A fully mechanized coal mining face roof disaster early warning method and system based on multi-source data

Abstract

The application provides a fully mechanized working face roof disaster early warning method and system based on multi-source data. The method comprises the following steps: collecting mine pressure data, environmental data and geological data of the fully mechanized working face; identifying a periodic pressure stage according to hydraulic support resistance data to obtain pressure cycle information; processing the hydraulic support resistance data, roof separation data, borehole stress data, gas concentration data and water inflow data to extract multi-source feature data; associating and fusing the multi-source feature data with the pressure cycle information to generate a fusion feature data set; inputting the fusion feature data set into a pre-trained disaster risk identification model to obtain a disaster risk probability index; and issuing an early warning signal when the risk probability index is greater than a preset early warning threshold. The technical scheme provided by the application improves the utilization rate of underground coal mine data and the accuracy of roof disaster risk early warning through multi-source data deep fusion analysis and dynamic threshold early warning.

Description

Technical Field

[0001] This application relates to the field of early warning technology for underground coal mine disasters, and in particular to a method and system for early warning of roof disasters in fully mechanized mining faces based on multi-source data. Background Technology

[0002] In coal mining, roof falls in fully mechanized mining faces are one of the main types of disasters threatening safe production. Roof falls are characterized by high frequency of occurrence, numerous triggering factors, suddenness, and difficulty in prediction. Once they occur, they can easily cause casualties and equipment damage. Therefore, accurate early warning of roof falls has significant engineering value and practical importance.

[0003] Currently, early warning of roof collapse in fully mechanized coal mining faces mainly relies on mine pressure monitoring systems. These systems collect data such as support resistance, roof delamination, and anchor bolt stress by installing sensors on hydraulic supports, roof delamination meters, and borehole stress gauges. Existing early warning methods typically analyze historical monitoring data and set static risk warning thresholds. When the monitored data exceeds a preset threshold, an early warning signal is issued. However, existing technologies have the following shortcomings: First, low data utilization. Existing early warning methods mainly use mine pressure data from a single source, such as relying solely on hydraulic support resistance or roof delamination, ignoring the inherent correlation between environmental data such as gas concentration and water inflow and roof collapse, resulting in a large amount of monitoring data not being effectively utilized. Second, insufficient accuracy of single-parameter early warning. Roof collapses are often the result of multiple factors working together; a single parameter cannot fully reflect the complex process of roof instability, easily leading to missed or false alarms. Third, poor adaptability of static thresholds. Different working faces have varying geological conditions, mining techniques, and support methods. Furthermore, the roof activity patterns at different mining stages within the same working face also differ. Using a uniform static threshold cannot accurately cover the roof disaster risk characteristics under different mining environments, making it difficult to guarantee the accuracy of early warnings. Therefore, there is an urgent need to propose a solution that can fully utilize multi-source monitoring data to establish a dynamic early warning model adapted to different mining conditions, thereby improving the accuracy and reliability of roof disaster early warnings. Summary of the Invention

[0004] This application provides a method and system for early warning of roof disasters in fully mechanized mining faces based on multi-source data, in order to at least solve the technical problems of existing coal mine roof disaster early warning methods, which are unable to achieve accurate early warning due to low data utilization, insufficient accuracy of single parameter early warning, and static thresholds that cannot adapt to different mining environments.

[0005] The first aspect of this application proposes a method for early warning of roof disasters in fully mechanized mining faces based on multi-source data, the method comprising:

[0006] Collect mining pressure data, environmental data, and geological data from the fully mechanized mining face. The mining pressure data includes hydraulic support resistance data, roof delamination data, and borehole stress data. The environmental data includes gas concentration data and water inflow data. Based on the hydraulic support resistance data, the periodic pressure-increasing phase is identified to obtain pressure-increasing cycle information, which includes the early pressure-increasing phase, the middle pressure-increasing phase, the late pressure-increasing phase, and the stable phase. The hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data, and water inflow data are processed respectively to extract multi-source feature data, which includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features, and water inflow variation features. The multi-source feature data is correlated and fused with the pressure cycle information to generate a fused feature dataset; The fused feature dataset is input into a pre-trained disaster risk identification model to obtain a disaster risk probability index; A warning signal is issued when the danger probability index is greater than a preset warning threshold.

[0007] Preferably, the hydraulic support resistance data is processed, including: Perform trend decomposition on the time series data of hydraulic support resistance to separate the trend term, periodic term and residual term; The start and end points of each pressure cycle are identified based on the cycle term, and the pressure step size and pressure duration are calculated. The long-term trend of the resistance of the hydraulic support at the working face is determined based on the trend terms. The residual items are used to filter out invalid data caused by sensor malfunctions or communication interruptions. Based on the identified pressure inflow cycle, the statistical characteristics within each pressure inflow cycle are calculated. The statistical characteristics include the maximum resistance value, average resistance, resistance non-uniformity coefficient, and dynamic load coefficient. Based on the hydraulic support resistance data after filtering out invalid data, the location of the point of application of the resultant force of the hydraulic support group at the working face at the same moment is calculated, and the high resistance zone and low resistance zone are identified.

[0008] Preferably, the processing of the top slab delamination data includes: Continuous monitoring data from shallow and deep roof delamination instruments are obtained. The shallow roof delamination data reflects the deformation of the shallow loosened zone of the surrounding rock, and the deep roof delamination data reflects the settlement of the overlying key layer. Differential operations were performed on shallow and deep roof delamination time series data to calculate the delamination rate, and a second differential operation was performed on the delamination rate to calculate the delamination acceleration, which was used to evaluate the dynamic trend and acceleration characteristics of roof delamination development. Based on the delamination data of the roof at different locations of the working face, a spatial interpolation method is used to construct a simulated field of roof delamination data of the working face, calculate the correlation coefficient between delamination data at different locations, and analyze the spatial coordination of roof delamination to determine whether roof fracture occurs synchronously. Based on the changing trends of shallow and deep delamination amounts, a time series curve of the ratio of shallow to deep delamination amounts is established. When the shallow delamination amount increases faster than the deep delamination amount, it is judged as a shallow loosening zone of the surrounding rock. When the deep delamination amount increases faster than the shallow delamination amount, it is judged as instability of the overlying key layer. This is used to identify potential modes of roof instability. Based on the delamination rate and delamination acceleration, abrupt changes in delamination amount are identified. When the delamination rate exceeds a preset threshold, it is marked as a delamination anomaly event, which is used to provide early warning of rapid expansion of roof delamination.

