A coal mine safety monitoring and disaster prediction system and method based on multi-source data fusion

Abstract

This invention relates to the field of coal mine safety technology, specifically providing a coal mine safety monitoring and disaster prediction system and method based on multi-source data fusion. The system dynamically divides multi-source data into a first dataset and a second dataset. The first dataset contains multi-source data from key areas corresponding to the mine area, while the second dataset contains multi-source data from non-key areas corresponding to the mine area. First-level data fusion processing is performed on both datasets to obtain a first-level fusion result for the first dataset and a second-level fusion result for the second dataset. Second-level feature fusion processing is then performed on both results to obtain a second-level feature fusion result. Finally, a third-level decision-level fusion is performed on the second-level feature fusion result and the real-time personnel location data of the mine area to obtain a comprehensive assessment result for the mine area. This improves the accuracy of coal mine safety monitoring and disaster early warning, ensuring the overall safety level of coal mines.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety technology, specifically providing a coal mine safety monitoring and disaster prediction system and method that integrates multi-source data. Background Technology

[0002] With the development of the coal mining industry and technologies such as artificial intelligence and the Internet of Things, coal mine safety monitoring has broken through traditional monitoring methods. Its aim is to improve the accuracy of coal mine safety predictions and provide safety guarantees for daily mine production and personnel safety. Traditional coal mine safety monitoring suffers from low accuracy and slow response times. Furthermore, existing single sensors are susceptible to interference from environmental factors such as dust and electromagnetic noise, leading to inaccuracies in coal mine safety monitoring. Single data points, such as gas concentration, are insufficient for accurate and timely disaster prediction; comprehensive safety monitoring combining data from multiple sources, including geological structure, ventilation structure, and real-time environmental changes, is necessary. Without data fusion and processing across different monitoring systems, such as those for gas, roof support, and ventilation, a comprehensive assessment of disaster risk cannot be achieved, resulting in inaccurate safety monitoring and disaster early warning systems. Summary of the Invention

[0003] This invention provides a coal mine safety monitoring and disaster prediction system and method that integrates multi-source data to solve the problem of inaccurate safety monitoring and disaster early warning.

[0004] In a first aspect, embodiments of this application provide a method for coal mine safety monitoring and disaster prediction based on multi-source data fusion, including: Acquire multi-source data on the mining area to determine its safety level; The multi-source data is dynamically divided into a first dataset and a second dataset. The first dataset is multi-source data of the key area corresponding to the mining area, and the second dataset is multi-source data of the non-key area corresponding to the mining area. Perform first-level data fusion processing on the first dataset and the second dataset respectively to obtain the first-level data fusion result corresponding to the first dataset and the second-level data fusion result corresponding to the second dataset; The first-level fusion result of the first data and the first-level fusion result of the second data are subjected to second-level feature fusion processing to obtain the second-level feature fusion result. The secondary feature fusion results and the real-time personnel location of the mining area are fused at the third-level decision level to obtain the comprehensive evaluation results of the mining area; Based on the comprehensive assessment results, the safety level of the mining area is output.

[0005] In some embodiments, the step of performing first-level data fusion processing on the first dataset and the second dataset respectively to obtain a first-level data fusion result corresponding to the first dataset and a second-level data fusion result corresponding to the second dataset includes: Obtain the gas concentration of the first dataset; The gas concentration of the first dataset is standardized using Z-score to obtain the standardized gas concentration, and the standardized gas concentration is used as the first data level fusion result corresponding to the first dataset. Obtain wind speed, temperature, and dust data from the second dataset; The first step involves using the isolated forest algorithm to identify and remove outlier data from the wind speed, temperature, and dust data in the second dataset, thereby obtaining the normal data for wind speed, temperature, and dust in the second dataset. The second step involves processing the normal data of wind speed, temperature, and dust in the second dataset using an adaptive weighted average method to obtain standardized environmental parameters of the second dataset. These standardized environmental parameters are then used as the first-level fusion result of the second dataset.

[0006] In some embodiments, the step of performing secondary feature fusion processing on the first data primary fusion result and the second data primary fusion result to obtain a secondary feature fusion result includes: Obtain the pre-defined safety feature database of the coal mine; The first-level fusion result of the first data and the first-level fusion result of the second data are used to perform feature extraction and attention weighting using a convolutional neural network (CNN) algorithm, and the associated feature vector is output. The associated feature vector is matched with the preset security feature library, and the fusion result is output. The fusion result is the first data level fusion result and the association strength of the second data level fusion result. The association strength is used as the second-level feature fusion result. The Long Short-Term Memory (LSTM) network algorithm is used to jointly analyze and predict the first-level fusion result of the first data and the second-level fusion result of the second data to obtain the predicted value of the gas concentration of the first dataset. The predicted value of the gas concentration of the first dataset is used as the second-level feature fusion result.

