Coal and gas outburst dynamic monitoring and intelligent early warning method and device

Abstract

The present application relates to the field of data processing technology, and discloses a method and device for dynamic monitoring and intelligent early warning of coal and gas outbursts. The method comprises: collecting and preprocessing multi-source sensor data set in the target area to obtain a multidimensional data set; performing multivariate time series analysis on the multidimensional data set to obtain coal seam gas outburst hazard prediction data; based on the coal seam gas outburst hazard prediction data, performing dynamic risk assessment and spatial interpolation processing on the multidimensional data set to obtain real-time risk distribution data of the target area; performing multidimensional threshold analysis on the real-time risk distribution data to obtain preliminary early warning data; performing multimodal information fusion on the preliminary early warning data to obtain target early warning data; performing intelligent decision tree analysis on the target early warning data to obtain a target early warning scheme. The present application improves the accuracy and reliability of dynamic monitoring and intelligent early warning of coal and gas outbursts.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method and device for dynamic monitoring and intelligent early warning of coal and gas outbursts. Background Art

[0002] Coal and gas outbursts are common major safety hazards in the coal mining process, seriously threatening the lives of miners and coal mine production. Traditional coal and gas outburst monitoring methods mainly rely on single sensor data and simple threshold judgments, such as gas concentration monitoring and coal seam stress monitoring. Although these methods can reflect the state of coal seams to a certain extent, it is difficult to fully capture the complex dynamic process of coal and gas outbursts. With the development of technology, some studies have begun to use multi-source data fusion and machine learning algorithms to improve the accuracy of early warning, such as using neural network models to conduct comprehensive analysis of multiple parameters, or using support vector machines for coal and gas outburst risk assessment.

[0003] However, the existing coal and gas outburst monitoring and early warning technologies still have some shortcomings. First, most methods fail to fully utilize the spatiotemporal correlation of multi-source heterogeneous data, resulting in insufficient information utilization. Secondly, the data preprocessing and feature extraction processes are often relatively simple, failing to effectively remove noise and extract key features. Furthermore, most of the existing prediction models are static and difficult to adapt to the dynamic changes in coal seam conditions. In addition, in terms of risk assessment and early warning decision-making, there is a lack of effective processing and dynamic adjustment mechanisms for uncertainty, which affects the accuracy and reliability of early warnings. Finally, the existing methods are not deep enough in the research of multimodal information fusion and intelligent decision-making, making it difficult to achieve efficient and accurate intelligent early warnings. Summary of the invention

[0004] The present application provides a method and device for dynamic monitoring and intelligent early warning of coal and gas outbursts, which are used to improve the accuracy and reliability of dynamic monitoring and intelligent early warning of coal and gas outbursts.

[0005] In the first aspect, the present application provides a method for dynamic monitoring and intelligent early warning of coal and gas outbursts, which includes: collecting and preprocessing multi-source sensor data set in the target area to obtain a multidimensional data set; performing multivariate time series analysis on the multidimensional data set to obtain coal seam gas outburst hazard prediction data; based on the coal seam gas outburst hazard prediction data, performing dynamic risk assessment and spatial interpolation processing on the multidimensional data set to obtain real-time risk distribution data of the target area; performing multidimensional threshold analysis on the real-time risk distribution data to obtain preliminary warning data; performing multimodal information fusion on the preliminary warning data to obtain target warning data; performing intelligent decision tree analysis on the target warning data to obtain a target warning plan.

[0006] In a second aspect, the present application provides a coal and gas outburst dynamic monitoring and intelligent early warning device, the coal and gas outburst dynamic monitoring and intelligent early warning device comprising:

[0007] A processing module, used for collecting and preprocessing the multi-source sensor data set in the target area to obtain a multi-dimensional data set;

[0008] A prediction module, used for performing multivariate time series analysis on the multidimensional data set to obtain coal seam gas outburst hazard prediction data;

[0009] An interpolation module is used to perform dynamic risk assessment and spatial interpolation processing on the multidimensional data set based on the coal seam gas outburst hazard prediction data to obtain real-time risk distribution data of the target area;

[0010] An analysis module, used to perform multi-dimensional threshold analysis on the real-time risk distribution data to obtain preliminary warning data;

[0011] A fusion module, used for performing multi-modal information fusion on the preliminary warning data to obtain target warning data;

[0012] The decision module is used to perform intelligent decision tree analysis on the target warning data to obtain a target warning solution.

[0013] In the technical solution provided by the present application, through information gain calculation, the role of different attributes in predicting gas outburst can be clarified, and the attributes can be sorted based on this, effectively screening out the most influential decision factors. This process greatly improves the training efficiency of the early warning model, and also improves the prediction accuracy of the model, so that the model can identify high-risk factors more quickly. The technical features of decision tree construction and pruning based on attribute importance sorting avoid the overfitting phenomenon that is prone to occur in traditional decision tree models. Through the construction and pruning of the initial decision tree, the solution can obtain a more streamlined and more generalized decision tree model. This means that the model not only performs well on training data, but also maintains stable performance in actual monitoring, thereby effectively responding to complex and changeable data inputs in coal mine environments. The pruned decision tree model reduces unnecessary nodes and branches, improves the prediction efficiency of the model, and also ensures that scientific and reasonable early warning judgments can still be made under new data. Through path analysis, the complex decision process can be presented in the form of rules, which allows early warning personnel to intuitively understand the logic behind each early warning conclusion. Rule conflict detection and extraction of consistent rule subsets further ensure the logical consistency and decision reliability of the model, avoid misjudgment caused by rule contradictions, and thus improve the practical operability of the early warning system. The technical features of rule matching and confidence evaluation based on consistent rule subsets enable the model to quickly find the most suitable rules and evaluate their reliability based on real-time data in practical applications, and then focus on high-confidence decision paths. This processing method can more effectively identify high-risk events and take targeted preventive measures when facing diverse and complex field data, thereby improving the timeliness of early warning response. The technical features of processing high-confidence decision paths through case reasoning can make full use of the experience of similar historical cases, so that the model not only relies on data rules, but also optimizes the early warning plan in combination with actual experience. In the early warning of coal and gas outburst events, this case-based reasoning method enables the system to quickly give a reasonable response plan when facing uncertain emergencies, thereby greatly improving the accuracy and scientificity of the early warning plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0015] Figure 1 A schematic diagram of an embodiment of a method for dynamic monitoring and intelligent early warning of coal and gas outbursts in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an embodiment of a device for dynamic monitoring and intelligent early warning of coal and gas outbursts in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The embodiments of the present application provide a method and device for dynamic monitoring and intelligent early warning of coal and gas outbursts. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the method for dynamic monitoring and intelligent early warning of coal and gas outburst includes:

[0019] Step S101, collecting and preprocessing multi-source sensor data set in the target area to obtain a multi-dimensional data set;

[0020] Step S102, performing multivariate time series analysis on the multidimensional data set to obtain coal seam gas outburst hazard prediction data;

[0021] Step S103: Based on the coal seam gas outburst hazard prediction data, dynamic risk assessment and spatial interpolation processing are performed on the multidimensional data set to obtain real-time risk distribution data of the target area;

[0022] Step S104: Perform multi-dimensional threshold analysis on the real-time risk distribution data to obtain preliminary warning data;

[0023] Step S105: performing multi-modal information fusion on the preliminary warning data to obtain target warning data;

[0024] Step S106: Perform intelligent decision tree analysis on the target warning data to obtain a target warning solution.

