Deep well thick coal seam gob-side entry retaining overburden deformation dynamic prediction method

Abstract

The application discloses a kind of deep well thick coal seam gob-side entry retaining overburden deformation dynamic prediction methods, it is related to coal mining technical field, comprising: through the multiple door type support of deployment in gob-side entry retaining, multiple source bearing data are collected;Based on the construction step of artificial support structure, the multiple source bearing data are grouped and divided, and feature extraction is carried out based on grouping and division result, obtains overburden movement state vector group;Combining the historical overburden movement state vector group of last control cycle, overburden deformation identification based on space consistency is carried out to the overburden movement state vector group;According to overburden deformation identification result, overburden deformation prediction is carried out in combination with the prior lithology data of target entry retaining area, obtains overburden deformation prediction result and feedback.The technical problem that the technical problem that the overburden deformation identification precision is insufficient and false alarm rate is high due to the deformation of overburden abnormal movement is ignored in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of coal mining technology, specifically to a method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf. Background Technology

[0002] Goaf retention is a method of eliminating or significantly reducing coal pillars in certain sections through scientific roadway layout and support, enabling continuous pillarless mining between working faces, thereby improving coal resource recovery and reducing resource waste. In deep, thick coal seams, due to intense overburden movement and high ground stress levels, the deformation of the surrounding rock in goaf retention becomes increasingly severe. Portal supports are the roadway support devices. Goaf retention undergoes different service stages, including installation, load bearing, concrete curing, and dismantling, with significant differences in the overburden load transfer mechanism and deformation characteristics at each stage.

[0003] However, traditional methods for monitoring the deformation of overburden along the goaf lack the dynamic characteristics of different support service stages. In the identification of deformation anomalies, they ignore the deformation caused by abnormal movement of overburden and are prone to misjudging inherent differences in lithology as abnormal deformation. Summary of the Invention

[0004] This invention provides a dynamic prediction method for overburden deformation along the goaf in deep, thick coal seams, which solves the technical problem in the prior art that the deformation recognition accuracy of overburden is insufficient and the false alarm rate is high due to the neglect of abnormal overburden movement.

[0005] This invention provides a method for dynamically predicting the deformation of overburden in deep, thick coal seams along the goaf, the method comprising: Multi-source load data is collected through an IoT monitoring network deployed on multiple portal supports along the goaf. Based on the construction steps of artificially supported structures, the multi-source bearing data are grouped and divided, and feature extraction is performed based on the grouping results to obtain the overburden movement state vector group. Combining the historical overburden movement state vector set from the previous control cycle, the overburden movement state vector set is used to identify overburden deformation based on spatial consistency. Based on the overburden deformation identification results, combined with the prior lithological data of the target roadway area, the overburden deformation is predicted, and the prediction results are obtained and fed back.

[0006] Optionally, multi-source load data can be collected through an IoT monitoring network deployed on multiple portal frames along the goaf, including: Sensor units are deployed on multiple portal supports arranged along the direction of the empty lane to construct an Internet of Things monitoring network covering the empty lane area; The IoT monitoring network collects multi-source monitoring data of each of the portal supports during the overburden bearing process, wherein the multi-source monitoring data includes at least one of pressure data, displacement data, and attitude data; The multi-source monitoring data is preprocessed and time-synchronized to obtain the multi-source carrying data.

[0007] Optionally, sensing units are deployed on multiple portal frames arranged along the direction of the tunnel to construct an Internet of Things (IoT) monitoring network covering the tunnel area, including: A first pressure sensing unit for collecting load data of the overburden acting on the top beam is deployed on the bearing surface of the top beam of the portal frame. A second pressure sensing unit is deployed in each column of the gantry frame to collect data on the axial force borne by each column. An attitude sensing unit for collecting tilt angle data of the top beam is deployed on the top beam portion of the portal frame. A displacement sensing unit for collecting displacement data of the base plate is deployed in the base portion of the portal frame. A ranging sensing unit for collecting distance change data between the support and a fixed reference point is deployed on the gantry frame; Each of the aforementioned sensing units is connected to the mine wired transmission network via a mine wireless communication module to construct the Internet of Things monitoring network.

[0008] Optionally, based on the construction progress of artificially supported structures, the multi-source load-bearing data is grouped and divided, including: Obtain the individual dimensional parameters and construction plan of the artificial support structure; Based on the individual unit size parameters, determine the number of portal frames required to support a single artificial support structure, and divide the number of portal frames corresponding to that number into one measuring station; The construction step is determined according to the construction sequence in the construction plan. The construction step is used to characterize the construction time interval of individual artificial support structures between adjacent stations. Based on the multiple stations obtained and the construction steps, the multi-source carrier data is divided into multiple sets of station carrier data corresponding to each station.

[0009] Optionally, feature extraction is performed based on the grouping results to obtain a set of overburden motion state vectors, including: Based on prior knowledge of overburden deformation monitoring, a list of key features is determined, which includes at least load-related features, displacement-related features, and correlation-coupling-related features. Based on the grouping results, a subset of the key feature list corresponding to each station is matched in the key feature list; By combining a subset of key feature lists and setting a sliding window, multi-parameter correlation features of stations are extracted from the station bearing data of the grouping results to generate overburden movement state vectors; By traversing multiple stations, the overburden motion state vectors of each station are summarized into the overburden motion state vector group.

[0010] Optionally, based on the grouping results, matching a subset of the key feature list corresponding to each station in the key feature list further includes: Based on the grouping results, extract the station carrying data corresponding to each station; The data carried by the station is subjected to stage identification based on preset rules to obtain the first stage identification result; The station load data is subjected to stage identification based on data features to obtain the second stage identification result. The data features include at least two combinations of load level, load change trend, displacement level, displacement change rate, load fluctuation amplitude, and off-center load coefficient. Change point detection is performed on the data carried by the station to obtain the stage transition auxiliary calibration results; By integrating the first-stage identification results, the second-stage identification results, and the stage transition auxiliary calibration results, the current support service stage of the station is determined. Based on the service stage of the support structure, the corresponding key features of the stage are matched and extracted from the key feature list, and the output is a subset of the key feature list.

[0011] Optionally, combining the historical overburden movement state vector set from the previous control cycle, overburden deformation identification based on spatial consistency is performed on the overburden movement state vector set, including: Taking the current control cycle as the scope of analysis, according to the support service stage, the overburden movement state vector of each station in the overburden movement state vector group is matched and aligned with the historical overburden movement state vector of the stations in the same support service stage in the historical overburden movement state vector group of the previous control cycle. Based on the matching and alignment results, a multi-dimensional comparison analysis is performed, wherein the multi-dimensional comparison includes at least a combination of two of the following: feature value level comparison, feature change trend comparison, feature fluctuation pattern comparison, and correlation consistency comparison. The results of multi-dimensional comparative analysis are fused and calculated to obtain the station phase consistency index for each station in the corresponding support service phase. By integrating the station-stage consistency indicators of multiple stations under the current control cycle, a spatial consistency map is generated; Extract a preset number of historical spatial consistency maps from previous control cycles and serialize them to form a historical spatial consistency sequence; By combining the historical spatial consistency sequence and the spatial consistency map, trend analysis and deformation identification are performed to locate abnormal overburden deformation stations and normal overburden deformation stations, and the output is the overburden deformation identification result.

[0012] Optionally, based on the overburden deformation identification results, combined with prior lithological data of the target roadway area, overburden deformation prediction is performed, and the overburden deformation prediction results are obtained and fed back, including: For the abnormal overburden deformation stations in the overburden deformation identification results, the spatial consistency map is corrected for anomalies by combining the lithological calibration parameters in the prior lithological data. Based on the spatial consistency map after anomaly correction, trend analysis and deformation identification are performed again, and pseudo-anomaly stations caused by lithological differences are removed. The removed pseudo-anomaly stations are then remarked as normal overburden deformation stations. For the corrected abnormal overburden deformation stations, the most unfavorable trend information of each comparison dimension is extracted and the trend is extrapolated respectively; The trend extrapolation results are weighted by combining the lithological weight coefficients in the prior lithological data to obtain the first overburden deformation prediction results, and the results are fed back in real time.

[0013] Optionally, based on the overburden deformation identification results and combined with prior lithological data of the target roadway area, overburden deformation prediction is performed, and the overburden deformation prediction results are obtained and fed back. This also includes: For the normal overburden deformation stations in the overburden deformation identification results, combined with the historical spatial consistency sequence, regression extrapolation based on confidence constraints is performed to obtain the second overburden deformation prediction results; The second overburden deformation prediction result is periodically fed back according to the preset feedback cycle.