[0009] Preferably, the processing of borehole stress data includes: Data from borehole stress gauges distributed at different locations on the working face are obtained. The borehole stress data includes the pressure distribution data of the advance support pressure in front of the mining area, the lateral support pressure, and the stress distribution data behind the goaf. Based on borehole stress data at different locations on the working face, numerical interpolation is used to invert the stress field distribution in the mining area and generate a stress contour map of the mining area, which is used to assess the overall stress state of the mining area and the location of the stress peak area. The stress gradient is obtained by calculating the ratio of the stress difference between adjacent measuring points to the distance between measuring points. The gradient is divided into low gradient, medium gradient and high gradient regions according to the magnitude of the gradient change, which is used to identify stress change zones and potential failure areas. Combining geological structure distribution data, working face and roadway layout, a stress propagation path analysis algorithm is used to calculate the propagation path of stress from the mining area to the surrounding coal and rock mass, identify stress concentration areas affected by geological structures, and determine the degree of influence of geological structures such as faults and folds on the stress in the mining area. Establish a borehole stress time-series variation curve, calculate the stress change rate, and mark it as a stress anomaly event when the stress change rate exceeds a preset threshold, which is used to provide early warning of sudden stress changes in the mining area.

[0010] Preferably, the processing of gas concentration data includes: Data from gas sensor monitoring points arranged in the fully mechanized mining face and roadway are acquired, and the gas concentration data includes continuous concentration monitoring values ​​over multiple complete production cycles. Calculate the maximum, minimum, average, variation range, and variance of gas concentration throughout the complete production cycle to assess the overall level and fluctuation of gas emission. Sliding window analysis is performed on the time series data of gas concentration to calculate the rate of change, fluctuation, and duration of concentration exceeding the limit in each window. When the rate of change exceeds a preset threshold, it is marked as an abnormal gas outburst event, which is used to identify the precursor characteristics of abnormal gas outbursts. By combining the spatial distribution of sensors, a spatial interpolation method is used to construct a model of the gas concentration change trend at the working face, generate a gas concentration contour map, analyze the direction of the gas concentration gradient, and infer the location of the gas emission source, which is used to analyze the spatial evolution law of gas concentration as the working face advances. By coupling gas concentration time-series data with the working face advance, a gas concentration-advancement relationship curve is established. The periodic variation law of gas concentration with the mining process is analyzed to identify the correlation between abnormal gas outbursts and roof pressure.

[0011] Preferably, the processing of the water inflow data includes: Acquire water inflow monitoring data of the longwall mining face during multiple complete production cycles, including instantaneous flow rate, cumulative flow rate, and water quality sensor data; Sliding window analysis is performed on instantaneous flow time series data to calculate the instantaneous flow rate change rate and fluctuation coefficient, identify abrupt changes and abnormal fluctuations in instantaneous flow, and capture sudden changes in water inflow. Trend fitting is performed on the cumulative flow time series data to establish the cumulative flow growth curve and calculate the cumulative flow growth rate. When the cumulative flow growth rate exceeds the preset threshold, it is marked as an abnormal water inflow event to identify the trend of continuous increase in water inflow. By combining water quality sensor data, changes in water quality parameters are analyzed. When water quality parameters change abruptly, they are marked as water source anomalies to determine whether the source of the water inflow has changed. The water inflow data is coupled with the working face advance to establish a water inflow-advance relationship curve. The change in water inflow per unit advance distance is analyzed to identify the evolution of water inflow during the mining process. Establish a time-series prediction model for water inflow, predict future water inflow trends based on historical water inflow data, and mark water inflow risk events when the predicted value exceeds a preset threshold for early warning of water inflow disasters.

[0012] Preferably, the step of associating and fusing the multi-source feature data with the pressure cycle information to generate a fused feature dataset includes: Establish an index structure with the complete advance cycle of the working face as the unit of time and the sensor position as the unit of space; The extracted multi-source feature data are correlated and compared according to the stage labels of the early stage of pressure, the middle stage of pressure, the late stage of pressure and the stable stage, to generate fused feature data containing temporal features, spatial distribution features and pressure stage labels.

[0013] Preferably, the training process of the disaster risk identification model includes: Acquire historical data, including historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, historical water inflow data, geological data, and historical roof disaster case data; Based on the historical hydraulic support resistance data, identify the historical periodic pressure phases to obtain historical pressure cycle information; The historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, and historical water inflow data are respectively subjected to in-depth processing to extract historical multi-source feature data; A unified temporal and spatial index is established to associate and fuse the historical multi-source feature data with the historical pressure cycle information to generate a historical fused feature dataset, and disaster precursor labels are marked on the data samples before the disaster occurs; A strategy combining feature importance ranking and recursive feature elimination is used to select a subset of key features from the historical fused feature dataset; A hybrid learning model is trained using the aforementioned key feature subset. The hybrid learning model includes a time-series learning module and a state discrimination module. It is trained using historical normal data and case data labeled with disaster precursors, so that the model outputs a disaster risk probability index between 0 and 1.

[0014] Preferably, the preset threshold includes a multi-level warning threshold: a blue warning is issued when the danger probability index is greater than 0.6, a yellow warning is issued when it is greater than 0.75, an orange warning is issued when it is greater than 0.85, and a red warning is issued when it is greater than 0.95. When an early warning signal is issued, an audible and visual alarm is triggered at the monitoring center, and the early warning information is pushed through the industrial internet platform. The early warning information includes the risk level, disaster type, abnormal indicators, and recommended measures.

[0015] The second aspect of this application proposes a roof disaster early warning system for fully mechanized mining faces based on multi-source data, comprising: A roof disaster early warning system for fully mechanized mining faces based on multi-source data, characterized in that the system comprises: The data acquisition module is used to collect mining pressure data, environmental data, and geological data of the fully mechanized mining face. The mining pressure data includes hydraulic support resistance data, roof delamination data, and borehole stress data. The environmental data includes gas concentration data and water inflow data. The acquisition module is used to identify the periodic pressure-increasing phase based on the hydraulic support resistance data and obtain pressure-increasing cycle information, wherein the pressure-increasing cycle includes the early pressure-increasing phase, the middle pressure-increasing phase, the late pressure-increasing phase, and the stable phase. The processing module is used to process the hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data and water inflow data respectively, and extract multi-source feature data, wherein the multi-source feature data includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features and water inflow variation features. The fusion module is used to associate and fuse the multi-source feature data with the pressure cycle information to generate a fused feature dataset; The identification module is used to input the fused feature dataset into a pre-trained disaster risk identification model to obtain a disaster risk probability index; The early warning module is used to issue an early warning signal when the danger probability index is greater than a preset early warning threshold.

