless connection, and the cloud server communicates with the host control platform system. The system uses multiple independent gas inspection robot systems to detect and process gas at the field, transmitting the data to the host control platform system for aggregation and analysis, thereby achieving control of the cluster of multiple independent field systems. However, this patent application still suffers from insufficient intelligence in the gas inspection robot system, requiring further improvement in the detection accuracy and stability of the gas inspection robots.

[0007] To solve the above problems, there is an urgent need for a system that can achieve efficient and accurate gas inspection. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a collaborative decision-making system and method for coal mine gas inspection based on multi-agent reinforcement learning, solving the problems of low efficiency, poor accuracy, and lack of diverse data feature capture and intelligent analysis capabilities in existing coal mine gas inspection technologies. This system can autonomously complete gas monitoring under complex geological conditions, possesses good adaptability and predictive capabilities, and can promptly detect and warn of potential safety hazards. Simultaneously, the system also has intelligent data processing and decision-making capabilities, effectively utilizing historical data information, capturing diverse data features, and enabling multi-robot collaborative operations to improve inspection efficiency and accuracy.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] Solution 1: A collaborative decision-making system for coal mine gas inspection based on multi-agent reinforcement learning. The system adopts a three-level architecture of "cloud-edge-device", which includes a perception layer, a transmission layer, an edge computing layer, a cloud platform, and an auxiliary system.

[0011] The perception layer comprises a distributed inspection robot cluster and fixed monitoring nodes. The distributed inspection robot cluster is deployed throughout the coal mine roadways to collect roadway structural parameters and methane concentration. The fixed monitoring nodes are deployed at key locations within the roadways and include multi-parameter sensors for methane, wind speed, and temperature. The inspection robots are equipped with laser methane sensors, infrared CO2 sensors, and UWB positioning modules.

[0012] The transport layer is used to realize data transmission between the sensing layer, edge computing layer, cloud platform, and auxiliary systems. It employs a 5G private network, transmitting data in URLLC mode, and serves as the backbone transmission channel. An industrial ring network is used as a redundant channel, supporting Modbus / TCP protocols.

[0013] The edge computing layer includes a data preprocessing unit and local decision nodes. The data preprocessing unit is used to preprocess the data collected by the perception layer (such as Kalman filtering for noise reduction and feature extraction). The local decision nodes are equipped with NVIDIA Jetson AGX Orin processors, run an improved multi-agent deep deterministic policy gradient (MADDPG) algorithm, analyze the preprocessed data, and implement dynamic task allocation based on the cooperative reward function.

[0014] The cloud platform includes a digital twin system and an emergency command center. The digital twin system uses a grey relational analysis module to assess environmental risk coefficients, performs dynamic path planning based on a 3D digital twin model of the tunnel, and constructs a 3D dynamic gas diffusion prediction model based on an LSTM network. The emergency command center integrates audible and visual alarms and power-off linkage functions.

[0015] The auxiliary system includes a power supply and distribution system and a positioning reference station; the power supply and distribution system uses explosion-proof lithium batteries and intrinsically safe power supplies to provide dual-circuit power to the electrical equipment on each floor; the positioning reference station is used to provide centimeter-level RTK calibration services for the distributed inspection robot.

[0016] Furthermore, the Kalman filter state equation used in the Kalman filter noise reduction process is:

[0017] x k =F k x k-1 +B k u k +N(0,Q k )and

[0018] Where, x k U represents the state at the current time k. k F represents the system input vector. k B represents the state transition matrix. k F represents the control matrix. k and B k The dynamic equations were generated by inverting the three-dimensional twin model of the tunnel, and the model parameters were fitted by laser scanning and geological borehole data; Q k The process noise covariance is represented and calibrated based on historical gas emission data, where... V represents wind pressure. drift The wind speed drift is indicated by actual measurement using a mining ultrasonic velocimeter. The pressure change rate is taken from the barometric pressure sensor array.

[0019] Furthermore, the improved MADDPG algorithm employs a bi-objective optimization function, expressed as:

[0020]

[0021] Among them, E i (t) represents the energy consumption of the i-th robot, and n represents the number of robots; The response delay is indicated by m, which represents the number of time-stamped logs based on the robot's UWB positioning system. W1 and W2 represent weighting coefficients, which are dynamically adjusted by a fuzzy PID controller. The initial values ​​are set according to the mine's gas level (0.7 for high-gas mines and 0.3 for W2), and then updated adaptively based on the real-time gas concentration.