[0014] Optionally, based on the lithological calibration parameters in the prior lithological data, anomaly correction is performed on the spatial consistency map, including: The prior lithological data includes a lithological deformation adjustment coefficient matrix established based on historical monitoring data and statistical analysis; Based on the lithological deformation adjustment coefficient matrix, the station stage consistency index is re-compared and calculated, and the spatial consistency map is updated accordingly. The lithological deformation adjustment coefficient matrix represents the mapping relationship of overburden deformation between stations with different lithological conditions that are in the same service phase during adjacent control cycles.

[0015] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a method for dynamic prediction of overburden deformation in deep, thick coal seams with roadways along the goaf. First, multi-source bearing capacity data is collected and preprocessed and synchronized to provide reliable foundational data for subsequent feature extraction and deformation identification. Second, the construction steps of artificial support structures are used for segmentation and feature extraction to identify the service stages of the supports, overcoming the deficiency of insufficient feature extraction accuracy due to ignoring differences in service stages. Third, spatial consistency is compared using historical overburden movement state vector groups to generate a spatial consistency map, enabling the detection of overburden deformation. Furthermore, an anomaly correction is performed on the spatial consistency map using a lithological deformation adjustment coefficient matrix established through statistical analysis of historical monitoring data, reducing the false alarm rate of deformation identification.

[0016] In summary, a technical solution was established that integrates multi-source bearing data acquisition, grouping and feature extraction, spatial consistency identification, lithology correction, and hierarchical prediction. For abnormal stations, the most unfavorable trend information was extracted, extrapolated, and fed back in real time. For normal stations, regression extrapolation was performed based on confidence constraints, and feedback was provided periodically. This solution enables multi-dimensional fusion prediction of overburden deformation and improves the reliability of safe production in deep, thick coal seams with roadways along the goaf. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for dynamically predicting the deformation of overburden in deep, thick coal seams along the goaf, provided by the present invention. Figure 2 This is a logical schematic diagram of a method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf provided by the present invention. Figure 3 This is a table showing the correspondence between different lithological sections and lithological deformation adjustment coefficients in a dynamic prediction method for overburden deformation along roadways in deep, thick coal seams provided by this invention. Detailed Implementation

[0018] This invention provides a dynamic prediction method for overburden deformation along the goaf in deep, thick coal seams, which solves the technical problem in the prior art that the deformation recognition accuracy of overburden is insufficient and the false alarm rate is high due to the neglect of abnormal overburden movement.

[0019] The present invention will now be described in detail with reference to the accompanying drawings.

[0020] In one embodiment, such as Figure 1 , Figure 2 As shown, this invention provides a method for dynamic prediction of overburden deformation in deep, thick coal seams with roadways along the goaf, the method comprising: S100: Collects multi-source load data through an IoT monitoring network deployed on multiple portal supports along the goaf. In this embodiment of the invention, the portal frame refers to the support structure used to support the overburden load in the goaf retention roadway; goaf retention roadway is a mining engineering technology that maintains the original mining roadway along the edge of the goaf after the coal mining face and realizes pillarless mining.

[0021] Specifically, multiple sensing units are deployed on each portal frame support. Each sensing unit is connected to the mine's wired transmission network via a mine-use wireless communication module, constructing an IoT monitoring network covering the entire roadway area. Through the sensing units within the IoT monitoring network, multi-source monitoring data of the portal frame supports during the overburden bearing process are collected in real time. The collected raw data is preprocessed and time-synchronized for calibration to obtain multi-source bearing data.

[0022] Step S100 in the method of this embodiment of the invention includes: Sensor units are deployed on multiple portal supports arranged along the direction of the empty lane to construct an Internet of Things monitoring network covering the empty lane area; The IoT monitoring network collects multi-source monitoring data of each of the portal supports during the overburden bearing process, wherein the multi-source monitoring data includes at least one of pressure data, displacement data, and attitude data; The multi-source monitoring data is preprocessed and time-synchronized to obtain the multi-source carrying data.

[0023] In this embodiment of the invention, during the engineering implementation of the goaf retention roadway in a deep, thick coal seam, multiple sets of portal supports are sequentially arranged along the goaf retention roadway direction, and each set of supports is rotated and put into service as the working face advances.

[0024] First, sensor units are deployed on multiple portal frames arranged along the goaf-retention tunnel direction to construct an Internet of Things (IoT) monitoring network covering the goaf-retention area. The sensor units are physical quantity sensing devices deployed at key locations on the portal frames, used to measure physical quantities such as pressure, displacement, angle, and distance.

[0025] Specifically, a first pressure sensing unit is deployed on the bearing surface of the top beam of the portal frame to collect load data of the overburden acting on the top beam; a second pressure sensing unit is deployed on each column to collect axial force data of each column; an attitude sensing unit is deployed on the top beam to collect tilt angle data of the top beam; a displacement sensing unit is deployed on the base to collect displacement data of the base plate; and a distance measuring sensing unit is deployed on the support body to collect distance change data between the support and a fixed reference point. Each sensing unit is connected to the mine's wired transmission network via mine wireless communication equipment, constructing an Internet of Things (IoT) monitoring network covering the entire tunnel area.

[0026] Secondly, the overburden bearing process refers to the complete time process from when the gantry crane bears the initial overburden load to when the load reaches a stable state or changes. During the service life of the gantry crane, the overburden load is transmitted to the crane through the top beam, and various parts of the crane generate corresponding mechanical and deformation responses. The load borne by the top beam changes with the movement of the overburden, the axial force of the column changes with the load distribution, the tilt angle of the top beam changes with eccentric loading or uneven deformation, the displacement of the base plate changes with the degree of bottom bulging, and the distance between the crane and the fixed reference point changes with the overall movement of the crane.

[0027] Therefore, through an IoT monitoring network, multi-source monitoring data of each portal frame support during the overburden bearing process are collected in real time at a preset sampling frequency, such as 10Hz. The multi-source monitoring data includes at least one of pressure data, displacement data, and attitude data; preferably, all three types of data are collected simultaneously to achieve multi-parameter collaborative monitoring.

[0028] Finally, the multi-source monitoring data is preprocessed and time-synchronized for calibration to obtain multi-source bearer data. Preprocessing includes outlier detection and removal, missing value imputation, and noise reduction. Outlier detection and removal identify and remove values ​​that significantly deviate from the normal range due to sensor malfunction or external interference; missing data caused by transmission packet loss is reasonably estimated and filled in through missing value imputation; and random noise in the signal is eliminated through noise reduction.

[0029] Specifically, during preprocessing, wavelet thresholding is used to eliminate random noise in the signal. For example, the db4 wavelet is used to decompose the random noise into five levels. The threshold coefficient for wavelet denoising needs to be preset and can be obtained using a general threshold calculation formula. Outliers are identified by analyzing the numerical range of multi-source monitoring data in the physical constraint rule engine, and values ​​exceeding the range are removed, retaining only normal multi-source monitoring data. A KNN imputation method based on Pearson correlation between adjacent stations is used to estimate and fill in missing data. The significance threshold for the Pearson correlation coefficient is preset to 0.7. Only when the absolute value of the correlation coefficient between two adjacent stations on historical data of the same period is greater than 0.7 is the two considered sufficiently similar, allowing the use of each other's data for missing value imputation. During imputation, the number of neighbors for KNN imputation is set to K=3, and the top three adjacent stations with the highest correlation to the target station are selected as imputation reference sources.

[0030] In a preferred embodiment, if the difference between the current stress state of the interpolation reference source and the target station exceeds a preset range, interpolation is abandoned and the data for that time period is marked as missing. Simultaneously, time synchronization calibration is performed on the multi-source monitoring data. Using the ground master clock as a reference, the data is transmitted step-by-step to the downhole master station clock via the PTP / NTP protocol, and then synchronized to the local memory of each sensor node via a wireless gateway / substation using wireless beacon time synchronization, ensuring that all data points are marked with a unified timestamp. After preprocessing and time synchronization calibration, high-quality multi-source data is obtained.

