Intelligent monitoring and early warning method and system for coal spontaneous combustion risk in coal mine goaf

Abstract

The invention provides a coal mine goaf coal spontaneous combustion risk intelligent monitoring and early warning method and system, and relates to the technical field of coal mine safety management. A dot-matrix wireless sensor network is deployed in a target area, temperature and gas concentration data are collected in real time, and data processing is carried out through edge calculation. A dynamic characteristic matrix is generated based on transfer learning and a space-time diagram convolutional network, a comprehensive risk index is calculated, four early warning levels are divided, efficient and accurate coal spontaneous combustion risk monitoring and early warning are achieved, and an innovative solution is provided for mine safety management.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine safety management, and in particular to a method and system for intelligent monitoring and early warning of coal spontaneous combustion risks in coal mine goaf areas. Background Art

[0002] Spontaneous combustion of coal in coal mine goafs is one of the main causes of mine fires. Existing monitoring technologies primarily rely on temperature sensors and gas analyzers, collecting environmental data through methods such as fiber optic temperature measurement and bundled tube monitoring, and combining them with threshold methods for early warning. In recent years, wireless sensor networks (WSNs) and deep learning technologies have been gradually applied to coal spontaneous combustion monitoring. For example, lattice sensor networks are used to construct temperature and gas seepage fields, and multi-source data fusion is combined to provide graded early warning.

[0003] With the development of technologies like edge computing and reinforcement learning, coal spontaneous combustion monitoring systems are evolving toward higher real-time performance, lower energy consumption, and stronger model generalization capabilities. Furthermore, multimodal data fusion and adaptive early warning mechanisms have become key research areas for improving monitoring accuracy.

[0004] Existing deep learning models have poor adaptability to dynamic environments and a high false alarm rate. In addition, the existing early warning mechanism is single and relies on static graded warnings that change in color and lack the ability to comprehensively analyze multi-dimensional parameters. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide an intelligent monitoring and early warning method and system for coal spontaneous combustion risk in coal mine goaf areas, which effectively improves the real-time, accuracy and comprehensive analysis capabilities of coal mine spontaneous combustion risk monitoring, and provides an innovative solution for mine safety management.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] An intelligent monitoring and early warning method for coal spontaneous combustion risk in coal mine goaf areas, comprising:

[0008] Deploy a dot matrix wireless sensor network in the target goaf; the dot matrix wireless sensor network includes a plurality of sensor nodes; the sensor nodes are integrated with temperature, CO, CH4, O2 and humidity detection modules;

[0009] The sensor nodes are used to collect ambient temperature concentration, gas concentration data and humidity data, and the edge computing nodes are used to perform real-time filtering and outlier removal to obtain pre-processed multimodal data;

[0010] Extracting features from the multimodal data based on a transfer learning model to generate a dynamic feature matrix;

[0011] The dynamic feature matrix is ​​modeled using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf and obtain prediction results;

[0012] Calculate a comprehensive risk index based on the prediction results; the comprehensive risk index incorporates temperature gradient, gas concentration change rate, and humidity correlation factors;

[0013] Four warning levels are divided according to the comprehensive risk index, and graded warnings are carried out according to the warning levels.

[0014] Preferably, it also includes:

[0015] Combined with preset historical disaster data, the parameters of the spatiotemporal graph convolutional network are dynamically updated through the Bayesian optimization algorithm;

[0016] The objective function of the Bayesian optimization algorithm is:

[0017]

[0018] Wherein, θ is the parameter to be optimized of the spatiotemporal graph convolutional network; the parameters to be optimized include: convolution kernel weight and graph attention coefficient; is a historical disaster dataset containing spatiotemporal data of time series; EI(θ) is the expected improvement function, Measure the parameter θ compared to the current optimal value The improvement potential of historical data is: β is the historical experience weight coefficient, which is used to control the contribution of historical consistency terms; λ is the time decay factor, which determines the exponential decay rate of the influence of historical data. is the parameter similarity function, Where σ is the Gaussian kernel scale parameter, is the optimal parameter obtained by Bayesian optimization at time τ.

[0019] Preferably, the warning levels include:

[0020] Green level: the comprehensive risk index is ≤0.3;

[0021] Yellow level: 0.3<the comprehensive risk index≤0.6;

[0022] Orange level: 0.6<the comprehensive risk index≤0.8;

[0023] Red level: the comprehensive risk index is greater than 0.8.

[0024] Preferably, it also includes:

[0025] When the sensor node fails, reconstructing the network topology of the sensor node based on the ant colony algorithm;

[0026] After the red level warning is triggered, the mine ventilation system and fire extinguishing equipment are automatically linked, and a disaster avoidance path is generated and pushed to the terminal.

[0027] Preferably, feature extraction is performed on the multimodal data based on a transfer learning model to generate a dynamic feature matrix, including:

[0028] Performing data preprocessing on the multimodal data; the data preprocessing includes standardizing the data collected by each sensor node to ensure comparability of different data sources;

[0029] Selecting a source domain and a target domain in a transfer learning model; the source domain includes a set of labeled training data; the target domain is real-time collected data for feature extraction;

[0030] Using convolutional neural networks as the infrastructure for transfer learning models, by fine-tuning model parameters, the features of the source domain are effectively transferred to the target domain to adapt to the distribution characteristics of real-time data.

