Gas outburst monitoring and early warning method and system based on improved ConvLSTM model

Abstract

The invention discloses a gas outburst monitoring and early warning method and system based on an improved ConvLSTM model, and relates to the technical field of gas monitoring, and the key points of the technical scheme are that the method comprises the following steps: constructing a mine gas data set; cleaning the gas data set, converting unstructured data into structured data, and performing data cleaning, denoising and standardization on the structured data; performing high-order derived feature extraction on the preprocessed gas data set; an improved ConvLSTM model is constructed; using a historical gas data set to train the improved ConvLSTM model; identifying a potential gas high-risk area; based on an analysis result of the improved ConvLSTM model, identifying a potential gas outburst ignition risk area; and constructing a monitoring platform. By improving the ConvLSTM model, efficient collection and processing of gas data are achieved, monitoring accuracy and real-time performance are improved, and powerful support is provided for mine safety management.

Description

Technical Field

[0001] The present invention relates to the field of gas monitoring technology, and more specifically, to a gas outburst monitoring and early warning method and system based on an improved ConvLSTM model. Background Art

[0002] Gas outburst monitoring and early warning have become an important part of mine safety management. Existing gas outburst monitoring and early warning methods mainly rely on traditional sensor data acquisition and simple data analysis technology, which have many limitations and cannot meet the high requirements of modern mining for real-time and accuracy.

[0003] First, the existing gas outburst monitoring and early warning methods lack accuracy in data processing and risk assessment. Traditional methods usually use basic data analysis and statistical methods, lack effective intelligent algorithm support, and cannot fully capture the data characteristics in complex mine environments, resulting in inaccurate monitoring results. These methods often rely on a single sensor or a limited sensor network, and the breadth and depth of data collection are insufficient to fully reflect the safety status of the mine. At the same time, the data processing efficiency is low, and it is impossible to quickly obtain accurate risk assessment results under large-scale data, which affects the real-time and effectiveness of monitoring.

[0004] Secondly, the existing monitoring and early warning system also has problems in data collection frequency and layout design. The sensor layout of the traditional system is often not reasonable enough, resulting in incomplete data collection, affecting the accuracy of subsequent analysis and risk assessment. Due to the low frequency of data collection, the system cannot obtain the latest mine safety data in a timely manner, resulting in monitoring lag. In addition, there is a lot of noise and inconsistency in the raw data, and it has not been effectively pre-processed, resulting in low data quality, which further affects the reliability of the monitoring results.

[0005] In response to these problems, the improved ConvLSTM model provides a new solution for gas outburst detection. Although the existing particle swarm algorithm performs well in optimization problems, its efficiency and accuracy still need to be improved when processing complex gas outburst monitoring and early warning data. When processing complex gas outburst early warning data, the traditional particle swarm algorithm faces the challenge of high-dimensional data and large computational complexity, and is prone to fall into local optimality, affecting the overall monitoring effect.

[0006] Therefore, the present invention aims to provide a gas outburst monitoring and early warning method and system based on an improved ConvLSTM model to solve the above problems. Summary of the invention

[0007] The purpose of the present invention is to provide a gas outburst monitoring and early warning method and system based on an improved ConvLSTM model. By improving the ConvLSTM model, the present invention realizes efficient data collection and processing, improves the accuracy and real-time performance of monitoring, and provides strong support for mine safety production.

[0008] The above technical objectives of the present invention are achieved through the following technical solutions: A gas outburst monitoring and early warning method based on an improved ConvLSTM model, comprising the following steps:

[0009] S1. Various sensors are deployed in key areas of the mine to collect gas concentration, rock vibration signals, temperature data, humidity data and air flow pressure to build a gas data set;

[0010] S2. Clean and preprocess the gas data set, convert the unstructured data into structured data, and then clean, denoise and standardize the structured data to retain the node location information of each data in the gas data set;

[0011] S3, extracting high-order derived features from the preprocessed gas data set to extract time domain features, frequency domain features, and time-frequency domain features of the gas data set;

[0012] S4. Based on the time domain features, frequency domain features and time-frequency domain features, combined with the mine geographical location and adjacent relationship of the sensor, an improved ConvLSTM model is constructed. The goal of the improved ConvLSTM model is to monitor five levels of gas risks;

[0013] S5. Use the historical gas dataset to train the improved ConvLSTM model;

[0014] S6. Input the preprocessed gas data set into the trained improved ConvLSTM model for analysis. The improved ConvLSTM model monitors the gas outburst risk according to the input data and generates analysis results.

[0015] S7. Based on the analysis results of the improved ConvLSTM model, potential gas risk areas are identified, the risk levels of the identified risk areas are evaluated, and corresponding early warning information is generated. When high-risk areas are detected, alarms are issued in a timely manner to notify relevant personnel and systems, and preventive measures are taken. When data trends within a preset range are monitored, corresponding monitoring feedback measures are taken;

[0016] S8. Based on monitoring feedback information and monitoring early warning system, build a monitoring platform and provide a visual interface so that managers can view the current monitoring status, historical data trends and early warning information.