[0007] In some embodiments, the secondary feature fusion result and the real-time personnel location of the mining area are fused at a third-level decision level to obtain a comprehensive evaluation result of the mining area, including: The secondary feature fusion results and the real-time personnel location of the mining area are combined using a Bayesian network algorithm to perform a three-level decision-level fusion analysis, resulting in risk level classification, risk probability output, and risk-related factor labeling. The risk level classification, risk probability output, and risk-related factor labeling are then used as the comprehensive evaluation results of the mining area.

[0008] Secondly, the present invention also provides a coal mine safety monitoring and disaster prediction system based on multi-source data fusion, comprising: Data acquisition module: Acquires multi-source data from the mine; Data processing module: data preprocessing and data fusion processing; Data fusion module: includes primary data fusion, secondary feature fusion processing, and tertiary decision-level fusion; Safety monitoring module: Outputs safety level warnings.

[0009] The coal mine safety monitoring and disaster prediction system and method provided in this application embodiment integrates multiple types of data such as gas, water hazards, fire, roof collapse, dust, and temperature, which can improve the accuracy of monitoring and disaster early warning, thereby improving the overall safety level of coal mines and personnel safety. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the coal mine safety monitoring and disaster prediction method based on multi-source data fusion provided by the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0013] The following is combined with Figure 1 The illustrated embodiments describe the technical solution of the present invention: S110: Obtain multi-source data for the mining area.

[0014] In some embodiments, the implementation of step S110 (acquiring multi-source data of the mining area) may include: It should be noted that multi-source data fusion in coal mine safety monitoring involves various types of data, such as gas concentration, roof pressure, and dust, which can improve the accuracy of coal mine safety monitoring. Based on historical and real-time monitoring data, potential future safety hazards can be predicted, providing scientific early warning information for mine management and ultimately ensuring the safety of workers.

[0015] Sensors using Internet of Things (IoT) technology collect multi-source data from the mine in real time; the mine uses GPS, total station, 3D laser scanning and other measurement methods to determine the mining area.

[0016] Specifically, catalytic combustion and infrared absorption sensors are installed at the coal face to monitor methane concentration in real time. Hydraulic pressure sensors detect pressure changes in real time to obtain roof pressure data. Gas sensors collect methane concentration data; impeller-type anemometers collect wind speed and volume data; infrared thermometers detect temperature data; optical dust sensors collect coal dust particle concentration in real time; and ultra-wideband (UWB) positioning technology is used to locate personnel in real time, with tags attached to personnel to monitor their movement trajectories. This process acquires data on methane concentration, roof pressure, equipment status, temperature, dust concentration, personnel location, and geological structure.

[0017] Geological mapping, physical exploration, and remote sensing technologies were used to obtain geological data for the mine. It should be noted that when acquiring multi-source data from a mine, it is also necessary to obtain real-time information about the mine's regional conditions. Monitoring multi-source data in coal mines requires combining it with regional information to accurately and dynamically analyze the mine's status, thereby ensuring the overall safety of the mine. For example, if sensor data exceeds the standard, it is necessary to know the specific location of the sensor in order to take appropriate measures.

[0018] It should be noted that coal mines are mainly divided into underground coal mines and open-pit coal mines; roof collapse is a serious hazard in underground coal mines. Based on the mine type and geographical location, a sensor network should be rationally deployed, with more sensors deployed at key locations to collect data. Simultaneously, an efficient data transmission network must be built to ensure real-time and accurate transmission to the data center. Different coal mines have different data sources, and even within the same mine, data types may vary. There may be incompatibility issues between different types of data. For example, data such as gas concentration, roof pressure, wind speed, temperature, dust concentration, and personnel location are different types of data and require preprocessing to eliminate their impact on data fusion.

[0019] S120: Dynamically divide the multi-source data into a first dataset and a second dataset. The first dataset is multi-source data of the key area corresponding to the mining area, and the second dataset is multi-source data of the non-key area corresponding to the mining area.