[0025] It is understandable that the execution subject of the present application may be a coal and gas outburst dynamic monitoring and intelligent early warning device, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0026] Specifically, in the field of coal mine safety production, the multi-source sensor data set in the target area is collected and preprocessed to obtain a multidimensional data set. In this step, the multi-source sensors include gas concentration sensors, coal seam stress sensors, temperature sensors, etc., through which various parameter data of the target area are collected in real time. The preprocessing process includes data cleaning, denoising and standardization to ensure data quality and consistency. Subsequently, multivariate time series analysis is performed on the multidimensional data set to obtain coal seam gas outburst hazard prediction data. This step uses time series analysis techniques, such as autoregressive moving average model (ARIMA) or long short-term memory network (LSTM), to analyze the historical data of various parameters and identify potential gas outburst risk patterns and trends.

[0027] Next, based on the coal seam gas outburst hazard prediction data, dynamic risk assessment and spatial interpolation processing are performed on the multidimensional data set to obtain real-time risk distribution data for the target area. The dynamic risk assessment adopts a multi-factor weighted scoring method to consider the impact of changes in factors such as gas concentration, coal seam stress, and temperature on the outburst risk. The spatial interpolation processing uses the Kriging interpolation method to expand the discrete monitoring point data to the entire target area to form a continuous risk distribution map. Multidimensional threshold analysis is performed on the real-time risk distribution data to obtain preliminary warning data. This step sets thresholds for multiple risk indicators, such as gas concentration thresholds, stress change rate thresholds, etc. When a certain indicator or a combination of multiple indicators exceeds the preset threshold, a preliminary warning is triggered.

[0028] Then, the preliminary warning data is subjected to multimodal information fusion to obtain the target warning data. Multimodal information fusion adopts Dempster-Shafer evidence theory, comprehensively considers the reliability and uncertainty of various sensor data, and improves the accuracy and credibility of the warning. Finally, the target warning data is subjected to intelligent decision tree analysis to obtain the target warning plan. The decision tree analysis is based on historical data and expert experience, and automatically generates corresponding warning levels and response measures according to the current risk situation.

[0029] For example, in a coal mine working face, a multi-source sensor network monitors gas concentration, coal seam stress and temperature changes in real time. After data preprocessing, a standardized multidimensional data set is obtained. Time series analysis found that the gas concentration is on an upward trend, and it is predicted that it may reach a dangerous level within 24 hours. Dynamic risk assessment combines the increase in gas concentration and the increase in coal seam stress to calculate the risk index of each area of ​​the working face. Multidimensional threshold analysis shows that the gas concentration and stress change rate in a certain area are close to the warning threshold. Multimodal information fusion further confirms the high-risk status of the area. Finally, the intelligent decision tree analysis generated a targeted early warning plan, including measures such as personnel evacuation, enhanced ventilation and local area closure, which effectively prevented possible coal and gas outburst accidents.

[0030] In the embodiment of the present application, through information gain calculation, the role of different attributes in predicting gas outburst can be clarified, and the attributes can be sorted based on this, effectively screening out the most influential decision factors. This process greatly improves the training efficiency of the early warning model, and also improves the prediction accuracy of the model, so that the model can identify high-risk factors more quickly. The technical features of decision tree construction and pruning based on attribute importance sorting avoid the overfitting phenomenon that is prone to occur in traditional decision tree models. Through the construction and pruning of the initial decision tree, the scheme can obtain a more streamlined and more generalized decision tree model. This means that the model not only performs well on training data, but also maintains stable performance in actual monitoring, thereby effectively responding to complex and changeable data inputs in coal mine environments. The pruned decision tree model reduces unnecessary nodes and branches, improves the prediction efficiency of the model, and also ensures that scientific and reasonable early warning judgments can still be made under new data. Through path analysis, the complex decision process can be presented in the form of rules, which allows early warning personnel to intuitively understand the logic behind each early warning conclusion. Rule conflict detection and extraction of consistent rule subsets further ensure the logical consistency and decision reliability of the model, avoid misjudgment caused by rule contradictions, and thus improve the practical operability of the early warning system. The technical features of rule matching and confidence evaluation based on consistent rule subsets enable the model to quickly find the most suitable rules and evaluate their reliability based on real-time data in practical applications, and then focus on high-confidence decision paths. This processing method can more effectively identify high-risk events and take targeted preventive measures when facing diverse and complex field data, thereby improving the timeliness of early warning response. The technical features of processing high-confidence decision paths through case reasoning can make full use of the experience of similar historical cases, so that the model not only relies on data rules, but also optimizes the early warning plan in combination with actual experience. In the early warning of coal and gas outburst events, this case-based reasoning method enables the system to quickly give a reasonable response plan when facing uncertain emergencies, thereby greatly improving the accuracy and scientificity of the early warning plan.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] (1) Collecting gas pressure data in the target area in real time to obtain raw gas pressure data, and filtering the raw gas pressure data to obtain filtered gas pressure data;

[0033] (2) Collecting coal seam stress data in the target area in real time to obtain original coal seam stress data, and standardizing the original coal seam stress data to obtain standardized coal seam stress data;

[0034] (3) Real-time acquisition of microseismic signal data in the target area to obtain raw microseismic signal data, and denoising the raw microseismic signal data through wavelet transform to obtain denoised microseismic signal data;

[0035] (4) Collect gas concentration data in the target area in real time to obtain raw gas concentration data, and perform mean filtering on the raw gas concentration data to obtain smoothed gas concentration data;

[0036] (5) Collect temperature data in the target area in real time to obtain raw temperature data, and process the raw temperature data through the Kalman filter algorithm to obtain corrected temperature data;

[0037] (6) Performing time synchronization processing on the filtered gas pressure data, standardized coal seam stress data, denoised microseismic signal data, smoothed gas concentration data, and corrected temperature data to obtain time-synchronized multi-source data;

[0038] (7) Perform outlier detection on the time-synchronized multi-source data to obtain candidate multi-source data, and perform interpolation processing on the candidate multi-source data to obtain the completed multi-source data;

[0039] (8) Extract features from the completed multi-source data to obtain a feature vector set, and perform dimensionality reduction processing on the feature vector set to obtain feature data after dimensionality reduction;

[0040] (9) Performing data fusion processing on the feature data after dimension reduction to obtain fused feature data, and normalizing the fused feature data to obtain normalized fused feature data;

[0041] (10) The normalized fused feature data is processed using the principal component analysis algorithm to obtain a multidimensional data set.

[0042] Specifically, the gas pressure data in the target area is collected in real time to obtain the original gas pressure data, and the noise is eliminated through filtering to improve the data quality. The filtering process uses the median filtering algorithm, which can effectively remove sudden interference and outliers and retain the true change trend of the data. Then, the coal seam stress data is collected and standardized in real time. The standardized process converts stress data of different dimensions into a unified scale to facilitate subsequent analysis and comparison.