[0022] Furthermore, the expression for the collaborative reward function is:

[0023]

[0024] Where, r coop Represents the collaborative reward function, N robot Indicates the number of robots, ΔC gas The concentration change is represented by data acquired in real time by a laser methane sensor and denoised using a Kalman filter; d gas α represents the spatial Euclidean distance between the robot and the gas source (calculated using the UWB positioning module); α represents the urgency of the associated task; λ represents the control distance decay rate. The initial values ​​of α and λ are determined through offline training of the MADDPG algorithm and are dynamically updated every 5 minutes in the online phase based on the results of grey relational analysis.

[0025] Furthermore, the grey relational analysis module specifically calculates the grey relational degree between gas concentration, tunnel structural parameters, and environmental risk, providing a basis for risk assessment and decision-making. The grey relational degree γ is calculated using an improved Dumbledore relational degree algorithm, and its expression is:

[0026]

[0027] Where ρ is the resolution coefficient, which is preset according to the roadway risk level and ranges from 0.1 to 0.5 (ρ = 0.1 in high-gas areas and ρ = 0.5 in low-gas areas), and is dynamically optimized through the entropy weight-TOPSIS algorithm; x0(k) is the reference sequence, x i (k) is a ratio sequence taken from the sensor's historical database (such as methane concentration, temperature, vibration). The data is standardized using z-score processing to eliminate the influence of dimensions.

[0028] Furthermore, the expression for the gas diffusion prediction model is:

[0029]

[0030] Where D is the diffusion coefficient, determined by coal sample permeability test; V is the wind speed vector; R sourceThe source term is calculated using the absolute gas emission formula, and the contribution of depressurized gas from adjacent layers is added. The concentration gradient term is solved by the gradient descent algorithm combined with finite element mesh discretization.

[0031] Option 2: A collaborative decision-making method for coal mine gas inspection based on multi-agent reinforcement learning, specifically including the following steps:

[0032] S1: Real-time data acquisition: Various types of inspection robots deployed in coal mine roadways collect real-time data on gas concentration and roadway structural parameters through sensors.

[0033] S2: Data Processing and Analysis: The edge computing layer receives data collected by the inspection robot and fixed monitoring points. First, the collected data is preprocessed, and then the improved MADDPG algorithm is used for data processing and analysis to achieve dynamic task allocation.

[0034] S3: Risk Assessment and Route Planning: The cloud platform's digital twin system uses the grey relational analysis module to assess environmental risk coefficients, performs dynamic route planning based on the three-dimensional digital twin model of the tunnel, and constructs a three-dimensional dynamic gas diffusion prediction model based on the LSTM network.

[0035] S4: Decision Execution and Feedback: The digital twin system distributes the decision results to the inspection robot and the emergency command center for execution, and corrects the robot's pose and environmental parameters in real time through Kalman filtering.

[0036] The beneficial effects of this invention are as follows:

[0037] 1) This invention adopts a perception layer that combines a distributed inspection robot cluster with fixed monitoring nodes, which realizes comprehensive and efficient collection of roadway structural parameters and gas concentration, overcomes the problems of high intensity and low efficiency of traditional manual inspection, and significantly improves the safety and accuracy of gas inspection.

[0038] 2) This invention achieves intelligent data preprocessing and decision analysis at the edge computing layer through the improved MADDPG algorithm, realizes dynamic task allocation, improves the collaborative operation efficiency of inspection robots, and solves the problem of low efficiency of multi-robot collaborative operation in the prior art.

[0039] 3) This invention combines a grey relational analysis module to conduct environmental risk assessment on a cloud platform and performs dynamic path planning based on a digital twin system, thereby achieving intelligent adaptation and accurate prediction of complex geological conditions and overcoming the shortcomings of existing technologies in effectively capturing and intelligently analyzing diverse data features.

[0040] 4) This invention constructs a three-dimensional dynamic gas diffusion prediction model through an LSTM network, which realizes accurate prediction of gas concentration change trends, avoids response delays or judgment errors caused by insufficient training data, and improves the system's early warning capability.

[0041] 5) This invention uses electronic fence technology to ensure a long dwell time at the detection point, effectively reducing the rate of missed inspections and improving the quality of inspections, thus solving the problem of uncontrollable inspection quality in the existing technology.

[0042] 6) By integrating multiple sensors and intelligent devices, this invention enables real-time monitoring of multiple indicators such as gas concentration, wind speed, and temperature, and constructs a complete monitoring system for complex mine environments, overcoming the shortcomings of existing technologies that rely on a single monitoring indicator.