[0017] The present invention is further configured as follows: the method for constructing the gas data set in step S1 comprises the following steps:

[0018] S11. Arrange gas concentration sensors, rock vibration signal sensors, temperature sensors, humidity sensors and airflow pressure sensors in key areas of the mine to collect gas concentration G(t), rock vibration signal S(t), temperature data T(t), humidity data H(t) and airflow pressure A(t), respectively;

[0019] S12. Summarize the collected data to construct a gas data set D:

[0020] D={(t, G(t), S(t), T(t), H(t), A(t), L)};

[0021] Among them, t represents the collection time, and L represents the geographical location information of the sensor.

[0022] The present invention is further configured as follows: the pre-processing method in step S2 comprises the following steps:

[0023] S21, performing preliminary cleaning on the gas data set D to convert unstructured data into structured data;

[0024] S22. Clean the structured gas data set to remove outliers and incomplete data points:

[0025] If |d i -μ D |>σ D , then d i It is an outlier and needs to be removed;

[0026] Among them, d i is a single data point in the gas data set, μ D is the mean of the gas data set D, σ D is the standard deviation of the gas data set D;

[0027] S23, performing denoising processing on the cleaned gas data set, using a filtering algorithm to remove noise components in the data, and obtaining smooth mine environment data;

[0028] S24, performing standardization processing on the denoised gas data set, converting data of different dimensions into the same dimension range;

[0029] S25. Keep the node location information L of each data point in the gas data set to ensure the consistency and traceability of the data in geographic space:

[0030] D final ={(t, P norm(t), S norm (t), T norm (t), H norm (t), A norm (t), L)};

[0031] Among them, D final To preserve the final gas data set after the node location information is retained, P norm (t), S norm (t), T norm (t), H norm (t), A norm (t) are the standardized gas concentration, rock vibration signal, temperature data, humidity data and air flow pressure, respectively.

[0032] The present invention is further configured as follows: the method in step S4, based on the time domain characteristics, frequency domain characteristics and time-frequency domain characteristics, combined with the mine geographical location and neighboring relationship of the sensor, comprises the following steps:

[0033] S41. According to the gas data set D final The time domain features, frequency domain features and time-frequency domain features are extracted to extract the corresponding feature matrix F:

[0034] F={F t ,F f ,F tf};

[0035] Among them, F t represents the time domain feature matrix, F f represents the frequency domain feature matrix, F tf Represents the time-frequency domain feature matrix;

[0036] S42. Combine the mine geographical location L and the adjacent relationship of the sensor to construct a geographical location matrix G to describe the spatial relationship between the sensors:

[0037] G = {g ij};

[0038] Among them, g ij represents the geographical proximity relationship between sensors i and j. If sensors i and j are adjacent, then g ij =1, otherwise g ij =0;

[0039] S43. Based on the feature matrix F and the geographic location matrix G, an improved ConvLSTM model is constructed. The model includes a convolutional layer, a graph convolutional network layer, a long short-term memory network layer, and a self-attention mechanism layer.

[0040] The present invention is further configured as follows: the method for constructing the improved ConvLSTM model in step S43 comprises the following steps:

[0041] S431. In the convolution layer, a multi-scale convolution kernel structure is used to capture the characteristics of gas concentration G(t), rock vibration signal S(t), temperature data T(t), humidity data H(t) and airflow pressure A(t) at different spatial and temporal scales:

[0042]

[0043] in, represents the output feature value of the lth convolutional layer at position (x, y), represents the (m,n)th weight of the convolution kernel in the first layer, b (l) represents the bias of the lth layer, σ represents the activation function, and D final (x+m, y+n) represents the value of the standardized gas data set at the position (x+m, y+n), φ(m, n) represents the newly introduced multi-scale convolution kernel adjustment function, which is used to adjust the feature extraction of different scales and capture the multi-scale features related to gas prominence;

[0044] S432, introduce an improved graph convolutional network, fuse the geographic location matrix G with the feature matrix, and capture the spatial relationship and correlation between sensors by introducing a new node relationship modeling method. The graph convolution operation is expressed as:

[0045]

[0046] in, represents the output feature of the k-th layer graph convolutional network at node v, represents the set of neighbor nodes of node v, c vu represents the normalization constant, G vu represents the weight between nodes v and u in the geographic location matrix, represents the output feature of the k-1th layer convolutional network at node u, represents the weight matrix of the k-th layer graph convolutional network, represents the bias of the k-th layer graph convolutional network, ψ(v,u) represents the new node relationship modeling function, which is used to model the spatial relationship between sensor nodes and optimize the spatial correlation analysis of gas outburst monitoring;

[0047] S433, after the graph convolutional network layer, an adaptive time-frequency analysis mechanism is introduced to dynamically extract and weight the features of different time periods and frequency ranges to generate comprehensive time-frequency features:

[0048]