[0043] For microseismic signal data, wavelet transform is used for denoising. Wavelet transform can analyze signals in both time and frequency domains and effectively separate noise and useful signals. By selecting appropriate wavelet basis functions and decomposition layers, fine denoising of microseismic signals can be achieved and key feature information can be retained. Gas concentration data is processed by mean filtering to smooth data fluctuations. Mean filtering takes the average value of data in the sliding window, which can effectively suppress random noise and highlight the overall change trend of the data. The temperature data is processed by the Kalman filter algorithm, which continuously optimizes the estimated value through two steps of prediction and update to improve the accuracy and reliability of temperature data.

[0044] The processed multi-source data is time synchronized to ensure the time consistency of the data and provide a basis for subsequent analysis. Time synchronization uses the timestamp alignment method to correspond the data of different sensors to a unified time series. Outlier detection is performed on the synchronized data, and statistical methods such as the 3σ criterion are used to identify and mark abnormal data points. For missing or abnormal data points, interpolation processing methods such as linear interpolation or spline interpolation are used to complete them to ensure the continuity and integrity of the data.

[0045] The feature extraction step extracts key features from multi-source data, including time domain features (such as mean, variance) and frequency domain features (such as main frequency, energy distribution). The extracted feature vector set is subjected to dimensionality reduction processing, and methods such as principal component analysis (PCA) or linear discriminant analysis (LDA) are used to reduce data dimensions and improve computational efficiency. Data fusion processing comprehensively considers feature data from different sources, and uses Dempster-Shafer evidence theory or fuzzy comprehensive evaluation method to obtain fused feature data. Normalization processing unifies the fused feature data to the [0,1] interval to eliminate the dimension effect. Finally, the normalized fused feature data is further reduced in dimension and feature extracted through the principal component analysis algorithm to obtain the final multidimensional data set.

[0046] For example, in a coal mine working face, gas pressure, coal seam stress, microseismic, gas concentration and temperature sensors are arranged. The original data of gas pressure fluctuates in the range of 0-10MPa. After median filtering, the sudden interference is eliminated and the data curve is smoother. The original range of coal seam stress data is 0-50MPa. After standardization, it is converted into dimensionless data in the interval [-1,1]. The original data of microseismic signal contains high-frequency noise. After wavelet transform denoising, the low-frequency characteristics are more obvious, and microseismic events are effectively identified. The original range of gas concentration data is 0-5%. After mean filtering, short-term fluctuations are reduced and long-term trends are clearer. The original range of temperature data is 20-40℃. After Kalman filtering, the measurement error is significantly reduced and the temperature change curve is smoother. After time synchronization, outlier detection and interpolation processing, a complete multi-source data set is obtained. The feature extraction step extracts 10 features from each data, a total of 50 features. Dimensionality reduction processing reduces the number of features to 20. After data fusion and normalization, fused feature data between 0 and 1 were obtained. Finally, principal component analysis reduced the data to 5 principal components, which contained 85% of the information of the original data, forming the final multidimensional data set, providing a high-quality data foundation for subsequent coal and gas outburst risk assessment and early warning.

[0047] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0048] (1) Divide the multidimensional data set into time windows to obtain time series subsets, and perform autocorrelation analysis on the time series subsets to obtain autocorrelation feature data;

[0049] (2) Perform trend decomposition on the autocorrelation characteristic data to obtain trend component data, and perform seasonal analysis on the trend component data to obtain seasonal characteristic data;

[0050] (3) The seasonal characteristic data is processed by the long short-term memory network algorithm to obtain the time series prediction results, and the residual analysis of the time series prediction results is performed to obtain the residual characteristic data;

[0051] (4) Perform frequency domain conversion on the residual feature data to obtain frequency domain feature data, and process the frequency domain feature data using the support vector regression algorithm to obtain regression prediction results;

[0052] (5) The time series prediction results and regression prediction results are weighted and fused to obtain fused prediction data, which are then processed through Bayesian inference to obtain coal seam gas outburst hazard prediction data.

[0053] Specifically, the multidimensional data set is divided into time windows, and the continuous data stream is divided into time series subsets of fixed length, so that the changing characteristics of the data at different time scales can be captured. The length of the time window is set according to the actual situation of the coal mine, and can usually be selected as 4 hours, 8 hours or 12 hours. Autocorrelation analysis is performed on each time series subset, and the autocorrelation coefficients of different time lags are calculated to obtain autocorrelation characteristic data. Autocorrelation analysis can reveal the inherent periodicity and dependence of the data, providing an important basis for subsequent predictions. Next, the autocorrelation characteristic data is trend decomposed to decompose the time series into trend components, seasonal components and random components. Trend decomposition uses the STL (Seasonal and Trend decomposition using Loess) method, which can effectively handle nonlinear trends and complex seasonal patterns. Seasonal analysis is performed on the decomposed trend component data to identify the periodic change pattern in the data and obtain seasonal characteristic data. Seasonal analysis includes periodicity identification and intensity assessment, which helps to understand the periodic change laws of parameters such as gas pressure and coal seam stress.

[0054] Then, the seasonal characteristic data is processed by the long short-term memory network (LSTM) algorithm. LSTM is a special recurrent neural network that can effectively capture long-term dependencies and is particularly suitable for processing time series data. The input layer of the LSTM network receives seasonal characteristic data, the hidden layer contains multiple LSTM units, and the output layer gives the time series prediction results. The training process of LSTM includes forward propagation and backpropagation, and the prediction error is minimized by adjusting the network weights. Residual analysis is performed on the time series prediction results of LSTM to calculate the difference between the predicted value and the actual value to obtain the residual characteristic data. Residual analysis helps to evaluate the prediction accuracy of the model and provide a basis for subsequent optimization. The residual characteristic data is converted into the frequency domain, and the fast Fourier transform (FFT) is used to convert the time domain signal into the frequency domain representation to obtain the frequency domain characteristic data. Frequency domain analysis can reveal the implicit periodic components and frequency characteristics in the data, providing a new perspective for identifying potential abnormal patterns. The frequency domain characteristic data is processed by the support vector regression (SVR) algorithm, a nonlinear regression model is constructed, and the regression prediction results are obtained. The SVR algorithm maps the input space to a high-dimensional feature space, achieving nonlinear regression while ensuring generalization ability, and is particularly suitable for processing high-dimensional feature data.

[0055] Finally, the time series prediction results of LSTM and the regression prediction results of SVR are weighted and fused, and the advantages of the two models are comprehensively considered to obtain the fused prediction data. The weight coefficient can be verified and determined by historical data to minimize the error of the fused prediction. The fused prediction data is processed by Bayesian inference, and prior knowledge and uncertainty quantification are introduced to obtain the final coal seam gas outburst hazard prediction data. Bayesian inference can combine expert experience and historical data to give the probability distribution of outburst hazard and provide more comprehensive information support for decision-making.

[0056] For example, in a coal mine working face, multi-source monitoring data were collected for 30 consecutive days, including parameters such as gas pressure, coal seam stress, microseismic signals, gas concentration and temperature. First, the data was divided into 8-hour time windows to obtain 90 time series subsets. Autocorrelation analysis of the gas pressure data showed that there was a significant positive correlation at a 24-hour lag, with an autocorrelation coefficient of 0.75, indicating that the gas pressure had obvious daily variation characteristics. After trend decomposition, an upward trend in gas pressure was identified, with a monthly growth rate of about 2%. Seasonal analysis showed that there were periodic differences in gas pressure on weekdays and weekends, with the average pressure on weekends being 3% higher than that on weekdays. The processed data was input into the LSTM network, and the network structure consisted of an input layer, two LSTM hidden layers containing 64 neurons, and an output layer. After 1,000 rounds of iterative training, the root mean square error (RMSE) of the model on the validation set dropped to 0.15 MPa.

