Mine high-temperature heat damage monitoring system based on multi-mode Internet of Things data

Abstract

The invention relates to the technical field of mine safety monitoring, and discloses a mine high-temperature heat damage monitoring system based on multi-mode Internet of Things data, and the system comprises a prediction and control integrated system architecture module which constructs a prediction and control integrated system architecture, and builds a three-layer PCIS architecture of a perception layer, a prediction layer and a control layer; the thermal field and flow field combined optimizer module is used for constructing a thermal field and flow field combined optimizer and realizing direct coupling of a prediction model and a control algorithm; the self-adaptive anti-interference control mechanism module is used for implementing a self-adaptive anti-interference control mechanism and realizing system fault tolerance through a fault mode library and dynamic reconstruction; the performance self-optimization module is used for implementing system performance self-optimization and continuously improving the system performance through incremental learning and knowledge migration; according to the method, through multi-modal data fusion and the space-time diagram convolutional network, the heat damage prediction precision is improved, the prediction time range is expanded, and more sufficient response time is provided for heat damage prevention and control.

Description

Technical Field

[0001] The present invention relates to the technical field of mine safety monitoring, and more specifically, to a mine high-temperature heat hazard monitoring system based on multimodal Internet of Things data. Background Art

[0002] With the continuous increase in mining depth and the increasing complexity of mine structures, high-temperature heat damage in mines has become an increasingly prominent problem, posing a serious threat to miner health and production safety. Therefore, mine heat damage monitoring and control technology has become a key research direction in the field of mine safety and production. Currently, mine heat damage monitoring systems generally adopt a "prediction-control" separated architecture. This involves collecting environmental parameters through a sensing layer, using a prediction system to analyze heat damage trends, and then using the prediction results to formulate ventilation and other control measures.

[0003] However, the existing technology still has many deficiencies in practical applications, which are mainly reflected in the following aspects: In the existing system, the information flow efficiency between the prediction system and the control system is low. Since the two belong to different functional modules, delays and losses are prone to occur during the information transmission process, resulting in control decisions lagging behind the development of heat damage and difficulty in responding to the rapid changes in the mine environment in a timely manner; there is a contradiction between prediction accuracy and control timeliness. High-precision heat damage prediction usually relies on complex models and a large amount of computing resources, resulting in a long prediction process, while the control system requires a rapid response to prevent and control heat damage risks. This contradiction makes it difficult for the system to take into account both prediction accuracy and real-time control; in the existing system, the prediction results are often unable to directly guide control decisions. The heat damage development trend output by the prediction system usually requires manual analysis and decision conversion to form a specific control strategy. This process is not only time-consuming but also requires manual analysis and decision conversion to form a specific control strategy. It is also easy to introduce subjective judgment errors, affecting the control effect; the existing technology lacks an adaptive coordination mechanism, the prediction system and the control system are optimized independently, lacking a unified optimization target and coordinated adjustment mechanism, and cannot be adaptively optimized according to the overall operation effect of the system, affecting the overall performance of the system; the traditional mine heat hazard monitoring system has a complex structure, usually including a five-layer architecture of "perception-analysis-prediction-decision-control". Too many layers lead to high system complexity and high maintenance costs, and interface compatibility problems between layers occur frequently, affecting the stability and scalability of the system; the existing system has poor anti-interference ability. When faced with emergencies such as tunnel collapse and ventilation equipment failure, the system lacks rapid response and automatic adjustment capabilities, which can easily lead to failure of heat hazard monitoring and pose a major safety hazard.

[0004] In summary, the existing mine heat hazard monitoring and control technology has problems that need to be improved in terms of information flow, prediction and control coordination, adaptive optimization, system structure simplification and anti-interference ability. There is a need for a new mine heat hazard monitoring system that can achieve deep coupling of heat hazard prediction and ventilation control and has efficient coordination and adaptive capabilities. Summary of the Invention

[0005] The present invention provides a mine high-temperature heat damage monitoring system based on multimodal Internet of Things data, which solves the technical problems of delayed heat damage response, complex system architecture and poor anti-interference ability caused by the separation of prediction and control in related technologies.

[0006] The present invention provides a mine high temperature heat damage monitoring system based on multimodal Internet of Things data, including: Predictive control integrated system architecture module, builds the predictive control integrated system architecture, and establishes a three-layer PCIS architecture of perception layer, prediction layer and control layer; The thermal field and flow field joint optimizer module builds a thermal field and flow field joint optimizer based on the multimodal data output by the perception layer, realizing the direct coupling of the prediction model and the control algorithm; The adaptive anti-disturbance control mechanism module implements the adaptive anti-disturbance control mechanism based on the thermal field prediction results generated by the prediction layer and the feedback information during the execution of the control layer, and realizes system fault tolerance through the fault mode library and dynamic reconstruction; The performance self-optimization module implements system performance self-optimization based on historical data and real-time performance feedback accumulated during system operation, and continuously improves system performance through incremental learning and knowledge transfer.

