Adaptive control system and method based on coal wall spalling risk assessment

Abstract

The application provides an adaptive control system and method based on coal wall spalling risk assessment, relates to the technical field of intelligent control, and the system comprises: a data acquisition module that acquires multi-source monitoring data of a fully mechanized coal face; a target identification module that identifies risk factors when detecting a spalling event; an interval statistical module that performs interval statistics on hydraulic support working resistance data; a statistical analysis module that performs statistical analysis on actual mining height data; a risk assessment module that inputs each risk factor into a risk assessment model; and a control strategy execution module that executes hierarchical adaptive control according to the coal wall spalling risk level. The application solves the technical problem in the prior art that due to single spalling risk monitoring dimension and lack of hierarchical progression of control strategies, hidden spalling signs cannot be accurately identified, and the safety of coal mining operations is affected, realizes advanced and accurate early warning and differentiated and flexible prevention and control of coal wall spalling risks, and improves the safety of coal mining operations.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology, specifically to an adaptive control system and method based on coal wall spalling risk assessment. Background Technology

[0002] Currently, the main method used in engineering sites for monitoring coal face risk is a combination of single-parameter threshold alarms for hydraulic support working resistance and manual video inspections, supplemented by fixed support force maintenance and passive shutdown control measures. However, this method relies excessively on a single physical quantity and cannot effectively capture multi-source latent signs, resulting in a severe underutilization of early signs of coal face collapse. Risk assessment results are characterized by significant lag and a high rate of misjudgment. Due to incomplete monitoring information and the lack of a multi-level progressive control mechanism, existing control strategies generally adopt a trigger-based, black-and-white response mode, i.e., only performing shutdowns or full-load support operations when coal face collapse has clearly occurred. Conservative operations easily lead to over-support and frequent shutdowns, severely restricting the efficiency of face advancement; while risky operations, due to insufficient intervention intensity, cannot effectively curb the deterioration of coal wall stability, easily inducing roof collapse and support collapse accidents, thereby affecting the safety and production efficiency of coal mining operations.

[0003] In summary, existing technologies suffer from several technical problems. Due to the single dimension of risk monitoring for coal spalling and the lack of hierarchical progression in control strategies, they are unable to accurately identify hidden signs of coal spalling, which can easily lead to over-intervention or under-intervention, further affecting the safety and production efficiency of coal mining operations. Summary of the Invention

[0004] The purpose of this application is to provide an adaptive control system and method based on coal wall spalling risk assessment, in order to solve the technical problems in the prior art, which are that due to the single dimension of spalling risk monitoring and the lack of hierarchical progression of control strategies, it is impossible to accurately identify hidden spalling signs, which easily leads to over-intervention or under-intervention, and further affects the safety and production efficiency of coal mining operations.

[0005] To achieve the above objectives, this application provides an adaptive control system and method based on coal wall spalling risk assessment.

[0006] Firstly, this application provides an adaptive control system based on coal face spalling risk assessment. The adaptive control system includes: a data acquisition module for acquiring multi-source monitoring data from the fully mechanized mining face, including coal face image data, hydraulic support working resistance data, and actual mining height data; a target recognition module for preprocessing and identifying targets in the coal face image data, and outputting an image recognition risk factor based on the area ratio of the spalling region when a spalling event is detected; and an interval statistics module for performing interval statistics on the hydraulic support working resistance data, and outputting pressure anomaly risk based on the proportion of supports in the low-pressure interval and the proportion of supports in the high-pressure over-limit interval. The system includes: a factor; a statistical analysis module for statistically analyzing the actual mining height data and outputting mining height risk factors based on the mining height deviation rate and the degree of mining height fluctuation exceeding limits; a risk assessment module for inputting the image recognition risk factor, the pressure anomaly risk factor, and the mining height risk factor into a risk assessment model, and outputting the coal wall spalling risk level of the fully mechanized mining face after inference by the risk assessment model; and a control strategy execution module for executing a graded adaptive control strategy based on the coal wall spalling risk level. The graded adaptive control strategy includes: executing a first control measure when the risk level is level one, executing a second control measure when the risk level is level two, and executing a third control measure when the risk level is level three.