[0057] The residual analysis of the LSTM prediction results showed that the standard deviation of the residuals was 0.08MPa, indicating that the prediction accuracy was high. The residual data was transformed by FFT, and two significant frequency peaks were found at 0.04Hz and 0.08Hz, corresponding to periodic changes twice and four times a day, respectively. The SVR model uses the RBF kernel function, and optimizes the parameters C=10 and gamma=0.01 through grid search, and the coefficient of determination R² on the test set reaches 0.92. The prediction results of LSTM and SVR are fused with weights of 0.6 and 0.4 respectively, and the prediction error after fusion is reduced by 15% compared with that of a single model. Finally, through Bayesian inference, combined with historical outburst accident data and expert experience, it is concluded that the probability of coal and gas outbursts in the next 24 hours is 3.5%. Compared with the conventional threshold method, the warning accuracy is increased by 20% and the missed reporting rate is reduced by 30%.

[0058] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0059] (1) The coal seam gas outburst hazard prediction data is divided into risk levels to obtain risk level data, and the risk level data is subjected to spatial correlation analysis to obtain spatial correlation characteristics;

[0060] (2) Based on the spatial correlation characteristics, spatial interpolation processing is performed on the multidimensional data set to obtain interpolated grid data, and risk weights are assigned to the interpolated grid data to obtain weighted risk data;

[0061] (3) Perform spatial cluster analysis on the weighted risk data to obtain risk clustering results, and identify the boundaries of the risk clustering results to obtain risk area boundary data;

[0062] (4) Perform morphological processing on the risk area boundary data to obtain an optimized risk area contour, and calculate the risk intensity of the optimized risk area contour to obtain risk intensity distribution data;

[0063] (5) The risk intensity distribution data is processed using the Kriging interpolation algorithm to obtain the real-time risk distribution data of the target area.

[0064] Specifically, the coal seam gas outburst hazard prediction data is subjected to risk level classification, spatial interpolation processing, risk weight allocation, spatial cluster analysis, morphological processing, and Kriging interpolation, aiming to form accurate real-time risk distribution data to enhance the monitoring and early warning capabilities of coal mine gas disasters. Risk level data is obtained by classifying the coal seam gas outburst hazard prediction data into risk levels. The risk level classification is based on specific evaluation indicators, and the predicted data is divided into different risk levels, such as high, medium, and low, according to parameters such as gas concentration and pressure. This level classification facilitates subsequent risk assessment and analysis, can effectively highlight the severity of risk points, and guide the next step of data processing. For the risk level data, spatial correlation analysis is further performed to obtain spatial correlation characteristics. Spatial correlation analysis calculates the similarity and correlation of data between different measuring points to reveal the spatial distribution law of data. For example, the autocorrelation coefficient is calculated using the following formula:

[0065] ;

[0066] in, Indicates The measurement points and The spatial correlation between the measurement points, and Respectively represent and The measuring point is The gas concentration value at the time, and is the mean value of each measurement point, is the number of measuring points. By calculating the correlation between each measuring point, it is possible to determine which areas in space have similar gas concentration change trends, thereby generating spatial correlation features. Next, based on the spatial correlation features obtained above, the multidimensional data set is spatially interpolated to generate interpolated grid data. Spatial interpolation is to estimate the data values ​​of unknown points between known measuring points, thereby generating a continuous spatial data field. In this scheme, the commonly used inverse distance weighted (IDW) algorithm is used for interpolation calculation, and the formula is as follows:

[0067] ;

[0068] in, Indicates location The interpolation of It is The data value of the known measurement point, It's location To The distance between the measuring points, is a weight index used to adjust the effect of distance on the interpolation result. Through this formula, the data originally scattered at multiple measuring points can be converted into continuous grid data, so as to conduct a more detailed risk assessment of the target area. After the interpolation grid data is generated, it is necessary to assign risk weights to it to obtain weighted risk data. The risk weight allocation is based on a comprehensive analysis of the predicted data and risk level data of each interpolation point, and different risk weights are assigned to each grid point to highlight high-risk areas. By calculating the weight of each point, the generated weighted risk data can more accurately reflect the gas risk distribution in different areas, which is convenient for subsequent risk clustering analysis.

[0069] Afterwards, the weighted risk data is subjected to spatial clustering analysis to obtain the risk clustering results, and the risk clustering results are subjected to boundary identification to generate risk area boundary data. Spatial clustering analysis forms relatively independent risk areas by aggregating grid points with similar risk characteristics. The K-means algorithm is often used in this step, and its goal is to minimize the sum of the square distances between the cluster center and the data point. Through such clustering processing, the concentrated areas of gas outburst risk can be more clearly identified. Further boundary identification is performed on these clustering results, and the outer contour of the risk area is extracted to generate risk area boundary data. For the identified risk area boundary data, morphological processing methods are used for optimization to obtain the optimized risk area contour. Morphological processing is an image processing method that is usually used to smooth boundaries, fill small holes, etc. In this scheme, morphological operations such as dilation and erosion help to optimize the coherence of regional boundaries, reduce noise interference, and make the shape of the risk area more regular and close to the actual gas distribution. Afterwards, the risk intensity calculation is performed on the optimized risk area contour to evaluate the degree of gas outburst risk in each area.

[0070] Finally, the risk intensity distribution data is processed by the Kriging interpolation algorithm to obtain the real-time risk distribution data of the target area. Kriging interpolation is an interpolation method in geostatistics. Its characteristic is that it can consider the spatial autocorrelation between measurement points in the estimation process, thereby improving the accuracy of the interpolation results. The mathematical expression of Kriging interpolation is:

[0071] ;

[0072] in, Indicates location The estimated value at is the weight coefficient, obtained by solving the Kriging equations, For measuring point Observed values. Kriging interpolation ensures optimal interpolation in space by solving the variogram, making the interpolation result more accurate. The real-time risk distribution data finally generated can intuitively show the spatial distribution of gas outburst risk in the target area and guide the implementation of early warning and response measures. For example, suppose that several gas concentration sensors are distributed in a certain mine area. After real-time monitoring, gas concentration data of multiple sensor points are collected. Through the spatial correlation analysis in the above steps, it is found that the gas concentration change trends of sensor A and sensor B are similar, and their correlation coefficient is 0.85, while the correlation with sensor C is lower, only 0.2. This shows that the area where sensors A and B are located has a strong gas risk association in space. Based on this correlation, the IDW algorithm is applied to interpolate this area to generate grid data of gas concentration. Assume that after interpolation, the grid data shows that the gas concentration reaches 2.8% at a certain location between sensors A and B, which is much higher than the average concentration in other areas. After dividing the risk level of this area and assigning a higher weight, the K-means clustering algorithm was used to analyze and identify that this high-concentration area and several surrounding grid points formed an obvious high-risk area. After that, the boundary of the high-risk area was smoothed by morphological operations, and the real-time risk distribution data of the area was obtained through Kriging interpolation. The results showed that the risk intensity of the area was 3.2%, indicating a high-risk trend of gas accumulation. This data will trigger the early warning system in time and guide the mine to take targeted gas emission measures to reduce the risk of gas outbursts.