[0031] In step S100 of the method of this embodiment of the invention, sensing units are deployed on multiple portal supports arranged along the direction of the empty lane to construct an Internet of Things monitoring network covering the lane area, including: A first pressure sensing unit for collecting load data of the overburden acting on the top beam is deployed on the bearing surface of the top beam of the portal frame. A second pressure sensing unit is deployed in each column of the gantry frame to collect data on the axial force borne by each column. An attitude sensing unit for collecting tilt angle data of the top beam is deployed on the top beam portion of the portal frame. A displacement sensing unit for collecting displacement data of the base plate is deployed in the base portion of the portal frame. A ranging sensing unit for collecting distance change data between the support and a fixed reference point is deployed on the gantry frame; Each of the aforementioned sensing units is connected to the mine wired transmission network via a mine wireless communication module to construct the Internet of Things monitoring network.

[0032] In this embodiment of the invention, during the engineering implementation of gob-side roadway retention in deep, thick coal seams, multiple sets of portal supports are sequentially deployed along the gob-side roadway direction, with each set of supports rotating in service as the working face advances. Therefore, it is necessary to deploy various types of sensing units on each set of portal supports deployed along the gob-side roadway direction.

[0033] First, a first pressure sensing unit is deployed on the bearing surface of the top beam of the portal frame to collect load data on the overburden acting on the top beam. This first pressure sensing unit is specifically deployed in the central area and key stress-bearing locations on both sides of the bearing surface of the top beam, and is used to collect load data on the overburden acting on the top beam. The sensing unit includes at least a pressure gauge, a displacement sensor, and an inclinometer. The portal frame adopts a two-column supported integral crossbeam structure, with its top beam directly contacting the roadway roof and bearing the entire load transferred from the overburden.

[0034] Secondly, the axial force data of the gantry frame directly reflects the stress state of each support point. Second pressure sensing units are deployed in each column of the gantry frame to collect the axial force data of each column.

[0035] Secondly, since the top beam of the portal frame may tilt, this tilting could lead to uneven settlement of the top plate, or even a decrease in the load-bearing capacity of the frame and instability. Therefore, an attitude sensing unit is deployed on the top beam of the portal frame to collect tilt angle data.

[0036] Furthermore, due to mining activities, the tunnel floor may experience heave or subsidence deformation, which directly affects the stability and support effectiveness of the gantry crane. Therefore, displacement sensing units are deployed in the base of the gantry crane to collect floor displacement data.

[0037] Furthermore, since the portal frame is subjected to vertical overburden loads in the goaf-side roadway, horizontal displacement along the roadway's strike or dip may occur. Therefore, a distance measuring sensor unit is deployed on the portal frame to collect data on the distance changes between the frame and a fixed reference point.

[0038] Finally, each sensing unit is connected to the mine's wired transmission network via a mining wireless communication module such as explosion-proof Wi-Fi to build an Internet of Things (IoT) monitoring network.

[0039] In this embodiment of the invention, an IoT monitoring network on multiple portal supports enables synchronous acquisition and time synchronization calibration of multi-source data. Subsequently, each sensing unit connects to the mine's wired transmission network via a mine-use wireless communication module, constructing an IoT monitoring network covering the tunnel area. This adapts to the relocation requirements of support rotation and provides basic data for subsequent operations.

[0040] S200: Based on the construction steps of artificial support structures, the multi-source bearing data is grouped and divided, and feature extraction is performed based on the grouping results to obtain the overburden movement state vector group; In this embodiment of the invention, the artificial support structure is an artificial structure used to replace the coal pillar in the goaf-retaining roadway, which is usually a flexible concrete wall; the construction step is the construction progress rhythm of the individual artificial support structure along the goaf-retaining roadway.

[0041] Specifically, since the size of an artificially supported structure may not be supported by a single support frame, but rather by multiple supports for a single structure, it is necessary to divide the multi-source load data into multiple sets of load data corresponding to different monitoring stations.

[0042] First, the individual dimensional parameters and construction plan of the artificially supported structure are obtained. Based on the individual dimensional parameters, the number of portal frames required to support each artificially supported structure is determined, and each corresponding number of portal frames is designated as a monitoring station. The construction step is determined according to the construction sequence in the construction plan. Based on the multiple monitoring stations and construction steps obtained, the multi-source bearing data is divided into multiple sets of station bearing data corresponding to each station. According to the grouping results, a subset of the key feature list corresponding to each station is matched in the key feature list. A sliding window is set based on this subset, and multi-parameter correlation features of the stations are extracted from the station bearing data of the grouping results to generate overburden motion state vectors for each station. After traversing all stations, the vectors are summarized into an overburden motion state vector group.

[0043] Step S200 in the method of this embodiment of the invention includes: Obtain the individual dimensional parameters and construction plan of the artificial support structure; Based on the individual unit size parameters, determine the number of portal frames required to support a single artificial support structure, and divide the number of portal frames corresponding to that number into one measuring station; The construction step is determined according to the construction sequence in the construction plan. The construction step is used to characterize the construction time interval of individual artificial support structures between adjacent stations. Based on the multiple stations obtained and the construction steps, the multi-source carrier data is divided into multiple sets of station carrier data corresponding to each station.

[0044] In this embodiment of the invention, firstly, since the artificial support structure is constructed segment by segment as the working face advances, each segment is a single unit. The dimensional parameters of these units may differ between different mines and working faces; therefore, it is necessary to obtain the dimensional parameters of the individual artificial support structures and the construction plan. The dimensional parameters of a single artificial support structure refer to its geometric dimensions along the goaf-retention direction, perpendicular to the goaf-retention direction, and in the height direction. The length of the unit along the goaf-retention direction determines the number of portal frames required to support it. The construction plan is a construction organization design document for the artificial support structure, which may include, for example, the construction sequence along the goaf-retention direction, the construction time for each unit, and the construction sequence relationship between adjacent units.

[0045] Secondly, since the portal frames are evenly spaced along the goaf direction, and the spacing between adjacent frames is fixed, the number of portal frames required to support a single artificial support structure can be determined by combining the individual structure's dimensional parameters. For example, if the length of a single artificial support structure along the goaf direction is L, and the spacing between adjacent frames is d, the number of portal frames is L / d rounded up. The portal frames corresponding to this number are then divided into monitoring stations, sequentially along the goaf direction, with each monitoring station corresponding to one artificial support structure. A monitoring station refers to a monitoring unit along the goaf direction, consisting of a set of portal frames supporting the same artificial support structure and its corresponding monitoring data.

[0046] Secondly, based on the construction sequence in the construction plan, the construction step is determined. The construction step is used to characterize the construction interval of individual artificial support structures between adjacent stations. Specifically, artificial support structures are constructed segment by segment along the goaf roadway. After each individual structure is completed, the construction team advances to the next location to construct the next individual structure. The construction interval between the completion of two adjacent individual artificial support structures is the construction step. For example, if one individual structure is constructed per day, the construction step is 1 day. The construction step determines the construction interval of individual artificial support structures when adjacent stations enter the same service stage. For example, station A enters the bearing stage one construction step earlier than station B.

[0047] Finally, based on the station division results, each station is assigned a portal frame with a specific number, and the multi-source load data is assigned to the corresponding station according to the frame number. Subsequently, according to the construction progress, the data of each station is marked with its temporal position relative to the construction start point. For example, the data of station A starts from day 1, and the data of station B starts from day 2. Finally, the multi-source load data is divided into multiple sets of station load data corresponding to each station.

[0048] Step S200 in the method of this embodiment of the invention further includes: Based on prior knowledge of overburden deformation monitoring, a list of key features is determined, which includes at least load-related features, displacement-related features, and correlation-coupling-related features. Based on the grouping results, a subset of the key feature list corresponding to each station is matched in the key feature list; By combining a subset of key feature lists and setting a sliding window, multi-parameter correlation features of stations are extracted from the station bearing data of the grouping results to generate overburden movement state vectors; By traversing multiple stations, the overburden motion state vectors of each station are summarized into the overburden motion state vector group.

[0049] In this embodiment of the invention, firstly, based on prior knowledge of overburden deformation monitoring, a list of key features is determined, wherein the list of key features includes at least load-related features, displacement-related features, and correlation-coupling-related features.