[0007] Furthermore, the construction of the predictive control integrated system architecture includes: Build a multimodal data perception layer, deploying temperature sensor arrays, airflow sensors, pressure difference sensors, humidity sensors, and gas composition sensors. Use edge computing units for data preprocessing and anomaly detection, and use convergent processing units to achieve spatiotemporal alignment and quality assessment of multimodal data. Construct a heat damage prediction layer and establish a mine heat damage prediction model based on a spatiotemporal graph convolutional network; Build a ventilation control layer to optimize ventilation parameters and implement control strategies based on prediction results.

[0008] Furthermore, the construction of the thermal field and flow field joint optimizer includes: Construct a thermal field and flow field mapping model, and establish a bidirectional mapping relationship between temperature distribution and airflow distribution based on the principles of computational fluid dynamics; Construct a response matrix to describe the influence of the change of control variables on the thermal field and flow field; Build a unified predictive control framework to directly map thermal field prediction results into ventilation control parameters; Realize multi-time scale control strategy generation, generate short-term, medium-term and long-term control strategies respectively.

[0009] Furthermore, the implementation of the adaptive anti-disturbance control mechanism includes: Build a fault mode library to store characteristic patterns of common fault types and predefined response strategies; Implement dynamic control reconstruction algorithm to quickly reconstruct the control strategy when system anomalies are detected; Build a self-calibration mechanism for the prediction model to automatically adjust model parameters by comparing the prediction results with the actual measurement data; Implement an incremental learning module to continuously learn from operational data and optimize system performance.

[0010] Furthermore, the implementation of system performance self-optimization includes: Construct a performance evaluation index system, including prediction accuracy index, control effect index and system stability index; Implement incremental learning and parameter optimization, regularly updating prediction models and control algorithms based on newly collected data; Build a knowledge accumulation and transfer learning mechanism to use historical experience to quickly adapt to new environments and new tasks.

[0011] Furthermore, the spatiotemporal graph convolutional network of the heat damage prediction layer includes: The spatial graph convolution module processes the spatial topological relationship of the mine and extracts spatial features; the temporal convolution module uses causal convolution to process time series data and maintain temporal relationships; the spatiotemporal attention mechanism focuses on important spatial areas and key time points respectively; and the multi-scale feature extraction module captures multi-time scale patterns through different convolution kernel sizes.

[0012] Furthermore, the thermal field and flow field mapping model is based on a simplified computational fluid dynamics model, which discretizes the mine space into a node network. The thermal field and flow field are respectively represented as node temperature values ​​and edge flow values, and mass conservation constraints, energy conservation constraints and physical boundary constraints are established.

[0013] Furthermore, the predictive control unified framework adopts a differential dynamic programming algorithm to construct a multi-objective optimization function, comprehensively considering the heat damage reduction effect, energy consumption and ventilation balance to generate an optimal ventilation control strategy.

[0014] Furthermore, the dynamic control reconstruction algorithm includes: Fault detection and diagnosis: real-time monitoring of system operating status and detection of potential faults; Control constraint reconfiguration, reconfiguring control constraints based on fault conditions; The control strategy is reconstructed in real time and re-optimized based on the updated system model and constraints.

[0015] The present invention provides a computer storage medium for executing the above-mentioned mine high-temperature heat damage monitoring system based on multimodal Internet of Things data, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned mine high-temperature heat damage monitoring system based on multimodal Internet of Things data.

[0016] The beneficial effects of the present invention are: through multimodal data fusion and spatiotemporal graph convolutional networks, the accuracy of heat damage prediction is improved, the prediction time range is expanded, and more sufficient response time is provided for heat damage prevention and control; The integrated predictive control architecture shortens the entire process from data acquisition to control execution, reducing system response time from minutes in traditional systems to seconds, enabling rapid response to heat damage. The system hierarchy has been simplified from the traditional five-layer "perception-analysis-prediction-decision-control" to the three-layer "perception-prediction-control". The computing process has been changed from serial to parallel, which has improved system operating efficiency and reduced maintenance costs. When faced with emergencies (such as tunnel collapse and ventilation equipment failure), the system can quickly and automatically reconfigure the control strategy to maintain most system stability, which is superior to the performance of traditional systems in fault conditions; This system can predict heat damage events in advance and take preventive measures, which improves the efficiency of heat damage reduction and reduces the incidence and impact of heat damage accidents; Through multi-objective optimization and predictive control, the system can more reasonably allocate ventilation resources and reduce ventilation system energy consumption while ensuring safety, thus achieving the dual goals of safety and energy saving. Through incremental learning and self-optimization mechanisms, the system continuously accumulates experience during operation, the prediction model and control strategy are continuously optimized, and the system performance improves with the extension of operation time. After three months, the overall performance of the system has improved compared to the initial state. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a mine high temperature heat damage monitoring system based on multimodal Internet of Things data in the present invention; Figure 2 It is a bar chart comparing the prediction accuracy and response time of the system of the present invention and the traditional system; Figure 3 It is a line graph showing the expansion of the prediction time range of the system of the present invention compared with the traditional system; Figure 4 It is a pie chart showing the distribution of optimization effects of the system of the present invention; Figure 5It is a radar chart comparing the performance of the system of the present invention and the traditional system in five key performance indicators. DETAILED DESCRIPTION