[0031] Extracting feature vectors of each layer through forward propagation of the transfer learning model, and merging the feature vectors into an initial dynamic feature matrix; wherein the initial dynamic feature matrix represents the environmental state information at the current moment;

[0032] The extracted initial dynamic feature matrix is ​​subjected to dimensionality reduction processing to reduce computational complexity and obtain the final dynamic feature matrix.

[0033] Preferably, the dynamic feature matrix is ​​modeled using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf, and the prediction results are obtained, including:

[0034] Constructing a topological structure of a spatiotemporal graph convolutional network; the topological structure of the spatiotemporal graph convolutional network includes the definition of nodes and edges; wherein the nodes represent monitoring locations within the goaf; and the edges represent the spatial relationship between different monitoring locations;

[0035] Integrating the temporal features and spatial features in the dynamic feature matrix to form a spatiotemporal feature matrix to meet the input requirements of the spatiotemporal graph convolutional network;

[0036] Passing the spatiotemporal feature matrix to the graph convolution layer of the spatiotemporal graph convolution network at the input layer of the spatiotemporal graph convolution network to extract the spatial dependency relationship between nodes and aggregate neighborhood information through a graph convolution operation;

[0037] Passing the output of the convolutional layer of the spatiotemporal graph convolutional network to a time recursive layer, and modeling the time series data of the time recursive layer using a long short-term memory structure to capture the time dependence of gas diffusion trends and temperature changes;

[0038] The final output of the spatiotemporal graph convolutional network is mapped to the temperature field and gas concentration prediction value of the goaf through the fully connected layer of the spatiotemporal graph convolutional network to obtain the prediction result;

[0039] The error between the prediction result and the actual monitoring data is calculated using a loss function, and the parameters of the spatiotemporal graph convolutional network are updated and adjusted using a preset optimization algorithm.

[0040] Preferably, the comprehensive risk index is calculated based on the prediction results, including:

[0041] According to the formula Calculate the temperature gradient factor; where N is the number of valid sensor nodes; represents the temperature gradient of the i-th sensor node; T max and T min are the highest and lowest temperatures monitored in the current goaf respectively;

[0042] According to the formula Calculate the gas concentration change rate factor; where k represents the gas type index, k is 1 for CO, k is 2 for CH4, k is 3 for O2, and k is 4 for humidity; w k is the weight coefficient of the kth gas, satisfying ∑w k =1, represents the instantaneous rate of change of the kth gas concentration;

[0043] According to the formula Calculate the humidity correlation factor; where M is the number of humidity sensor nodes; H j is the relative humidity measurement value of the jth sensor node; H sat is the saturated humidity at the current temperature; I TH is the temperature-humidity coupling coefficient: Where T opt is the optimal temperature for coal spontaneous combustion, the preset value of which is 60℃, σ T is the standard deviation of temperature distribution

[0044] According to the formula R = γ1·sigmoid(α T )+γ2·tanh(α G )+γ3·α H Calculate the comprehensive risk index R; where γ1, γ2, and γ3 are the weight coefficients of each factor, γ1+γ2+γ3=1, is the normalization function, and tanh(·) is the hyperbolic tangent function.

[0045] An intelligent monitoring and early warning system for coal spontaneous combustion risk in coal mine goaf areas, comprising:

[0046] A network deployment unit is used to deploy a dot matrix wireless sensor network in the target goaf; the dot matrix wireless sensor network includes a plurality of sensor nodes; the sensor nodes are integrated with temperature, CO, CH4, O2 and humidity detection modules;

[0047] A data processing unit is used to collect ambient temperature concentration, gas concentration data and humidity data using the sensor nodes, and perform real-time filtering and outlier removal through edge computing nodes to obtain pre-processed multimodal data;

[0048] A feature extraction unit, configured to extract features from the multimodal data based on a transfer learning model to generate a dynamic feature matrix;

[0049] A result prediction unit is used to model the dynamic feature matrix using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf and obtain a prediction result;

[0050] An index calculation unit, configured to calculate a comprehensive risk index based on the prediction results; the comprehensive risk index incorporates temperature gradient, gas concentration change rate, and humidity correlation factors;

[0051] The hierarchical warning unit is used to divide the comprehensive risk index into four warning levels and to perform hierarchical warning according to the warning levels.

[0052] The present invention discloses the following technical effects:

[0053] The intelligent monitoring and early warning method for the risk of spontaneous combustion of coal in coal mine goafs proposed in this invention realizes the real-time and efficient collection and processing of environmental data by adopting a dot matrix wireless sensor network and edge computing technology, significantly improving the flexibility and accuracy of coal spontaneous combustion monitoring. Compared with the existing technology, this method uses transfer learning and spatiotemporal graph convolutional networks for dynamic feature extraction and trend prediction, which enhances the adaptability of the system in dealing with complex and dynamic environmental changes and reduces the false alarm rate. In addition, the calculation method of the comprehensive risk index takes into account the influencing factors of multiple dimensions and improves the accuracy and reliability of the early warning by dynamically adjusting the early warning mechanism, thereby providing more accurate protection for the safe production of mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 A flow chart of a method provided by an embodiment of the present invention;

[0056] Figure 2 A schematic diagram of the system structure provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] The purpose of the present invention is to provide a method and system for intelligent monitoring and early warning of coal spontaneous combustion risk in coal mine goaf areas, which effectively improves the real-time, accuracy and comprehensive analysis capabilities of coal mine spontaneous combustion risk monitoring and provides an innovative solution for mine safety management.