[0050] Specifically, load-related features are calculated based on pressure sensing data and include: load transfer coefficient, eccentric load coefficient, load level, load fluctuation index, and load change rate. The load transfer coefficient is the ratio of column pressure to top beam pressure, reflecting force transfer efficiency; the eccentric load coefficient is the ratio of the absolute value of the left-right pressure difference to the sum of the left-right pressures, reflecting spatial load non-uniformity; the load level is the ratio of top beam pressure to design bearing capacity; the load fluctuation index is the coefficient of variation within the sliding window; and the load change rate is the derivative of pressure with respect to time. Displacement-related features are calculated based on displacement sensing data and include: base plate deformation, base plate deformation rate, base plate deformation acceleration, and top beam relative displacement. The base plate deformation is a cumulative deformation value; the base plate deformation acceleration reflects whether instability is accelerated; and the top beam relative displacement is the change in laser ranging. Correlation and coupling-related features are calculated by integrating multiple types of sensor data, reflecting the correlation between different physical quantities, and include: pressure-displacement coupling coefficient, left-right symmetry deviation, and attitude change rate. Among them, the pressure-displacement coupling coefficient is the correlation coefficient between the pressure increment and the displacement increment within the sliding window; the attitude change rate is the rate of change of the tilt angle.

[0051] Secondly, since gantry cranes go through different stages throughout their complete service life, the overburden load transfer mechanism and deformation characteristics differ significantly at each stage, and the key features to focus on also differ at each stage. Therefore, it is necessary to identify the current stage of each station, match and extract the corresponding stage key features from the key feature list based on the grouping results, and output a subset of the key feature list for each station.

[0052] Next, by combining a subset of the key feature list, a sliding window is set to extract multi-parameter correlation features of the stations from the station bearing data of the grouped division results, generating the overburden movement state vector. Here, the sliding window refers to a data truncation window that slides forward with a fixed width on the time series data, used to extract features within a specific time interval from continuous station bearing data.

[0053] Specifically, a sliding window method is used, setting the window width and sliding step size, such as a window width of 6 hours and a sliding step size of 1 hour. Within each sliding window, the specific values ​​of each feature item in the key feature list subset are calculated from the station's load data. For example, the average load of the top beam, the rate of load change, and the eccentric load coefficient are calculated within the sliding window. After each feature is calculated, they are arranged in a preset order to form the overburden motion state vector for that station and that window. The sliding window slides forward along the time axis, generating an overburden motion state vector for each window.

[0054] Finally, multiple monitoring stations were divided along the direction of the open tunnel. By traversing all the monitoring stations, the overburden motion state vectors of each monitoring station were summarized into an overburden motion state vector group.

[0055] Step S200 of the method in this embodiment of the invention, which involves matching a subset of the key feature list corresponding to each station in the key feature list based on the grouping and partitioning results, further includes: Based on the grouping results, extract the station carrying data corresponding to each station; The data carried by the station is subjected to stage identification based on preset rules to obtain the first stage identification result; The station load data is subjected to stage identification based on data features to obtain the second stage identification result. The data features include at least two combinations of load level, load change trend, displacement level, displacement change rate, load fluctuation amplitude, and off-center load coefficient. Change point detection is performed on the data carried by the station to obtain the stage transition auxiliary calibration results; By integrating the first-stage identification results, the second-stage identification results, and the stage transition auxiliary calibration results, the current support service stage of the station is determined. Based on the service stage of the support structure, the corresponding key features of the stage are matched and extracted from the key feature list, and the output is a subset of the key feature list.

[0056] In this embodiment of the invention, firstly, based on the grouping results, for the target station that needs to be identified in stages, the station carrying data corresponding to the station is extracted from the multiple groups of station carrying data that have been divided.

[0057] Secondly, the data carried by the monitoring station is subjected to stage identification based on preset rules to obtain the first stage identification results. The preset rules refer to the stage determination logic rules pre-established based on prior knowledge of the service life cycle of the portal frame, such as determining the service stage of the monitoring station based on the difference between the installation time, concrete pouring time, nominal curing period and the current time.

[0058] Specifically, the service stage of the monitoring station is determined based on the known installation time, concrete pouring time, and nominal curing period. A preliminary judgment is made using a rule engine: if the difference between the current time and the installation time is less than a first preset time threshold, the station is determined to be in the first service stage; if the current time is earlier than the concrete pouring time, the station is determined to be in the second service stage; if the current time is between the concrete pouring time and the concrete pouring time plus the actual curing period, the station is determined to be in the third service stage; if the current time is between the curing end time (the result of the concrete pouring time plus the actual curing period) and the curing end time plus the second preset time threshold, the station is determined to be in the fourth service stage. If none of the above conditions are met, the station is determined to be in the fifth service stage.

[0059] The nominal curing period needs to be adjusted according to the type of concrete and the ambient temperature. For example, the nominal curing period for flexible formwork concrete is usually 7 days during winter construction, but can be shortened to 3 days during summer construction. The first and second preset time thresholds are preset time thresholds. The first preset time threshold can be determined based on the time required for the roof delamination to stabilize after the portal frame is installed, or the time required for the initial support force to reach the design value, combined with statistical analysis of on-site mine pressure monitoring data. For example, based on historical data, if the roof delamination stabilizes within 6 hours after the portal frame is installed in a certain mine, then the first preset time threshold is set to 6 hours. The second preset time threshold can be set based on the allowable waiting time before the support is removed after concrete curing, or the connection cycle of subsequent support procedures, according to engineering experience and construction organization arrangements; for example, it can be set to 24 hours.

[0060] Furthermore, since the data characteristics of the station's load data exhibit significantly different patterns under different service stages, it is necessary to perform stage identification based on data characteristics to obtain the second stage identification results. The data characteristics include at least two combinations of load level, load change trend, displacement level, displacement change rate, load fluctuation amplitude, and off-center load coefficient.

[0061] Specifically, the process begins with detailed classification based on data characteristics: historical monitoring data from stations that have completed a full service cycle are selected as training samples. These training samples are then labeled with stage tags according to their actual service phase. For example, 50 sets of historical monitoring data for gantry supports that have completed a full service cycle within the past six months at a certain mine are selected. Engineering technicians, based on construction logs and on-site records, label each set of data with its corresponding service phase. For each service phase, data characteristics are extracted from the training samples to construct a stage feature fingerprint. This stage feature fingerprint includes at least two or more combinations of load level statistical characteristics, load change trend characteristics, displacement level statistical characteristics, displacement change rate characteristics, load fluctuation amplitude characteristics, and off-center load coefficient characteristics. The load fluctuation amplitude characteristic can be represented as the standard deviation or variance of the load, and the load change trend characteristic can be, for example, the first derivative or the slope of the fitted area within the window.

[0062] Subsequently, based on the stage feature fingerprint, a random forest classifier is trained using the data features of the training samples as input and the corresponding stage label as the prediction target to generate a stage identification model. Random forest is an ensemble learning algorithm that improves classification accuracy by constructing multiple decision trees and combining their prediction results. The current station's load data is input into the trained stage identification model, and the model outputs the service stage of the current station, serving as the second stage identification result. For example, inputting the current load data of station 2 into the random forest classifier, the model outputs a prediction of the main stress stage with a confidence level of 92%.

[0063] Furthermore, change point detection is performed on the data carried by the stations to obtain auxiliary calibration results for the phase transition. Change point detection refers to detecting key transition points in the time series.

[0064] Specifically, online change point detection is performed on the load time series in the station's load data to identify the moments when the statistical characteristics of the load time series change significantly, and these moments are used as auxiliary calibration bases for phase transitions. These moments include at least the starting point when the load jumps from the zero value range to the rapid rise phase, corresponding to the first transition point from the first service phase to the second service phase; the turning point when the load transitions from the rapid rise phase to the stable fluctuation phase, corresponding to the second transition point from the second service phase to the third service phase; the starting point when the load transitions from the high value stable phase to the slow decline phase, corresponding to the third transition point from the third service phase to the fourth service phase; and the ending point when the load transitions from the low value stable phase to the stable dismantling phase, corresponding to the fourth transition point from the fourth service phase to the fifth service phase.

[0065] Furthermore, by integrating the identification results of the first stage, the identification results of the second stage, and the stage transition auxiliary calibration results, the current support service stage of the station is determined.