[0018] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a mine high temperature heat damage monitoring system based on multimodal Internet of Things data, such as Figure 1 As shown, including: Predictive control integrated system architecture module, builds the predictive control integrated system architecture, and establishes a three-layer PCIS architecture of perception layer, prediction layer and control layer; According to the embodiments of this application, a three-layer predictive control integrated architecture is constructed, including a perception layer, a prediction layer, and a control layer, to achieve deep coupling between heat damage prediction and ventilation control. Specifically, it includes: Step 1.1, build a multimodal data perception layer; A multimodal sensor network is deployed in the mine environment to collect environmental parameter data such as temperature, humidity, air velocity, air pressure, and gas composition. The sensors include: Temperature sensor array: Temperature sensors are placed at key nodes of the mine working face, tunnels, and ventilation systems to collect spatial temperature distribution data; Airflow sensors: Install airflow speed and direction sensors at key locations along the ventilation path to monitor the operating status of the ventilation system; Air pressure differential sensor: Install air pressure differential sensors between ventilation network nodes to monitor changes in ventilation system resistance; Humidity sensor: monitors the air humidity in mines, which is an important parameter for heat damage assessment; Gas composition sensor: monitors the concentration of gases such as CO, CO2, and CH4 for heat damage type identification and safety monitoring.

[0020] Alternatively, in some embodiments, the temperature sensor can utilize a distributed fiber optic temperature sensing system to enable continuous temperature monitoring of long-distance tunnels. Furthermore, infrared thermal imaging cameras can be added to high-risk areas to provide more intuitive visualization of heat distribution.

[0021] The perception layer adopts a layered data processing architecture, including edge computing units and convergence processing units: Edge computing unit: Located at the sensor node or sensor gateway, responsible for data preprocessing, anomaly detection and preliminary feature extraction, reducing the amount of data transmission; Convergence processing unit: Integrates multimodal data, performs spatiotemporal alignment, data completion, and quality assessment to provide high-quality input data for the prediction layer.

[0022] It should be understood that in some embodiments, the edge computing unit can be implemented using a low-power ARM architecture processor or a dedicated ASIC chip to meet the low power consumption and explosion-proof requirements of mines.

[0023] Step 1.2, construct the heat damage prediction layer; According to the embodiments of the present application, a mine heat damage prediction model is constructed based on a spatiotemporal graph convolutional network to achieve accurate prediction of spatiotemporal dynamic changes in the thermal field.

[0024] like Figure 2 As shown, the system's prediction accuracy is 40% higher than that of traditional systems, and its response time is reduced from minutes (approximately 180 seconds) to seconds (approximately 5 seconds), achieving rapid response throughout the entire process from data acquisition to control execution. This performance improvement is primarily attributed to the innovative design of the integrated predictive control architecture, which eliminates the information transmission delay between the prediction and control modules in traditional systems. The comparative data in the figure shows that by adopting a spatiotemporal graph convolutional network and a deeply coupled predictive control mechanism, the system achieves a 36-fold increase in response speed while maintaining high prediction accuracy, providing strong technical support for the timely prevention and control of mine heat hazards.

[0025] like Figure 3 As shown, the prediction timeframe has been extended from 10 to 15 minutes (average 12.5 minutes) for conventional systems to 30 to 45 minutes (average 37.5 minutes) for this system, providing more time for response to heat damage. This threefold increase in prediction timeframe is due to the deep learning capabilities of the spatiotemporal graph convolutional network for complex spatiotemporal patterns, as well as the information gain from multimodal data fusion. This longer prediction timeframe enables the system to implement true preventive control rather than reactive post-event response, fundamentally enhancing the proactive protection capabilities of mine heat damage monitoring.

[0026] Specifically include: Mine space modeling: Based on the mine's three-dimensional terrain data, build a mine space topology map: ; in Represents the spatial topology of the mine; represents a node set, representing spatial sampling points; Represents an edge set, which represents the physical connection relationship between nodes; Spatiotemporal graph convolutional network structure: Construct a deep neural network consisting of temporal convolution modules and spatial convolution modules to capture the spatiotemporal patterns of thermal field evolution. The network structure includes: Spatial graph convolution layer: processes spatial topological relationships and extracts spatial features; Temporal convolution layer: processes time series data and captures temporal evolution patterns; Attention mechanism: highlighting the impact of key areas and time points; Multi-scale feature fusion: Fusion of features at different spatiotemporal scales to improve prediction accuracy.

[0027] The spatiotemporal graph convolutional network in this application is a deep learning model designed specifically for mine heat damage prediction tasks. Its specific implementation is as follows: Input layer: receives multimodal sensor data in the form of a time series graph data structure: ; in Represents the input data tensor; represents the field of real numbers; is the number of sensor nodes; is the characteristic dimension (such as temperature, humidity, airflow, etc.); is the length of the historical time window; Spatial graph convolution module: Use the graph convolution algorithm to process the spatial topological relationship of the mine. The calculation formula is: The spatial graph convolution module in this application is used to process the information transmission in the spatial topological relationship of the mine. The processing process of this module is as follows: First, the system obtains the node feature matrix of the current layer, which contains the feature information of each spatial sampling point; then, the system constructs an enhanced adjacency matrix, which ensures that the node's own information will not be lost during the transmission process by adding self-connections (that is, the connection between the node and itself); then, the system calculates the normalized form of the adjacency matrix. The specific method is to use the inverse of the square root of the diagonal matrix and multiply it with the adjacency matrix. This step ensures that the information will not produce scale deviations due to the different number of node connections during the transmission process; then, the system multiplies the normalized adjacency matrix with the node feature matrix of the current layer to realize the transmission of features in the spatial topological structure; finally, the system multiplies the transferred feature matrix with the learnable weight matrix and processes it through the activation function to obtain the node feature representation of the next layer. This process realizes the effective transmission and fusion of spatial information in the mine topological structure.