[0043] In summary, this invention integrates discrete detection equipment, emergency systems, and management processes into an organically collaborative intelligent entity, achieving a qualitative leap from "passive response" to "proactive prevention," and greatly improving the efficiency and accuracy of coal mine gas inspection.

[0044] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0046] Figure 1 This is a schematic diagram of the collaborative decision-making system for coal mine gas inspection based on multi-agent reinforcement learning provided in Example 1.

[0047] Figure 2 The flowchart is provided for Example 2, which describes a collaborative decision-making method for coal mine gas inspection based on multi-agent reinforcement learning. Detailed Implementation

[0048] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0049] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0050] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0051] Example 1:

[0052] Please see Figure 1 This invention provides a collaborative decision-making system for coal mine gas inspection based on multi-agent reinforcement learning. The system adopts a three-level architecture of "cloud-edge-device", specifically including a perception layer, a transmission layer, an edge computing layer, a cloud platform, and an auxiliary system.

[0053] (1) The perception layer includes a distributed inspection robot cluster and fixed monitoring nodes. The distributed inspection robot cluster consists of eight wheeled robots deployed in different areas of the coal mine roadway to collect roadway structural parameters and methane concentration. Fixed monitoring nodes are deployed at key locations in the roadway and include four methane sensors, three wind speed sensors, and two temperature sensors. The inspection robots are equipped with a laser methane sensor, an infrared CO2 sensor (model MQ-7A), a UWB positioning module (model UWB-103), and a high-definition camera.

[0054] (2) The transport layer uses a 5G private network to transmit data in URLLC mode and serves as the backbone transmission channel. An industrial ring network is used as a redundant channel, supporting Modbus / TCP protocol.

[0055] (3) The edge computing layer includes a data preprocessing unit and local decision nodes. The data preprocessing unit performs Kalman filtering noise reduction and feature extraction on the data collected by the perception layer. The features include basic parameters such as gas concentration, wind speed, and temperature.

[0056] The local decision node, equipped with an NVIDIA Jetson AGX Orin processor, runs an improved MADDPG algorithm to analyze preprocessed data and dynamically allocate tasks based on a collaborative reward function.

[0057] (4) The cloud platform includes a digital twin system, an emergency command center, and a visual interface.

[0058] The digital twin system utilizes a grey relational analysis module to assess environmental risk coefficients, performs dynamic path planning based on a 3D digital twin model of a real coal mine roadway, and constructs a 3D dynamic gas diffusion prediction model based on an LSTM network. The grey relational degree is calculated using an improved Dumbledore relational degree algorithm, with a resolution coefficient ρ of 0.3. The collaborative reward function expression is as follows:

[0059]

[0060] Where, r coop Let ΔC represent the collaborative reward function. gas The concentration change is represented by data acquired in real time by a laser methane sensor and denoised using a Kalman filter; d gas α represents the Euclidean distance between the robot and the gas source (calculated using the UWB positioning module); α represents the urgency of the associated task; and λ represents the control distance decay rate.

[0061] The gas diffusion prediction model expression is:

[0062]

[0063] Wherein, the diffusion coefficient D is taken as 0.1, the wind speed vector V is measured in real time by the wind speed sensor, and the source term R... source Calculated using the absolute gas emission formula, and the contribution of gas from adjacent layers under pressure relief is added.

[0064] The emergency command center integrates audible and visual alarms with power-off linkage functions.

[0065] The visual interface provides intuitive data display and decision support tools.

[0066] (5) Auxiliary system, including power supply and distribution system and positioning reference station; the power supply and distribution system adopts explosion-proof lithium battery + intrinsically safe power supply to power the electrical equipment on each floor in a dual circuit; the positioning reference station is used to provide centimeter-level RTK calibration service for distributed inspection robots.

[0067] Example 2:

[0068] This embodiment provides a collaborative decision-making method for coal mine gas inspection based on multi-agent reinforcement learning. (See also...) Figure 2 Specifically, it includes the following steps:

[0069] S1: Real-time data acquisition: The inspection robot collects data such as gas concentration, wind speed, and temperature in the tunnel in real time through sensors, and obtains its own location information through the UWB positioning module.

[0070] S2: Data Processing and Analysis: The edge computing layer uses the improved MADDPG algorithm for data processing and analysis, and implements dynamic task allocation based on the collaborative reward function.