[0049] Among them, F tf_adaptive (t, f) represents the adaptive time-frequency feature at time t and frequency f, ατv (t, f) shows the dynamic weight at time τ and frequency v, F gcn (τ, v) represents the output features of the graph convolutional network at time τ and frequency v, and γ(τ, v) represents a new time-frequency weighting function, which is used to dynamically adjust the weights of different time and frequency features, and optimize and improve the sensitivity of the ConvLSTM model to gas salient features at different time and frequency distributions;

[0050] S434. In the long short-term memory network layer, the spatial relationship between sensors is modeled by combining the feature matrix processed by the graph convolutional network:

[0051]

[0052] in, represents the output feature of the h-th layer long short-term memory network at time t, represents the weight matrix input to the hidden layer, represents the weight matrix from hidden layer to hidden layer, H( h - 1 ) represents the value of the hidden state of the h-1th layer at time t-1, b( h ) represents the bias of the hth layer, λ(t) represents the time adjustment function;

[0053] S435. Introduce the self-attention mechanism, focus on the time features by weighted calculation of the features of different time steps, and optimize and improve the ConvLSTM model's ability to predict gas outbursts at critical moments:

[0054]

[0055] Among them, F attention (t) represents the attention weighted feature at time t, β i (t) represents the attention weight at time step i, F lstm (i) represents the output features of the LSTM network at time step i, and θ(i) represents the attention weighting function, which is used to focus on the features of the key time steps;

[0056] S436, based on the output of the position perception long short-term memory network layer and the self-attention mechanism layer, combined with the multi-layer perceptron, the gas outburst risk level is predicted, and five levels of gas outburst risk level R are defined, namely, no risk, r 1 , low risk 2 , medium risk 3 , medium and high risk 4 and high risk 5 .

[0057] The present invention is further configured as follows: the method for identifying potential gas risk areas based on the analysis results of the improved ConvLSTM model in step S7 comprises the following steps:

[0058] S71. Based on the analysis results of the improved ConvLSTM model, potential gas outburst risk areas are identified, and the model outputs the gas outburst risk score matrix R score :

[0059] R score = {r ij};

[0060] Among them, r ij represents the gas outburst risk score at location (i, j);

[0061] S72. Evaluate the identified risk areas and determine their risk level R level The risk levels are classified as follows:

[0062]

[0063] Among them, θ 1 ,θ 2 ,θ 3 ,θ 4 They are the thresholds for different risk levels;

[0064] S73, according to the risk level R level Generate corresponding warning information W, which includes risk level, specific location and recommended measures;

[0065] S74, when a high-risk area is detected 5 When an alarm occurs, the system will promptly issue an alarm A, notify relevant personnel and systems, and take preventive measures;

[0066] S75, when the data trend T within the preset range is monitored trend When the system takes corresponding monitoring feedback measures, the data trend monitoring formula is:

[0067] T trend ={t|f(t)>φ};

[0068] Wherein, f(t) represents the data trend function, φ represents the preset trend threshold, and when the trend function f(t) exceeds the threshold φ, corresponding monitoring feedback measures are taken;

[0069] S76. Record the warning information W and the alarm A in the system log.

[0070] The present invention also provides a gas outburst monitoring and early warning system based on an improved ConvLSTM model, which is characterized by comprising a data acquisition module, a data preprocessing module, a feature extraction module, an improved ConvLSTM model module, a model training module, a risk assessment and early warning module and a monitoring platform module;

[0071] The data acquisition module is used to collect gas concentration, rock mass vibration signal, temperature data, humidity data and air flow pressure to generate a gas data set;

[0072] The data preprocessing module is used to clean the collected gas data set, convert the unstructured data into structured data, and then clean, denoise and standardize the structured data to retain the node position information of each data in the gas data set;

[0073] The feature extraction module is used to extract high-order derived features from the preprocessed gas data set, extract the time domain features, frequency domain features and time-frequency domain features of the gas data set, and construct a feature matrix and a geographical location matrix in combination with the mine geographical location and adjacent relationship of the sensor;

[0074] The improved ConvLSTM model module is based on the feature matrix and the geographic location matrix, and is used to construct an improved ConvLSTM model, wherein the model includes a convolution layer, a graph convolution network layer, a long short-term memory network layer, and a self-attention mechanism layer, and is used to monitor five levels of gas outburst risks;

[0075] The model training module uses the historical gas data set to improve the ConvLSTM model for training and optimizes the model parameters so that it can accurately predict the risk level of gas outburst;

[0076] The risk assessment and early warning module is based on the analysis results of the improved ConvLSTM model, and is used to identify potential gas outburst risk areas, evaluate the risk level of the identified risk areas, generate corresponding early warning information, and issue an alarm in time when a high-risk area is detected, notify relevant personnel and systems, take preventive measures, and take corresponding monitoring feedback measures when a data trend within a preset range is monitored;

[0077] The monitoring platform module provides a visual interface for management personnel to view the current monitoring status, historical data trends and warning information in real time, and record warning information and alarm information.