[0007] Optionally, the filtering preprocessing unit is used to perform filtering preprocessing on the coal wall image data to obtain filtered image data; the frame-by-frame inference unit is used to perform frame-by-frame inference on the filtered image data to obtain the target coal, wherein the target coal includes flared coal and piled coal; the flared event determination unit is used to determine the existence of a flared event when the number of detection frames of the detected flared coal reaches a preset number threshold and the detection frame of the piled coal is detected at the same time, and the bounding rectangle of all detection frames is taken as the flared area; the area ratio calculation unit is used to calculate the ratio of the area of ​​the flared region to the area of ​​the entire image as the flared area ratio; and the image recognition risk factor determination unit is used to match the preset interval to which the flared area ratio belongs and output the image recognition risk factor.

[0008] Optionally, the working range division unit is used to obtain a predetermined normal working range and divide the working resistance value of the stent based on the predetermined normal working range to obtain a low-pressure range and a high-pressure over-limit range; the low-pressure ratio statistics unit is used to count the proportion of stents in the low-pressure range during the current monitoring period as a support insufficiency indicator; the high-pressure ratio statistics unit is used to count the proportion of stents in the high-pressure over-limit range as a pressure intensity indicator; and the pressure anomaly risk factor determination unit is used to output the pressure anomaly risk factor based on the combination of the support insufficiency indicator and the pressure intensity indicator.

[0009] Optionally, the deviation rate calculation unit is used to calculate the deviation rate of the average mining height of the working face relative to the designed mining height and the degree of exceedance of the mining height fluctuation amplitude relative to the preset fluctuation threshold within the current monitoring period; the mining height risk factor determination unit is used to output the mining height risk factor based on the combination of the deviation rate and the degree of exceedance.

[0010] Optionally, the risk assessment model is a machine learning classification model trained based on historical monitoring data and corresponding actual measurement results of the face slope. It is trained using cross-validation and configured with an incremental learning interface. The risk assessment model also includes a physical constraint verification mechanism, which comprises: calculating the coal wall stability coefficient based on the current working face's geomechanical parameters while executing the machine learning classification model's inference to output the risk level. The coal wall stability coefficient is positively correlated with the roof pressure and mining height. The trend of the coal wall stability coefficient's change is compared with the trend of the risk level output by the machine learning classification model. If the trends are inconsistent, the current model inference result is determined to be abnormal, triggering the model to re-infer and output the risk level corrected by physical constraints. If the trends are consistent, the model inference result is directly output.

[0011] Optionally, the first control measure includes: increasing the support force of the hydraulic support side plate and maintaining the existing cutting parameters of the coal mining machine.

[0012] Optionally, the second control measures include: increasing the support force of the side guard plate and controlling the side guard plate to move closer to the coal wall side to increase the coal wall confinement pressure, while increasing the initial support force replenishment frequency and replenishment target value of the hydraulic support, and reducing the cutting speed of the coal mining machine.

[0013] Optionally, the third control measure includes: further reducing the cutting height of the coal mining machine to reduce the exposed height of the coal wall based on the second control measure, and activating the underground high-flow-rate horizontal long-hole regional fracturing device pre-embedded in the hard roof to inject high-pressure liquid into the directional long borehole to form a three-dimensional fracture network.

[0014] Optionally, the fully mechanized mining face division module is used to divide the fully mechanized mining face into a first section, a second section, and a third section along the dip direction when the fully mechanized mining face is an inclined coal seam; the control strategy execution module is used to assess the coal wall spalling risk level of each of the first, second, and third sections and execute an appropriate adaptive control strategy, wherein a higher control intensity level is configured for the first and second sections where the roof pressure is severe; the joint verification module is used to perform joint verification by combining the pressure sensor data and height sensor data of the hydraulic support corresponding to the monitoring area when a spalling event is determined to exist in the monitoring area of ​​any image acquisition device, so as to reduce the false alarm rate of image recognition.