[0016] The technical solutions provided by the embodiments of this application have at least the following beneficial effects: This application proposes a method and system for early warning of roof disasters in fully mechanized mining faces based on multi-source data. The method includes: collecting mine pressure data, environmental data, and geological data of the fully mechanized mining face; the mine pressure data includes hydraulic support resistance data, roof delamination data, and borehole stress data; the environmental data includes gas concentration data and water inflow data; identifying periodic pressure stages based on the hydraulic support resistance data to obtain pressure cycle information, wherein the pressure cycle includes early pressure stage, middle pressure stage, late pressure stage, and stable period; processing the hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data, and water inflow data respectively to extract multi-source feature data, wherein the multi-source feature data includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features, and water inflow variation features; associating and fusing the multi-source feature data with the pressure cycle information to generate a fused feature dataset; inputting the fused feature dataset into a pre-trained disaster hazard identification model to obtain a disaster hazard probability index; and issuing an early warning signal when the hazard probability index is greater than a preset early warning threshold. The technical solution proposed in this application improves the utilization rate of underground coal mine data and the accuracy of roof disaster risk warning through deep fusion analysis of multi-source data and dynamic threshold early warning.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a method for early warning of roof disasters in fully mechanized mining faces based on multi-source data, according to an embodiment of this application. Figure 2 This is a structural diagram of a roof disaster early warning system for fully mechanized mining faces based on multi-source data, according to an embodiment of this application. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0020] This application proposes a method and system for early warning of roof disasters in fully mechanized mining faces based on multi-source data. The method includes: collecting mine pressure data, environmental data, and geological data of the fully mechanized mining face; the mine pressure data includes hydraulic support resistance data, roof delamination data, and borehole stress data; the environmental data includes gas concentration data and water inflow data; identifying periodic pressure stages based on the hydraulic support resistance data to obtain pressure cycle information, wherein the pressure cycle includes early pressure stage, middle pressure stage, late pressure stage, and stable period; processing the hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data, and water inflow data respectively to extract multi-source feature data, wherein the multi-source feature data includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features, and water inflow variation features; associating and fusing the multi-source feature data with the pressure cycle information to generate a fused feature dataset; inputting the fused feature dataset into a pre-trained disaster hazard identification model to obtain a disaster hazard probability index; and issuing an early warning signal when the hazard probability index is greater than a preset early warning threshold. The technical solution proposed in this application improves the utilization rate of underground coal mine data and the accuracy of roof disaster risk warning through deep fusion analysis of multi-source data and dynamic threshold early warning.

[0021] The following description, with reference to the accompanying drawings, illustrates an embodiment of a method and system for early warning of roof disasters in a fully mechanized mining face based on multi-source data.

[0022] Example 1 Figure 1 This is a flowchart illustrating a method for early warning of roof disasters in a fully mechanized mining face based on multi-source data, according to an embodiment of this application. Figure 1 As shown, the method includes: Step 1: Collect mining pressure data, environmental data, and geological data of the fully mechanized mining face. The mining pressure data includes hydraulic support resistance data, roof delamination data, and borehole stress data. The environmental data includes gas concentration data and water inflow data. It should be noted that the data acquisition process in step 1 of this embodiment is as follows: By using hydraulic support resistance sensors, the initial support force, working resistance data, and cycle end resistance data of the hydraulic support at the working face are continuously collected at high frequency to ensure that the data covers multiple complete coal mining cycles or cycles.

[0023] Collect continuous roof displacement and borehole stress variation curves throughout the entire production cycle, and analyze the changes in delamination amount, delamination rate, mining-induced stress field distribution, borehole stress value, and stress gradient. Simultaneously, acquire and record data such as working face excavation footage, working face height, sensor locations, information on the roof and floor strata, and geological structure information of the working face.

[0024] Obtain unparsed raw mine pressure monitoring data from the corresponding Kafka or FTP, such as borehole stress data, roof delamination data, and working face hydraulic support resistance data. Upload the raw fully mechanized hydraulic support data, borehole stress data, anchor cable stress data, and upper and lower roadway roof delamination data.

[0025] Historical data from gas sensor monitoring points located in the longwall mining face and roadway were analyzed to obtain data such as the maximum, minimum, and variation range of gas concentration in the longwall mining face over multiple consecutive complete production cycles.

[0026] The historical water inflow data of the longwall mining face during multiple complete production cycles were analyzed to obtain data such as instantaneous flow fluctuations, cumulative flow fluctuations, and water quality changes in different time periods.

[0027] Step 2: Identify the periodic pressure-increasing phase based on the hydraulic support resistance data to obtain pressure-increasing cycle information. The pressure-increasing cycle includes the early pressure-increasing phase, the middle pressure-increasing phase, the late pressure-increasing phase, and the stable phase. It should be noted that the process of obtaining the pressure cycle in step 2 of this embodiment is as follows: Organize the periodic pressure data of the working face top plate, and analyze the starting point, peak point and ending point values ​​of each periodic pressure by combining the resistance data change pattern of the working face hydraulic support in multiple production cycles, and label the data with "pressure status".

[0028] A database is established with the location of hydraulic supports and other sensors on the working face as the spatial unit, based on the complete advance cycle of the working face as the unit of time. The data on mine pressure, roof layer volume, gas concentration, and water inflow extracted in steps three to seven are integrated. The data are correlated and compared with the "early stage of pressure", "middle stage of pressure", "late stage of pressure" and "stable stage" of the periodic pressure on the working face obtained in step eight. The distribution of mine pressure, stratum stability, and evolution trends of mine pressure, gas concentration and water inflow at different production stages of the mining face are analyzed.

[0029] Step 3: Process the hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data, and water inflow data respectively to extract multi-source feature data, which includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features, and water inflow variation features. In this embodiment of the disclosure, the processing of the hydraulic support resistance data includes: Perform trend decomposition on the time series data of hydraulic support resistance to separate the trend term, periodic term and residual term; The start and end points of each pressure cycle are identified based on the cycle term, and the pressure step size and pressure duration are calculated. The long-term trend of the resistance of the hydraulic support at the working face is determined based on the trend terms. The residual items are used to filter out invalid data caused by sensor malfunctions or communication interruptions. Based on the identified pressure inflow cycle, the statistical characteristics within each pressure inflow cycle are calculated. The statistical characteristics include the maximum resistance value, average resistance, resistance non-uniformity coefficient, and dynamic load coefficient. Based on the hydraulic support resistance data after filtering out invalid data, the location of the point of application of the resultant force of the hydraulic support group at the working face at the same moment is calculated, and the high resistance zone and low resistance zone are identified.

[0030] It should be noted that the process of extracting the pressure periodic features in step 3 of this embodiment is as follows: Data on hydraulic supports at coal mining faces during historical roof collapses were analyzed to separate trend terms, periodic terms, and residual terms (noise and abnormal signals).

[0031] Calculate the statistical characteristics for each pressure cycle: maximum resistance value, average resistance, resistance non-uniformity coefficient (standard deviation / mean), and dynamic load coefficient (resistance at the end of the cycle / time-weighted average resistance).