[0073] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0074] (1) Extract statistical features from real-time risk distribution data to obtain risk statistical features, and perform normality tests on risk statistical features to obtain distribution characteristic data;

[0075] (2) Based on the distribution characteristic data, anomaly detection is performed on the real-time risk distribution data to obtain an abnormal risk point set, and density estimation is performed on the abnormal risk point set to obtain the risk density distribution;

[0076] (3) Calculate the threshold of risk density distribution to obtain a multidimensional threshold set, and dynamically adjust the multidimensional threshold set to obtain adaptive threshold data;

[0077] (4) Based on the adaptive threshold data, the real-time risk distribution data is segmented by threshold to obtain the risk area division results, and the connectivity analysis of the risk area division results is performed to obtain the high-risk connected areas;

[0078] (5) Conduct spatiotemporal evolution analysis on high-risk connected areas to obtain preliminary warning data.

[0079] Specifically, statistical feature extraction is performed on the real-time risk distribution data to obtain risk statistical features. The purpose of statistical feature extraction is to analyze the main characteristics of real-time data, including mean, standard deviation, skewness, kurtosis and other indicators. Specifically, assuming that the real-time data of gas concentration in the monitoring area is , these indicators can be calculated by the following formula:

[0080] Mean calculation:

[0081] ;

[0082] in, represents the mean of the data set, It is Gas concentration data of each measuring point

[0083] Standard deviation calculation:

[0084] ;

[0085] It is used to describe the degree of discreteness of gas concentration data, which helps to identify the volatility of the data. After extracting these statistical features, a normality test is performed on them to obtain distribution characteristic data. The normality test is to determine whether the real-time risk distribution data obeys the normal distribution. Commonly used test methods include the Shapiro-Wilk test or the Kolmogorov-Smirnov test. Taking the Shapiro-Wilk test as an example, its test result will generate a p-value. When the p-value is greater than the set significance level, it can be considered that the data approximately obeys the normal distribution. The results of the normality test will directly affect the selection of subsequent anomaly detection methods. Based on the distribution characteristic data, anomaly detection is performed on the real-time risk distribution data to obtain an abnormal risk point set. The goal of anomaly detection is to identify measurement points where the gas concentration deviates significantly from the normal range. A standard deviation-based method is usually used to set a threshold. ,in is a constant used to adjust the abnormal range. If the data of a certain measurement point exceeds this range, it is considered an abnormal point. Assuming that the mean value of a certain monitoring point is 2.5% and the standard deviation is 0.3%, set ( =3), the range of anomaly detection is [1.6%, 3.4%], and gas concentration data outside this range will be identified as abnormal data. These abnormal data points are grouped into an abnormal risk point set, and density estimation is performed on them.

[0086] For the abnormal risk point set, density estimation is performed using the kernel density estimation (KDE) method to obtain the risk density distribution. Kernel density estimation is a common non-parametric estimation method used to estimate the probability density function of a random variable. Its formula is as follows:

[0087] ;

[0088] in, For location The density estimate at , is the sample size, is the smoothing parameter (bandwidth), is the kernel function. Through KDE, we can draw a density distribution map of risk points in space, so as to intuitively show the distribution of high-density risk points. After obtaining the risk density distribution, the threshold calculation is performed to generate a multidimensional threshold set, and the multidimensional threshold set is dynamically adjusted to obtain adaptive threshold data. The purpose of threshold calculation is to determine the boundary of the high-risk area based on the density estimation results. The quantile method can be used, such as selecting data points with density values ​​in the top 5% as high-risk thresholds. At the same time, since the monitoring data changes dynamically, the threshold needs to be adjusted in real time. The dynamic adjustment adopts an adaptive filtering method. By combining the threshold of the previous moment and the current risk density data, the threshold change trend is adjusted to adapt to the dynamic changes of the risk. This ensures the flexibility of the threshold so that it can cope with fluctuations in gas concentration.

[0089] Based on the adaptive threshold data, the real-time risk distribution data is threshold segmented to obtain the risk area division results, and the risk area division results are subjected to connectivity analysis to obtain high-risk connected areas. Threshold segmentation refers to dividing areas with risk density higher than the threshold into high-risk areas and areas with risk density lower than the threshold into low-risk areas. Next, adjacent high-risk points are identified through connectivity analysis and merged into connected risk areas. Connectivity analysis usually uses algorithms such as breadth-first search (BFS) or depth-first search (DFS) to identify grid points with adjacent relationships. After obtaining the high-risk connected areas, a spatiotemporal evolution analysis is performed on them to obtain preliminary warning data. The goal of spatiotemporal evolution analysis is to track the changing trends of high-risk areas in time and space to predict the direction and speed of risk diffusion. This analysis combines time series data and spatial location data, and uses regression models or ARIMA models to predict future trends in gas concentration, thereby generating preliminary warning data.

[0090] For example, suppose that several gas concentration sensors are arranged at monitoring points in a coal mine area, and these sensors collect data once a minute. At a certain moment, the concentration values ​​of sensors A, B, and C are 3.1%, 2.8%, and 3.5%, respectively, while the mean value of the area is 2.5% and the standard deviation is 0.3%. Based on the set anomaly detection threshold [1.6%, 3.4%], the data of sensor C is detected as an outlier, forming an abnormal risk point set. Then, the density of these outliers is estimated by kernel density estimation. Assume that the estimation result shows that a high-density area is formed between sensors A and C, and its estimated value is 0.15. This result shows that the density of abnormal concentration values ​​in this area is high. Based on this density distribution, the high-risk threshold is further calculated. Assuming that the 95th percentile of the density value is selected as 0.14, the area with a density higher than 0.14 is considered a high-risk area. Due to the volatility of real-time monitoring data, the threshold is adjusted by adaptive filtering, and the adjusted adaptive threshold is 0.135. Then, the risk density distribution is threshold segmented, and the results show the distribution range of the high-risk area. Next, connectivity analysis was used to connect adjacent high-density areas to form a larger high-risk area. Finally, through spatiotemporal evolution analysis, it was predicted that the high-risk area might spread 100 meters to the southeast in the next 10 minutes, and the concentration would increase significantly, thus generating early warning data to guide on-site personnel to take timely gas emission measures to avoid potential gas outburst risks.

[0091] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0092] (1) Perform feature decomposition on the preliminary warning data to obtain a multimodal feature set, and perform correlation analysis on the multimodal feature set to obtain a feature correlation matrix;

[0093] (2) Based on the feature correlation matrix, feature selection is performed on the multimodal feature set to obtain a key feature subset, and the key feature subset is normalized to obtain standardized feature data;

[0094] (3) Perform multimodal alignment on the standardized feature data to obtain an aligned feature sequence, and process the aligned feature sequence through a deep belief network to obtain a fused feature vector;

[0095] (4) Calculate the confidence of the fused feature vector to obtain feature confidence data, and weight the fused feature vector based on the feature confidence data to obtain a weighted fused feature;

[0096] (5) The weighted fusion features are processed through the fuzzy inference system to obtain target warning data.