[0015] Secondly, this application also provides an adaptive control method based on coal face spalling risk assessment. The adaptive control system based on coal face spalling risk assessment includes: acquiring multi-source monitoring data from the fully mechanized mining face, including coal face image data, hydraulic support working resistance data, and actual mining height data; preprocessing and target identification of the coal face image data; outputting an image identification risk factor based on the area ratio of the spalling region when a spalling event is detected; performing interval statistics on the hydraulic support working resistance data; and outputting a pressure anomaly based on the proportion of supports in the low-pressure range and the proportion of supports in the high-pressure over-limit range. Risk factors; statistical analysis is performed on the actual mining height data, and mining height risk factors are output based on the mining height deviation rate and the degree of mining height fluctuation exceeding the limit; the image recognition risk factor, the pressure anomaly risk factor, and the mining height risk factor are input into the risk assessment model, and the coal wall spalling risk level of the fully mechanized mining face is output after reasoning by the risk assessment model; a graded adaptive control strategy is executed according to the coal wall spalling risk level, the graded adaptive control strategy includes: executing a first control measure when the risk level is the first level, executing a second control measure when the risk level is the second level, and executing a third control measure when the risk level is the third level.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: The data acquisition module acquires multi-source monitoring data from the fully mechanized mining face, including coal face image data, hydraulic support working resistance data, and actual mining height data. The target recognition module preprocesses and identifies targets in the coal face image data, outputting an image recognition risk factor based on the area ratio of the detected spalling event. The interval statistics module performs interval statistics on the hydraulic support working resistance data, outputting a pressure anomaly risk factor based on the proportion of supports in the low-pressure interval and the proportion of supports in the high-pressure over-limit interval. The statistical analysis module performs statistical analysis on the actual mining height data. The system outputs mining height risk factors based on the mining height deviation rate and the degree of mining height fluctuation exceeding limits. A risk assessment module inputs the image recognition risk factor, the pressure anomaly risk factor, and the mining height risk factor into a risk assessment model, and outputs the coal wall spalling risk level of the fully mechanized mining face after inference by the risk assessment model. A control strategy execution module executes a graded adaptive control strategy based on the coal wall spalling risk level. The graded adaptive control strategy includes: executing a first control measure when the risk level is level one, executing a second control measure when the risk level is level two, and executing a third control measure when the risk level is level three. In other words, by constructing a multi-source data fusion assessment model based on coal wall image recognition, hydraulic support pressure monitoring, and mining process data, and executing a graded adaptive progressive control strategy, advanced and accurate early warning and differentiated flexible intelligent control of coal wall spalling risk are achieved, improving the safety and production efficiency of coal mining operations.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the adaptive control system based on coal wall spalling risk assessment in this application.

[0020] Figure 2This is a flowchart illustrating the adaptive control method based on coal wall spalling risk assessment proposed in this application.

[0021] Figure labeling: Data acquisition module 11, target identification module 12, interval statistics module 13, statistical analysis module 14, risk assessment module 15, control strategy execution module 16. Detailed Implementation

[0022] This application provides an adaptive control system and method based on coal wall spalling risk assessment. It addresses the technical problems in existing technologies where the single dimension of spalling risk monitoring and the lack of hierarchical progression in control strategies lead to inaccurate identification of hidden spalling signs, resulting in over-intervention or under-intervention, which further impacts the safety and production efficiency of coal mining operations. By constructing a multi-source data fusion assessment model integrating coal wall image recognition, hydraulic support pressure monitoring, and mining process data, and implementing a hierarchical adaptive progressive control strategy, it achieves advanced and accurate early warning and differentiated flexible intelligent control of coal wall spalling risks, thereby improving the safety and production efficiency of coal mining operations.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides an adaptive control system based on coal wall spalling risk assessment, wherein the adaptive control system based on coal wall spalling risk assessment is used to implement the steps of an adaptive control method based on coal wall spalling risk assessment, and the adaptive control system based on coal wall spalling risk assessment includes: The data acquisition module 11 is used to acquire multi-source monitoring data of the fully mechanized mining face, including coal face image data, hydraulic support working resistance data and actual mining height data.