[0032] The key analysis focuses on the resistance distribution of the hydraulic support group at the working face at the same moment, calculating the location of the point of application of the resultant force of the support resistance and the resistance distribution along the length of the working face, and identifying "island" high-resistance or low-resistance areas.

[0033] In this embodiment of the disclosure, processing the top plate delamination data includes: Continuous monitoring data from shallow and deep roof delamination instruments are obtained. The shallow roof delamination data reflects the deformation of the shallow loosened zone of the surrounding rock, and the deep roof delamination data reflects the settlement of the overlying key layer. Differential operations were performed on shallow and deep roof delamination time series data to calculate the delamination rate, and a second differential operation was performed on the delamination rate to calculate the delamination acceleration, which was used to evaluate the dynamic trend and acceleration characteristics of roof delamination development. Based on the delamination data of the roof at different locations of the working face, a spatial interpolation method is used to construct a simulated field of roof delamination data of the working face, calculate the correlation coefficient between delamination data at different locations, and analyze the spatial coordination of roof delamination to determine whether roof fracture occurs synchronously. Based on the changing trends of shallow and deep delamination amounts, a time series curve of the ratio of shallow to deep delamination amounts is established. When the shallow delamination amount increases faster than the deep delamination amount, it is judged as a shallow loosening zone of the surrounding rock. When the deep delamination amount increases faster than the shallow delamination amount, it is judged as instability of the overlying key layer. This is used to identify potential modes of roof instability. Based on the delamination rate and delamination acceleration, abrupt changes in delamination amount are identified. When the delamination rate exceeds a preset threshold, it is marked as a delamination anomaly event, which is used to provide early warning of rapid expansion of roof delamination.

[0034] It should be noted that the process of extracting the delamination dynamic features in step 3 of this embodiment is as follows: The data on roof delamination in the shallow and deep parts of the fully mechanized mining face are processed to obtain the roof delamination rate and delamination acceleration data during the working cycle of the mining face.

[0035] Data on roof delamination at different locations on the working face were analyzed to construct a roof delamination data simulation field and analyze the correlation of the roof delamination data. Since the roof fracture process is not synchronous, the roof bearing capacity and motion were analyzed by analyzing the roof delamination rate and roof delamination acceleration at different locations, and their coordinated change trend was evaluated.

[0036] By analyzing the data on the deep and shallow delamination of the roof of the working face, the development trend and depth of the delamination can be determined, and it can be judged whether the delamination is caused by the shallow loosening zone of the surrounding rock or the instability and rotation of the key overlying layer of the working face.

[0037] In this embodiment of the disclosure, the processing of borehole stress data includes: Data from borehole stress gauges distributed at different locations on the working face are obtained. The borehole stress data includes the pressure distribution data of the advance support pressure in front of the mining area, the lateral support pressure, and the stress distribution data behind the goaf. Based on borehole stress data at different locations on the working face, numerical interpolation is used to invert the stress field distribution in the mining area and generate a stress contour map of the mining area, which is used to assess the overall stress state of the mining area and the location of the stress peak area. The stress gradient is obtained by calculating the ratio of the stress difference between adjacent measuring points to the distance between measuring points. The gradient is divided into low gradient, medium gradient and high gradient regions according to the magnitude of the gradient change, which is used to identify stress change zones and potential failure areas. Combining geological structure distribution data, working face and roadway layout, a stress propagation path analysis algorithm is used to calculate the propagation path of stress from the mining area to the surrounding coal and rock mass, identify stress concentration areas affected by geological structures, and determine the degree of influence of geological structures such as faults and folds on the stress in the mining area. Establish a borehole stress time-series variation curve, calculate the stress change rate, and mark it as a stress anomaly event when the stress change rate exceeds a preset threshold, which is used to provide early warning of sudden stress changes in the mining area.

[0038] It should be noted that the process of extracting stress distribution features in step 3 of this embodiment is as follows: Based on a comprehensive analysis of borehole stress and force-measuring anchor cable stress data distributed at different locations on the working face, the stress field distribution of coal and rock mass within a certain range of the mining area is inverted to infer the stress field distribution. Combined with the overlying strata strength, geological structure distribution data, and the layout of the working face and roadways, the stress propagation path and stress concentration zone distribution are calculated.

[0039] Based on the stress data obtained at different locations, the stress change gradient data between adjacent measuring points is calculated and classified and labeled according to the magnitude of the change.

[0040] In this embodiment of the disclosure, the processing of gas concentration data includes: Data from gas sensor monitoring points arranged in the fully mechanized mining face and roadway are acquired, and the gas concentration data includes continuous concentration monitoring values ​​over multiple complete production cycles. Calculate the maximum, minimum, average, variation range, and variance of gas concentration throughout the complete production cycle to assess the overall level and fluctuation of gas emission. Sliding window analysis is performed on the time series data of gas concentration to calculate the rate of change, fluctuation, and duration of concentration exceeding the limit in each window. When the rate of change exceeds a preset threshold, it is marked as an abnormal gas outburst event, which is used to identify the precursor characteristics of abnormal gas outbursts. By combining the spatial distribution of sensors, a spatial interpolation method is used to construct a model of the gas concentration change trend at the working face, generate a gas concentration contour map, analyze the direction of the gas concentration gradient, and infer the location of the gas emission source, which is used to analyze the spatial evolution law of gas concentration as the working face advances. By coupling gas concentration time-series data with the working face advance, a gas concentration-advancement relationship curve is established. The periodic variation law of gas concentration with the mining process is analyzed to identify the correlation between abnormal gas outbursts and roof pressure.

[0041] It should be noted that the process of extracting gas emission features in step 3 of this embodiment is as follows: Historical data from gas sensor monitoring points located in the longwall mining face and roadway were analyzed to obtain data such as the maximum, minimum, and variation range of gas concentration in the longwall mining face over multiple consecutive complete production cycles.

[0042] The data from multiple gas monitoring points in the fully mechanized mining face were analyzed to obtain secondary data such as the rate of concentration change, fluctuation amount, and duration of abnormality in different time periods.

[0043] By combining the obtained gas monitoring data with the location of the monitoring points, a model of the gas concentration change trend during the complete production cycle of the working face is constructed.