[0020] In some embodiments, the implementation of step S120 (dynamically dividing multi-source data into a first dataset and a second dataset, wherein the first dataset is multi-source data of key areas corresponding to the mining area, and the second dataset is multi-source data of non-key areas corresponding to the mining area) may include: 3DMine 3D modeling software was used to convert geological data, gas data, and ventilation data of the mine area into 3D models, which were used to divide areas such as coal mining faces, tunneling faces, and roadways. Specifically, 3DMine 3D modeling software is used to provide spatial coordinate references for data division. Then, the multi-source data of the mine is associated with the spatial coordinates of the 3D model to divide different areas.

[0021] It should be noted that, based on a comprehensive analysis of the mine's geological structure, production process, and equipment status, the mine is divided into different areas; Specifically, based on historical coal mine accidents, critical areas are those with a high probability of danger, including coal mining faces and tunneling faces; non-critical areas are other areas besides the critical areas, such as roadways. Therefore, the mine area is divided into critical and non-critical areas.

[0022] It should be noted that coal mining faces and tunneling faces are the core areas of mine production, directly related to worker safety and coal output, and therefore require real-time safety monitoring as critical areas. While the gas and wind speed parameters in the secondary roadway area are relatively stable, their safety monitoring priority is lower than that of critical areas. However, the ventilation stability of the main intake and return air roadways and the boundary roadways directly affects mine safety, thus classifying them as critical areas. Furthermore, when a fault occurs in a roadway, that area should also be classified as a critical area.

[0023] It should be noted that as the work on the coal face progresses, the geological conditions of the mine will change, and previously non-critical areas may temporarily become high-risk areas, i.e., critical areas. Non-critical areas may temporarily become critical areas. The scope of critical areas can be adjusted based on real-time mine geological data to achieve dynamic division of critical and non-critical areas in the mine.

[0024] It should be noted that microseismic sensor arrays are deployed in key locations underground and on the surface, such as in front of the mining face, near fault zones, areas with remaining coal pillars, and key layers of overlying strata, which are potentially high-risk areas. These arrays continuously collect vibration signals generated by rock mass fracturing and then filter and remove noise from the collected signals.

[0025] The ratio of short-time average to long-time average is used to identify and extract microseismic events in the second dataset. If potential active fault areas or shock risk areas are detected in the second dataset, the multi-source data corresponding to the potential active fault areas or shock risk areas in the second dataset are classified as the first dataset.

[0026] Specifically, the ratio of short-time average to long-time average is used to identify and extract microseismic events from the second dataset. The ratio of the short-term energy average to the long-term energy average of the signal is calculated. When the ratio exceeds the threshold, it is determined that a microseismic event has started.

[0027] Based on the velocity models of P-waves and S-waves and their time differences in arrival at different sensors, the Geiger method is used to calculate the spatial location (X, Y, Z) of the source of microseismic events, and to calculate the energy and magnitude released by each microseismic event. Then, all the located microseismic events within a certain period of time are visualized to form a microseismic activity cloud map. b-value analysis and statistical analysis of the proportion of earthquakes of different magnitudes are performed.

[0028] It should be noted that a significant decrease in the b-value indicates a high degree of stress concentration, which is a precursor to a major event.

[0029] If microseismic events are detected to be densely distributed along a certain linear structure, it is determined that there is a potential activated fault in the area; if a cluster of high-energy, low-b-value microseismic events is detected in a certain area in a short period of time, it is determined that there is a high risk of rockburst in the area; the newly monitored fault area or rockburst risk area will be passively classified as a critical area.

[0030] It should be noted that the scope of the critical area is adjusted based on the results of seismic monitoring. If new faults or impact risks are detected, the area will be designated as a critical area.

[0031] It should be noted that critical areas are high-risk areas in mines, such as coal mining faces. Due to the dense concentration of equipment and personnel, a roof collapse at the coal face can directly lead to production stoppage. The sudden increase in gas concentration can also cause casualties. Therefore, coal mining faces are considered critical areas. Because of the high risk factor in critical areas, monitoring frequency should be increased, sensor nodes should be densely distributed, and disaster prediction updates should be rapid.

[0032] It should be noted that coal mine safety monitoring methods, through multi-source data fusion, can comprehensively analyze the correlations between early warnings of disasters such as gas and roof collapse, enabling more accurate safety monitoring and disaster early warning, thereby improving the overall safety of coal mines. Specifically, multi-source data fusion includes primary data fusion, secondary feature fusion, and tertiary decision-level fusion.