[0059] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0060] Figure 1 A flow chart of the method provided in the embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides an intelligent monitoring and early warning method for the risk of spontaneous combustion of coal in coal mine goaf, comprising:

[0061] Step 100: deploying a dot matrix wireless sensor network in the target goaf; the dot matrix wireless sensor network includes multiple sensor nodes; the sensor nodes are integrated with temperature, CO, CH4, O2 and humidity detection modules;

[0062] Step 200: Using sensor nodes to collect ambient temperature concentration, gas concentration data, and humidity data, and performing real-time filtering and outlier removal through edge computing nodes to obtain pre-processed multimodal data;

[0063] Step 300: Extract features from multimodal data based on the transfer learning model to generate a dynamic feature matrix;

[0064] Step 400: Modeling the dynamic feature matrix using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf and obtain prediction results;

[0065] Step 500: Calculate a comprehensive risk index based on the prediction results; the comprehensive risk index incorporates temperature gradient, gas concentration change rate, and humidity correlation factors;

[0066] Step 600: Divide the warning levels into four levels according to the comprehensive risk index, and perform graded warnings according to the warning levels. For example, the warning levels in this embodiment include:

[0067] Green level: the comprehensive risk index is ≤0.3;

[0068] Yellow level: 0.3<the comprehensive risk index≤0.6;

[0069] Orange level: 0.6<the comprehensive risk index≤0.8;

[0070] Red level: the comprehensive risk index is greater than 0.8.

[0071] Specifically, when deploying a dot matrix wireless sensor network in the target goaf, this embodiment first conducts on-site investigation and analysis to determine the optimal installation location for each sensor node. The selection of these locations will be comprehensively designed based on a variety of factors such as the spatial layout of the coal mine goaf, airflow path, geological structure, location of past spontaneous combustion accidents, and thermodynamic characteristics of the area. Through reasonable layout, it is ensured that the sensor nodes can cover the key monitoring points of the entire goaf area, including potential spontaneous combustion risk hotspots. Each sensor node integrates temperature, CO, CH4, O2 and humidity detection modules, and can collect and record relevant environmental data in real time. This sensor node is interconnected based on a low-power wide area network or other wireless communication protocols to form a reliable data transmission network to ensure that under the complex and changeable conditions of the goaf environment, the collected data can be accurately and quickly transmitted to the edge computing node for subsequent processing.

[0072] In the process of selecting and arranging sensor nodes, the first thing to do is to select products that are adaptable to the special environment of the mine. The sensor needs to have a wide temperature measurement range and good gas detection sensitivity to meet the monitoring needs under different working conditions in the goaf area. In addition, the humidity detection module of this embodiment has high measurement accuracy and can work stably in a high humidity environment. All sensor nodes adopt explosion-proof and moisture-proof design to ensure their stability and reliability in extreme mine environments. After completing the review and layout of the sensor nodes, they are installed in the preset position and the system is debugged to verify the effectiveness of wireless communication between sensors. After debugging is completed, ensure that each sensor node can accurately and quickly collect and transmit data. These data will provide a high-quality information basis for subsequent deep learning models and risk prediction systems, and enhance the overall effect of coal mine spontaneous combustion risk monitoring.

[0073] Optionally, in this embodiment, the sensor nodes collect environmental temperature, gas concentration, and humidity data through an integrated detection module. Each sensor node sends the real-time acquired data to the edge computing node via a wireless communication protocol. During the data collection process, the temperature sensor provides current ambient temperature information, the gas sensor (such as CO, CH4, O2 sensors) measures changes in gas concentration, and the humidity sensor records the relative humidity level. To ensure the accuracy and reliability of the data, the sensor nodes continuously collect data multiple times within a certain time interval. The data collected by each sensor node will be numbered and stored together with a timestamp to form a time series dataset.

[0074] After the edge computing node receives the data from each sensor node, this embodiment first performs real-time filtering on the data. The filtering algorithm uses technologies such as Kalman filtering or moving average method, the purpose of which is to eliminate noise interference in the data and improve the smoothness and accuracy of the data. After the filtering process is completed, the system will perform outlier detection on the acquired data. By setting a reasonable threshold range, the edge computing node can identify and eliminate data points that exceed the normal range to prevent abnormal data from affecting subsequent data analysis and risk prediction results. Finally, after filtering and outlier removal, the pre-processed multimodal data output by the edge computing node will be stored and used for subsequent feature extraction and risk assessment. This process ensures the accuracy and reliability of the data used, thereby laying a solid foundation for the realization of intelligent monitoring of coal spontaneous combustion risks.

[0075] Preferably, it also includes:

[0076] Combined with preset historical disaster data, the parameters of the spatiotemporal graph convolutional network are dynamically updated through the Bayesian optimization algorithm;

[0077] The objective function of the Bayesian optimization algorithm is:

[0078]

[0079] Wherein, θ is the parameter to be optimized of the spatiotemporal graph convolutional network; the parameters to be optimized include: convolution kernel weight and graph attention coefficient; is a historical disaster dataset containing spatiotemporal data of time series; EI(θ) is the expected improvement function, Measure the parameter θ compared to the current optimal value The improvement potential of historical data is: β is the historical experience weight coefficient, which is used to control the contribution of historical consistency terms; λ is the time decay factor, which determines the exponential decay rate of the influence of historical data. is the parameter similarity function, Where σ is the Gaussian kernel scale parameter, is the optimal parameter obtained by Bayesian optimization at time τ.