[0044] In this embodiment of the disclosure, the processing of the water inflow data includes: Acquire water inflow monitoring data of the longwall mining face during multiple complete production cycles, including instantaneous flow rate, cumulative flow rate, and water quality sensor data; Sliding window analysis is performed on instantaneous flow time series data to calculate the instantaneous flow rate change rate and fluctuation coefficient, identify abrupt changes and abnormal fluctuations in instantaneous flow, and capture sudden changes in water inflow. Trend fitting is performed on the cumulative flow time series data to establish the cumulative flow growth curve and calculate the cumulative flow growth rate. When the cumulative flow growth rate exceeds the preset threshold, it is marked as an abnormal water inflow event to identify the trend of continuous increase in water inflow. By combining water quality sensor data, changes in water quality parameters are analyzed. When water quality parameters change abruptly, they are marked as water source anomalies to determine whether the source of the water inflow has changed. The water inflow data is coupled with the working face advance to establish a water inflow-advance relationship curve. The change in water inflow per unit advance distance is analyzed to identify the evolution of water inflow during the mining process. Establish a time-series prediction model for water inflow, predict future water inflow trends based on historical water inflow data, and mark water inflow risk events when the predicted value exceeds a preset threshold for early warning of water inflow disasters.

[0045] It should be noted that the process of extracting the water inrush change characteristics in step 3 of this embodiment is as follows: The historical water inflow data of the longwall mining face during multiple complete production cycles were analyzed to obtain data such as instantaneous flow fluctuations, cumulative flow fluctuations, and water quality changes in different time periods.

[0046] By coupling the water inflow data of the working face within a complete production cycle with the working face advance, a model of the changing trend of water inflow of the working face during the production process is obtained.

[0047] Step 4: Associate and fuse the multi-source feature data with the pressure cycle information to generate a fused feature dataset; In this embodiment of the disclosure, step 4 specifically includes: Establish an index structure with the complete advance cycle of the working face as the unit of time and the sensor position as the unit of space; The extracted multi-source feature data are correlated and compared according to the stage labels of the early stage of pressure, the middle stage of pressure, the late stage of pressure and the stable stage, to generate fused feature data containing temporal features, spatial distribution features and pressure stage labels.

[0048] It should be noted that step 4 specifically includes: Organize the periodic pressure data of the working face top plate, and analyze the starting point, peak point and ending point values ​​of each periodic pressure by combining the resistance data change pattern of the working face hydraulic support in multiple production cycles, and label the data with "pressure status".

[0049] A database was established with the location of hydraulic supports and other sensors on the working face as the spatial unit, based on the complete advance cycle of the working face as the unit of time. The extracted data on mine pressure, roof layer volume, gas concentration, and water inflow of the fully mechanized mining face were integrated and compared according to the "early stage of pressure", "middle stage of pressure", "late stage of pressure" and "stable stage" of the periodic pressure on the working face. The distribution of mine pressure, stratum stability, and the evolution trends of mine pressure, gas concentration and water inflow of the mining face at different production stages were analyzed.

[0050] A database structure based on unit time is constructed to generate fused feature data based on the unit working time of the mining face, including statistical, temporal, and spatial characteristics of sensor data such as hydraulic support, roof delamination, and borehole stress, as well as the periodic pressure stage to which the current data belongs.

[0051] Step 5: Input the fused feature dataset into the pre-trained disaster risk identification model to obtain the disaster risk probability index; In this embodiment of the disclosure, the training process of the disaster risk identification model includes: Acquire historical data, including historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, historical water inflow data, geological data, and historical roof disaster case data; Based on the historical hydraulic support resistance data, identify the historical periodic pressure phases to obtain historical pressure cycle information; The historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, and historical water inflow data are respectively subjected to in-depth processing to extract historical multi-source feature data; A unified temporal and spatial index is established to associate and fuse the historical multi-source feature data with the historical pressure cycle information to generate a historical fused feature dataset, and disaster precursor labels are marked on the data samples before the disaster occurs; A strategy combining feature importance ranking and recursive feature elimination is used to select a subset of key features from the historical fused feature dataset; A hybrid learning model is trained using the aforementioned key feature subset. The hybrid learning model includes a time-series learning module and a state discrimination module. It is trained using historical normal data and case data labeled with disaster precursors, so that the model outputs a disaster risk probability index between 0 and 1.

[0052] It should be noted that step 5 specifically includes: Feature comparison and early warning model training are conducted. The fused feature data obtained in the preceding steps undergoes importance ranking and weighting analysis, as well as data correlation mining analysis, to calculate multi-dimensional feature parameters, including temporal statistical features, spatial distribution features, and composite features based on the mine pressure manifestation pattern. A strategy combining feature importance ranking and recursive feature elimination is employed to select the key feature subset with the highest correlation to roof hazard risk from the multi-dimensional feature parameters. Examples include combinations of "periodic pressure stage identifiers, support resistance non-uniformity coefficients, roof delamination acceleration, and borehole stress change gradients." This constructs a feature vector characterizing the roof state, and different weights are assigned according to their importance.

[0053] A hybrid learning model was trained using the selected fusion feature dataset. This model was used to learn the dynamic evolution of roof disaster precursors in mining faces over time (e.g., how the amount of delamination accelerates over time) and the static risk status under the combined effect of multiple factors per unit time. Through the above steps, a large amount of historical normal data and known disaster cases were labeled and analyzed for training to distinguish between conventional mine pressure manifestations and actual roof disaster precursors.

[0054] The trained roof hazard risk analysis model is used to perform inference analysis on the real-time mine pressure monitoring data transmitted from the coal mine working face, and outputs a hazard risk probability index between 0 and 1.

[0055] It should be noted that the training data can be historical hydraulic support, roof delamination instrument, and borehole stress gauge data from working faces that have experienced roof collapses under similar geological conditions within the region. Store cases of fully mechanized mining faces that have experienced general or above-level roof collapses under similar geological and production conditions in the mining area, and record the geological conditions (such as coal seam thickness, coal seam dip angle, roof and floor lithology, geological structure within the mining area, etc.) and production conditions (coal mining, roadway support technology, etc.) of the mining face where the collapse occurred in the corresponding case mine.

[0056] Extract complete data on hydraulic supports, roof delamination, borehole stress, and anchor cable stress from the working face within 1-3 coal mining cycles prior to the disaster, and record the time of the disaster, the extent of the disaster, the depth of roof collapse, and the shape of the collapse zone.

[0057] Step 6: Issue a warning signal when the danger probability index is greater than the preset warning threshold.

[0058] It should be noted that the preset threshold includes multi-level warning thresholds: a blue warning is issued when the danger probability index is greater than 0.6, a yellow warning is issued when it is greater than 0.75, an orange warning is issued when it is greater than 0.85, and a red warning is issued when it is greater than 0.95. When an early warning signal is issued, an audible and visual alarm is triggered at the monitoring center, and the early warning information is pushed through the industrial internet platform. The early warning information includes the risk level, disaster type, abnormal indicators, and recommended measures.

[0059] It should be noted that step 6 specifically includes: Based on the risk probability index output by the model and combined with historical case verification, multi-level warning thresholds are set (e.g., blue warning: probability > 0.6; yellow warning: probability > 0.75; orange warning: probability > 0.85; red warning: probability > 0.95).