[0025] Specifically, this involves acquiring multi-source monitoring data from the fully mechanized longwall face, which refers to the work area in coal mining where all five core processes—coal breaking, loading, transportation, support, and goaf treatment—are fully mechanized. A mine-use intrinsically safe high-definition camera is fixedly installed in front of the top beam of each hydraulic support in the longwall face, ensuring its field of view completely covers the coal face surface. The camera transmits the video stream in real time to the industrial computer at the face control center via fiber optic or wireless network. Simultaneously, mine-use digital pressure transmitters are installed in the lower cavity of the left and right columns of each hydraulic support, continuously collecting column working resistance data at a sampling frequency of no less than 1Hz and uploading it to the data acquisition substation via the RS485 communication protocol.

[0026] An inertial navigation device and a rocker arm tilt sensor are installed in the middle of the coal mining machine to calculate the current cutting height in real time. Simultaneously, an ultrasonic ranging sensor is installed below the hydraulic support top beam to directly measure the distance between the roof and floor. The two data sources are cross-checked, and the average is taken as the actual cutting height, which is then uploaded to the central control center at a frequency of no less than 0.5Hz. The data acquisition server at the central control center performs time-series alignment and timestamp synchronization on all data to ensure that the three types of data are stored under a unified time reference.

[0027] Multi-source monitoring data is a collection of monitoring information from different sensors and different physical dimensions. Coal wall image data consists of real-time video streams or frame images of the coal wall surface collected by mine explosion-proof high-definition cameras installed on the hydraulic support top beam or scraper conveyor cable trough. This data is used for visual observation of morphological changes in the coal wall, such as crack development and spalling. Hydraulic support working resistance data is the liquid pressure value inside the lower cavity of the hydraulic support column. It is collected in real-time by mine pressure sensors installed on the column, reflecting the active supporting force of the support on the roof and indirectly characterizing the roof load. Actual mining height data is the actual vertical height of the coal wall at the working face after the coal mining machine drum has cut it. It is usually calculated by combining the tilt angle sensor installed on the coal mining machine rocker arm with the machine's attitude, or directly measured by an ultrasonic ranging sensor installed on the hydraulic support.

[0028] The target recognition module 12 is used to preprocess and recognize the target in the coal wall image data. When a rockfall event is detected, the module outputs an image recognition risk factor based on the area ratio of the rockfall region.

[0029] Furthermore, the target recognition module 12 in the adaptive control system based on coal wall spalling risk assessment is also configured to: a filtering preprocessing unit, used to perform filtering preprocessing on the coal wall image data to obtain filtered image data; a frame-by-frame inference unit, used to perform frame-by-frame inference on the filtered image data to obtain target coal, wherein the target coal includes spalling coal and piled coal; a spalling event determination unit, used to determine the existence of a spalling event when the number of detection frames of the detected spalling coal reaches a preset threshold and the detection frame of the piled coal is detected at the same time, and to use the bounding rectangle of all detection frames as the spalling area; an area ratio calculation unit, used to calculate the ratio of the area of ​​the spalling area to the area of ​​the entire image as the spalling area ratio; and an image recognition risk factor determination unit, used to match the preset interval to which the spalling area ratio belongs and output the image recognition risk factor.

[0030] Specifically, the coal face image data undergoes filtering preprocessing. Median filtering is performed for noise reduction using a 5×5 pixel sliding window. The grayscale values ​​of all pixels within the window are sorted by size, and the median value is used to replace the original value of the center pixel, eliminating isolated bright or dark spots caused by underground dust particles obscuring the camera. After median filtering, Gaussian filtering is performed for smoothing. A 3×3 Gaussian kernel with σ=0.8 is used to convolve the coal face image data, further suppressing Gaussian noise while preserving high-frequency features such as crack edges and blocky textures of the coal face. After filtering, the image resolution remains unchanged. A combination of median and Gaussian filtering is used. Median filtering (e.g., using a 5×5 sliding window) removes isolated impulse noise, such as bright or dark spots caused by dust particles, while Gaussian filtering smooths the image, suppresses Gaussian noise, and preserves the texture and edge information of the coal face.