[0071] The MADDPG algorithm employs a dual-objective optimization function to balance energy consumption and response time. A Critic network is used to evaluate the value of actions generated by the policy network. A distributed training mechanism supports parallel learning and training of multiple agents, improving the algorithm's convergence speed and performance.

[0072] The MADDPG algorithm uses a bi-objective optimization function:

[0073]

[0074] Among them, E i (t) represents the energy consumption of the i-th robot, and n represents the number of robots; The parameter represents the task response delay; m represents the number of time-stamped logs based on the robot's UWB positioning system, with a value of 20; W1 and W2 represent weighting coefficients, dynamically adjusted by a fuzzy PID controller, with initial values ​​set according to the mine's gas level (W1 initial value is 0.7, W2 is 0.3), and then adaptively updated based on real-time gas concentration. S3: Risk Assessment and Path Planning: Environmental risk coefficients are assessed using a grey relational analysis module, path planning is performed based on a digital twin model, and a three-dimensional dynamic gas diffusion prediction model is constructed based on an LSTM network.

[0075] The grey relational analysis module specifically calculates the grey relational degree between gas concentration, tunnel structure parameters, and environmental risk, providing a basis for risk assessment and decision-making. The grey relational degree is calculated using an improved Dumbledore relational degree algorithm, expressed as:

[0076]

[0077] Where ρ is the resolution coefficient, preset to 0.2 based on the roadway risk level, and dynamically optimized using the entropy-weighted TOPSIS algorithm; x0(k) is the reference sequence, x i (k) is a ratio sequence taken from the sensor's historical database (such as methane concentration, temperature, vibration). The data is standardized using z-score processing to eliminate the influence of dimensions.

[0078] The expression for the gas diffusion prediction model is:

[0079]

[0080] Where D is the diffusion coefficient, with a value of 0.15; V is the wind speed vector; R source The source term is calculated using the absolute gas emission formula, and the contribution of depressurized gas from adjacent layers is added. The concentration gradient term is solved by the gradient descent algorithm combined with finite element mesh discretization.

[0081] S4: Decision Execution and Feedback: Distribute decision results to inspection robots and emergency command centers for execution, and use Kalman filtering modules to correct robot pose and environmental parameters in real time, thereby improving data accuracy and system robustness.

[0082] The Kalman filter module includes: a state equation, a mathematical model describing the dynamic characteristics of the system; an observation equation, a mathematical model that links the observed data with the system state; and a filtering algorithm, which iteratively approximates the true state.

[0083] The state equation for the Kalman filter is:

[0084]

[0085] Where, x k U represents the state at the current time k. k F represents the system input vector. k B represents the state transition matrix. k F represents the control matrix. k and B k The dynamic equations were generated by inverting the three-dimensional twin model of the tunnel, and the model parameters were fitted by laser scanning and geological borehole data; Q k The process noise covariance is represented and calibrated based on historical gas emission data, where... V represents wind pressure. drift The wind speed drift is indicated by actual measurement using a mining ultrasonic velocimeter. The pressure change rate is taken from the barometric pressure sensor array.

[0086] Example 3:

[0087] This embodiment provides a collaborative decision-making system for coal mine gas inspection based on multi-agent reinforcement learning, with the following workflow:

[0088] 1) After the system starts, the distributed inspection robot cluster performs inspections according to the preset path, while the fixed monitoring node begins to collect environmental parameters.

[0089] 2) After receiving the data, the edge computing layer performs preliminary processing, using the Kalman filter algorithm to reduce noise and extract feature information;

[0090] 3) The local decision node calculates the current task allocation of each robot and optimizes the inspection path based on the collaborative reward function of the improved MADDPG algorithm;

[0091] 4) The digital twin system dynamically plans inspection routes based on the latest environmental risk assessment results and distributes tasks through the cloud platform;

[0092] 5) After receiving the task, the robot moves along the new path while monitoring the surrounding environmental parameters in real time;

[0093] 6) When gas levels exceed limits or environmental anomalies are detected, the system automatically triggers the emergency response mechanism, issues a warning, and automatically cuts off power;

[0094] 7) The system updates the digital twin model based on real-time data to ensure the timeliness and accuracy of the model.

[0095] Through the above-mentioned technical solution, this invention realizes the intelligent, automated and collaborative nature of coal mine gas inspection, significantly improving inspection efficiency and accuracy, reducing labor and maintenance costs, and enhancing the safety and reliability of the system.