[0060] The data collected and preprocessed in real time is input into the trained fusion model for online inference, which calculates the probability of danger in the current and future periods in real time.

[0061] When the probability value exceeds the threshold, the system automatically issues an audible and visual alarm at the monitoring center and pushes early warning information through the industrial internet platform. The information includes the risk level, the type of disaster that may occur, the main abnormal indicators, and recommended measures (such as "strengthen the support of the area from support XX to support XX").

[0062] In summary, the roof disaster early warning method for fully mechanized mining faces based on multi-source data proposed in this embodiment improves the utilization rate of underground coal mine data and the accuracy of roof disaster risk early warning through in-depth fusion analysis of multi-source data and dynamic threshold early warning.

[0063] Example 2 Figure 2 This is a structural diagram of a roof disaster early warning system for fully mechanized mining faces based on multi-source data, according to an embodiment of this application. Figure 2 As shown, the system includes: The data acquisition module 100 is used to acquire mining pressure data, environmental data and geological data of the fully mechanized mining face. The mining pressure data includes hydraulic support resistance data, roof delamination data and borehole stress data. The environmental data includes gas concentration data and water inflow data. The acquisition module 200 is used to identify the periodic pressure-increasing stage based on the hydraulic support resistance data and obtain pressure-increasing cycle information, wherein the pressure-increasing cycle includes the early pressure-increasing stage, the middle pressure-increasing stage, the late pressure-increasing stage, and the stable stage. Processing module 300 is used to process the hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data and water inflow data respectively, and extract multi-source feature data, wherein the multi-source feature data includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features and water inflow variation features. The fusion module 400 is used to associate and fuse the multi-source feature data with the pressure cycle information to generate a fused feature dataset; The identification module 500 is used to input the fused feature dataset into a pre-trained disaster risk identification model to obtain a disaster risk probability index; The early warning module 600 is used to issue an early warning signal when the danger probability index is greater than a preset early warning threshold.

[0064] In this embodiment of the disclosure, the processing module 300 is further configured to: Perform trend decomposition on the time series data of hydraulic support resistance to separate the trend term, periodic term and residual term; The start and end points of each pressure cycle are identified based on the cycle term, and the pressure step size and pressure duration are calculated. The long-term trend of the resistance of the hydraulic support at the working face is determined based on the trend terms. The residual items are used to filter out invalid data caused by sensor malfunctions or communication interruptions. Based on the identified pressure inflow cycle, the statistical characteristics within each pressure inflow cycle are calculated. The statistical characteristics include the maximum resistance value, average resistance, resistance non-uniformity coefficient, and dynamic load coefficient. Based on the hydraulic support resistance data after filtering out invalid data, the location of the point of application of the resultant force of the hydraulic support group at the working face at the same moment is calculated, and the high resistance zone and low resistance zone are identified.

[0065] In this embodiment of the disclosure, the processing module 300 is further configured to: Continuous monitoring data from shallow and deep roof delamination instruments are obtained. The shallow roof delamination data reflects the deformation of the shallow loosened zone of the surrounding rock, and the deep roof delamination data reflects the settlement of the overlying key layer. Differential operations were performed on shallow and deep roof delamination time series data to calculate the delamination rate, and a second differential operation was performed on the delamination rate to calculate the delamination acceleration, which was used to evaluate the dynamic trend and acceleration characteristics of roof delamination development. Based on the delamination data of the roof at different locations of the working face, a spatial interpolation method is used to construct a simulated field of roof delamination data of the working face, calculate the correlation coefficient between delamination data at different locations, and analyze the spatial coordination of roof delamination to determine whether roof fracture occurs synchronously. Based on the changing trends of shallow and deep delamination amounts, a time series curve of the ratio of shallow to deep delamination amounts is established. When the shallow delamination amount increases faster than the deep delamination amount, it is judged as a shallow loosening zone of the surrounding rock. When the deep delamination amount increases faster than the shallow delamination amount, it is judged as instability of the overlying key layer. This is used to identify potential modes of roof instability. Based on the delamination rate and delamination acceleration, abrupt changes in delamination amount are identified. When the delamination rate exceeds a preset threshold, it is marked as a delamination anomaly event, which is used to provide early warning of rapid expansion of roof delamination.

[0066] In this embodiment of the disclosure, the processing module 300 is further configured to: Data from borehole stress gauges distributed at different locations on the working face are obtained. The borehole stress data includes the pressure distribution data of the advance support pressure in front of the mining area, the lateral support pressure, and the stress distribution data behind the goaf. Based on borehole stress data at different locations on the working face, numerical interpolation is used to invert the stress field distribution in the mining area and generate a stress contour map of the mining area, which is used to assess the overall stress state of the mining area and the location of the stress peak area. The stress gradient is obtained by calculating the ratio of the stress difference between adjacent measuring points to the distance between measuring points. The gradient is divided into low gradient, medium gradient and high gradient regions according to the magnitude of the gradient change, which is used to identify stress change zones and potential failure areas. Combining geological structure distribution data, working face and roadway layout, a stress propagation path analysis algorithm is used to calculate the propagation path of stress from the mining area to the surrounding coal and rock mass, identify stress concentration areas affected by geological structures, and determine the degree of influence of geological structures such as faults and folds on the stress in the mining area. Establish a borehole stress time-series variation curve, calculate the stress change rate, and mark it as a stress anomaly event when the stress change rate exceeds a preset threshold, which is used to provide early warning of sudden stress changes in the mining area.

[0067] In this embodiment of the disclosure, the processing module 300 is further configured to: Data from gas sensor monitoring points arranged in the fully mechanized mining face and roadway are acquired, and the gas concentration data includes continuous concentration monitoring values ​​over multiple complete production cycles. Calculate the maximum, minimum, average, variation range, and variance of gas concentration throughout the complete production cycle to assess the overall level and fluctuation of gas emission. Sliding window analysis is performed on the time series data of gas concentration to calculate the rate of change, fluctuation, and duration of concentration exceeding the limit in each window. When the rate of change exceeds a preset threshold, it is marked as an abnormal gas outburst event, which is used to identify the precursor characteristics of abnormal gas outbursts. By combining the spatial distribution of sensors, a spatial interpolation method is used to construct a model of the gas concentration change trend at the working face, generate a gas concentration contour map, analyze the direction of the gas concentration gradient, and infer the location of the gas emission source, which is used to analyze the spatial evolution law of gas concentration as the working face advances. By coupling gas concentration time-series data with the working face advance, a gas concentration-advancement relationship curve is established. The periodic variation law of gas concentration with the mining process is analyzed to identify the correlation between abnormal gas outbursts and roof pressure.