[0097] Specifically, the preliminary warning data is feature decomposed to obtain a multimodal feature set. A multimodal feature set refers to feature information extracted from multiple data sources, including different types of data such as gas concentration, temperature, humidity, and mine pressure. In the coal mine monitoring environment, these multimodal features jointly reflect the state of the mine environment. The feature decomposition process will classify and extract these data to form a multimodal feature set to ensure the diversity of data sources. The introduction of multimodal features allows the analysis to take into account more dimensional information, which helps to identify risks more comprehensively in subsequent analysis. Next, a correlation analysis is performed on the multimodal feature set to generate a feature correlation matrix. The feature correlation matrix is ​​used to show the linear relationship between different features. By analyzing these relationships, some redundant or highly correlated features can be identified. Correlation analysis can not only reduce the number of features, but also remove features that have little effect on the prediction effect, thereby providing a basis for subsequent feature selection. The purpose of this is to ensure that the features finally retained can play the greatest role in the prediction, thereby improving the accuracy of the warning.

[0098] Based on the feature correlation matrix, feature selection is performed on the multimodal feature set to obtain a key feature subset. Feature selection is a screening process that aims to select the features that have the greatest impact on coal mine risk warning from a large number of features. In the coal mine environment, gas concentration, temperature, and pressure may be important features that affect the outburst phenomenon. Through feature selection methods, such as methods based on correlation analysis, these key features can be determined and redundant features can be eliminated, so that the data dimension can be simplified while retaining the most important information. The advantage of this treatment is that it reduces the redundancy of the amount of data in the subsequent processing process, improves the efficiency of calculation, and also improves the interpretability of the model. After obtaining the key feature subset, it is normalized to obtain standardized feature data. The purpose of normalization is to eliminate the impact caused by different dimensions between different features and make them analyzed on the same scale. For example, the range of gas concentration may be between 0.1% and 5%, while the range of mine pressure may be between several atmospheres. If these features are used directly for analysis, the results will be affected by different scales. Through normalization, the data of different features can be adjusted to the same range, so as to ensure that subsequent data processing can be carried out more smoothly.

[0099] Next, the standardized feature data is multimodally aligned to obtain an aligned feature sequence. Multimodal alignment is to synchronize features from different data sources in time or space to ensure that the feature data is consistent at the same time step. For example, a gas concentration sensor may collect data once a minute, while a pressure sensor collects data once every five minutes. In order to analyze these data at the same time point, the data of the two need to be interpolated or otherwise aligned to obtain a feature sequence with a consistent time step. This ensures that subsequent data fusion will not cause information loss or distortion due to different data sampling frequencies. After aligning the feature sequence, it is processed using a deep belief network (DBN) to obtain a fused feature vector. A deep belief network is a multi-layer deep learning model that can automatically learn high-dimensional feature representations from data. It is trained layer by layer to extract deeper correlations from multimodal features and ultimately form a comprehensive fused feature vector. For coal and gas outburst warning, DBN can extract more critical potential risk patterns through a comprehensive analysis of multiple features such as gas concentration, temperature, and pressure. The generation of fused feature vectors makes the subsequent risk analysis more comprehensive and accurate.

[0100] After obtaining the fused feature vector, the confidence calculation is performed on it to evaluate the contribution of each feature in the early warning, and further weighted processing is performed according to the confidence. The confidence calculation obtains the importance of each feature by analyzing the relationship between the feature and the risk label. For example, through models such as logistic regression, the importance of a feature to risk prediction can be evaluated and used as the basis for confidence. Weighted processing of fused features based on confidence data can highlight those features that contribute more to risk prediction, so that the final early warning result is more accurate. Finally, the weighted fused features are processed by the fuzzy inference system to obtain the target early warning data. The fuzzy inference system is a method suitable for processing uncertain information. It processes the input weighted features through preset fuzzy rules to generate early warning results that conform to the actual situation. In coal mine monitoring, these fuzzy rules can be defined based on experience and historical data, such as "If the gas concentration is high and the pressure increases, the risk of outburst is higher". By analyzing the weighted fusion features through the fuzzy inference system, the target early warning data finally obtained can intuitively display the potential risks in the mine environment, thereby providing a basis for decision-making.

[0101] For example, suppose that during the monitoring process of a mine, the gas concentration gradually increases from 0.8% to 2.5%, while the mine pressure also increases from 2.1 atmospheres to 3.0 atmospheres. Through feature decomposition, these data are extracted as a multimodal feature set. After correlation analysis, it is identified that there is a strong correlation between gas concentration and pressure, so it is selected as a key feature. Then, these features are normalized, and the data range of gas concentration is adjusted to between 0 and 1, and the pressure data is also normalized. After alignment, feature sequences synchronized by minutes are generated. The DBN model processes these aligned feature sequences and identifies potential outburst risk patterns in the case of rapid increase in gas concentration and pressure. Through confidence calculation, the influence of gas concentration is set to 70%, while the influence of pressure is 30%, so the fused feature vector will give a higher weight to gas concentration when calculating. Finally, based on these weighted features and combined with preset rules, the fuzzy inference system determines that the risk of gas outburst is high in the current environment and generates corresponding warning signals.

[0102] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0103] (1) Divide the target warning data into attributes to obtain a decision attribute set, and calculate the information gain of the decision attribute set to obtain the attribute importance ranking;

[0104] (2) Based on the attribute importance ranking, a decision tree is constructed for the target warning data to obtain an initial decision tree structure, and the initial decision tree structure is pruned to obtain an optimized decision tree;

[0105] (3) Perform path analysis on the optimized decision tree to obtain a decision rule set, and perform conflict detection on the decision rule set to obtain a consistent rule subset;

[0106] (4) Based on the consistent rule subset, the target warning data is matched with rules to obtain a matching result set, and the confidence of the matching result set is evaluated to obtain a high-confidence decision path;

[0107] (5) The high-confidence decision path is processed through case-based reasoning to obtain the target early warning plan.

[0108] Specifically, the target warning data is attributed and a decision attribute set is obtained. The target warning data usually includes multiple data types, such as gas concentration, rock formation pressure, temperature and geological parameters. In the attribute division stage, these multidimensional data are divided into their own decision attributes for subsequent processing. Next, the information gain of these decision attribute sets needs to be calculated. Information gain is a measure of the amount of information provided by an attribute in the classification of the target variable. In this stage, the importance of the attribute is determined and sorted by calculating the ability of each attribute to distinguish certain categories in the warning data. This step helps to screen out the key attributes that have greater predictive power for coal and gas outburst events in the warning model. Based on the importance ranking of the attributes, a decision tree is constructed for the target warning data. A decision tree is a tree structure in which each node represents an attribute, each branch represents a possible value on the attribute, and each leaf node represents the decision result. When constructing the initial decision tree, attributes are gradually selected as division nodes according to the size of the information gain, thereby obtaining a preliminary tree structure. However, the initial decision tree often has an overfitting problem, which means that the model may perform well on the training data but perform poorly when processing new data. Therefore, it is necessary to prune the initial decision tree. Pruning removes some redundant nodes or branches and simplifies the tree structure to obtain an optimized decision tree, thereby improving its generalization ability in practical applications.