[0078] In summary, the present invention has the following beneficial effects:

[0079] 1. The improved ConvLSTM model of the present invention adopts a multi-scale convolution kernel structure in the convolution layer, which can capture the characteristics of gas concentration, rock vibration signal, temperature data, humidity data and airflow pressure at different spatial and temporal scales, and improve the feature extraction capability of complex gas outburst monitoring and early warning data. An improved graph convolutional network is introduced to fuse the geographic location matrix with the feature matrix, and the spatial relationship and correlation between sensors are captured through a new node relationship modeling method, which optimizes the model's ability to analyze the spatial relationship between sensor nodes. After the graph convolutional network layer, an adaptive time-frequency analysis mechanism is introduced to dynamically extract and weight the features of different time periods and frequency ranges to generate comprehensive time-frequency features, thereby improving the model's sensitivity and prediction accuracy to gas outburst features at different time-frequency distributions. A self-attention mechanism is introduced to focus on important time features by weighted calculation of features of different time steps, thereby optimizing the model's ability to predict gas outbursts at critical moments. This mechanism significantly improves the model's attention and recognition ability to key features, and enhances the timeliness and accuracy of early warning. By combining convolutional neural networks, graph convolutional networks, long short-term memory networks, and self-attention mechanisms, a multi-level, multi-scale, and multi-dimensional comprehensive gas outburst monitoring model is constructed. It not only improves the processing capability and analysis accuracy of complex gas outburst monitoring and early warning data, but also enhances the robustness and stability of the model, and can achieve efficient and accurate risk assessment under large-scale data;

[0080] 2. Based on the analysis results of the improved ConvLSTM model, the system of the present invention can identify potential gas outburst risk areas in real time, evaluate the risk level of the identified risk areas, and generate corresponding early warning information. When a high-risk area is detected, the system will promptly issue an alarm to notify relevant personnel and systems to take preventive measures to effectively prevent disasters. The system has built an integrated monitoring platform that provides a visual interface, allowing managers to view the current monitoring status, historical data trends, and early warning information in real time. By recording and analyzing early warning information and alarm information, managers can make effective decisions and interventions to improve the overall level of mine safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a flow chart of a gas outburst monitoring and early warning method based on an improved ConvLSTM model in Example 1 of the present invention;

[0082] Figure 2 It is a structural diagram of an improved ConvLSTM model of a gas outburst monitoring and early warning method based on an improved ConvLSTM model in Example 1 of the present invention. DETAILED DESCRIPTION

[0083] The following is combined with Figure 1-2The present invention is described in further detail.

[0084] Embodiment 1: A gas outburst monitoring and early warning method based on an improved ConvLSTM model comprises the following steps:

[0085] S1. Various types of sensors are deployed in key areas of the mine to collect gas concentration, rock vibration signals, temperature data, humidity data and air flow pressure to build a gas data set.

[0086] The method for constructing a gas data set in step S1 of this embodiment includes the following steps:

[0087] S11. Arrange gas concentration sensors, rock vibration signal sensors, temperature sensors, humidity sensors and airflow pressure sensors in key areas of the mine to collect gas concentration G(t), rock vibration signal S(t), temperature data T(t), humidity data H(t) and airflow pressure A(t), respectively;

[0088] S12. Summarize the collected data to construct a gas data set D:

[0089] D={(t,G(t),S(t),T(t),H(t),A(t),L)};

[0090] Among them, t represents the collection time, and L represents the geographical location information of the sensor.

[0091] S2. Clean and preprocess the gas data set, convert the unstructured data into structured data, and then clean, denoise and standardize the structured data to retain the node location information of each data in the gas data set.

[0092] The pre-processing method in step S2 of this embodiment includes the following steps:

[0093] S21, performing preliminary cleaning on the gas data set D to convert unstructured data into structured data;

[0094] S22. Clean the structured gas data set to remove outliers and incomplete data points:

[0095] If |d i -μ D |>σ D , then d i It is an outlier and needs to be removed;

[0096] Among them, d i is a single data point in the gas data set, μ D is the mean of the gas data set D, σ D is the standard deviation of the gas data set D;

[0097] S23, performing denoising processing on the cleaned gas data set, using a filtering algorithm to remove noise components in the data, and obtaining smooth mine environment data;

[0098] S24, performing standardization processing on the denoised gas data set, converting data of different dimensions into the same dimension range;

[0099] S25. Keep the node location information L of each data point in the gas data set to ensure the consistency and traceability of the data in geographic space:

[0100] D final ={(t, P norm (t), S norm (t), T norm (t), H norm (t), A norm (t), L)};

[0101] Among them, D final To preserve the final gas data set after the node location information is retained, P norm (t), S norm (t), T norm (t), H norm (t), A norm (t) are the standardized gas concentration, rock vibration signal, temperature data, humidity data and air flow pressure, respectively.

[0102] S3. Perform high-order derived feature extraction on the preprocessed gas data set to extract time domain features, frequency domain features, and time-frequency domain features of the gas data set.