[0068] In this embodiment of the disclosure, the processing module 300 is further configured to: Acquire water inflow monitoring data of the longwall mining face during multiple complete production cycles, including instantaneous flow rate, cumulative flow rate, and water quality sensor data; Sliding window analysis is performed on instantaneous flow time series data to calculate the instantaneous flow rate change rate and fluctuation coefficient, identify abrupt changes and abnormal fluctuations in instantaneous flow, and capture sudden changes in water inflow. Trend fitting is performed on the cumulative flow time series data to establish the cumulative flow growth curve and calculate the cumulative flow growth rate. When the cumulative flow growth rate exceeds the preset threshold, it is marked as an abnormal water inflow event to identify the trend of continuous increase in water inflow. By combining water quality sensor data, changes in water quality parameters are analyzed. When water quality parameters change abruptly, they are marked as water source anomalies to determine whether the source of the water inflow has changed. The water inflow data is coupled with the working face advance to establish a water inflow-advance relationship curve. The change in water inflow per unit advance distance is analyzed to identify the evolution of water inflow during the mining process. Establish a time-series prediction model for water inflow, predict future water inflow trends based on historical water inflow data, and mark water inflow risk events when the predicted value exceeds a preset threshold for early warning of water inflow disasters.

[0069] In this embodiment of the disclosure, the fusion module 400 is further configured to: Establish an index structure with the complete advance cycle of the working face as the unit of time and the sensor position as the unit of space; The extracted multi-source feature data are correlated and compared according to the stage labels of the early stage of pressure, the middle stage of pressure, the late stage of pressure and the stable stage, to generate fused feature data containing temporal features, spatial distribution features and pressure stage labels.

[0070] It should be noted that the training process of the disaster risk identification model includes: Acquire historical data, including historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, historical water inflow data, geological data, and historical roof disaster case data; Based on the historical hydraulic support resistance data, identify the historical periodic pressure phases to obtain historical pressure cycle information; The historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, and historical water inflow data are respectively subjected to in-depth processing to extract historical multi-source feature data; A unified temporal and spatial index is established to associate and fuse the historical multi-source feature data with the historical pressure cycle information to generate a historical fused feature dataset, and disaster precursor labels are marked on the data samples before the disaster occurs; A strategy combining feature importance ranking and recursive feature elimination is used to select a subset of key features from the historical fused feature dataset; A hybrid learning model is trained using the aforementioned key feature subset. The hybrid learning model includes a time-series learning module and a state discrimination module. It is trained using historical normal data and case data labeled with disaster precursors, so that the model outputs a disaster risk probability index between 0 and 1.

[0071] It should be noted that the preset threshold includes multi-level warning thresholds: a blue warning is issued when the danger probability index is greater than 0.6, a yellow warning is issued when it is greater than 0.75, an orange warning is issued when it is greater than 0.85, and a red warning is issued when it is greater than 0.95. When an early warning signal is issued, an audible and visual alarm is triggered at the monitoring center, and the early warning information is pushed through the industrial internet platform. The early warning information includes the risk level, disaster type, abnormal indicators, and recommended measures.

[0072] In summary, the roof disaster early warning system for fully mechanized mining faces proposed in this embodiment improves the utilization rate of underground coal mine data and the accuracy of roof disaster risk early warning through in-depth fusion analysis of multi-source data and dynamic threshold early warning.

[0073] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0074] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0075] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for early warning of roof disasters in fully mechanized mining faces based on multi-source data, characterized in that, The method includes: Collect mining pressure data, environmental data, and geological data from the fully mechanized mining face. The mining pressure data includes hydraulic support resistance data, roof delamination data, and borehole stress data. The environmental data includes gas concentration data and water inflow data. Based on the hydraulic support resistance data, the periodic pressure-increasing phase is identified to obtain pressure-increasing cycle information, which includes the early pressure-increasing phase, the middle pressure-increasing phase, the late pressure-increasing phase, and the stable phase. The hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data, and water inflow data are processed respectively to extract multi-source feature data, which includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features, and water inflow variation features. The multi-source feature data is correlated and fused with the pressure cycle information to generate a fused feature dataset; The fused feature dataset is input into a pre-trained disaster risk identification model to obtain a disaster risk probability index; A warning signal is issued when the danger probability index is greater than a preset warning threshold.

2. The method as described in claim 1, characterized in that, Processing the hydraulic support resistance data includes: Perform trend decomposition on the time series data of hydraulic support resistance to separate the trend term, periodic term and residual term; The start and end points of each pressure cycle are identified based on the cycle term, and the pressure step size and pressure duration are calculated. The long-term trend of the resistance of the hydraulic support at the working face is determined based on the trend terms. The residual items are used to filter out invalid data caused by sensor malfunctions or communication interruptions. Based on the identified pressure inflow cycle, the statistical characteristics within each pressure inflow cycle are calculated. The statistical characteristics include the maximum resistance value, average resistance, resistance non-uniformity coefficient, and dynamic load coefficient. Based on the hydraulic support resistance data after filtering out invalid data, the location of the point of application of the resultant force of the hydraulic support group at the working face at the same moment is calculated, and the high resistance zone and low resistance zone are identified.

3. The method as described in claim 1, characterized in that, Processing the delamination data of the roof slab includes: Continuous monitoring data from shallow and deep roof delamination instruments are obtained. The shallow roof delamination data reflects the deformation of the shallow loosened zone of the surrounding rock, and the deep roof delamination data reflects the settlement of the overlying key layer. Differential operations were performed on shallow and deep roof delamination time series data to calculate the delamination rate, and a second differential operation was performed on the delamination rate to calculate the delamination acceleration, which was used to evaluate the dynamic trend and acceleration characteristics of roof delamination development. Based on the delamination data of the roof at different locations of the working face, a spatial interpolation method is used to construct a simulated field of roof delamination data of the working face, calculate the correlation coefficient between delamination data at different locations, and analyze the spatial coordination of roof delamination to determine whether roof fracture occurs synchronously. Based on the changing trends of shallow and deep delamination amounts, a time series curve of the ratio of shallow to deep delamination amounts is established. When the shallow delamination amount increases faster than the deep delamination amount, it is judged as a shallow loosening zone of the surrounding rock. When the deep delamination amount increases faster than the shallow delamination amount, it is judged as instability of the overlying key layer. This is used to identify potential modes of roof instability. Based on the delamination rate and delamination acceleration, abrupt changes in delamination amount are identified. When the delamination rate exceeds a preset threshold, it is marked as a delamination anomaly event, which is used to provide early warning of rapid expansion of roof delamination.