[0066] Specifically, the scaffolding service phase includes the following five typical stages: The first service phase is the initial installation phase. In this phase, the gantry crane is newly installed and has not yet borne overburden loads. The pressure on the top beam and columns is close to zero or at extremely low levels. Initial adjustment of the base plate displacement occurs, and the tilt angle of the top beam is close to zero. This phase preferably lasts from several hours to one day. The second service phase is the initial bearing phase. In this phase, the gantry crane begins to bear overburden pressure, and the load increases rapidly. The base plate displacement begins to increase due to high stress, and the rate of change and fluctuation of various monitoring parameters are high. The artificially supported structure has not yet been poured or has just been poured and has not yet reached bearing strength. This phase preferably lasts from one to three days. The third service phase is the main bearing phase. In this phase, the gantry crane continuously bears overburden loads, and the artificially supported structure is in the curing process. The first stage is the most sensitive stage for monitoring overburden deformation, characterized by a high load level accompanied by periodic fluctuations in mine pressure, a slow and continuous increase in displacement, and a stabilization or periodic change pattern in the fluctuations of various monitoring parameters. The duration of this stage is preferably 3 to 7 days. The second stage is the attenuation transition stage, during which the artificial support structure reaches its design strength, the overburden load gradually shifts from the portal frame to the artificial support structure, the support load begins to decrease slowly, the displacement growth rate slows down and tends to stabilize, and the duration of this stage is preferably 1 to 3 days. The third stage is the dismantling preparation stage, during which most of the overburden load has been transferred to the artificial support structure, the portal frame load drops to a low level, the displacement tends to stabilize and no longer changes significantly, and the support awaits rotation instructions and is about to be dismantled and moved to a new location. The duration of this stage is preferably several hours to 1 day.

[0067] Finally, based on the service stage of the support structure, the corresponding key features for each stage are matched and extracted from the key feature list, and a subset of the key feature list is output. The key feature list includes load-related features, displacement-related features, and correlation and coupling-related features, and the subset of the key feature list is different for different service stages.

[0068] In this embodiment of the invention, the number of portal frames required to support a single unit is determined and monitoring stations are divided. The construction progress is determined according to the construction sequence, enabling multi-source bearing data to be grouped by monitoring station. A list of key features is determined based on prior knowledge of overburden deformation monitoring. A sliding window is set to extract multi-parameter correlation features and generate an overburden motion state vector. The first-stage identification result is obtained by performing stage identification based on preset rules on the station bearing data, and the second-stage identification result is obtained based on data feature stage identification. Change point detection is performed to obtain stage transition auxiliary calibration results. These results are then fused to determine the current support service stage of the monitoring station. A subset of the key feature list is extracted to achieve accurate identification and feature matching of the support service stage, providing data input for subsequent overburden deformation identification based on spatial consistency.

[0069] S300: Combine the historical overburden movement state vector group from the previous control cycle to identify overburden deformation based on spatial consistency. In this embodiment of the invention, the control period is a time unit for unified analysis and evaluation of the deformation state of the overburden along the goaf; the historical overburden movement state vector group is a collection of overburden movement state vectors of each station extracted and stored in the previous control period using the same method.

[0070] Specifically, in the actual engineering practice of roadway retention along the goaf, gantry supports are used in rotation. The oldest set of supports is removed and then installed at the newest forward position. The rotation cycle depends on the concrete curing time, and the number of supports in service depends on the length of the active zone in the roadway. Taking the current control cycle as the analysis scope, according to the support service stage, the overburden movement state vectors of each station in the current cycle's overburden movement state vector group are matched and aligned with the historical overburden movement state vectors of stations in the same support service stage in the previous control cycle's historical overburden movement state vector group. Multi-dimensional comparative analysis is performed based on the matching and alignment results. The results of the multi-dimensional comparative analysis are then fused and calculated to obtain the station stage consistency index for each station in the corresponding support service stage. Subsequently, the station stage consistency indices of multiple stations under the current control cycle are fused to generate a spatial consistency map. Finally, trend analysis and deformation identification are performed by combining the historical spatial consistency sequence and the current spatial consistency map to locate stations with abnormal and normal overburden deformation, and the output is the overburden deformation identification result.

[0071] Step S300 in the method of this embodiment of the invention includes: Taking the current control cycle as the scope of analysis, according to the support service stage, the overburden movement state vector of each station in the overburden movement state vector group is matched and aligned with the historical overburden movement state vector of the stations in the same support service stage in the historical overburden movement state vector group of the previous control cycle. Based on the matching and alignment results, a multi-dimensional comparison analysis is performed, wherein the multi-dimensional comparison includes at least a combination of two of the following: feature value level comparison, feature change trend comparison, feature fluctuation pattern comparison, and correlation consistency comparison. The results of multi-dimensional comparative analysis are fused and calculated to obtain the station phase consistency index for each station in the corresponding support service phase. By integrating the station-stage consistency indicators of multiple stations under the current control cycle, a spatial consistency map is generated; Extract a preset number of historical spatial consistency maps from previous control cycles and serialize them to form a historical spatial consistency sequence; By combining the historical spatial consistency sequence and the spatial consistency map, trend analysis and deformation identification are performed to locate abnormal overburden deformation stations and normal overburden deformation stations, and the output is the overburden deformation identification result.

[0072] In this embodiment of the invention, firstly, taking the current control period as the analysis scope, according to the support service stage, the overburden movement state vector of each station in the overburden movement state vector group is matched and aligned with the historical overburden movement state vector of the stations in the same support service stage in the historical overburden movement state vector group of the previous control period.

[0073] Specifically, the overburden motion state vectors corresponding to two consecutive control cycles are obtained, namely the previous control cycle and the current control cycle. For each service phase, the service level ranges from the first to the fifth service phase. The stations in the service phase of the previous control cycle and the stations in the service phase of the current control cycle are obtained respectively, and a pairing relationship is established: (stations in the service phase of the previous control cycle, stations in the service phase of the current control cycle). When there are multiple stations in the same service phase, for example, when there are many in-service supports, multiple stations may be in the same phase at the same time. The pair of stations with the closest rotation time is selected as the effective paired stations for that service phase.

[0074] Secondly, based on the matching and alignment results, a multi-dimensional comparison analysis is performed. The multi-dimensional comparison includes at least two of the following: feature value level comparison, feature change trend comparison, feature fluctuation pattern comparison, and correlation consistency comparison.

[0075] Specifically, eigenvalue level comparison directly compares the values ​​of each characteristic item of paired stations under the same service phase, calculating the eigenvalue level deviation, which is the ratio of the difference between two eigenvalue level values ​​to one of the eigenvalue level values. For example, the average load characteristic value of the top beam is extracted for each paired station. Characteristic change trend comparison compares the direction and rate of characteristic changes of paired stations under the same service phase. For example, comparing the load change trends of paired stations during their respective third service phases: if both show a trend of first rising and then stabilizing, the trend direction is consistent. Characteristic fluctuation pattern comparison compares the characteristic fluctuation period, amplitude, and phase of paired stations under the same service phase. For example, overburden pressure during the third service phase may cause periodic load fluctuations. The dominant frequency of load fluctuations at the stations is extracted through spectral analysis. If the dominant frequencies of load fluctuations of paired stations are close, it indicates that the characteristic fluctuation patterns are consistent. Correlation consistency comparison compares the coupling relationships between multiple characteristics of paired stations under the same service phase. For example, calculating the load-displacement coupling degree; under normal conditions, as the load increases, the displacement should increase accordingly.

[0076] Next, the results of the multi-dimensional comparative analysis are fused and calculated, i.e., normalized, and weighted by setting corresponding weight coefficients for each dimension to obtain the station-stage consistency index for each station in the corresponding support service phase. The station-stage consistency index characterizes the overall similarity between the current station and its paired stations in the same service phase in the previous cycle; a higher index value indicates better consistency, while a lower value indicates greater deviation. The weight coefficients are set according to the influence of the corresponding dimension comparative analysis results on the station-stage consistency index. The sum of the weight coefficients is always 1; the greater the influence, the larger the corresponding weight coefficient.

[0077] Specifically, the mean deviation of monitoring stations identified as having abnormal deformations over the past 12 control cycles is statistically analyzed across all dimensions. The weighting coefficients for each dimension are then obtained by normalizing the mean deviations. For example, the dimensions include both eigenvalue level comparison and eigenvalue trend comparison. Compared to eigenvalue level comparison, eigenvalue trend comparison has a greater impact on the station's phase consistency index. If the historical average deviation for the eigenvalue trend comparison dimension is 0.3, and for the eigenvalue level comparison dimension it is 0.2, then their weighting coefficients are 0.3 / (0.3+0.2)=0.6 and 0.2 / (0.3+0.2)=0.4, respectively. This indicates that the eigenvalue trend comparison dimension has a greater impact on the consistency index, and the weighting coefficients are set to 0.6 and 0.4.

[0078] Furthermore, by integrating the station-stage consistency indicators of multiple stations under the current control cycle, a spatial consistency map is generated. The spatial consistency map is a visual mapping of the position of each station along the goaf-retention direction and its corresponding consistency indicator. The horizontal axis represents the spatial position of the station in the goaf-retention direction, and the vertical axis represents the station-stage consistency indicator value of each station.