[0103] S4. Based on the time domain characteristics, frequency domain characteristics and time-frequency domain characteristics, combined with the mine geographical location and adjacent relationship of the sensor, an improved ConvLSTM model is constructed. The goal of the improved ConvLSTM model is to monitor five levels of gas risks.

[0104] In step S4 of this embodiment, the method of combining the mine geographical location and the neighboring relationship of the sensor according to the time domain characteristics, the frequency domain characteristics and the time-frequency domain characteristics includes the following steps:

[0105] S41. According to the gas data set D final The time domain features, frequency domain features and time-frequency domain features are extracted to extract the corresponding feature matrix F:

[0106] F={F t , F f , F tf};

[0107] Among them, F trepresents the time domain feature matrix, F f represents the frequency domain feature matrix, F tf Represents the time-frequency domain feature matrix;

[0108] S42. Combine the mine geographical location L and the adjacent relationship of the sensor to construct a geographical location matrix G to describe the spatial relationship between the sensors:

[0109] G={gi j};

[0110] Among them, g ij represents the geographical proximity relationship between sensors i and j. If sensors i and j are adjacent, then g ij =1, otherwise g ij =0;

[0111] S43. Based on the feature matrix F and the geographic location matrix G, an improved ConvLSTM model is constructed. The model includes a convolutional layer, a graph convolutional network layer, a long short-term memory network layer, and a self-attention mechanism layer.

[0112] The method for constructing an improved ConvLSTM model in step S43 of this embodiment includes the following steps:

[0113] S431. In the convolution layer, a multi-scale convolution kernel structure is used to capture the characteristics of gas concentration G(t), rock vibration signal S(t), temperature data T(t), humidity data H(t) and airflow pressure A(t) at different spatial and temporal scales:

[0114]

[0115] in, represents the output feature value of the lth convolutional layer at position (x, y), represents the (m,n)th weight of the convolution kernel of the lth layer, b (l) represents the bias of the lth layer, σ represents the activation function, and D final (x+m, y+n) represents the value of the standardized gas data set at the position (x+m, y+n), φ(m, n) represents the newly introduced multi-scale convolution kernel adjustment function, which is used to adjust the feature extraction of different scales and capture the multi-scale features related to gas prominence;

[0116] S432, introduce an improved graph convolutional network, fuse the geographic location matrix G with the feature matrix, and capture the spatial relationship and correlation between sensors by introducing a new node relationship modeling method. The graph convolution operation is expressed as:

[0117]

[0118] in, represents the output feature of the k-th layer graph convolutional network at node v, represents the set of neighbor nodes of node v, c vu represents the normalization constant, G vu represents the weight between nodes v and u in the geographic location matrix, represents the output feature of the k-1th layer convolutional network at node u, represents the weight matrix of the k-th layer graph convolutional network, represents the bias of the k-th layer graph convolutional network, ψ(v,u) represents the new node relationship modeling function, which is used to model the spatial relationship between sensor nodes and optimize the spatial correlation analysis of gas outburst monitoring;

[0119] S433, after the graph convolutional network layer, an adaptive time-frequency analysis mechanism is introduced to dynamically extract and weight the features of different time periods and frequency ranges to generate comprehensive time-frequency features:

[0120]

[0121] Among them, F tf_adaptive (t, f) represents the adaptive time-frequency feature at time t and frequency f, α τv (t, f) shows the dynamic weight at time τ and frequency v, F gcn (τ, v) represents the output features of the graph convolutional network at time τ and frequency v, and γ(τ, v) represents a new time-frequency weighting function, which is used to dynamically adjust the weights of different time and frequency features, and optimize and improve the sensitivity of the ConvLSTM model to gas salient features at different time and frequency distributions;

[0122] S434. In the long short-term memory network layer, the spatial relationship between sensors is modeled by combining the feature matrix processed by the graph convolutional network:

[0123]

[0124] in, represents the output feature of the h-th layer long short-term memory network at time t, represents the weight matrix input to the hidden layer, represents the weight matrix from hidden layer to hidden layer, H (h-1) represents the value of the hidden state of the h-1th layer at time t-1, b (h) represents the bias of the hth layer, λ(t) represents the time adjustment function;

[0125] S435. Introduce the self-attention mechanism, focus on the time features by weighted calculation of the features of different time steps, and optimize and improve the ConvLSTM model's ability to predict gas outbursts at critical moments:

[0126]

[0127] Among them, F attention (t) represents the attention weighted feature at time t, β i (t) represents the attention weight at time step i, F lstm (i) represents the output features of the LSTM network at time step i, and θ(i) represents the attention weighting function, which is used to focus on the features of the key time steps;

[0128] S436, based on the output of the position perception long short-term memory network layer and the self-attention mechanism layer, combined with the multi-layer perceptron, the gas outburst risk level is predicted, and five levels of gas outburst risk level R are defined, namely, no risk, r 1 , low risk 2 , medium risk 3 , medium and high risk 4 and high risk 5 .

[0129] S5. Use the historical gas dataset to train the improved ConvLSTM model.