4. The method as described in claim 1, characterized in that, Processing borehole stress data includes: Data from borehole stress gauges distributed at different locations on the working face are obtained. The borehole stress data includes the pressure distribution data of the advance support pressure in front of the mining area, the lateral support pressure, and the stress distribution data behind the goaf. Based on borehole stress data at different locations on the working face, numerical interpolation is used to invert the stress field distribution in the mining area and generate a stress contour map of the mining area, which is used to assess the overall stress state of the mining area and the location of the stress peak area. The stress gradient is obtained by calculating the ratio of the stress difference between adjacent measuring points to the distance between measuring points. The gradient is divided into low gradient, medium gradient and high gradient regions according to the magnitude of the gradient change, which is used to identify stress change zones and potential failure areas. Combining geological structure distribution data, working face and roadway layout, a stress propagation path analysis algorithm is used to calculate the propagation path of stress from the mining area to the surrounding coal and rock mass, identify stress concentration areas affected by geological structures, and determine the degree of influence of geological structures such as faults and folds on the stress in the mining area. Establish a borehole stress time-series variation curve, calculate the stress change rate, and mark it as a stress anomaly event when the stress change rate exceeds a preset threshold, which is used to provide early warning of sudden stress changes in the mining area.

5. The method as described in claim 1, characterized in that, Processing gas concentration data includes: Data from gas sensor monitoring points arranged in the fully mechanized mining face and roadway are acquired, and the gas concentration data includes continuous concentration monitoring values ​​over multiple complete production cycles. Calculate the maximum, minimum, average, variation range, and variance of gas concentration throughout the complete production cycle to assess the overall level and fluctuation of gas emission. Sliding window analysis is performed on the time series data of gas concentration to calculate the rate of change, fluctuation, and duration of concentration exceeding the limit in each window. When the rate of change exceeds a preset threshold, it is marked as an abnormal gas outburst event, which is used to identify the precursor characteristics of abnormal gas outbursts. By combining the spatial distribution of sensors, a spatial interpolation method is used to construct a model of the gas concentration change trend at the working face, generate a gas concentration contour map, analyze the direction of the gas concentration gradient, and infer the location of the gas emission source, which is used to analyze the spatial evolution law of gas concentration as the working face advances. By coupling gas concentration time-series data with the working face advance, a gas concentration-advancement relationship curve is established. The periodic variation law of gas concentration with the mining process is analyzed to identify the correlation between abnormal gas outbursts and roof pressure.

6. The method as described in claim 1, characterized in that, The processing of the water inflow data includes: Acquire water inflow monitoring data of the longwall mining face during multiple complete production cycles, including instantaneous flow rate, cumulative flow rate, and water quality sensor data; Sliding window analysis is performed on instantaneous flow time series data to calculate the instantaneous flow rate change rate and fluctuation coefficient, identify abrupt changes and abnormal fluctuations in instantaneous flow, and capture sudden changes in water inflow. Trend fitting is performed on the cumulative flow time series data to establish the cumulative flow growth curve and calculate the cumulative flow growth rate. When the cumulative flow growth rate exceeds the preset threshold, it is marked as an abnormal water inflow event to identify the trend of continuous increase in water inflow. By combining water quality sensor data, changes in water quality parameters are analyzed. When water quality parameters change abruptly, they are marked as water source anomalies to determine whether the source of the water inflow has changed. The water inflow data is coupled with the working face advance to establish a water inflow-advance relationship curve. The change in water inflow per unit advance distance is analyzed to identify the evolution of water inflow during the mining process. Establish a time-series prediction model for water inflow, predict future water inflow trends based on historical water inflow data, and mark water inflow risk events when the predicted value exceeds a preset threshold for early warning of water inflow disasters.

7. The method as described in claim 1, characterized in that, The step of associating and fusing the multi-source feature data with the pressure cycle information to generate a fused feature dataset includes: Establish an index structure with the complete advance cycle of the working face as the unit of time and the sensor position as the unit of space; The extracted multi-source feature data are correlated and compared according to the stage labels of the early stage of pressure, the middle stage of pressure, the late stage of pressure and the stable stage, to generate fused feature data containing temporal features, spatial distribution features and pressure stage labels.

8. The method as described in claim 1, characterized in that, The training process of the disaster risk identification model includes: Acquire historical data, including historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, historical water inflow data, geological data, and historical roof disaster case data; Based on the historical hydraulic support resistance data, identify the historical periodic pressure phases to obtain historical pressure cycle information; The historical hydraulic support resistance data, historical roof delamination data, historical borehole stress data, historical gas concentration data, and historical water inflow data are respectively subjected to in-depth processing to extract historical multi-source feature data; A unified temporal and spatial index is established to associate and fuse the historical multi-source feature data with the historical pressure cycle information to generate a historical fused feature dataset, and disaster precursor labels are marked on the data samples before the disaster occurs; A strategy combining feature importance ranking and recursive feature elimination is used to select a subset of key features from the historical fused feature dataset; A hybrid learning model is trained using the aforementioned key feature subset. The hybrid learning model includes a time-series learning module and a state discrimination module. It is trained using historical normal data and case data labeled with disaster precursors, so that the model outputs a disaster risk probability index between 0 and 1.

9. The method as described in claim 1, characterized in that, The preset thresholds include multi-level warning thresholds: a blue warning is issued when the danger probability index is greater than 0.6, a yellow warning is issued when it is greater than 0.75, an orange warning is issued when it is greater than 0.85, and a red warning is issued when it is greater than 0.

95. When an early warning signal is issued, an audible and visual alarm is triggered at the monitoring center, and the early warning information is pushed through the industrial internet platform. The early warning information includes the risk level, disaster type, abnormal indicators, and recommended measures.

10. A roof disaster early warning system for fully mechanized mining faces based on multi-source data, characterized in that, The system includes: The data acquisition module is used to collect mining pressure data, environmental data, and geological data of the fully mechanized mining face. The mining pressure data includes hydraulic support resistance data, roof delamination data, and borehole stress data. The environmental data includes gas concentration data and water inflow data. The acquisition module is used to identify the periodic pressure-increasing phase based on the hydraulic support resistance data and obtain pressure-increasing cycle information, wherein the pressure-increasing cycle includes the early pressure-increasing phase, the middle pressure-increasing phase, the late pressure-increasing phase, and the stable phase. The processing module is used to process the hydraulic support resistance data, roof delamination data, borehole stress data, gas concentration data and water inflow data respectively, and extract multi-source feature data, wherein the multi-source feature data includes pressure cycle features, delamination dynamic features, stress distribution features, gas emission features and water inflow variation features. The fusion module is used to associate and fuse the multi-source feature data with the pressure cycle information to generate a fused feature dataset; The identification module is used to input the fused feature dataset into a pre-trained disaster risk identification model to obtain a disaster risk probability index; The early warning module is used to issue an early warning signal when the danger probability index is greater than a preset early warning threshold.

Images