[0079] Furthermore, a preset number of historical spatial consistency maps from preceding control periods are extracted and serialized to form a historical spatial consistency sequence. The preset number can be set according to actual needs, such as extracting the first 5, 10, or 20 control periods from the historical spatial consistency maps. The extracted historical spatial consistency maps are then arranged in chronological order to obtain the historical spatial consistency sequence.

[0080] Finally, by combining historical spatial consistency sequences and spatial consistency maps, trend analysis and deformation identification are performed to locate abnormal overburden deformation stations and normal overburden deformation stations, and the output is the overburden deformation identification result.

[0081] Specifically, for each paired station, the multi-dimensional comprehensive deviation degree of its overlying motion state vector is calculated to obtain the single-comparison anomaly degree of that paired station. The single-comparison anomaly degree is the multi-dimensional comprehensive deviation degree. The multi-dimensional comprehensive deviation degree is calculated by fusing the multi-dimensional comparison analysis results, and is the deviation relative to the corresponding average value of the multi-dimensional comparison analysis results. The multi-dimensional comprehensive deviation degree is calculated as: (Comparison result value for that dimension - Statistical mean of the comparison results for that dimension) / Statistical mean of the comparison results for that dimension. The average value of the multi-dimensional comparison analysis results is obtained by summing the results of each multi-dimensional comparison analysis and dividing each sum by the number of multi-dimensional comparison analysis results.

[0082] Based on the comparison results of the single comparison anomaly degree with the preset anomaly degree threshold, the following overburden deformation identification and judgment are performed: When the anomaly score in a single comparison exceeds a preset anomaly threshold, it is determined that there is an overburden deformation anomaly at the current location corresponding to the paired station. The preset anomaly threshold can be determined based on statistical analysis of historical data; for example, twice the standard deviation of the mean of the station stage consistency index in the historical consistency map can be used as the preset anomaly threshold. Assuming the preset anomaly threshold is 0.25, it means that an overburden deformation anomaly is determined to exist when the station stage consistency index is below 0.75. Assuming that the station stage consistency index of station 5 is 0.45 and the single comparison anomaly score is 0.55, exceeding the preset anomaly threshold of 0.25, then it is determined that there is an overburden deformation anomaly at the location of station 5.

[0083] When the anomaly degree of a single comparison at the same spatial location shows a continuous increasing trend across multiple cycle periods, it is determined that the overburden deformation at that location is showing a trend of deterioration; when anomalies in overburden deformation occur simultaneously at adjacent locations along the goaf direction, it is determined that the range of anomalies is expanding; when the anomaly degree of a single comparison shows a sudden and sharp increase, it is determined that a sudden overburden failure event may occur. Based on the above judgment results, the overburden deformation identification results for the current control period are output.

[0084] In this embodiment of the invention, the overburden movement state vectors of each station in the current control cycle are matched and aligned with the historical overburden movement state vectors of the same support service stage in the previous control cycle. Based on the matching and alignment results, multi-dimensional comparative analysis is performed, and station stage consistency indicators are obtained through fusion calculation and spatial consistency maps are generated. By combining the historical spatial consistency sequence with the current spatial consistency map, trend analysis and deformation identification are performed, so as to accurately distinguish and locate stations with abnormal overburden deformation and stations with normal overburden deformation. By utilizing the control cycle generated by the rotation service of portal supports, the incomparability problem caused by stage differences is solved, and the accuracy of overburden deformation anomaly identification is improved.

[0085] S400: Based on the overburden deformation identification results, combined with the prior lithology data of the target roadway area, overburden deformation is predicted, and the prediction results are obtained and fed back.

[0086] In this embodiment of the invention, the prior lithological data are the pre-survey data of the distribution of overburden lithology in different sections along the goaf direction of the target roadway area and its corresponding deformation response parameters, including a lithological deformation adjustment coefficient matrix established based on historical monitoring data through statistical analysis. The sources of prior lithological information include: geological exploration profiles, measured calibration of constructed sections, geophysical supplementation, and borehole inspection. The lithological calibration parameters are correction coefficients that quantify the differences in overburden deformation response under different lithological conditions.

[0087] Specifically, due to the inherent differences in lithological properties in different sections along the goaf, different lithologies will exhibit different deformation responses under the same load conditions. Without lithological correction, these inherent lithological differences can easily be misjudged as abnormal deformation.

[0088] Therefore, for anomalous overburden deformation stations identified in the overburden deformation identification results, the spatial consistency map is corrected for anomalies by combining lithological calibration parameters from prior lithological data. Subsequently, trend analysis and deformation identification are re-performed based on the anomaly-corrected spatial consistency map, and pseudo-anomaly stations are removed and re-marked as normal overburden deformation stations. For the confirmed anomalous overburden deformation stations after correction, the most unfavorable trend information for each comparison dimension is extracted. The trend extrapolation results are weighted by combining lithological weight coefficients from prior lithological data to obtain the first overburden deformation prediction result, which is fed back in real time. Simultaneously, for normal overburden deformation stations identified in the overburden deformation identification results, regression extrapolation based on confidence constraints is performed using historical spatial consistency sequences to obtain the second overburden deformation prediction result, which is periodically fed back according to a preset feedback cycle.

[0089] Step S400 in the method of this embodiment of the invention includes: For the abnormal overburden deformation stations in the overburden deformation identification results, the spatial consistency map is corrected for anomalies by combining the lithological calibration parameters in the prior lithological data. Based on the spatial consistency map after anomaly correction, trend analysis and deformation identification are performed again, and pseudo-anomaly stations caused by lithological differences are removed. The removed pseudo-anomaly stations are then remarked as normal overburden deformation stations. For the corrected abnormal overburden deformation stations, the most unfavorable trend information of each comparison dimension is extracted and the trend is extrapolated respectively; The trend extrapolation results are weighted by combining the lithological weight coefficients in the prior lithological data to obtain the first overburden deformation prediction results, and the results are fed back in real time.

[0090] In this embodiment of the invention, during actual engineering implementation, the lithology of the overburden strata in different sections along the goaf-retention tunnel direction may vary significantly. Under the same load conditions, the deformation responses of different lithologies differ. Therefore, the abnormal station stage consistency index of abnormal overburden deformation monitoring stations may be caused by two reasons. If spatial consistency maps are not used for anomaly correction, the false alarm rate will be too high.

[0091] First, for stations with abnormal overburden deformation identified in the overburden deformation identification results, anomaly corrections are performed on the spatial consistency maps by combining lithological calibration parameters from prior lithological data. Specifically, based on prior lithological calibration, anomaly corrections based on historical experience are performed on stations with abnormal overburden deformation using statistical or regression analysis. The station-stage consistency index is adjusted according to the lithological calibration parameters, and the product of the lithological calibration parameters and the station-stage consistency index is used as the corrected station-stage consistency index, eliminating the consistency index deviation caused by lithological differences.

[0092] Secondly, based on the spatial consistency map after anomaly correction, trend analysis and deformation identification are performed again. The corrected consistency index is compared with the preset anomaly threshold to determine whether it is still below the preset anomaly threshold. If the corrected consistency index of a station exceeds the preset anomaly threshold, it indicates that the original low consistency of the station was mainly caused by lithological differences, and the station is a pseudo-anomaly station. In this case, the station is removed from the anomaly list and remarked as a normal overburden deformation station. If the corrected consistency index is still below the preset anomaly threshold, it is confirmed as a true anomaly station.

[0093] Next, for the corrected anomalous overburden deformation monitoring stations, the most unfavorable trend information for each comparison dimension is extracted, and trend extrapolation is performed separately. The most unfavorable trend information refers to the dimension that deviates most severely from the normal range and exhibits the most unfavorable development trend among the dimensions analyzed in the multi-dimensional comparison. Trend extrapolation refers to the process of inferring the future development trend of overburden deformation based on currently observed trends, using mathematical models such as linear regression and exponential smoothing.

[0094] Specifically, trend information for each comparison dimension is extracted from the multi-dimensional comparative analysis results, including trend data for dimensions such as eigenvalue level deviation, eigenvalue change trend deviation, eigenvalue fluctuation pattern deviation, and correlation consistency deviation. The dimension with the largest deviation and the most dangerous change trend is identified from the trend information of each comparison dimension as the most unfavorable trend information. Trend extrapolation is then performed on the trend data of each most unfavorable dimension, and a linear regression mathematical model is used to predict the change trend and value of that dimension in the future based on current and historical data.