[0033] S130: Perform first-level data fusion processing on the first dataset and the second dataset respectively to obtain the first-level data fusion result corresponding to the first dataset and the second-level data fusion result corresponding to the second dataset.

[0034] In some embodiments, the implementation of step S130 (performing first-level data fusion processing on the first dataset and the second dataset respectively to obtain the first-level data fusion result corresponding to the first dataset and the second-level data fusion result corresponding to the second dataset) may include: It should be noted that for similar types of sensor data, such as gas sensor data, a single sensor cannot fully reflect the complex environment underground. The output of primary data fusion can be used as input for secondary fusion, allowing for further correlation analysis to uncover potential hidden risks. Preprocessing data from the same sensors in the same area can reduce discrepancies; fusing data from different sensors in the same area can further uncover safety hazards.

[0035] It should be noted that without data integration and analysis, it is impossible to extract valuable feature information for safety monitoring from massive amounts of monitoring data, and thus impossible to accurately predict and locate potential risks and disasters.

[0036] It should be noted that methane concentration is directly related to explosion risk and is a crucial indicator for coal mine safety monitoring. Monitoring of methane concentration should focus on high-risk areas such as mining faces and return airways, requiring real-time monitoring. Methane concentration may also be correlated with ventilation, dust levels, and equipment status.

[0037] It should be noted that Kalman filtering can suppress data noise and remove abnormal data from sensors, thereby improving data quality; it also reduces input errors in subsequent decision-level fusion, thus improving the accuracy of coal mine safety monitoring.

[0038] It should be noted that the gas concentration and roof pressure data are fused using extended Kalman filtering to obtain gas target state estimates and roof target state estimates, which can be used as the first-level fusion results.

[0039] The gas concentration data of the first dataset and the wind speed, temperature and dust data of the second dataset were obtained using sensors.

[0040] The gas concentration in the first dataset is standardized using Z-score to obtain the standardized gas concentration. The standardized gas concentration is then used as the first-level data fusion result corresponding to the first dataset.

[0041] It should be noted that temporary sensor malfunctions and noise can affect the data fusion results. Therefore, abnormal data processing is performed first, and then the data in the second dataset is processed using an adaptive weighted average method. The adaptive weighted average method can dynamically adjust the weights of indicators. For example, the impact of wind speed on gas diffusion will change under different working conditions.

[0042] Specifically, the isolated forest algorithm is first used to identify abnormal data (for example, if the temperature sensor suddenly exceeds the normal range, it indicates that the sensor itself is malfunctioning and cannot collect data correctly, rather than that the data is actually abnormal) and the abnormal data is removed to obtain the normal data of wind speed, temperature and dust in the second dataset. Then, the adaptive weighted average method is used to process the normal data to obtain the standardized environmental parameters of the second dataset. The standardized environmental parameters of the second dataset are used as the second data first-level fusion result corresponding to the second dataset.

[0043] It should be noted that the results of the first-level data fusion are used as input for the second-level feature fusion, and the results of the first-level data fusion are transformed into standardized feature vectors. The purpose is to analyze the correlation between different parameters, so as to uncover hidden disaster risks and issue early warnings as early as possible.

[0044] S140: Perform secondary feature fusion processing on the first-level fusion result of the first data and the first-level fusion result of the second data to obtain the secondary feature fusion result.

[0045] In some embodiments, the implementation of step S140 (performing secondary feature fusion processing on the first data primary fusion result and the second data primary fusion result to obtain a secondary feature fusion result) may include: It should be noted that, based on the results of the first-level data fusion, feature information is extracted. There is a certain correlation between the gas concentration and the roof pressure in the first dataset (key area). For example, whether the sudden increase in gas concentration and the sudden increase in roof pressure occurred in the same mining area and simultaneously. Also, roof collapse can lead to a sudden gas outburst. The second-level feature fusion is used to extract coupling features and perform correlation analysis on different parameter data to discover potential hidden safety hazards.

[0046] Secondary fusion is the process of analyzing the correlation indicators of data such as gas concentration, temperature, and dust, and outputting the secondary feature fusion results through correlation analysis.

[0047] It should be noted that statistical features, such as the gradient rate of change and variance of gas concentration, and the pressure change acceleration of roof pressure, are extracted from the target state estimate.

[0048] It should be noted that during the secondary fusion of coal mine safety monitoring, correlation analysis is required for data such as gas concentration, temperature, and dust. Multiple sensor data points exist within the mine, necessitating the establishment of multiple monitoring points.