[0109] After obtaining the optimized decision tree, it is necessary to perform path analysis on it to extract a decision rule set that can explain the prediction process. Path analysis refers to extracting rules from the root node of the decision tree to each leaf node. These rules describe how to make warning decisions under specific conditions. Due to the complexity and diversity of data, the extracted rule set may contain conflicting rules. In order to ensure the reliability of the warning model, it is necessary to perform conflict detection on the rule set. By detecting and excluding inconsistent or contradictory rules, a consistent rule subset is obtained. This subset only retains those rules that are logical and accurate in most cases. After obtaining the consistent rule subset, the target warning data is matched based on these rules. Through rule matching, the newly monitored data can be compared with each rule in the rule set to obtain a matching result set. The matching result set reflects the matching between each input data and the rule. However, different matching results have different importance for decision making, so it is necessary to perform confidence assessment on the matching result set. Confidence assessment can help identify which rule matching results are more reliable, and thus obtain high-confidence decision paths. These high-confidence paths can effectively guide the next decision process.

[0110] Based on the high-confidence decision path, the decision path is further processed through case-based reasoning (CBR). Case-based reasoning is a method of solving new problems by storing, retrieving, matching and adjusting historical cases. At this stage, the final target warning plan is obtained by comparing with previous similar cases and combining the results of the high-confidence decision path. Case-based reasoning can make reasonable inferences and decisions on the current situation based on the experience of handling similar historical events, thereby improving the accuracy of the warning plan.

[0111] For example, in a mine, real-time monitoring shows that the gas concentration gradually increases and the rock formation pressure fluctuates greatly. Through attribute division, these monitoring data are classified into decision attribute sets, and information gain calculation is performed. It is found that the change in gas concentration has the greatest impact on coal and gas outbursts. Therefore, gas concentration is preferred as the division node when building the decision tree. After pruning, a simplified decision tree is obtained. The rule set extracted by path analysis contains the rule that "there is a high risk when the gas concentration exceeds the critical value and the rock formation pressure changes rapidly". In practical applications, the monitoring data matches the rule with a high confidence level. Therefore, through case reasoning and combined with the gas outburst phenomenon under similar conditions in history, the target early warning plan of "immediate measures need to be taken for evacuation and gas discharge" is finally obtained. Through this early warning method, the potential losses caused by coal and gas emergencies can be effectively reduced.

[0112] The above describes the method for dynamic monitoring and intelligent early warning of coal and gas outburst in the embodiment of the present application. The following describes the device for dynamic monitoring and intelligent early warning of coal and gas outburst in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the coal and gas outburst dynamic monitoring and intelligent early warning device includes:

[0113] The processing module 201 is used to collect and pre-process the multi-source sensor data set in the target area to obtain a multi-dimensional data set;

[0114] A prediction module 202 is used to perform multivariate time series analysis on the multidimensional data set to obtain coal seam gas outburst hazard prediction data;

[0115] An interpolation module 203 is used to perform dynamic risk assessment and spatial interpolation processing on the multidimensional data set based on the coal seam gas outburst hazard prediction data to obtain real-time risk distribution data of the target area;

[0116] An analysis module 204 is used to perform multi-dimensional threshold analysis on the real-time risk distribution data to obtain preliminary warning data;

[0117] A fusion module 205 is used to perform multi-modal information fusion on the preliminary warning data to obtain target warning data;

[0118] The decision module 206 is used to perform intelligent decision tree analysis on the target warning data to obtain a target warning solution.

[0119] Through the cooperation of the above components and the calculation of information gain, the role of different attributes in predicting gas outburst can be clarified, and the attributes can be sorted based on this to effectively screen out the most influential decision factors. This process greatly improves the training efficiency of the early warning model, and also improves the prediction accuracy of the model, so that the model can identify high-risk factors more quickly. The technical characteristics of decision tree construction and pruning based on attribute importance sorting avoid the overfitting phenomenon that is prone to occur in traditional decision tree models. Through the construction and pruning of the initial decision tree, the scheme can obtain a more streamlined and more generalized decision tree model. This means that the model not only performs well on training data, but also maintains stable performance in actual monitoring, thereby effectively coping with complex and changeable data inputs in coal mine environments. The pruned decision tree model reduces unnecessary nodes and branches, improves the prediction efficiency of the model, and also ensures that scientific and reasonable early warning judgments can still be made under new data. Through path analysis, the complex decision process can be presented in the form of rules, which allows early warning personnel to intuitively understand the logic behind each early warning conclusion. Rule conflict detection and extraction of consistent rule subsets further ensure the logical consistency and decision reliability of the model, avoid misjudgment caused by rule contradictions, and thus improve the practical operability of the early warning system. The technical features of rule matching and confidence evaluation based on consistent rule subsets enable the model to quickly find the most suitable rules and evaluate their reliability based on real-time data in practical applications, and then focus on high-confidence decision paths. This processing method can more effectively identify high-risk events and take targeted preventive measures when facing diverse and complex field data, thereby improving the timeliness of early warning response. The technical features of processing high-confidence decision paths through case reasoning can make full use of the experience of similar historical cases, so that the model not only relies on data rules, but also optimizes the early warning plan in combination with actual experience. In the early warning of coal and gas outburst events, this case-based reasoning method enables the system to quickly give a reasonable response plan when facing uncertain emergencies, thereby greatly improving the accuracy and scientificity of the early warning plan.

[0120] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for dynamic monitoring and intelligent early warning of coal and gas outburst, characterized in that: The coal and gas outburst dynamic monitoring and intelligent early warning method comprises: Collecting and preprocessing multi-source sensor data set in the target area to obtain a multi-dimensional data set; Performing multivariate time series analysis on the multidimensional data set to obtain coal seam gas outburst hazard prediction data; Based on the coal seam gas outburst hazard prediction data, dynamic risk assessment and spatial interpolation processing are performed on the multidimensional data set to obtain real-time risk distribution data of the target area; Performing multidimensional threshold analysis on the real-time risk distribution data to obtain preliminary warning data, including: extracting statistical features from the real-time risk distribution data to obtain risk statistical features, and performing normality test on the risk statistical features to obtain distribution characteristic data; based on the distribution characteristic data, performing anomaly detection on the real-time risk distribution data to obtain an abnormal risk point set, and performing density estimation on the abnormal risk point set to obtain a risk density distribution; performing threshold calculation on the risk density distribution to obtain a multidimensional threshold set, and dynamically adjusting the multidimensional threshold set to obtain adaptive threshold data; based on the adaptive threshold data, performing threshold segmentation on the real-time risk distribution data to obtain a risk area division result, and performing connectivity analysis on the risk area division result to obtain a high-risk connected area; performing spatiotemporal evolution analysis on the high-risk connected area to obtain the preliminary warning data; Performing multimodal information fusion on the preliminary warning data to obtain target warning data, including: performing feature decomposition on the preliminary warning data to obtain a multimodal feature set, and performing correlation analysis on the multimodal feature set to obtain a feature correlation matrix; performing feature selection on the multimodal feature set based on the feature correlation matrix to obtain a key feature subset, and performing normalization processing on the key feature subset to obtain standardized feature data; performing multimodal alignment on the standardized feature data to obtain an aligned feature sequence, and processing the aligned feature sequence through a deep belief network to obtain a fused feature vector; performing confidence calculation on the fused feature vector to obtain feature confidence data, and weighting the fused feature vector based on the feature confidence data to obtain a weighted fused feature; processing the weighted fused feature through a fuzzy inference system to obtain the target warning data; An intelligent decision tree analysis is performed on the target warning data to obtain a target warning plan.