[0095] Assuming that the load rise rate deviation is the largest and the trend is the most dangerous among the characteristic trend deviations, load time series data from the past control cycle of this station are extracted, and a linear regression model is used to fit the load change trend over time. The fitted relationship between load P and time t is assumed to be P = 280 + 22t, where t is in hours and load is in MPa. The linear regression mathematical model is used to extrapolate the load predictions for the next 24, 48, and 72 hours.

[0096] Finally, the trend extrapolation results are weighted by combining the lithological weight coefficients in the prior lithological data, that is, the trend extrapolation results are multiplied by the lithological weight coefficients to obtain the first overburden deformation prediction results, and then fed back to the ground monitoring center in real time.

[0097] Specifically, weighting coefficients are used to weight the extrapolation results of future trends, reflecting the degree of risk of overburden deformation under different lithological conditions. Generally, weaker rock layers have higher weighting coefficients and are more prone to sudden failure under the same load. For example, based on engineering rock mass classification standards and prior borehole data, the overburden can be divided into several engineering geological rock groups, and the measured values ​​of uniaxial compressive strength and elastic modulus of each rock group can be obtained through indoor rock mechanics tests. The measured value of elastic modulus is reduced by the rock mass integrity coefficient to obtain rock mass mechanics parameters that reflect the actual occurrence state on site. The mechanical parameters of the weakest rock layer, such as mudstone, are used as the benchmark parameter value to obtain the parameter value of the current rock layer. The weighting coefficient = benchmark parameter value / current rock layer parameter value. Based on this, the corresponding weighting coefficients of each trend extrapolation result are obtained. Subsequently, the trend extrapolation results of each most unfavorable dimension are weighted and synthesized to obtain the first overburden deformation prediction result.

[0098] In this embodiment of the invention, step S400, based on the overburden deformation identification result and combined with the prior lithological data of the target roadway area, performs overburden deformation prediction, obtains the overburden deformation prediction result, and provides feedback, further includes: For the normal overburden deformation stations in the overburden deformation identification results, combined with the historical spatial consistency sequence, regression extrapolation based on confidence constraints is performed to obtain the second overburden deformation prediction results; The second overburden deformation prediction result is periodically fed back according to the preset feedback cycle.

[0099] In this embodiment of the invention, firstly, for the normal overburden deformation stations in the overburden deformation identification results, regression extrapolation based on confidence constraints is performed in conjunction with the historical spatial consistency sequence to obtain the second overburden deformation prediction result, and the result is periodically fed back according to a preset feedback cycle. Even though no anomalies are currently identified for normal stations, their future deformation trends still need to be predicted.

[0100] Specifically, historical spatial consistency sequences from multiple control periods of the monitoring station are extracted, such as the past 10 periods, and extrapolated using a regression model with confidence intervals. Using the historical consistency index sequence as input, linear regression or locally weighted regression methods are used to fit the changing trend of the station's phase consistency index over time. The confidence interval of the prediction results is calculated based on the standard deviation of the fitted residuals. Under confidence constraints, the predicted values ​​of the station's phase consistency index and their confidence intervals for several future control periods are extrapolated to generate the second overburden deformation prediction results. Here, the confidence constraint refers to the prediction results including confidence intervals; for example, a confidence constraint of 95% confidence interval is set.

[0101] Secondly, according to the preset feedback cycle, the prediction results of the second overburden deformation are periodically pushed to the monitoring system or relevant personnel. The preset feedback cycle refers to the pre-defined time interval for periodically feeding back the prediction results from normal monitoring stations, usually matching the control cycle or shift cycle, such as each control cycle. The feedback content may include: the current consistency index of each normal monitoring station, the predicted future trend value, the confidence interval, and whether there are signs of trend deterioration, etc.

[0102] Step S400 of the method in this embodiment of the invention, combining the lithological calibration parameters in the prior lithological data, performs anomaly correction on the spatial consistency map, including: The prior lithological data includes a lithological deformation adjustment coefficient matrix established based on historical monitoring data and statistical analysis; Based on the lithological deformation adjustment coefficient matrix, the station stage consistency index is re-compared and calculated, and the spatial consistency map is updated accordingly. The lithological deformation adjustment coefficient matrix represents the mapping relationship of overburden deformation between stations with different lithological conditions that are in the same service phase during adjacent control cycles.

[0103] In this embodiment of the invention, the prior lithological data first includes a lithological deformation adjustment coefficient matrix established through statistical analysis based on historical monitoring data. The historical monitoring data refers to overburden deformation monitoring data and their corresponding lithological annotation information collected and stored in previous control cycles.

[0104] Specifically, the lithological deformation adjustment coefficient matrix is ​​established based on a large amount of historical monitoring data through statistical analysis. From the historical monitoring data, pairs of monitoring stations in the same service stage but located in different lithological sections are selected, and the ratio of their overburden deformation is calculated as the lithological deformation adjustment coefficient. After multiple statistical regressions, the average deformation mapping relationship between different lithologies is obtained, making lithological correction calculable and repeatable. For example, historical monitoring station data from the target roadway area that has completed a full service cycle are collected, including roof lithology information at each station's location and actual deformation monitoring data for each service stage. Using the deformation of the hard rock section as a benchmark, the lithological deformation adjustment coefficient is set to 1. The ratio of deformation under the same load level and service stage under different lithological conditions is statistically analyzed. Assuming that the statistics show the average relative displacement of the roof and floor in the mudstone section during the main bearing stage is 45 mm, while the average displacement in the fine sandstone section of the same area during the same stage is 18 mm, then the lithological deformation adjustment coefficient for mudstone relative to fine sandstone is 45 / 18 = 2.5. The final matrix of lithological deformation adjustment coefficients is formed. For example, the correspondence between different lithological sections and the lithological deformation adjustment coefficients is shown below. Figure 3 As shown.

[0105] Secondly, based on the lithological deformation adjustment coefficient matrix, the station-stage consistency index is re-compared and recalculated. Specifically, based on the lithological deformation adjustment coefficient matrix, the lithological deformation adjustment coefficient is multiplied by the original station-stage consistency index to obtain the recalculated station-stage consistency index. This recalculated index replaces the original one, thus updating the spatial consistency map. The lithological deformation adjustment coefficient matrix represents the mapping relationship of overburden deformation between stations in the same service stage but with different lithological conditions during adjacent control cycles. Specifically, it is a two-dimensional matrix with lithological categories as rows and columns, and lithological deformation adjustment coefficients between overburden deformation under different lithological conditions as elements.

[0106] In this embodiment of the invention, anomaly correction is performed on identified overburden deformation monitoring stations using lithological calibration parameters from prior lithological data. False anomaly stations are eliminated and re-marked as normal overburden deformation monitoring stations. Then, the most unfavorable trend information is extracted from the corrected true anomaly stations for trend extrapolation to calculate the first overburden deformation prediction result, achieving accurate early warning for anomaly monitoring stations. Simultaneously, a second overburden deformation prediction result is obtained by performing confidence-constrained regression extrapolation on normal overburden deformation monitoring stations using historical spatial consistency sequences. This result is periodically fed back according to a preset feedback cycle, eliminating interference from inherent geological differences on anomaly identification and achieving accurate prediction of overburden deformation.

[0107] Through the above specific implementation methods, the embodiments of the present invention achieve the following technical effects: In this embodiment of the invention, firstly, the synchronous acquisition and time synchronization calibration of multi-source data are achieved through an IoT monitoring network on multiple portal supports. Subsequently, each sensing unit connects to the mine's wired transmission network via a mine-use wireless communication module, constructing an IoT monitoring network covering the roadway area. This adapts to the relocation requirements of support rotation and provides basic data for subsequent operations.

[0108] Secondly, the number of portal frames required to support the individual structures was determined, and monitoring stations were divided. The construction sequence was determined based on the construction progress, enabling the multi-source bearing data to be grouped by monitoring station. A list of key features was determined by combining prior knowledge of overburden deformation monitoring. A sliding window was set to extract multi-parameter correlation features and generate an overburden motion state vector. The first-stage identification results were obtained by performing stage identification based on preset rules on the station bearing data, and the second-stage identification results were obtained based on data feature stage identification. Change point detection was performed to obtain stage transition auxiliary calibration results. These results were then fused to determine the current support service stage of the monitoring station. A subset of the key feature list was then extracted to achieve accurate identification and feature matching of the support service stage, providing data input for subsequent overburden deformation identification based on spatial consistency.