[0028] Temporal convolution module: Use causal convolution to process time series data to avoid information leakage and maintain time series relationships. The calculation formula is: The temporal convolution module in this application uses the causal convolution method to process time series data, ensuring that the model does not use information from future time points when making predictions. The processing process of this module is as follows: For each time point in the input sequence , the system calculates the output value at that time point, specifically by taking the time point and before time points (where The input value (where is the convolution kernel size) is multiplied by the corresponding convolution filter parameters and then summed. This processing ensures that the output at each time point depends only on the current and past inputs, without using future information, thus maintaining the causal nature of time series prediction. By selecting convolution kernels of different sizes, the system can capture patterns at different time scales, enhancing the ability to extract time series features.

[0029] Spatiotemporal attention mechanism: A dual attention mechanism is introduced to focus on important spatial regions and key time points respectively: The spatiotemporal attention mechanism in this application consists of two parts: spatial attention and temporal attention: The spatial attention mechanism is used to evaluate the importance relationship between different spatial nodes. The specific implementation process is as follows: First, the system calculates the correlation score between each pair of nodes, which is achieved by inputting the node feature vector into the correlation evaluation function; then, the system applies the softmax function to all correlation scores for normalization, so that the sum of all attention weights is 1, and the node For Node This mechanism enables the system to automatically identify and focus on node pairs that have a large spatial impact on each other, improving the accuracy of heat damage propagation path prediction.

[0030] The temporal attention mechanism is used to identify key moments in time series. The specific implementation process is as follows: First, the system inputs the feature vector of each time point into a temporal importance evaluation function to obtain the importance score of that time point. Then, the system applies a softmax function to normalize the importance scores of all time points to obtain the attention weight for each time point. This mechanism enables the system to automatically identify and highlight key change points in time series, improving its ability to warn of sudden heat damage events.

[0031] By combining the attention mechanisms in both spatial and temporal dimensions, the system can accurately capture the spatiotemporal patterns of heat damage evolution in the complex environment of mines and achieve more accurate predictions.

[0032] Multi-scale feature extraction: Capture patterns at different time scales by using different convolution kernel sizes (e.g., 3×1, 5×1, 7×1) in parallel, and fuse features from different layers through skip connections to enhance the model's ability to perceive multi-scale spatiotemporal patterns. These different convolution kernel sizes correspond to different k_c values. Output layer: Generate the future The thermal field distribution prediction for each time step is as follows: ; in 、 、 Respectively represent the first time step, the second time step, and the The predicted value of the thermal field distribution of each sampling point in the mine space in time steps, is the number of future time steps to predict.

[0033] In the mine heat damage monitoring scenario, the specific application examples of this spatiotemporal graph convolutional network are as follows: Early identification of hotspots: The system analyzes historical data from the temperature sensor network to identify areas experiencing abnormally high temperatures. For example, in one application, the model successfully predicted that the temperature of a work surface would exceed a safety threshold in 35 minutes, triggering cooling measures in advance. Prediction of heat damage transmission paths: Based on spatial topology and airflow data, the system can predict the possible transmission paths of heat damage. In one coal seam spontaneous combustion incident, the system accurately predicted the direction of the heat flow, guiding timely adjustments to the ventilation system to prevent the heat flow from spreading to crowded areas. Dynamic Adaptation: As the mining face advances, the system automatically adjusts its prediction model based on topological changes. For example, when a new roadway is penetrated, the system learns the thermal flow distribution characteristics of the new ventilation network within a few hours, maintaining prediction accuracy.

[0034] Optionally, in some embodiments, the spatiotemporal graph convolutional network can adopt a hierarchical training strategy, first using global data for pre-training, and then using historical data of a specific mine for fine-tuning to improve the prediction accuracy of the model in a specific environment.

[0035] Step 1.3, construct the ventilation control layer; According to the embodiments of the present application, a ventilation optimization control system is constructed based on the thermal field prediction results to achieve precise control of ventilation parameters. Specifically, it includes: Ventilation network modeling: Establish a mine ventilation network model, including the characteristic parameters and constraints of components such as fans, dampers, and tunnels; Ventilation optimization objective function: Construct a multi-objective optimization function that comprehensively considers heat damage reduction effect, energy consumption, and ventilation balance: The ventilation optimization goal in this application is to optimize the ventilation system within the forecast time range (from the current moment To the future time, represents the number of future time steps predicted) minimizes the total cost function The total cost function consists of three parts: First, for each prediction moment (from 0 to ), calculate the heat damage state deviation cost, which is the weighted deviation between the heat damage state vector obtained by inputting the predicted temperature and control parameters into the deviation function and the ideal state; Secondly, the control cost is calculated, which is the weighted norm of the control vector (including parameters such as fan speed and damper opening), reflecting the energy consumption; Finally, the control change cost, which is the weighted norm of the control vector changes at adjacent moments, is calculated to reflect the control stability.