[0096] Finally, it should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A collaborative decision-making system for coal mine gas inspection based on multi-agent reinforcement learning, characterized in that, The system adopts a three-tier architecture of "cloud-edge-device", which specifically includes a perception layer, a transmission layer, an edge computing layer, and a cloud platform; The perception layer includes a distributed inspection robot cluster and fixed monitoring nodes; the distributed inspection robot cluster is deployed in coal mine roadways to collect roadway structural parameters and methane concentration in the roadways; the fixed monitoring nodes are deployed at key locations in the roadways and include multi-parameter sensors for methane, wind speed and temperature. The edge computing layer includes data preprocessing units and local decision nodes; The data preprocessing unit is used to preprocess the data collected by the perception layer, specifically by performing Kalman filtering for noise reduction and feature extraction. The Kalman filtering noise reduction process uses the following Kalman filtering state equation: and in, Indicates the current time k state, Represents the system input vector. Represents the state transition matrix. Represents the control matrix. and The dynamic equations were generated by inverting the three-dimensional twin model of the tunnel, and the model parameters were fitted by laser scanning and geological borehole data. The process noise covariance is represented and calibrated based on historical gas emission data, where... Indicates wind pressure. Indicates the amount of wind speed drift; This represents the rate of change of pressure. The local decision node uses an improved MADDPG algorithm to analyze the preprocessed data and dynamically allocates tasks based on a collaborative reward function; where MADDPG represents the multi-agent deep deterministic policy gradient; the improved MADDPG algorithm employs a bi-objective optimization function, expressed as: in, Indicates the first i Energy consumption of a robot n Indicates the number of robots; This indicates a delay in task response. m This indicates the number of spatiotemporal stamp logs based on the robot's UWB positioning system; , The weighting coefficients are dynamically adjusted using a fuzzy PID controller. The initial value is set based on the mine's gas level and then adaptively updated according to real-time gas concentration. The cloud platform includes a digital twin system and an emergency command center. The digital twin system uses a grey relational analysis module to assess environmental risk coefficients, performs dynamic path planning based on a three-dimensional digital twin model of the roadway, and constructs a three-dimensional dynamic gas diffusion prediction model based on an LSTM network. The expression for the gas diffusion prediction model is: in, The diffusion coefficient is denoted as . This is the wind speed vector; For source terms; For concentration gradient terms; The emergency command center integrates audible and visual alarms and linked power-off functions. The transport layer is used to realize data transmission between the perception layer, the edge computing layer, and the cloud platform.

2. The coal mine gas inspection collaborative decision-making system according to claim 1, characterized in that, The inspection robot is equipped with a laser methane sensor, an infrared CO2 sensor, and a UWB positioning module.

3. The coal mine gas inspection collaborative decision-making system according to claim 1, characterized in that, The expression for the collaborative reward function is: in, This represents the collaborative reward function. Indicates the number of robots. Indicates the amount of concentration change; This represents the Euclidean distance between the robot and the gas source. Indicates the urgency of related tasks. This indicates the control distance attenuation rate.

4. The coal mine gas inspection collaborative decision-making system according to claim 1, characterized in that, The grey relational analysis module specifically calculates the grey relational degree between gas concentration, tunnel structure parameters, and environmental risk; the grey relational degree is calculated using an improved Dumbledore relational degree algorithm.

5. The coal mine gas inspection collaborative decision-making system according to claim 1, characterized in that, The system also includes a positioning reference station to provide centimeter-level RTK calibration services for distributed inspection robots.

6. The coal mine gas inspection collaborative decision-making system according to claim 1, characterized in that, The decision-making method of this system specifically includes the following steps: S1: Real-time data acquisition: Various types of inspection robots distributed in coal mine roadways collect real-time data on methane concentration and roadway structural parameters through sensors. S2: Data Processing and Analysis: The edge computing layer receives data collected by the inspection robot and fixed monitoring points. First, the collected data is preprocessed, and then the improved MADDPG algorithm is used for data processing and analysis to achieve dynamic task allocation. S3: Risk Assessment and Route Planning: The digital twin system on the cloud platform uses the grey relational analysis module to assess the environmental risk coefficient, performs dynamic route planning based on the three-dimensional digital twin model of the tunnel, and constructs a three-dimensional dynamic gas diffusion prediction model based on the LSTM network. S4: Decision Execution and Feedback: The digital twin system distributes the decision results to the inspection robot and the emergency command center for execution, and corrects the robot's pose and environmental parameters in real time through Kalman filtering.

Images