[0109] Furthermore, the overburden movement state vectors of each station in the current control cycle are matched and aligned with the historical overburden movement state vectors of the same support service stage in the previous control cycle. Based on the matching and alignment results, multi-dimensional comparative analysis is performed, and station stage consistency indicators are obtained through fusion calculation to generate spatial consistency maps. By combining historical spatial consistency sequences with the current spatial consistency map, trend analysis and deformation identification are performed to accurately distinguish and locate stations with abnormal overburden deformation and stations with normal overburden deformation. By utilizing the control cycles generated by the rotation of portal supports, the incomparability caused by stage differences is resolved, and the accuracy of overburden deformation anomaly identification is improved.

[0110] Finally, the identified anomalous overburden deformation stations were corrected using lithological calibration parameters from prior lithological data. False anomaly stations were eliminated and re-marked as normal overburden deformation stations. Then, the most unfavorable trend information was extracted from the corrected true anomaly stations for trend extrapolation to calculate the first overburden deformation prediction result, achieving accurate early warning for anomalous stations. Simultaneously, the second overburden deformation prediction result was obtained by performing confidence-constrained regression extrapolation on normal overburden deformation stations using historical spatial consistency sequences. This result was periodically fed back according to a preset feedback cycle, eliminating the interference of inherent geological differences on anomaly identification and achieving accurate prediction of overburden deformation.

[0111] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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. Such 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 this application.

Claims

1. A method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf, characterized in that, include: Multi-source load data is collected through an IoT monitoring network deployed on multiple portal supports along the goaf. Based on the construction steps of artificially supported structures, the multi-source bearing data are grouped and divided, and feature extraction is performed based on the grouping results to obtain the overburden movement state vector group. Combining the historical overburden movement state vector set from the previous control cycle, the overburden movement state vector set is used to identify overburden deformation based on spatial consistency. Based on the overburden deformation identification results, combined with the prior lithological data of the target roadway area, the overburden deformation is predicted, and the prediction results are obtained and fed back.

2. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 1, characterized in that, Through an IoT monitoring network deployed on multiple portal frames along the goaf, multi-source load data is collected, including: Sensor units are deployed on multiple portal supports arranged along the direction of the empty lane to construct an Internet of Things monitoring network covering the empty lane area; The IoT monitoring network collects multi-source monitoring data of each of the portal supports during the overburden bearing process, wherein the multi-source monitoring data includes at least one of pressure data, displacement data, and attitude data; The multi-source monitoring data is preprocessed and time-synchronized to obtain the multi-source carrying data.

3. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 2, characterized in that, Sensor units are deployed on multiple portal frames arranged along the direction of the exit tunnel to construct an Internet of Things (IoT) monitoring network covering the exit tunnel area, including: A first pressure sensing unit is deployed on the bearing surface of the top beam of the portal frame to collect load data of the overburden acting on the top beam; A second pressure sensing unit is deployed in each column of the gantry frame to collect data on the axial force borne by each column. An attitude sensing unit for collecting tilt angle data of the top beam is deployed on the top beam portion of the portal frame. A displacement sensing unit for collecting displacement data of the base plate is deployed in the base portion of the portal frame. A ranging sensing unit for collecting distance change data between the support and a fixed reference point is deployed on the gantry frame; Each of the aforementioned sensing units is connected to the mine wired transmission network via a mine wireless communication module to construct the Internet of Things monitoring network.

4. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 1, characterized in that, Based on the construction progress of artificially supported structures, the multi-source load-bearing data is grouped and divided, including: Obtain the individual dimensional parameters and construction plan of the artificial support structure; Based on the individual unit size parameters, determine the number of portal frames required to support a single artificial support structure, and divide the number of portal frames corresponding to that number into one measuring station; The construction step is determined according to the construction sequence in the construction plan. The construction step is used to characterize the construction time interval of individual artificial support structures between adjacent stations. Based on the multiple stations obtained and the construction steps, the multi-source carrier data is divided into multiple sets of station carrier data corresponding to each station.

5. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 1, characterized in that, Feature extraction is performed based on the grouping results to obtain the overlying rock movement state vector set, including: Based on prior knowledge of overburden deformation monitoring, a list of key features is determined, which includes at least load-related features, displacement-related features, and correlation-coupling-related features. Based on the grouping results, a subset of the key feature list corresponding to each station is matched in the key feature list; By combining a subset of key feature lists and setting a sliding window, multi-parameter correlation features of stations are extracted from the station bearing data of the grouping results to generate overburden movement state vectors; By traversing multiple stations, the overburden motion state vectors of each station are summarized into the overburden motion state vector group.

6. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 5, characterized in that, Based on the grouping results, the subset of key features for each station is matched in the key feature list, which also includes: Based on the grouping results, extract the station carrying data corresponding to each station; The data carried by the station is subjected to stage identification based on preset rules to obtain the first stage identification result; The station load data is subjected to stage identification based on data features to obtain the second stage identification result. The data features include at least two combinations of load level, load change trend, displacement level, displacement change rate, load fluctuation amplitude, and off-center load coefficient. Change point detection is performed on the data carried by the station to obtain the stage transition auxiliary calibration results; By integrating the first-stage identification results, the second-stage identification results, and the stage transition auxiliary calibration results, the current support service stage of the station is determined. Based on the service stage of the support structure, the corresponding key features of the stage are matched and extracted from the key feature list, and the output is a subset of the key feature list.

7. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 1, characterized in that, Based on the historical overburden movement state vector set from the previous control cycle, overburden deformation identification is performed on the overburden movement state vector set based on spatial consistency, including: Taking the current control cycle as the scope of analysis, according to the support service stage, the overburden movement state vector of each station in the overburden movement state vector group is matched and aligned with the historical overburden movement state vector of the stations in the same support service stage in the historical overburden movement state vector group of the previous control cycle. Based on the matching and alignment results, a multi-dimensional comparison analysis is performed, wherein the multi-dimensional comparison includes at least a combination of two of the following: feature value level comparison, feature change trend comparison, feature fluctuation pattern comparison, and correlation consistency comparison. The results of multi-dimensional comparative analysis are fused and calculated to obtain the station phase consistency index for each station in the corresponding support service phase. By integrating the station-stage consistency indicators of multiple stations under the current control cycle, a spatial consistency map is generated; Extract a preset number of historical spatial consistency maps from previous control cycles and serialize them to form a historical spatial consistency sequence; By combining the historical spatial consistency sequence and the spatial consistency map, trend analysis and deformation identification are performed to locate abnormal overburden deformation stations and normal overburden deformation stations, and the output is the overburden deformation identification result.

8. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 1, characterized in that, Based on the overburden deformation identification results, and combined with the prior lithological data of the target roadway area, overburden deformation is predicted. The predicted overburden deformation results are obtained and fed back, including: For the abnormal overburden deformation stations in the overburden deformation identification results, the spatial consistency map is corrected for anomalies by combining the lithological calibration parameters in the prior lithological data. Based on the spatial consistency map after anomaly correction, trend analysis and deformation identification are performed again, and pseudo-anomaly stations caused by lithological differences are removed. The removed pseudo-anomaly stations are then remarked as normal overburden deformation stations. For the corrected abnormal overburden deformation stations, the most unfavorable trend information of each comparison dimension is extracted and the trend is extrapolated respectively; The trend extrapolation results are weighted by combining the lithological weight coefficients in the prior lithological data to obtain the first overburden deformation prediction results, and the results are fed back in real time.

9. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 8, characterized in that, Based on the overburden deformation identification results, combined with the prior lithological data of the target roadway area, overburden deformation is predicted, and the prediction results are obtained and fed back. This also includes: For the normal overburden deformation stations in the overburden deformation identification results, combined with the historical spatial consistency sequence, regression extrapolation based on confidence constraints is performed to obtain the second overburden deformation prediction results; The second overburden deformation prediction result is periodically fed back according to the preset feedback cycle.

10. The method for dynamic prediction of overburden deformation in deep, thick coal seams along the goaf as described in claim 7, characterized in that, Based on the lithological calibration parameters in the aforementioned prior lithological data, anomaly correction is performed on the spatial consistency map, including: The prior lithological data includes a lithological deformation adjustment coefficient matrix established based on historical monitoring data and statistical analysis; Based on the lithological deformation adjustment coefficient matrix, the station stage consistency index is re-compared and calculated, and the spatial consistency map is updated accordingly. The lithological deformation adjustment coefficient matrix represents the mapping relationship of overburden deformation between stations with different lithological conditions that are in the same service phase during adjacent control cycles.

Images