[0049] Obtain the pre-defined safety feature database of the coal mine; The first-level fusion result of the first data and the first-level fusion result of the second data are used to perform feature extraction and attention weighting using a convolutional neural network (CNN) algorithm, and the associated feature vector is output. The associated feature vectors are matched with a preset safety feature library, and the fusion result is output. The fusion result is the correlation strength between the first-level fusion result and the second-level fusion result of the data. The correlation strength is used as the second-level feature fusion result. If there is a strong positive correlation between the first-level fusion result of the data (standardized gas concentration) and the second-level fusion result of the data (standardized environmental parameters), it is determined that the direction of the airflow needs to be monitored.

[0050] For example, in a coal mine monitoring system, n monitoring points are set up, where n is 5. Each monitoring point collects data on gas concentration, temperature, and dust. The time step is set to 1 minute, and data is acquired continuously for 30 minutes. The input tensor T = [5 × 3 × 30] is constructed. This represents 5 measuring points, 3 parameters, and 30 time steps. The standardized gas concentration, temperature, and dust data are processed using a convolutional neural network (CNN) algorithm for feature extraction and attention weighting, and finally fused and output using a fully connected layer.

[0051] Specifically, three 1D convolutional kernels (3x3 kernel size) are used to convolve the parameter-time series of each measuring point, outputting local feature maps, such as the abrupt changes in gas concentration over 5-10 minutes. Next, attention weighting is applied, using a parameter attention module to calculate the weights of each parameter. If the current gas concentration fluctuates significantly, a weight of 0.6 is assigned to gas concentration, 0.2 to temperature, and 0.2 to dust. A spatial attention module is also used to calculate the weights of each measuring point; for example, downwind measuring points have a higher weight than upwind measuring points.

[0052] The attention-weighted feature map is flattened into a vector and input into a fully connected layer. The output is a correlated feature vector, such as the intensity value of cross-correlation between gas and temperature, gas and dust, and temperature and dust. The correlated feature vector is matched with a preset safety feature library, and the fusion result is output. For example, if there is a strong positive correlation between the gas concentration at measuring point 2 and the dust concentration at measuring point 5 (correlation strength of 0.8), then the direction of airflow needs to be monitored.

[0053] It should be noted that the preset safety feature library includes basic threshold features, multi-parameter correlation features, and spatiotemporal dynamic features. The basic threshold features define the safety thresholds for individual monitoring parameters, such as a safety threshold of 0.5% for methane concentration, 26℃ for temperature, 10mg / m³ for dust concentration, and 0.25-4m / s for wind speed. These also provide a basis for multi-parameter correlation analysis. For multi-parameter correlation features, when methane concentration and temperature both increase, the correlation strength threshold is 0.7. If the correlation strength is greater than 0.7, the level is considered dangerous. When dust concentration increases while wind speed decreases, the correlation strength threshold is 0.5-0.7; when methane concentration increases while dust concentration decreases, the correlation strength threshold is 0.4-0.6. Combining spatiotemporal dynamic features with time makes risk assessment more closely reflect actual mine conditions. For example, a 0.3% increase in methane concentration within 5 minutes indicates a dangerous level; when three adjacent monitoring points on the working face simultaneously experience temperature increases, if the monitoring distance between the monitoring points is less than or equal to 50 meters, the level is also considered dangerous.

[0054] It should be noted that using the Long Short-Term Memory (LSTM) algorithm can improve the prediction accuracy of correlation analysis, uncover the correlation between multi-source data, trace the cause of monitoring and early warning more specifically, and may uncover hidden relationships between multi-source data; it can analyze data changes over time, predict future trends, discover potential disaster risks, and promptly issue prediction and early warning information if anomalies occur.

[0055] The Long Short-Term Memory (LSTM) network algorithm is used to jointly analyze and predict the first-level fusion results of the first data (key areas) and the second-level fusion results of the second data (non-key areas) to obtain the gas concentration prediction value of the first dataset. The gas concentration prediction value of the first dataset is used as the second-level feature fusion result.

[0056] Specifically, the Long Short-Term Memory (LSTM) network algorithm is used to perform joint analysis and prediction on the gas concentration of the first-level fusion result (key area) of the first dataset and the standardized environmental parameters (wind speed, temperature and dust data of the second dataset (non-key area)). The fused feature vector is output to predict the risk of gas outburst in the next 2-4 hours and can provide early warning of high-risk coal mine accidents such as gas exceeding the limit.