[0080] Specifically, in the present invention, the preset historical disaster data is combined with the application of the Bayesian optimization algorithm to dynamically update the parameters of the spatiotemporal graph convolutional network. First, the construction of the historical disaster data set is based on the records of previous coal spontaneous combustion events in goafs. These data include the spatiotemporal information of time series, which provides rich background information for the parameter optimization of the spatiotemporal graph convolutional network. Using these historical data, the system can accurately capture the relationship between environmental changes and disaster occurrences, and then perform targeted optimization of the model parameters. During the optimization process, the Bayesian optimization algorithm takes historical disaster data as input, constructs an objective function, and evaluates the improvement potential of the parameters to be optimized, so as to find the best parameter combination that can improve the prediction performance of the model.

[0081] During the Bayesian optimization process, the algorithm dynamically updates the historical experience weight coefficients based on the results obtained at each iteration to control the contribution of different historical data to the model optimization process. The time decay factor is used to ensure that event data from recent days or weeks has a more significant impact on model optimization, thereby adapting to changes in the environment and conditions and further improving the accuracy of the model. In addition, the parameter similarity function uses the Gaussian kernel method to measure the similarity between parameters. By maximizing the expected improvement function, the convolution kernel weights and graph attention coefficients are effectively adjusted in each iteration to achieve the optimization of model parameters. This process provides strong support for the adaptive adjustment of the spatiotemporal graph convolutional network, enabling it to maintain a high degree of prediction accuracy and response flexibility in dynamic environments, thereby improving the intelligent monitoring and early warning capabilities of coal spontaneous combustion risks.

[0082] For example, the parameters to be optimized, such as the convolution kernel weight and the graph attention coefficient, are obtained through preliminary model training, and cross-validation technology is used to evaluate their performance on different data sets, thereby achieving preliminary parameter assignment. The construction of the historical disaster data set relies on the comprehensive collection and organization of past coal spontaneous combustion events to ensure that the time series data it covers is representative and diverse. In order to design the expected improvement function, this embodiment takes into account the potential contribution of parameter changes to the improvement of model performance, and uses historical experience weight coefficients to balance the impact of historical data on the current model optimization. At the same time, the time decay factor is set according to the timeliness of the event to ensure that recent data plays a dominant role in the optimization process, and the Gaussian kernel scale parameter in the parameter similarity function is selected through multiple experiments to eliminate redundant effects while retaining effective information, thereby improving optimization efficiency. Finally, the comprehensive setting of these parameters is systematically debugged and verified to ensure that it adapts to the needs of dynamic environments and spatiotemporal changes, providing a stable and accurate parameter basis for the model.

[0083] Preferably, it also includes:

[0084] When the sensor node fails, reconstructing the network topology of the sensor node based on the ant colony algorithm;

[0085] After the red level warning is triggered, the mine ventilation system and fire extinguishing equipment are automatically linked, and a disaster avoidance path is generated and pushed to the terminal.

[0086] Specifically, in the present invention, when a sensor node fails, in order to ensure the effectiveness and stability of the network, this embodiment adopts a network topology reconstruction method based on the ant colony algorithm to handle it. First, this embodiment will monitor the failed sensor node and automatically locate it in the network. The ant colony algorithm simulates the foraging behavior of ants and widely searches for feasible alternative paths and new nodes to rebuild the network topology. In this process, the system will take into account factors such as the connection strength between nodes, data transmission delay, and redundancy, so as to ensure that the newly constructed topology structure can maintain the efficiency and robustness of the network. Through iterative optimization, the system will continuously adjust the connection relationship between nodes until the optimal network structure is found, so that data can flow efficiently in the reconstructed network, enhancing the stability of the entire monitoring system.

[0087] When the monitoring system triggers a red-level warning, this embodiment will immediately link the mine's ventilation system and fire extinguishing equipment to reduce potential fire risks through effective control strategies. Specifically, this embodiment automatically determines the current air circulation conditions and adjusts the operating status of the ventilation system according to the warning level to improve air circulation and reduce the concentration of combustible gases. At the same time, this embodiment will activate the fire extinguishing device to ensure that it can respond quickly and suppress the occurrence of fire when the risk of spontaneous combustion is the highest. In addition, this embodiment will automatically generate an evacuation path and push it to the operator terminal to guide the on-site personnel to evacuate safely. Through these measures, rapid response and handling of sudden fire incidents can be achieved to ensure the safety of coal mine workers and the stable operation of coal mines.

[0088] Preferably, feature extraction is performed on the multimodal data based on a transfer learning model to generate a dynamic feature matrix, including:

[0089] Performing data preprocessing on the multimodal data; the data preprocessing includes standardizing the data collected by each sensor node to ensure comparability of different data sources;

[0090] Selecting a source domain and a target domain in a transfer learning model; the source domain includes a set of labeled training data; the target domain is real-time collected data for feature extraction;

[0091] Using convolutional neural networks as the infrastructure for transfer learning models, by fine-tuning model parameters, the features of the source domain are effectively transferred to the target domain to adapt to the distribution characteristics of real-time data.