[0036] These three parts of the cost are multiplied by the corresponding weight matrix and then summed to obtain the total cost function. The system achieves optimal ventilation control by finding the control sequence that minimizes the total cost.

[0037] The predicted temperature is Always looking to the future Temperature forecast at time is the prediction time step index, the value range is arrive , the control vector includes parameters such as fan speed and damper opening, and the weight matrix 、 and They are used to balance the importance of the three objectives of heat damage reduction, energy consumption and control stability. This function represents the deviation between temperature and control parameters. Its implementation involves first calculating the difference between the predicted temperature and the target safe temperature, then evaluating the heat damage mitigation effect using the control parameters, and outputting a heat damage state vector. This function maps the temperature prediction results and control parameters into the heat damage state space, enabling the system to directly optimize heat damage mitigation.

[0038] Control strategy generation: Differential Dynamic Programming (DDP) is used to solve the above optimization problem and generate the ventilation control strategy: The control strategy in this application is generated using a differential dynamic programming algorithm, which obtains the control increment through iterative optimization calculation. Specifically, at each control moment, the system first obtains the deviation between the current state and the target state , and then multiply the deviation by the feedback gain matrix Obtain state-related control adjustments; at the same time, the system also calculates a state-independent feedforward control term , used to compensate for the known dynamic characteristics of the system. Finally, these two control variables are added together to obtain the final control increment, which is used to adjust ventilation control parameters such as fan speed and damper opening. The feedforward term is pre-calculated based on the system model and predicted trajectory, while the feedback gain matrix is ​​obtained by solving the local linearized model of the system. This enables real-time correction of state deviations, improving the system's robustness to disturbances.

[0039] Optionally, in some implementations, control strategy generation can utilize a Model Predictive Control (MPC) algorithm instead of differential dynamic programming. This algorithm offers improved real-time performance, particularly when deployed on edge devices with limited computational resources. Model Predictive Control (MPC) is based on the principle of receding horizon optimization. During each control cycle, the system uses the current state as the initial condition and solves for the optimal control sequence within a finite horizon, executing only the first control action and then repeating the process for the next control cycle. This approach offers low computational complexity and the ability to explicitly handle control constraints, making it suitable for resource-constrained environments.

[0040] Control execution unit: sends the optimized control parameters to fan inverters, automatic dampers and other execution equipment to achieve automatic control of the ventilation system.

[0041] The thermal field and flow field joint optimizer module builds a thermal field and flow field joint optimizer based on the multimodal data output by the perception layer, realizing the direct coupling of the prediction model and the control algorithm; According to the embodiments of the present application, a thermal field-flow field joint optimizer is constructed to directly map the thermal field prediction into a ventilation control strategy, thereby achieving deep coupling of prediction and control, which is the core innovative module of this system.

[0042] like Figure 4 As shown, the distribution of system optimization effects demonstrates the comprehensive advantages of the technical solution of this application: the heat damage reduction effect is improved by 40%, the ventilation efficiency is improved by 35%, and the energy consumption is reduced by 25%, which comprehensively improves the overall performance of the mine heat damage monitoring system. This multi-dimensional performance improvement fully reflects the core value of the thermal field-flow field joint optimizer. By unifying the physical field prediction and control strategy generation into the same optimization framework, a comprehensive leap in system performance is achieved. The data in the figure shows that the joint optimizer not only improves the heat damage control effect, but also realizes the energy-saving operation of the ventilation system, providing an economical and efficient technical solution for mine safety production.

[0043] Step 2.1, construct the thermal field and flow field mapping model; Based on the principles of computational fluid dynamics (CFD), a bidirectional mapping relationship model between thermal field and flow field is constructed to achieve unified representation of the two physical fields. Specifically, it includes: Heat transfer equation: describes the heat transfer law in the mine space, considering the three heat transfer modes of convection, conduction and radiation; Fluid dynamics equations: describe the flow of air in the mine space, taking into account viscosity, pressure gradients and boundary conditions; Thermal-fluid coupling model: establishes the interaction between the thermal field and the flow field, including the influence of temperature on fluid density, the influence of airflow on heat transfer, etc.

[0044] The thermal field-flow field joint optimizer in this application is the core innovation of the system, and its specific implementation is as follows: Simplified CFD model: To achieve real-time calculation, a simplified computational fluid dynamics model is used to simplify the complex partial differential equations into a node network model. For the heat transfer equation, the following is used: The heat transfer model in this application is based on the principle of conservation of energy and describes the process of heat propagation in the mine space. The model takes into account four key factors: first, the rate of change of temperature with time, which represents the change of temperature at a certain point in unit time; second, convective heat transfer, which describes the phenomenon of heat transfer with the movement of air flow, which is represented by the product of air density, specific heat capacity, air flow velocity vector and temperature gradient; third, heat conduction, which describes the phenomenon of heat diffusion from high temperature area to low temperature area, which is represented by the divergence of thermal conductivity coefficient and temperature gradient; finally, the heat source term, which represents the area in space that generates or absorbs heat, such as equipment heating, coal seam spontaneous combustion, etc. By comprehensively considering these four factors, the system can accurately simulate the law of heat propagation in the complex environment of mines and provide a physical basis for heat damage prediction.