[0057] In summary, the correlation strength between the first-level fusion result of the first data and the first-level fusion result of the second data, along with the predicted gas concentration value of the first dataset, is used as the second-level feature fusion result.

[0058] It should be noted that the obtained fusion feature vector integrates the temporal correlation of three types of features in the key area: gas, roof, and standardized environmental parameters of the roadway. For example, the time of gas concentration change with roof change and the dynamic impact of environmental parameters on the key area.

[0059] Specifically, cross-regional multi-source data correlation analysis is performed. For example, the correlation between abnormal gas concentrations in critical areas and wind speeds in ventilation roadways in non-critical areas can be analyzed. When the gas concentration in the coal face of a critical area increases while the wind speed in the ventilation roadways of a non-critical area decreases, secondary feature fusion may identify a risk of insufficient ventilation leading to gas accumulation. When the gas concentration in the critical area of ​​the coal face increases while the wind speed in the ventilation roadways of a non-critical area does not increase, it is judged that the ventilation system may be malfunctioning. For example, the probability value of a gas concentration >1% and a wind speed <2m / s occurring simultaneously.

[0060] It should be noted that the results of secondary feature fusion provide information on potential environmental risks, such as the probability of exceeding gas concentration limits.

[0061] S150: Perform three-level decision-level fusion of the secondary feature fusion results and the real-time personnel location of the mine area to obtain the comprehensive assessment results of the mine area, and finally output the safety level of the mine area.

[0062] In some embodiments, the implementation of step S150 (performing a three-level decision-level fusion of the secondary feature fusion result and the real-time personnel location in the mine area to obtain a comprehensive assessment result of the mine area, and finally outputting the safety level of the mine area) may include: It should be noted that the output of the secondary feature fusion is a feature vector; the output of the secondary feature fusion is used as the input of the tertiary decision level. The tertiary decision level fusion generates disaster early warning instructions based on the feature vector, and finally performs a comprehensive risk assessment.

[0063] It should be noted that the three-level decision fusion needs to consider both environmental and personnel risks. The associated risks of environmental parameters such as gas and concentration, which are the results of the two-level feature fusion, are combined with real-time personnel location. Then, the three-level decision fusion is carried out through dynamic Bayesian network (DBN) to meet the needs of collaborative protection of the environment and personnel in the mine.

[0064] It should be noted that since the risks in the mine are dynamic, such as the gas concentration changing with ventilation and the personnel's position moving with the operation, it is necessary to capture the dynamics. DBN can use time slice modeling to capture these dynamics.

[0065] It should be noted that Bayesian networks can integrate the probability of environmental risk and the probability of people being in dangerous areas into a comprehensive risk probability of people being in danger, and output probabilistic early warnings.

[0066] The results of the secondary feature fusion and real-time personnel positioning are used as inputs to the tertiary decision level. The Bayesian network algorithm is then used to calculate the safety risk level of the mine and provide decision recommendations.

[0067] Specifically, based on the aforementioned secondary feature fusion results—namely, predicted gas concentration, the correlation strength between gas and environmental parameters, and real-time personnel location—a Bayesian network algorithm is used for decision-level fusion analysis. This algorithm outputs a quantifiable and gradable gas risk result, rather than a single concentration value; it yields risk level classification, risk probability output, and risk-related factor labeling. It also outputs the confidence level of the current risk, for example, a high risk with a confidence level of 92%; the risk primarily stems from excessive gas concentration (correlation strength 85%), and personnel being located in a high-risk area.

[0068] Risk levels are categorized as safe, low risk, medium risk, high risk, and extremely risky. Safety: Predicted gas concentration <0.5%; No personnel are located in high-risk areas; Low risk: 0.5% ≤ predicted gas concentration < 0.8%; or 1-2 people are near a high-risk area; Medium risk: 0.8% ≤ predicted gas concentration < 1.0%; or personnel may have entered the high-risk area (number ≤ 3 people). High risk: 1.0% ≤ predicted gas concentration < 1.5%; or more than 3 people are located in a high-risk area; Extremely dangerous: Predicted gas concentration ≥1.5%; or a sudden increase in gas concentration (≥0.3% increase within 10 minutes).

[0069] Based on the risk assessment results, the algorithm will further output practical suggestions for personnel management and safe operation, such as immediate evacuation of personnel from the work area and equipment shutdown.

[0070] For example, step 1: The input variables for the three-level decision-level fusion are defined, specifying the input nodes of the DBN model. The second-level feature fusion results are the predicted gas concentration and the correlation strength between gas and environmental parameters. Real-time personnel positioning can include: the distance between personnel and the risk area, the direction of personnel movement (relative to the risk area), and the number of personnel within the risk area.