[0092] Extracting feature vectors of each layer through forward propagation of the transfer learning model, and merging the feature vectors into an initial dynamic feature matrix; wherein the initial dynamic feature matrix represents the environmental state information at the current moment;

[0093] The extracted initial dynamic feature matrix is ​​subjected to dimensionality reduction processing to reduce computational complexity and obtain the final dynamic feature matrix.

[0094] Optionally, in the present invention, the process of extracting features from multimodal data using a transfer learning model first includes data preprocessing. Specifically, the primary step in the preprocessing phase is to standardize the data collected by each sensor node. This operation primarily eliminates numerical discrepancies caused by differences in measurement range, units, and sensitivity between sensors, ensuring comparability across different data sources. After standardization, the data is uniformly scaled, making subsequent analysis and feature extraction more efficient and accurate.

[0095] When building a transfer learning model, the first step is to select a source domain and a target domain. The source domain introduces a set of labeled training data that contains rich environmental characteristics and provides a solid foundation for feature learning. The target domain consists of data collected in real time. By effectively connecting the target domain with the source domain, the transfer learning model can quickly adapt to the new data distribution, and the features learned during training can be effectively transferred to real-time data. In this case, a convolutional neural network is used as the foundation of the transfer learning model. Leveraging its powerful feature extraction capabilities, fine-tuning the model parameters allows the features from the source domain to be effectively transferred to the target domain.

[0096] During feature extraction, this embodiment extracts feature vectors from each layer through forward propagation using a transfer learning model. These feature vectors describe the current environmental state, and the extracted feature vectors are then merged into an initial dynamic feature matrix. To further improve computational efficiency and reduce computational complexity, this embodiment performs dimensionality reduction on the extracted initial dynamic feature matrix to generate a final dynamic feature matrix. This dynamic feature matrix not only effectively summarizes changes in the current environment but also provides an important feature foundation for subsequent risk assessment and predictive analysis, thereby enhancing the intelligent monitoring capabilities for coal mine spontaneous combustion risks.

[0097] Preferably, the dynamic feature matrix is ​​modeled using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf, and the prediction results are obtained, including:

[0098] Constructing a topological structure of a spatiotemporal graph convolutional network; the topological structure of the spatiotemporal graph convolutional network includes the definition of nodes and edges; wherein the nodes represent monitoring locations within the goaf; and the edges represent the spatial relationship between different monitoring locations;

[0099] Integrating the temporal features and spatial features in the dynamic feature matrix to form a spatiotemporal feature matrix to meet the input requirements of the spatiotemporal graph convolutional network;

[0100] Passing the spatiotemporal feature matrix to the graph convolution layer of the spatiotemporal graph convolution network at the input layer of the spatiotemporal graph convolution network to extract the spatial dependency relationship between nodes and aggregate neighborhood information through a graph convolution operation;

[0101] Passing the output of the convolutional layer of the spatiotemporal graph convolutional network to a time recursive layer, and modeling the time series data of the time recursive layer using a long short-term memory structure to capture the time dependence of gas diffusion trends and temperature changes;

[0102] The final output of the spatiotemporal graph convolutional network is mapped to the temperature field and gas concentration prediction value of the goaf through the fully connected layer of the spatiotemporal graph convolutional network to obtain the prediction result;

[0103] The error between the prediction result and the actual monitoring data is calculated using a loss function, and the parameters of the spatiotemporal graph convolutional network are updated and adjusted using a preset optimization algorithm.

[0104] Specifically, in the present invention, a spatiotemporal graph convolutional network is used to model the dynamic feature matrix to predict the temperature field and gas diffusion trend of the goaf. First, when constructing the topological structure of the spatiotemporal graph convolutional network, this embodiment needs to clearly define the nodes and edges of the network. In this structure, the nodes represent the various monitoring locations within the goaf, and the edges represent the spatial relationship between different monitoring locations. Through such a topological design, this embodiment can effectively capture the correlation between the monitoring points, thereby providing a good foundation for subsequent modeling.

[0105] Next, the temporal and spatial features in the dynamic feature matrix are integrated to form a spatiotemporal feature matrix that meets the input requirements of the spatiotemporal graph convolutional network. This integration process ensures that the data modeling takes into account both the impact of temporal dynamics and the characteristics of spatial distribution. At the network's input layer, the spatiotemporal feature matrix is ​​passed to the graph convolution layer, which uses graph convolution operations to extract spatial dependencies between nodes and aggregate neighborhood information, thereby enhancing the model's sensitivity to local features.

[0106] Subsequently, this embodiment passes the convolutional layer output of the spatiotemporal graph convolutional network to the time recursive layer, and uses the long short-term memory structure to model the time series data to capture the time dependence of gas diffusion trends and temperature changes. Finally, the final output of the spatiotemporal graph convolutional network is mapped to the temperature field and gas concentration prediction value of the goaf through the fully connected layer. The prediction result can not only provide a decision-making basis for real-time monitoring, but also provide effective support for subsequent risk assessment. During the training process, this embodiment uses a loss function to calculate the error between the prediction result and the actual monitoring data, and updates and adjusts the parameters of the spatiotemporal graph convolutional network according to a preset optimization algorithm to ensure continuous optimization of the model and improve prediction accuracy.