[0045] Network representation: The mine space is discretized into a node network, each node represents an area in the space, and the connection between nodes represents the physical connectivity. The thermal field and flow field are represented by the node temperature value in the network representation. and the flow value on the edge .

[0046] Response matrix construction: Construct the control response matrix through offline simulation or online learning , describes the degree of influence of changes in control variables (such as fan speed, damper opening, etc.) on the thermal field and flow field, in the form of: ; in, Represents the temperature change vector of each node, Represents the change vector of the flow rate at each edge, Represents the control response matrix, which describes the influence coefficient of the control variable change on the system state. A vector representing the adjustment of control variables (such as fan speed and damper opening). This formula establishes a linear mapping between control adjustments and system response and is the mathematical foundation for joint thermal-flow field optimization.

[0047] The response matrix construction in this application aims to establish a mapping relationship between the adjustment of control variables and the change of system state. The system first defines a matrix, the rows of which represent the system state variables (including the temperature changes of each node and the flow changes of each edge), and the columns represent the control variables (such as fan speed, damper opening, etc.). Each element in the matrix represents the impact of a unit change in the corresponding control variable on the corresponding state variable. Through this matrix representation, the system can predict the comprehensive impact of any control adjustment combination on the entire thermal field and flow field, thereby supporting the optimization decision of the control strategy. The response matrix can be obtained through offline numerical simulation, or it can be continuously updated through online learning during system operation to adapt to the dynamic changes in the mine environment.

[0048] Thermal-fluid coupling constraints: Establish coupling constraints between the thermal field and the flow field to ensure the consistency of the two physical fields. Mainly includes: Mass conservation constraint: For each non-source / sink node in a mine ventilation network (i.e., a node that is not an air inlet or return vent), the algebraic sum of the flow rates on all adjacent connections to that node must equal zero. This constraint ensures that no gas can be created or destroyed in the system, reflecting the fundamental physical law of conservation of mass. In practice, this constraint ensures proper airflow distribution within the ventilation network and is a fundamental condition for optimizing ventilation control.

[0049] Energy conservation constraint: For each node in a mine ventilation network, the total energy entering the node (including heat flowing into it from neighboring nodes and the node's own heat source) must equal the total energy flowing out of the node. Specifically, the system calculates the sum of the heat flowing into each neighboring node (flow rate multiplied by specific heat capacity, multiplied by the neighboring node temperature) and the node's own heat source. This value must equal the sum of the heat flowing out (flow rate multiplied by specific heat capacity, multiplied by the node's temperature). This constraint ensures energy conservation in the system and is the physical basis for predicting thermal field evolution.

[0050] Physical boundary constraints: Physical quantities such as temperature, flow rate, and pressure in the system must meet certain boundary conditions. For example, temperature cannot fall below the ambient minimum, flow rate cannot exceed the maximum allowed by the device, and pressure must be within the device's tolerance range. These constraints reflect the practical limitations of the physical system and ensure the feasibility of the control strategy.

[0051] In the mine heat hazard monitoring scenario, the specific application examples of this thermal field-flow field joint optimizer are as follows: Local heat damage reduction: When an abnormal temperature rise is detected on a working face, the system can quickly calculate the optimal local ventilation adjustment plan. For example, in a coal mining machine overheating incident, the system automatically increased the air volume in the area and adjusted the damper openings in the upper and lower air lanes, reducing the working face temperature by 3.5°C within 8 minutes and avoiding production suspension. Global ventilation optimization: Based on predicted heat damage trends, the system optimizes the global ventilation network. For example, during the summer heatwave, the system automatically creates a "differentiated day and night ventilation plan," increasing airflow to key areas during daytime heat waves and reducing it to non-operating areas at night, ensuring safety while reducing energy consumption by 25%. Emergency Ventilation Plan Generation: When an abnormal heat source (such as a spontaneous combustion point in a coal seam) is detected, the system can generate an emergency ventilation plan in real time. In a simulation test, the system calculated the optimal damper control plan to isolate the heat source and protect the evacuation route in just 2.3 seconds, over 80% faster than manual decision-making.

[0052] Optionally, in some embodiments, the thermal field-flow field joint optimizer can integrate an expert knowledge base, combine data-driven methods with domain expert experience, and improve the decision-making reliability of the system in special circumstances.

[0053] Step 2.2, build a unified prediction-control framework; According to the embodiments of this application, a framework is constructed to unify thermal field prediction and ventilation control into the same optimization goal, achieving a seamless connection between prediction results and control decisions. Specifically, it includes: State space representation: thermal field prediction results and ventilation control parameters Uniformly represented as variables in the state space; Prediction-control joint optimization: Based on the predicted future thermal field state, ventilation control parameters are directly optimized, skipping the traditional decision-making process; Feedback correction mechanism: Automatically adjust the prediction model and control strategy based on the deviation between the actual control effect and the expected effect.