[0071] Step 2: Construct a dynamic Bayesian network (DBN) structure, setting the time slice according to the rate of change of mine risk (e.g., 1 minute / slice); the comprehensive risk level within the same time slice depends on all input variables. Based on the secondary feature fusion results and real-time personnel positioning, a Bayesian network algorithm is used for decision-level fusion analysis to map multi-source risk indicators to safety levels.

[0072] It should be noted that environmental changes affect coal mine monitoring, and real-time monitoring of the location and safety status of all personnel in each area is necessary to ensure their safety. The advantage of Bayesian network algorithms is that they integrate prior knowledge with real-time data, allowing for a comprehensive evaluation of monitoring results.

[0073] It should be noted that if a gas concentration in a certain area is predicted to exceed the limit, and the location data of staff indicates that workers are nearby, an immediate evacuation should be carried out to prevent danger. Based on time changes, the three-level decision-making system employs a dynamic Bayesian network model to ultimately output probabilistic early warning information.

[0074] It should be noted that the parameter correlation patterns in historical coal mine accidents are the result of secondary feature fusion. It is also necessary to combine real-time feature vectors, such as real-time environmental changes, geological condition changes, and the influence of time dimension, such as the change pattern of disaster risk over time, and the superposition effect of short-term gas concentration increase and long-term roof pressure accumulation.

[0075] It should be noted that a feedback mechanism can be further established to transmit the results of the three-level decision-level fusion back to the first-level data fusion, dynamically adjusting the filtering parameters (such as the noise covariance matrix of EKF), thereby improving the accuracy of disaster prediction.

[0076] It should be noted that by combining data on gas, roof structure, and temperature, a regional safety level (four-level warning: red / orange / yellow / blue) is generated. Furthermore, the time and location of anomalies are analyzed to facilitate subsequent rescue efforts and optimize safety monitoring methods.

[0077] It should be noted that a risk factor analysis report is also generated when outputting the safety level. Early warning information is promptly sent to coal mine safety management personnel and mine workers, including specific information such as the safety level and specific location.

[0078] Based on the same inventive concept, this application also provides a coal mine safety monitoring and disaster prediction system based on multi-source data fusion, including the following: The data acquisition module is used to acquire multi-source data of the mining area in real time in order to determine the safety level of the mining area; The data processing module is used to dynamically divide multi-source data into a first dataset and a second dataset. The first dataset is multi-source data of the key area corresponding to the mining area, and the second dataset is multi-source data of the non-key area corresponding to the mining area. The data fusion module is used to perform first-level data fusion processing on the first dataset and the second dataset respectively, to obtain the first-level fusion result of the first dataset and the second-level fusion result of the second dataset; to perform second-level feature fusion processing on the first-level fusion result of the first dataset and the multi-source data of the mine area to obtain the second-level fusion result of the first dataset; to perform second-level feature fusion processing on the second-level fusion result of the second dataset and the multi-source data of the mine area to obtain the second-level fusion result of the second dataset; and to perform third-level decision-level fusion processing on the first data result and the second data result with the real-time environmental changes of the mine area to obtain the comprehensive evaluation result of the mine area; the first data result includes the first-level fusion result of the first dataset and the second-level fusion result of the first dataset, and the second data result includes the second-level fusion result of the second dataset and the second-level fusion result of the second dataset. The safety monitoring module outputs the safety level of the mining area based on the comprehensive assessment results.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coal mine safety monitoring and disaster prediction through multi-source data fusion, characterized in that, include: Acquire multi-source data on the mining area to determine its safety level; The multi-source data is dynamically divided into a first dataset and a second dataset. The first dataset is multi-source data of the key area corresponding to the mining area, and the second dataset is multi-source data of the non-key area corresponding to the mining area. Perform first-level data fusion processing on the first dataset and the second dataset respectively to obtain the first-level data fusion result corresponding to the first dataset and the second-level data fusion result corresponding to the second dataset; The first-level fusion result of the first data and the first-level fusion result of the second data are subjected to second-level feature fusion processing to obtain the second-level feature fusion result. The secondary feature fusion results and the real-time personnel location of the mining area are fused at the third-level decision level to obtain the comprehensive evaluation results of the mining area; Based on the comprehensive assessment results, the safety level of the mining area is output.

2. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 1, characterized in that, The step of dynamically dividing the multi-source data into a first dataset and a second dataset includes: The multi-source data is converted into three-dimensional spatial coordinates of the mine area using 3DMine three-dimensional modeling software. The multi-source data is then associated with the three-dimensional spatial coordinates and divided into a first dataset and a second dataset. The first dataset is multi-source data of key areas corresponding to the mining area, and the multi-source data of key areas is multi-source data of coal mining faces and tunneling faces in the mining area; The second dataset consists of multi-source data of non-critical areas corresponding to the mine area, and the multi-source data of non-critical areas consists of multi-source data of the roadway areas of the mine area.

3. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 1, characterized in that, The step of dynamically dividing the multi-source data into a first dataset and a second dataset also includes: The microseismic events in the second dataset are identified and extracted using the ratio of short-time average to long-time average. If potential active fault areas or shock risk areas are detected in the second dataset, the multi-source data corresponding to the potential active fault areas or shock risk areas in the second dataset are divided into the first dataset.

4. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 1, characterized in that, The step of performing first-level data fusion processing on the first dataset and the second dataset respectively to obtain a first-level data fusion result corresponding to the first dataset and a second-level data fusion result corresponding to the second dataset includes: Obtain the gas concentration of the first dataset; The gas concentration in the first dataset is standardized using Z-score to obtain the standardized gas concentration, which is then used as the first data level fusion result corresponding to the first dataset.

5. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 1, characterized in that, The step of performing first-level data fusion processing on the first dataset and the second dataset respectively to obtain a first-level data fusion result corresponding to the first dataset and a second-level data fusion result corresponding to the second dataset further includes: Obtain wind speed, temperature, and dust data from the second dataset; The first step involves using the isolated forest algorithm to identify and remove outlier data from the wind speed, temperature, and dust data in the second dataset, thereby obtaining the normal data for wind speed, temperature, and dust in the second dataset. The second step involves processing the normal data of wind speed, temperature, and dust in the second dataset using an adaptive weighted average method to obtain standardized environmental parameters of the second dataset. These standardized environmental parameters are then used as the first-level fusion result of the second dataset.

6. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 1, characterized in that, The second-level feature fusion process, which involves fusing the first-level fusion result of the first data with the second-level fusion result of the second data, yields the second-level feature fusion result, including: Obtain the pre-set safety feature database of the coal mine; The first-level fusion result of the first data and the first-level fusion result of the second data are used to perform feature extraction and attention weighting using a convolutional neural network (CNN) algorithm, and the associated feature vector is output. The associated feature vector is matched with the preset security feature library, and the fusion result is output. The fusion result is the first data level fusion result and the association strength of the second data level fusion result. The association strength is used as the second-level feature fusion result.

7. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 1, characterized in that, The step of performing secondary feature fusion processing on the first-level data fusion result and the second-level data fusion result to obtain the secondary feature fusion result further includes: The Long Short-Term Memory (LSTM) network algorithm is used to jointly analyze and predict the first-level fusion result of the first data and the second-level fusion result of the second data to obtain the predicted value of the gas concentration of the first dataset. The predicted value of the gas concentration of the first dataset is used as the second-level feature fusion result.

8. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 1, characterized in that, The third-level decision-level fusion of the secondary feature fusion result and the real-time personnel location of the mining area yields a comprehensive evaluation result of the mining area, including: The secondary feature fusion results and the real-time personnel location of the mining area are combined using a Bayesian network algorithm to perform a three-level decision-level fusion analysis, resulting in risk level classification, risk probability output, and risk-related factor labeling. The risk level classification, risk probability output, and risk-related factor labeling are then used as the comprehensive evaluation results of the mining area.

9. The coal mine safety monitoring and disaster prediction method based on multi-source data fusion according to claim 8, characterized in that, Based on the comprehensive assessment results, the safety level of the mining area is output, including: Based on the comprehensive assessment results, the risk levels are divided into safe, low risk, medium risk, high risk, and extremely high risk; and the confidence level of the current risk is output. The risk association factor labeling should at least include the correlation strength that the risk is mainly caused by excessive gas concentration, and that the personnel are located in a high-risk area.

10. A coal mine safety monitoring and disaster prediction system based on multi-source data fusion, characterized in that, include: Data acquisition module: Acquires multi-source data from the mine; Data processing module: data preprocessing and data processing; Data fusion module: includes primary data fusion, secondary feature fusion processing, and tertiary decision-level fusion; Safety monitoring module: Outputs safety level warnings.

Images