[0130] S6. Input the preprocessed gas data set into the trained improved ConvLSTM model for analysis. The improved ConvLSTM model monitors the gas outburst risk according to the input data and generates analysis results.

[0131] S7. Based on the analysis results of the improved ConvLSTM model, potential gas risk areas are identified, the risk levels of the identified risk areas are evaluated, and corresponding early warning information is generated. When high-risk areas are detected, alarms are issued in a timely manner to notify relevant personnel and systems, and preventive measures are taken. When data trends within the preset range are monitored, corresponding monitoring feedback measures are taken.

[0132] In step S7 of this embodiment, based on the analysis result of the improved ConvLSTM model, the method for identifying potential gas risk areas includes the following steps:

[0133] S71. Based on the analysis results of the improved ConvLSTM model, potential gas outburst risk areas are identified, and the model outputs the gas outburst risk score matrix R score :

[0134] R score = {r ij};

[0135] Among them, r ij represents the gas outburst risk score at location (i, j);

[0136] S72. Evaluate the identified risk areas and determine their risk level R level The risk levels are classified as follows:

[0137]

[0138] Among them, θ 1 ,θ 2 ,θ 3 ,θ 4 They are the thresholds for different risk levels;

[0139] S73, according to the risk level R level Generate corresponding warning information W, which includes risk level, specific location and recommended measures;

[0140] S74, when a high-risk area is detected 5 When an alarm occurs, the system will promptly issue an alarm A, notify relevant personnel and systems, and take preventive measures;

[0141] S75, when the data trend T within the preset range is monitored trend When the system takes corresponding monitoring feedback measures, the data trend monitoring formula is:

[0142] T trend ={t|f(t)>φ};

[0143] Wherein, f(t) represents the data trend function, φ represents the preset trend threshold, and when the trend function f(t) exceeds the threshold φ, corresponding monitoring feedback measures are taken;

[0144] S76. Record the warning information W and the alarm A in the system log.

[0145] S8. Based on monitoring feedback information and monitoring early warning system, build a monitoring platform and provide a visual interface to enable managers to view the current monitoring status, historical data trends and early warning information.

[0146] Embodiment 2: A gas outburst monitoring and early warning system based on an improved ConvLSTM model, comprising a data acquisition module, a data preprocessing module, a feature extraction module, an improved ConvLSTM model module, a model training module, a risk assessment and early warning module and a monitoring platform module.

[0147] In this embodiment, the data acquisition module is used to collect gas concentration, rock mass vibration signal, temperature data, humidity data and air flow pressure to generate a gas data set.

[0148] The data preprocessing module is used to clean the collected gas data set, convert unstructured data into structured data, and then clean, denoise and standardize the structured data to retain the node location information of each data in the gas data set.

[0149] The feature extraction module is used to extract high-order derived features from the preprocessed gas data set, extract the time domain features, frequency domain features and time-frequency domain features of the gas data set, and construct the feature matrix and geographic location matrix in combination with the mine geographical location and adjacent relationship of the sensor.

[0150] The improved ConvLSTM model module is used to construct an improved ConvLSTM model based on the feature matrix and the geographic location matrix. The model includes a convolutional layer, a graph convolutional network layer, a long short-term memory network layer, and a self-attention mechanism layer, which is used to monitor five levels of gas outburst risks.

[0151] The model training module uses historical gas datasets to improve the ConvLSTM model for training and optimizes model parameters so that it can accurately predict the risk level of gas outburst.

[0152] The risk assessment and early warning module is based on the analysis results of the improved ConvLSTM model and is used to identify potential gas outburst risk areas, evaluate the risk levels of the identified risk areas, generate corresponding early warning information, and issue alarms in a timely manner when high-risk areas are detected to notify relevant personnel and systems and take preventive measures. When data trends within a preset range are monitored, corresponding monitoring feedback measures are taken.

[0153] The monitoring platform module provides a visual interface for managers to view the current monitoring status, historical data trends and warning information in real time, and record warning information and alarm information.

[0154] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A gas outburst monitoring and early warning method based on an improved ConvLSTM model, characterized by: The following steps are involved: S1. Various sensors are deployed in key areas of the mine to collect gas concentration, rock mass vibration signals, temperature data, humidity data and air flow pressure to build a gas data set; S2. Clean and preprocess the gas data set, convert the unstructured data into structured data, and then clean, denoise and standardize the structured data to retain the node location information of each data in the gas data set; S3, extracting high-order derived features from the preprocessed gas data set to extract time domain features, frequency domain features, and time-frequency domain features of the gas data set; S4. Based on the time domain features, frequency domain features and time-frequency domain features, combined with the mine geographical location and adjacent relationship of the sensor, an improved ConvLSTM model is constructed. The goal of the improved ConvLSTM model is to monitor five levels of gas risks; S5. Use the historical gas dataset to train the improved ConvLSTM model; S6. Input the preprocessed gas data set into the trained improved ConvLSTM model for analysis. The improved ConvLSTM model monitors the gas outburst risk according to the input data and generates analysis results. S7. Based on the analysis results of the improved ConvLSTM model, potential gas risk areas are identified, the risk levels of the identified risk areas are evaluated, and corresponding early warning information is generated. When high-risk areas are detected, alarms are issued in a timely manner to notify relevant personnel and systems, and preventive measures are taken. When data trends within a preset range are monitored, corresponding monitoring feedback measures are taken; S8. Based on monitoring feedback information and monitoring early warning system, build a monitoring platform and provide a visual interface to enable managers to view the current monitoring status, historical data trends and early warning information.