[0054] Step 2.3, realize the generation of multi-time scale control strategy; According to the embodiments of the present application, a hierarchical control strategy is generated based on the prediction results of different time scales to achieve the unification of short-term precise control and long-term optimization planning. Specifically, it includes: Short-term control strategy (1-5 minutes): Based on recent thermal field predictions, optimize local ventilation parameters to achieve rapid response to heat damage; Medium-term control strategy (5-30 minutes): Based on the medium-term thermal field evolution trend, adjust the regional ventilation layout to achieve preventive control of heat damage; Long-term control strategy (more than 30 minutes): Based on the long-term development trend of heat damage, plan the overall ventilation resource allocation to maximize energy efficiency.

[0055] The adaptive anti-disturbance control mechanism module implements the adaptive anti-disturbance control mechanism based on the thermal field prediction results generated by the prediction layer and the feedback information during the execution of the control layer, and realizes system fault tolerance through the fault mode library and dynamic reconstruction; According to the embodiments of the present application, an adaptive anti-disturbance control mechanism is constructed to improve the robustness and adaptability of the system when facing emergencies.

[0056] Step 3.1, build a failure mode library and response strategy; Based on historical data and expert knowledge, a library of common failure modes and corresponding response strategies is constructed to provide a knowledge base for responding to emergencies. Specifically, it includes: Fault pattern recognition: including characteristic patterns of common fault types such as sensor failure, fan failure, damper failure, and tunnel collapse; Predefined response strategies: Basic response strategies for different types of faults, such as sensor backup switching and ventilation path reconstruction.

[0057] Optionally, in some implementations, the fault mode library can be constructed in the form of a knowledge graph, containing multi-dimensional information such as fault type, characteristic manifestations, severity, typical scenarios, and countermeasures, to improve the accuracy of fault diagnosis and the pertinence of countermeasures.

[0058] Step 3.2, implement the dynamic control reconstruction algorithm; According to the embodiments of the present application, a dynamic control reconstruction algorithm is used to quickly reconstruct the control strategy when a system anomaly or failure is detected to ensure continuous operation of the system. Specifically, it includes: Fault detection and diagnosis: Real-time monitoring of system operating status, detection of potential faults and location diagnosis; Control constraint reconfiguration: Reconfigure control constraints based on fault conditions to avoid faulty components; Real-time reconstruction of control strategy: re-optimize the control strategy based on the updated system model and constraints.

[0059] Step 3.3, construct a self-calibration mechanism for the prediction model; According to the embodiments of the present application, a self-calibration mechanism for the prediction model is used to continuously optimize the accuracy of the prediction model by comparing the prediction results with the actual measurement data. Specifically, the following steps are involved: Forecast error analysis: calculate the deviation between the forecast result and the actual measurement value, and analyze the error distribution characteristics; Adaptive adjustment of model parameters: Automatically adjust the prediction model parameters based on error analysis results; Dynamic optimization of model structure: adjust the model structure when necessary to adapt to environmental changes.

[0060] Optionally, in some embodiments, the prediction model self-calibration can adopt a Bayesian learning framework to update not only the point estimates of the model parameters, but also the probability distribution of the parameters, thereby improving the robustness of the model in an uncertain environment.

[0061] The performance self-optimization module implements system performance self-optimization based on historical data and real-time performance feedback accumulated during system operation, and continuously improves system performance through incremental learning and knowledge transfer; According to the embodiments of the present application, a system performance self-optimization mechanism is constructed to continuously improve the overall system performance through continuous learning and feedback adjustment.

[0062] Step 4.1, build a performance evaluation index system; Establish a comprehensive performance evaluation index system to quantify the system's performance in heat damage prediction and control. Specifically include: Prediction accuracy indicators: such as mean absolute error (MAE, MeanAbsoluteError), root mean square error (RMSE, RootMeanSquareError), etc. Control effect indicators: such as temperature control accuracy, response time, energy efficiency, etc.; System stability indicators: such as anti-interference ability, fault recovery time, etc.

[0063] Step 4.2: Implement incremental learning and parameter optimization; According to the embodiments of the present application, an incremental learning mechanism is adopted to enable the system to continuously learn from operating data and continuously optimize model parameters and control strategies. Specifically, it includes: Online data collection: Continuously collect system operation data, including sensor data, control parameters, performance indicators, etc. Incremental model updates: Regularly update prediction models and control algorithms based on newly collected data; Automatic parameter tuning: Use algorithms such as Bayesian optimization to automatically tune key system parameters.

[0064] Optionally, in some embodiments, the incremental learning process can adopt a federated learning framework, with edge computing nodes in various areas of the mine as learning clients and the central server as a model aggregation node, to achieve mine-wide model optimization while protecting data privacy.

[0065] Step 4.3: Build a knowledge accumulation and transfer learning mechanism; According to the embodiments of this application, a knowledge accumulation and transfer learning mechanism is adopted to enable the system to quickly adapt to new environments or new tasks by leveraging historical experience. Specifically, it includes: Knowledge base construction: storing the experience and patterns acquired during system operation into the knowledge base; Transfer learning algorithms: Leverage knowledge accumulated in similar environments to accelerate model training and optimization in new environments; Model generalization enhancement: Improve the adaptability and generalization ability of the model in different mining environments.