2. The method for dynamic monitoring and intelligent early warning of coal and gas outburst according to claim 1 is characterized in that: The multi-source sensor data set in the target area is collected and preprocessed to obtain a multi-dimensional data set, including: The gas pressure data in the target area is collected in real time to obtain raw gas pressure data, and the raw gas pressure data is filtered to obtain filtered gas pressure data; Collecting coal seam stress data in the target area in real time to obtain original coal seam stress data, and performing standardization processing on the original coal seam stress data to obtain standardized coal seam stress data; Real-time acquisition of microseismic signal data in the target area to obtain original microseismic signal data, and denoising the original microseismic signal data by wavelet transform to obtain denoised microseismic signal data; The gas concentration data in the target area is collected in real time to obtain raw gas concentration data, and the raw gas concentration data is processed by mean filtering to obtain smoothed gas concentration data; Collecting temperature data in the target area in real time to obtain original temperature data, and processing the original temperature data through a Kalman filter algorithm to obtain corrected temperature data; Performing time synchronization processing on the filtered gas pressure data, the standardized coal seam stress data, the denoised microseismic signal data, the smoothed gas concentration data and the corrected temperature data to obtain time-synchronized multi-source data; Performing outlier detection on the time-synchronized multi-source data to obtain candidate multi-source data, and performing interpolation processing on the candidate multi-source data to obtain completed multi-source data; Performing feature extraction on the completed multi-source data to obtain a feature vector set, and performing dimensionality reduction processing on the feature vector set to obtain feature data after dimensionality reduction; Performing data fusion processing on the feature data after dimension reduction to obtain fused feature data, and performing normalization processing on the fused feature data to obtain normalized fused feature data; The normalized fused feature data is processed by a principal component analysis algorithm to obtain the multidimensional data set.

3. The method for dynamic monitoring and intelligent early warning of coal and gas outburst according to claim 1 is characterized in that: The multivariate time series analysis is performed on the multidimensional data set to obtain coal seam gas outburst hazard prediction data, including: Dividing the multidimensional data set into time windows to obtain time series subsets, and performing autocorrelation analysis on the time series subsets to obtain autocorrelation feature data; Performing trend decomposition on the autocorrelation characteristic data to obtain trend component data, and performing seasonal analysis on the trend component data to obtain seasonal characteristic data; The seasonal characteristic data is processed by a long short-term memory network algorithm to obtain a time series prediction result, and residual analysis is performed on the time series prediction result to obtain residual characteristic data; Performing frequency domain conversion on the residual feature data to obtain frequency domain feature data, and processing the frequency domain feature data using a support vector regression algorithm to obtain a regression prediction result; The time series prediction results and the regression prediction results are weightedly fused to obtain fused prediction data, and the fused prediction data are processed through Bayesian inference to obtain the coal seam gas outburst hazard prediction data.

4. The method for dynamic monitoring and intelligent early warning of coal and gas outburst according to claim 1 is characterized in that: Based on the coal seam gas outburst hazard prediction data, dynamic risk assessment and spatial interpolation processing are performed on the multidimensional data set to obtain real-time risk distribution data of the target area, including: The coal seam gas outburst hazard prediction data is classified into risk levels to obtain risk level data, and the risk level data is analyzed for spatial correlation to obtain spatial correlation characteristics; Based on the spatial correlation feature, spatial interpolation processing is performed on the multidimensional data set to obtain interpolation grid data, and risk weight allocation is performed on the interpolation grid data to obtain weighted risk data; Performing spatial cluster analysis on the weighted risk data to obtain risk clustering results, and performing boundary identification on the risk clustering results to obtain risk area boundary data; Performing morphological processing on the risk area boundary data to obtain an optimized risk area contour, and performing risk intensity calculation on the optimized risk area contour to obtain risk intensity distribution data; The risk intensity distribution data is processed by a Kriging interpolation algorithm to obtain real-time risk distribution data of the target area.

5. The method for dynamic monitoring and intelligent early warning of coal and gas outburst according to claim 1 is characterized in that: The intelligent decision tree analysis is performed on the target warning data to obtain a target warning solution, including: Performing attribute division on the target warning data to obtain a decision attribute set, and performing information gain calculation on the decision attribute set to obtain an attribute importance ranking; Based on the attribute importance ranking, a decision tree is constructed for the target warning data to obtain an initial decision tree structure, and the initial decision tree structure is pruned to obtain an optimized decision tree; Performing path analysis on the optimized decision tree to obtain a decision rule set, and performing conflict detection on the decision rule set to obtain a consistency rule subset; Based on the consistency rule subset, the target warning data is matched with rules to obtain a matching result set, and the matching result set is evaluated for confidence to obtain a high-confidence decision path; The high-confidence decision path is processed through case-based reasoning to obtain the target early warning solution.

6. A device for dynamic monitoring and intelligent early warning of coal and gas outburst, used to implement the method for dynamic monitoring and intelligent early warning of coal and gas outburst as described in any one of claims 1 to 5, characterized in that: The coal and gas outburst dynamic monitoring and intelligent early warning device comprises: A processing module, used for collecting and preprocessing the multi-source sensor data set in the target area to obtain a multi-dimensional data set; A prediction module, used for performing multivariate time series analysis on the multidimensional data set to obtain coal seam gas outburst hazard prediction data; An interpolation module is used to perform dynamic risk assessment and spatial interpolation processing on the multidimensional data set based on the coal seam gas outburst hazard prediction data to obtain real-time risk distribution data of the target area; An analysis module is used to perform multidimensional threshold analysis on the real-time risk distribution data to obtain preliminary warning data, including: extracting statistical features from the real-time risk distribution data to obtain risk statistical features, and performing normality test on the risk statistical features to obtain distribution characteristic data; based on the distribution characteristic data, performing anomaly detection on the real-time risk distribution data to obtain an abnormal risk point set, and performing density estimation on the abnormal risk point set to obtain a risk density distribution; performing threshold calculation on the risk density distribution to obtain a multidimensional threshold set, and dynamically adjusting the multidimensional threshold set to obtain adaptive threshold data; based on the adaptive threshold data, performing threshold segmentation on the real-time risk distribution data to obtain a risk area division result, and performing connectivity analysis on the risk area division result to obtain a high-risk connected area; performing spatiotemporal evolution analysis on the high-risk connected area to obtain the preliminary warning data; A fusion module is used to perform multimodal information fusion on the preliminary warning data to obtain target warning data, including: performing feature decomposition on the preliminary warning data to obtain a multimodal feature set, and performing correlation analysis on the multimodal feature set to obtain a feature correlation matrix; based on the feature correlation matrix, performing feature selection on the multimodal feature set to obtain a key feature subset, and normalizing the key feature subset to obtain standardized feature data; performing multimodal alignment on the standardized feature data to obtain an aligned feature sequence, and processing the aligned feature sequence through a deep belief network to obtain a fused feature vector; performing confidence calculation on the fused feature vector to obtain feature confidence data, and weighting the fused feature vector based on the feature confidence data to obtain a weighted fused feature; processing the weighted fused feature through a fuzzy inference system to obtain the target warning data; The decision module is used to perform intelligent decision tree analysis on the target warning data to obtain a target warning solution.

Images