[0031] Filtered image data is fed into an object detection model deployed in a frame-by-frame inference unit for inference. YOLOv8m is used as the basic detection architecture. This model has been trained offline using a coal wall dataset containing over 15,000 labeled images, covering diverse scenarios with varying lighting conditions, dust concentrations, and coal wall failure morphologies. Before inference, the original image size is proportionally scaled to the model's standard input size. During scaling, gray edges are padded around the image to maintain the original aspect ratio and avoid distortion caused by stretching. The pixel values ​​of the scaled image are then normalized to the 0-1 range and standardized along the RGB channels. After forward propagation, the object detection model outputs three feature maps at different scales, corresponding to downsampling by 8x, 16x, and 32x, respectively. Each grid point on each feature map predicts multiple candidate detection boxes and their class probabilities. The raw detection results output by the object detection model are post-processed, applying a confidence threshold for filtering. For example, with a confidence threshold of 0.65, only candidate detection boxes with a confidence score ≥ 0.65 are retained, while those below this threshold are directly filtered out. The retained detection boxes are then sorted from highest to lowest confidence score, and non-maximum suppression is performed. An intersection-union (IU) threshold of 0.45 is set to merge overlapping boxes that repeatedly predict the same target. Specifically, when the IU of two detection boxes exceeds 0.45, only the one with the higher confidence score is retained, and the other is deleted. After post-processing and filtering, the final retained detection boxes are the final output of this inference. Each detection box contains three core pieces of information: a category label, normalized coordinates, and a confidence score. The normalized coordinates are restored to pixel coordinates in the original image coordinate system, and all detection boxes are stored in memory as a structure array.

[0032] Two types of coal targets directly related to coal face spalling were identified and located from the images: spalled coal and coal piles. Spalled coal refers to coal blocks or groups that have detached from the coal face but remain suspended on the surface, in the process of falling, or have just detached and not yet reached the floor. Visually, spalled coal is typically located in the upper-middle part of the coal face, with obvious cracks or shadow lines separating it from the surrounding intact coal face. Its boundaries are relatively clear, and its shape is often irregular and blocky. Its presence directly reflects localized coal face failure. Coal piles are loosely accumulated coal bodies that have completely detached from the coal face and fallen onto the scraper conveyor or the working face floor. Visually, coal piles are located in the bottom area of ​​the image, with a loosely accumulated surface and obvious granular texture, contrasting sharply with the smooth, intact coal face. The presence of coal piles is strong evidence that spalling has occurred or is occurring, as spalling inevitably leads to the downward accumulation of detached coal.

[0033] For the current frame, count the number of coal spalling detection frames retained. If the number of coal spalling detection frames is greater than or equal to a preset threshold (e.g., 3), and at least one coal pile detection frame exists simultaneously, then a coal spalling event is determined to have occurred in the monitoring area corresponding to the current frame. The preset threshold is a pre-set minimum number of coal spalling detection frames, calibrated according to actual working conditions, and is usually set to 3 to 5. This is used to exclude accidental single-frame detections caused by noise or model false detections, ensuring that the determination of a coal spalling event is statistically significant.

[0034] The logic behind the simultaneous detection of coal piles in the judgment condition is that coal wall spalling inevitably leads to the accumulation of detached coal at the bottom. Therefore, the co-occurrence of spalling coal detection frames and coal pile detection frames in the same frame is strong evidence of spalling events, effectively eliminating false alarms caused by small-scale localized coal detachment from the coal wall surface. The circumscribed rectangle is the smallest rectangular area that can exactly contain all the detected spalling coal and coal pile detection frames after merging them in the image coordinate system.

[0035] Once a coal spalling event is detected, all coal spalling detection frames and coal pile detection frames in the current frame are extracted. The minimum x-coordinate and minimum y-coordinate of the top left corner of each detection frame, as well as the maximum x-coordinate and maximum y-coordinate of the bottom right corner, are taken to form a minimum bounding rectangle, which is used as the coal spalling region of the frame.