[0066] like Figure 5 As shown, the system outperforms traditional systems in key performance indicators such as prediction accuracy, response time, energy efficiency, prediction time range, and system stability, demonstrating the technical advantages of the integrated predictive control architecture. The radar chart clearly demonstrates the comprehensive superiority of the system in five core dimensions, with the most significant improvements in prediction accuracy and response time, reaching 40% and 36 times, respectively. This all-encompassing performance improvement is not the result of simple parameter optimization, but rather stems from fundamental innovation at the system architecture level. By deeply integrating prediction and control functions, a synergistic effect of 1+1>2 is achieved, setting a new benchmark for the development of mine heat hazard monitoring technology.

[0067] A computer storage medium for executing the above-mentioned mine high-temperature heat damage monitoring system based on multimodal Internet of Things data, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned mine high-temperature heat damage monitoring system based on multimodal Internet of Things data.

[0068] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A mine high temperature heat damage monitoring system based on multimodal Internet of Things data, characterized by: include: Predictive control integrated system architecture module, builds the predictive control integrated system architecture, and establishes a three-layer PCIS architecture of perception layer, prediction layer and control layer; The thermal field and flow field joint optimizer module builds a thermal field and flow field joint optimizer based on the multimodal data output by the perception layer, realizing the direct coupling of the prediction model and the control algorithm; The adaptive anti-disturbance control mechanism module implements the adaptive anti-disturbance control mechanism based on the thermal field prediction results generated by the prediction layer and the feedback information during the execution of the control layer, and realizes system fault tolerance through the fault mode library and dynamic reconstruction; The performance self-optimization module implements system performance self-optimization based on historical data and real-time performance feedback accumulated during system operation, and continuously improves system performance through incremental learning and knowledge transfer.

2. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 1 is characterized in that: The construction of the predictive control integrated system architecture includes: Build a multimodal data perception layer, deploying temperature sensor arrays, airflow sensors, pressure difference sensors, humidity sensors, and gas composition sensors. Use edge computing units for data preprocessing and anomaly detection, and use convergent processing units to achieve spatiotemporal alignment and quality assessment of multimodal data. Construct a heat damage prediction layer and establish a mine heat damage prediction model based on a spatiotemporal graph convolutional network; Build a ventilation control layer to optimize ventilation parameters and implement control strategies based on prediction results.

3. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 1 is characterized in that: The construction of the thermal field and flow field joint optimizer includes: Construct a thermal field and flow field mapping model, and establish a bidirectional mapping relationship between temperature distribution and airflow distribution based on the principles of computational fluid dynamics; Construct a response matrix to describe the influence of the change of control variables on the thermal field and flow field; Build a unified predictive control framework to directly map thermal field prediction results into ventilation control parameters; Realize multi-time scale control strategy generation, generate short-term, medium-term and long-term control strategies respectively.

4. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 1 is characterized in that: The implementation of the adaptive disturbance rejection control mechanism includes: Build a fault mode library to store characteristic patterns of common fault types and predefined response strategies; Implement dynamic control reconstruction algorithm to quickly reconstruct the control strategy when system anomalies are detected; Build a self-calibration mechanism for the prediction model to automatically adjust model parameters by comparing the prediction results with the actual measurement data; Implement an incremental learning module to continuously learn from operational data and optimize system performance.

5. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 1 is characterized in that: The implementation of system performance self-optimization includes: Construct a performance evaluation index system, including prediction accuracy index, control effect index and system stability index; Implement incremental learning and parameter optimization, regularly updating prediction models and control algorithms based on newly collected data; Build a knowledge accumulation and transfer learning mechanism to use historical experience to quickly adapt to new environments and new tasks.

6. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 2 is characterized in that: The spatiotemporal graph convolutional network of the heat damage prediction layer includes: The spatial graph convolution module processes the spatial topological relationship of the mine and extracts spatial features; the temporal convolution module uses causal convolution to process time series data and maintain temporal relationships; the spatiotemporal attention mechanism focuses on important spatial areas and key time points respectively; and the multi-scale feature extraction module captures multi-time scale patterns through different convolution kernel sizes.

7. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 3 is characterized in that: The thermal field and flow field mapping model is based on a simplified computational fluid dynamics model, which discretizes the mine space into a node network. The thermal field and flow field are represented as node temperature values ​​and edge flow values, respectively, and mass conservation constraints, energy conservation constraints and physical boundary constraints are established.

8. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 3 is characterized in that: The predictive control unified framework adopts a differential dynamic programming algorithm to construct a multi-objective optimization function, comprehensively considers the heat damage reduction effect, energy consumption and ventilation balance, and generates an optimal ventilation control strategy.

9. The mine high temperature heat damage monitoring system based on multimodal Internet of Things data according to claim 4 is characterized in that: The dynamic control reconstruction algorithm includes: Fault detection and diagnosis: real-time monitoring of system operating status and detection of potential faults; Control constraint reconfiguration, reconfiguring control constraints based on fault conditions; The control strategy is reconstructed in real time and re-optimized based on the updated system model and constraints.

10. A computer storage medium, characterized in that It includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement the mine high temperature heat hazard monitoring system based on multimodal Internet of Things data as described in any one of claims 1 to 9.

Images