[0107] Exemplarily, the optimization algorithms preset in the present invention include stochastic gradient descent, Adam optimization algorithm and RMSProp algorithm. The main steps in the implementation process of these optimization algorithms are as follows: First, stochastic gradient descent updates the model parameters to minimize the error by calculating the gradient of the loss function. Specifically, a small batch of data is randomly selected in each iteration to update the parameters, thereby improving the training speed. The Adam optimization algorithm combines the advantages of momentum and adaptive learning rate. In each iteration, the learning rate of each parameter is dynamically adjusted according to the mean and variance of the historical gradient, which can effectively deal with the problems of gradient disappearance and explosion. The RMSProp algorithm avoids the problem of manual adjustment of the learning rate by maintaining the adaptive learning rate of each parameter, making the algorithm more stable when dealing with non-stationary targets. These optimization algorithms continuously update the model parameters in an iterative manner to ensure that the spatiotemporal graph convolutional network gradually converges to a better solution during each training process, thereby improving the overall prediction performance of the model.

[0108] Preferably, the comprehensive risk index is calculated based on the prediction results, including:

[0109] According to the formula Calculate the temperature gradient factor; where N is the number of valid sensor nodes; represents the temperature gradient of the i-th sensor node; T max and T min are the highest and lowest temperatures monitored in the current goaf respectively;

[0110] According to the formula Calculate the gas concentration change rate factor; where k represents the gas type index, k is 1 for CO, k is 2 for CH4, k is 3 for O2, and k is 4 for humidity; w k is the weight coefficient of the kth gas, satisfying ∑w k =1, represents the instantaneous rate of change of the kth gas concentration;

[0111] According to the formula Calculate the humidity correlation factor; where M is the number of humidity sensor nodes; H j is the relative humidity measurement value of the jth sensor node; H sat is the saturated humidity at the current temperature; I TH is the temperature-humidity coupling coefficient: Where T opt is the optimal temperature for coal spontaneous combustion, the preset value of which is 60℃, σ T is the standard deviation of temperature distribution

[0112] The comprehensive risk index R is calculated according to the formula R = γ1·sigmoid(αT)+γ2·tanh(αG)+γ3·αH; where γ1, γ2, and γ3 are the weight coefficients of each factor, γ1+γ2+γ3=1, is the normalization function, and tanh(·) is the hyperbolic tangent function.

[0113] Specifically, in the present invention, the process of calculating the comprehensive risk index based on the prediction results first includes the calculation of the temperature gradient factor. The calculation of this factor depends on the number of valid sensor nodes and the temperature data monitored by each node. By quantifying the temperature gradient of each sensor node and obtaining the current maximum and minimum temperature values ​​of the goaf, this embodiment can effectively assess the severity of temperature changes. The level of the temperature gradient factor directly reflects the temperature stability of the environment, providing an important information basis for subsequent risk assessment.

[0114] Secondly, the calculation of the gas concentration change rate factor involves real-time monitoring data for different types of gases. This embodiment sets a gas type index and performs a weighted calculation of the instantaneous rate of change of each gas concentration based on the weight coefficients of different gases (such as carbon monoxide, methane, oxygen, and humidity). These rates of change can reflect the fluctuations in gas concentration over a certain period of time, comprehensively assess the diffusion trend of the gas and its potential threat to environmental safety, and thus provide dynamic data support for the calculation of the comprehensive risk index.

[0115] Finally, the calculation of the humidity correlation factor is also indispensable. By evaluating the number of humidity sensor nodes and the relative humidity value of each node, combined with the saturated humidity information at the current temperature, the degree of influence of humidity on the risk is calculated. In addition, this embodiment also sets the temperature-humidity coupling coefficient to quantify the degree of correlation between humidity and the risk of coal spontaneous combustion. The comprehensive risk index is calculated based on the weighted combination of the above factors. In addition, this embodiment will use the normalization function and the hyperbolic tangent function to further process the results to ensure that the final generated comprehensive risk index can effectively reflect the real-time risk level of the coal mine operating environment. This series of calculations provides a scientific basis for the accurate assessment and early warning of coal spontaneous combustion risks through a systematic processing flow.

[0116] Furthermore, in the implementation of the present invention, step 600 involves classifying the warning levels according to the comprehensive risk index. First, a four-level warning standard is set, and the comprehensive risk index is divided into four levels: low risk, medium risk, high risk, and extremely high risk according to the numerical range. Specifically, this embodiment defines the risk index threshold corresponding to each level based on historical data and real-time monitoring results. These thresholds are system-debugged and verified to ensure that they can accurately reflect changes in the coal mine operating environment. Once the comprehensive risk index is calculated, this embodiment compares the index with the preset risk level threshold in real time to determine the current warning level.

[0117] After determining the warning level, the system will take corresponding graded warning measures according to different warning levels. For low-risk levels, the system can take routine monitoring measures to remind operators to remain vigilant; for medium-risk levels, this embodiment will recommend increasing the monitoring frequency and requiring additional inspections of relevant areas; high-risk levels will trigger real-time warnings, automatically starting the ventilation system and other safety protection measures; and extremely high-risk levels will require the immediate implementation of emergency plans to ensure the safe evacuation of on-site personnel and initiate comprehensive emergency response procedures. Through this graded warning mechanism, this embodiment can effectively respond to potential coal spontaneous combustion incidents caused by different risk levels and achieve scientific supervision and safety assurance of coal mine operations.