2. A gas outburst monitoring and early warning method based on an improved ConvLSTM model according to claim 1, characterized in that: The method for constructing the gas data set in step S1 comprises the following steps: S11. Arrange gas concentration sensors, rock mass vibration signals, temperature sensors, humidity sensors and airflow pressure sensors in key areas of the mine to collect gas concentration G(t), rock mass vibration signals S(t), temperature data T(t), humidity data H(t) and airflow pressure A(t), respectively; S12. Summarize the collected data to construct a gas data set D: D={(t,G(t),S(t),T(t),H(t),A(t),L)}; Among them, t represents the collection time, and L represents the geographical location information of the sensor.

3. A gas outburst monitoring and early warning method based on an improved ConvLSTM model according to claim 1, characterized in that: The method for pre-processing in step S2 comprises the following steps: S21, performing preliminary cleaning on the gas data set D to convert unstructured data into structured data; S22. Clean the structured gas data set to remove outliers and incomplete data points: If |d i -μ D |>σ D , then d i It is an outlier and needs to be removed; Among them, d i is a single data point in the gas data set, μ D is the mean of the gas data set D, σ D is the standard deviation of the gas data set D; S23, performing denoising processing on the cleaned gas data set, using a filtering algorithm to remove noise components in the data, and obtaining smooth mine environment data; S24, performing standardization processing on the denoised gas data set, converting data of different dimensions into the same dimension range; S25. Keep the node location information L of each data point in the gas data set to ensure the consistency and traceability of the data in geographic space: D final ={(t,P norm (t),S norm (t),T norm (t),H norm (t),A norm (t),L)}; Among them, D final To preserve the final gas data set after the node location information is retained, P norm (t),S norm (t),T norm (t),H norm (t),A norm (t) are the standardized gas concentration, rock vibration signal, temperature data, humidity data and air flow pressure, respectively.

4. A gas outburst monitoring and early warning method based on an improved ConvLSTM model according to claim 1, characterized in that: The method in step S4, based on the time domain features, frequency domain features and time-frequency domain features, combining the mine geographical location and neighboring relationship of the sensor, comprises the following steps: S41. According to the gas data set D final The time domain features, frequency domain features and time-frequency domain features are extracted to extract the corresponding feature matrix F: F={F t ,F f ,F tf }; Among them, F t represents the time domain feature matrix, F f represents the frequency domain feature matrix, F tf Represents the time-frequency domain feature matrix; S42. Combine the mine geographical location L and the adjacent relationship of the sensor to construct a geographical location matrix G to describe the spatial relationship between the sensors: G={g ij }; Among them, g ij represents the geographical proximity relationship between sensors i and j. If sensors i and j are adjacent, then g ij =1, otherwise g ij =0; S43. Based on the feature matrix F and the geographic location matrix G, an improved ConvLSTM model is constructed. The model includes a convolutional layer, a graph convolutional network layer, a long short-term memory network layer, and a self-attention mechanism layer.