[0118] Corresponding to the above method, this embodiment also provides a coal mine goaf area coal spontaneous combustion risk intelligent monitoring and early warning system, such as Figure 2 Shown, including:

[0119] A network deployment unit is used to deploy a dot matrix wireless sensor network in the target goaf; the dot matrix wireless sensor network includes a plurality of sensor nodes; the sensor nodes are integrated with temperature, CO, CH4, O2 and humidity detection modules;

[0120] A data processing unit is used to collect ambient temperature concentration, gas concentration data and humidity data using the sensor nodes, and perform real-time filtering and outlier removal through edge computing nodes to obtain pre-processed multimodal data;

[0121] A feature extraction unit, configured to extract features from the multimodal data based on a transfer learning model to generate a dynamic feature matrix;

[0122] A result prediction unit is used to model the dynamic feature matrix using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf and obtain a prediction result;

[0123] An index calculation unit, configured to calculate a comprehensive risk index based on the prediction results; the comprehensive risk index incorporates temperature gradient, gas concentration change rate, and humidity correlation factors;

[0124] The hierarchical warning unit is used to divide the comprehensive risk index into four warning levels and to perform hierarchical warning according to the warning levels.

[0125] The beneficial effects of the present invention are as follows:

[0126] (1) The present invention significantly improves the accuracy and real-time performance of coal mine spontaneous combustion risk monitoring by introducing a multimodal data feature extraction method based on transfer learning. Compared with traditional methods, the present invention can fuse data from different sensors and utilize spatiotemporal graph convolutional network modeling to fully exploit the spatiotemporal features in environmental monitoring data, thereby more accurately predicting temperature fields and gas diffusion trends. This integrated approach enables the system to adapt to dynamically changing environments and enhances the intelligence level of the monitoring system.

[0127] (2) The present invention adopts an optimized comprehensive risk index calculation method that combines multiple factors such as temperature gradient, gas concentration change rate, and humidity correlation. By constructing a scientific evaluation model, the system can quantitatively analyze the safety risks of the coal mine operating environment and provide an effective basis for on-site decision-making. Traditional monitoring methods often find it difficult to fully consider the comprehensive impact of various factors, and the multi-dimensional risk assessment method of the present invention effectively fills this gap and improves the scientific nature of coal mine safety management.

[0128] (3) The present invention utilizes a hierarchical early warning system based on a four-level early warning mechanism, which enables the present invention to have a rapid response capability in practical applications. When the comprehensive risk index reaches a preset threshold, the system can automatically determine the risk level and take appropriate response measures, thereby effectively reducing the probability of accidents. This dynamic early warning mechanism greatly improves the response speed of coal mine safety production management and ensures the safety of workers.

[0129] (4) The implementation of this invention fully considers the particularity and complexity of coal mine operations. Through in-depth mining and analysis of spatiotemporal data, it achieves comprehensive monitoring and intelligent early warning of coal spontaneous combustion risks. This not only ensures safe mine operations, but also provides important technical references for the industry, promotes the development of coal mine safety monitoring technology, and has significant economic and social benefits.

[0130] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0131] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An intelligent monitoring and early warning method for the risk of spontaneous combustion of coal in coal mine goaf, characterized in that: include: Deploy a dot matrix wireless sensor network in the target goaf; The dot matrix wireless sensor network includes a plurality of sensor nodes; the sensor nodes are integrated with temperature, CO, CH4, O2 and humidity detection modules; The sensor nodes are used to collect ambient temperature concentration, gas concentration data and humidity data, and the edge computing nodes are used to perform real-time filtering and outlier removal to obtain pre-processed multimodal data; Extracting features from the multimodal data based on a transfer learning model to generate a dynamic feature matrix; The dynamic feature matrix is ​​modeled using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf and obtain prediction results; Calculate the comprehensive risk index based on the prediction results; The comprehensive risk index incorporates temperature gradient, gas concentration change rate and humidity correlation factors; Four warning levels are divided according to the comprehensive risk index, and graded warnings are carried out according to the warning levels.

2. The method for intelligent monitoring and early warning of coal spontaneous combustion risk in coal mine goaf according to claim 1 is characterized in that: Also includes: Combined with preset historical disaster data, the parameters of the spatiotemporal graph convolutional network are dynamically updated through the Bayesian optimization algorithm; The objective function of the Bayesian optimization algorithm is: Wherein, θ is the parameter to be optimized of the spatiotemporal graph convolutional network; the parameters to be optimized include: convolution kernel weight and graph attention coefficient; is a historical disaster dataset containing spatiotemporal data of time series; EI(θ) is the expected improvement function, Measure the parameter θ compared to the current optimal value The improvement potential; β is the historical experience weight coefficient, which is used to control the contribution of the historical consistency term; λ is the time decay factor, which determines the exponential decay rate of the influence of historical data. is the parameter similarity function, Where σ is the Gaussian kernel scale parameter, is the optimal parameter obtained by Bayesian optimization at time τ.

3. The method for intelligent monitoring and early warning of coal spontaneous combustion risk in coal mine goaf according to claim 1 is characterized in that: The warning levels include: Green level: the comprehensive risk index is ≤0.3; Yellow level: 0.3<the comprehensive risk index≤0.6; Orange level: 0.6<the comprehensive risk index≤0.8; Red level: the comprehensive risk index is greater than 0.

8.

4. The method for intelligent monitoring and early warning of coal spontaneous combustion risk in coal mine goaf according to claim 3 is characterized in that: Also includes: When the sensor node fails, reconstructing the network topology of the sensor node based on the ant colony algorithm; After the red level warning is triggered, the mine ventilation system and fire extinguishing equipment are automatically linked, and a disaster avoidance path is generated and pushed to the terminal.

5. The method for intelligent monitoring and early warning of coal spontaneous combustion risk in coal mine goaf according to claim 1 is characterized in that: Feature extraction is performed on the multimodal data based on the transfer learning model to generate a dynamic feature matrix, including: Performing data preprocessing on the multimodal data; the data preprocessing includes standardizing the data collected by each sensor node to ensure comparability of different data sources; Selecting a source domain and a target domain in a transfer learning model; the source domain includes a set of labeled training data; the target domain is real-time collected data for feature extraction; Using convolutional neural networks as the infrastructure for transfer learning models, by fine-tuning model parameters, the features of the source domain are effectively transferred to the target domain to adapt to the distribution characteristics of real-time data. Extracting feature vectors of each layer through forward propagation of the transfer learning model, and merging the feature vectors into an initial dynamic feature matrix; wherein the initial dynamic feature matrix represents the environmental state information at the current moment; The extracted initial dynamic feature matrix is ​​subjected to dimensionality reduction processing to reduce computational complexity and obtain the final dynamic feature matrix.

6. The method for intelligent monitoring and early warning of coal spontaneous combustion risk in coal mine goaf according to claim 1 is characterized in that: The dynamic feature matrix is ​​modeled using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend in the goaf, and the prediction results are obtained, including: Constructing a topological structure of a spatiotemporal graph convolutional network; the topological structure of the spatiotemporal graph convolutional network includes the definition of nodes and edges; wherein the nodes represent monitoring locations within the goaf; and the edges represent the spatial relationship between different monitoring locations; Integrating the temporal features and spatial features in the dynamic feature matrix to form a spatiotemporal feature matrix to meet the input requirements of the spatiotemporal graph convolutional network; Passing the spatiotemporal feature matrix to the graph convolution layer of the spatiotemporal graph convolution network at the input layer of the spatiotemporal graph convolution network to extract the spatial dependency relationship between nodes and aggregate neighborhood information through a graph convolution operation; Passing the output of the convolutional layer of the spatiotemporal graph convolutional network to a time recursive layer, and modeling the time series data of the time recursive layer using a long short-term memory structure to capture the time dependence of gas diffusion trends and temperature changes; The final output of the spatiotemporal graph convolutional network is mapped to the temperature field and gas concentration prediction value of the goaf through the fully connected layer of the spatiotemporal graph convolutional network to obtain the prediction result; The error between the prediction result and the actual monitoring data is calculated using a loss function, and the parameters of the spatiotemporal graph convolutional network are updated and adjusted using a preset optimization algorithm.

7. The method for intelligent monitoring and early warning of coal spontaneous combustion risk in coal mine goaf according to claim 1 is characterized in that: Calculate the comprehensive risk index based on the prediction results, including: According to the formula Calculate the temperature gradient factor; where N is the number of valid sensor nodes; represents the temperature gradient of the i-th sensor node; T max and T min are the highest and lowest temperatures monitored in the current goaf respectively; According to the formula Calculate the gas concentration change rate factor; where k represents the gas type index, k is 1 for CO, k is 2 for CH4, k is 3 for O2, and k is 4 for humidity; w k is the weight coefficient of the kth gas, satisfying represents the instantaneous rate of change of the kth gas concentration; According to the formula Calculate the humidity correlation factor; where M is the number of humidity sensor nodes; H j is the relative humidity measurement value of the jth sensor node; H sat is the saturated humidity at the current temperature; I TH is the temperature-humidity coupling coefficient: Where T opt is the optimal temperature for coal spontaneous combustion, the preset value of which is 60℃, σ T is the standard deviation of temperature distribution According to the formula R=γ1·sigmoid(α T )+γ2·tanh(α G )+γ3·α H Calculate the comprehensive risk index R; where γ1, γ2, and γ3 are the weight coefficients of each factor, γ1+γ2+γ3=1, is the normalization function, and tanh(·) is the hyperbolic tangent function.

8. An intelligent monitoring and early warning system for coal spontaneous combustion risk in coal mine goaf areas, characterized by: include: A network deployment unit, used for deploying a dot matrix wireless sensor network in a target goaf; The dot matrix wireless sensor network includes a plurality of sensor nodes; the sensor nodes are integrated with temperature, CO, CH4, O2 and humidity detection modules; A data processing unit is used to collect ambient temperature concentration, gas concentration data and humidity data using the sensor nodes, and perform real-time filtering and outlier removal through edge computing nodes to obtain pre-processed multimodal data; A feature extraction unit, configured to extract features from the multimodal data based on a transfer learning model to generate a dynamic feature matrix; A result prediction unit is used to model the dynamic feature matrix using a spatiotemporal graph convolutional network to predict the temperature field and gas diffusion trend of the goaf and obtain a prediction result; An index calculation unit, used to calculate a comprehensive risk index based on the prediction results; The comprehensive risk index incorporates temperature gradient, gas concentration change rate and humidity correlation factors; The hierarchical warning unit is used to divide the comprehensive risk index into four warning levels and to perform hierarchical warning according to the warning levels.

Images