5. A gas outburst monitoring and early warning method based on an improved ConvLSTM model according to claim 4, characterized in that: The method for constructing the improved ConvLSTM model in step S43 comprises the following steps: S431. In the convolution layer, a multi-scale convolution kernel structure is used to capture the characteristics of gas concentration G(t), rock vibration signal S(t), temperature data T(t), humidity data H(t) and airflow pressure A(t) at different spatial and temporal scales: in, represents the output feature value of the lth convolutional layer at position (x, y), represents the (m,n)th weight of the convolution kernel of the lth layer, b (l) represents the bias of the lth layer, σ represents the activation function, and D final (x+m, y+n) represents the value of the standardized gas data set at the position (x+m, y+n), and φ(m,n) represents the newly introduced multi-scale convolution kernel adjustment function, which is used to adjust the feature extraction of different scales and capture the multi-scale features related to gas prominence; S432, introduce an improved graph convolutional network, fuse the geographic location matrix G with the feature matrix, and capture the spatial relationship and correlation between sensors by introducing a new node relationship modeling method. The graph convolution operation is expressed as: in, represents the output feature of the k-th layer graph convolutional network at node v, represents the set of neighbor nodes of node v, c vu represents the normalization constant, G vu represents the weight between nodes v and u in the geographic location matrix, represents the output feature of the k-1th layer convolutional network at node u, represents the weight matrix of the k-th layer graph convolutional network, represents the bias of the k-th layer graph convolutional network, ψ(v,u) represents the new node relationship modeling function, which is used to model the spatial relationship between sensor nodes and optimize the spatial correlation analysis of gas outburst monitoring; S433, after the graph convolutional network layer, an adaptive time-frequency analysis mechanism is introduced to dynamically extract and weight the features of different time periods and frequency ranges to generate comprehensive time-frequency features: Among them, F tf_adaptive (t,f) represents the adaptive time-frequency feature at time t and frequency f, α τν (t,f) shows the dynamic weight at time τ and frequency ν, F gcn (τ, ν) represents the output features of the graph convolutional network at time τ and frequency ν, and γ(τ, ν) represents a new time-frequency weighting function, which is used to dynamically adjust the weights of different time and frequency features, and optimize and improve the sensitivity of the ConvLSTM model to gas salient features at different time and frequency distributions; S434. In the long short-term memory network layer, the spatial relationship between sensors is modeled by combining the feature matrix processed by the graph convolutional network: in, represents the output feature of the h-th layer long short-term memory network at time t, represents the weight matrix input to the hidden layer, represents the weight matrix from hidden layer to hidden layer, H (h-1) represents the value of the hidden state of the h-1th layer at time t-1, b (h) represents the bias of the hth layer, λ(t) represents the time adjustment function; S435. Introduce the self-attention mechanism, focus on the time features by weighted calculation of the features of different time steps, and optimize and improve the ConvLSTM model's ability to predict gas outbursts at critical moments: Among them, F attention (t) represents the attention weighted feature at time t, β i (t) represents the attention weight at time step i, F lstm (i) represents the output features of the LSTM network at time step i, and θ(i) represents the attention weighting function, which is used to focus on the features of the key time steps; S436. Based on the output of the position-aware long short-term memory network layer and the self-attention mechanism layer, the gas outburst risk level is predicted in combination with the multi-layer perceptron, and five levels of gas outburst risk level R are defined, namely no risk r1, low risk r2, medium risk r3, medium-high risk r4 and high risk r5.

6. The gas outburst monitoring and early warning method based on the improved ConvLSTM model according to claim 1 is characterized in that: The method for identifying potential gas risk areas based on the analysis results of the improved ConvLSTM model in step S7 includes the following steps: S71. Based on the analysis results of the improved ConvLSTM model, potential gas outburst risk areas are identified, and the model outputs the gas outburst risk score matrix R score : R score ＝{r ij}； Among them, r ij represents the gas outburst risk score at location (i, j); S72. Evaluate the identified risk areas and determine their risk level R level The risk levels are classified as follows: Among them, θ1, θ2, θ3, and θ4 are thresholds of different risk levels; S73, according to the risk level R level Generate corresponding warning information W, which includes risk level, specific location and recommended measures; S74, when a high-risk area r5 is detected, the system promptly issues an alarm A, notifies relevant personnel and systems, and takes preventive measures; S75, when the data trend T within the preset range is monitored trend When the system takes corresponding monitoring feedback measures, the data trend monitoring formula is: T trend ={t∣f(t)>φ}; Wherein, f(t) represents the data trend function, φ represents the preset trend threshold, and when the trend function f(t) exceeds the threshold φ, corresponding monitoring feedback measures are taken; S76. Record the warning information W and the alarm A in the system log.

7. A gas outburst monitoring and early warning system based on an improved ConvLSTM model, characterized by: It includes data acquisition module, data preprocessing module, feature extraction module, improved ConvLSTM model module, model training module, risk assessment and early warning module and monitoring platform module; The data acquisition module is used to collect gas concentration, rock mass vibration signal, temperature data, humidity data and air flow pressure to generate a gas data set; The data preprocessing module is used to clean the collected gas data set, convert the unstructured data into structured data, and then clean, denoise and standardize the structured data to retain the node position information of each data in the gas data set; The feature extraction module is used to extract high-order derived features from the preprocessed gas data set, extract the time domain features, frequency domain features and time-frequency domain features of the gas data set, and construct a feature matrix and a geographical location matrix in combination with the mine geographical location and adjacent relationship of the sensor; The improved ConvLSTM model module is based on the feature matrix and the geographic location matrix, and is used to construct an improved ConvLSTM model, wherein the model includes a convolution layer, a graph convolution network layer, a long short-term memory network layer, and a self-attention mechanism layer, and is used to monitor five levels of gas outburst risks; The model training module uses the historical gas data set to improve the ConvLSTM model for training and optimizes the model parameters so that it can accurately predict the risk level of gas outburst; The risk assessment and early warning module is based on the analysis results of the improved ConvLSTM model, and is used to identify potential gas outburst risk areas, evaluate the risk level of the identified risk areas, generate corresponding early warning information, and issue an alarm in time when a high-risk area is detected, notify relevant personnel and systems, take preventive measures, and take corresponding monitoring feedback measures when a data trend within a preset range is monitored; The monitoring platform module provides a visual interface for management personnel to view the current monitoring status, historical data trends and warning information in real time, and record warning information and alarm information.

Images