[0036] Multiply the pixel width of the bounding rectangle by its pixel height to obtain the pixel area of ​​the border region. The total pixel area of ​​the entire image = image width × image height. The border area ratio is the ratio of the border region's pixel area to the total pixel area.

[0037] The calculated percentage of spalled coal area is matched with preset area percentage ranges. For example, if range 1 is 0% ≤ spalled coal area ≤ 5%, the corresponding image recognition risk factor is 0.1; range 2 is 5% < spalled coal area ≤ 15%, the corresponding image recognition risk factor is 0.4; range 3 is 15% < spalled coal area ≤ 30%, the corresponding image recognition risk factor is 0.7; and range 4 is spalled coal area > 30%, the corresponding image recognition risk factor is 1.0. The matching logic adopts the left-closed, right-open interval principle and outputs the image recognition risk factor corresponding to the matching interval. If the current frame does not trigger the spalled coal event determination in step 3, i.e., the number of spalled coal detection frames is < 3 or there are no coal pile detection frames, the image recognition risk factor is directly output as 0, indicating that there is no spalled coal risk at present.

[0038] For example, suppose the original resolution of a coal seam image at a certain support is 1920×1080 pixels, then the area of ​​the entire image is 2,073,600 pixels. 2. The image is preprocessed by Gaussian filtering (kernel size 3×3, Sigma=1.0) and then input into the YOLOv5s model. The model inference result shows that a total of 4 coal spallation detection boxes are identified in the image, which meets the preset quantity threshold of ≥3, and their coordinates are [120,300,260,550], [280,320,400,580], [420,310,500,560], [520,290,610,570] respectively, and 1 coal pile detection box located at the bottom is identified at the same time, with coordinates of [100,800,900,1050]. It is determined that a spallation event exists. Then calculate the global coordinate extreme values of all detection boxes: x min =100, y min =290, x max =900, y max =1050. Thus, the area of the bounding rectangle of the spallation area is determined to be 608000pixel 2 . The calculated proportion of the spallation area is about 0.293. According to the preset interval rule, it belongs to the high-risk interval of 15%<R≤30%, and the final output image recognition risk factor is 0.7.

[0039] By introducing the dual determination logic of the coexistence of the number of spalled coal and coal pile, the anti-interference ability and accuracy of spallation event identification are significantly improved, the interference of humid and dark underground environment, water stains and shadows on visual recognition is effectively overcome, and the false alarm rate is greatly reduced. By quantifying the visual features into a specific area proportion and mapping it to the image recognition risk factor, the transformation from unstructured image data to structured risk indicators is realized.

[0040] The interval statistics module 13 is configured to perform interval statistics on the working resistance data of the hydraulic supports, and output the pressure abnormality risk factor according to the proportion of the number of supports in the low-pressure interval and the proportion of the number of supports in the high-pressure over-limit interval.

[0041] Further, the interval statistics module 13 in the adaptive control system based on coal wall spallation risk assessment is further configured with: a working interval division unit, configured to obtain a predetermined normal working interval, and divide the working resistance values of the supports based on the predetermined normal working interval to obtain a low-pressure interval and a high-pressure over-limit interval; a low-pressure proportion statistics unit, configured to count the proportion of the number of supports in the low-pressure interval in the current monitoring period as the characterization index of insufficient support; a high-pressure proportion statistics unit, configured to count the proportion of the number of supports in the high-pressure over-limit interval as the characterization index of weighting intensity; a pressure abnormality risk factor determination unit, configured to output the pressure abnormality risk factor according to the combination of the insufficient support characterization index and the weighting intensity characterization index.

[0042] Specifically, during the initialization phase, the hydraulic support model and its rated working resistance value are retrieved from the working face design file. Using the rated working pressure of the lower cavity of the support column as a benchmark, a predetermined normal working range is defined according to industry-standard rules. This range is based on the column pressure corresponding to the rated working resistance, with 60% to 85% of this pressure serving as the lower and upper limits. The lower limit is 0.6 times the rated working pressure, and the upper limit is 0.85 times the rated working pressure. A reading below the lower limit indicates insufficient initial support force or delamination of the top plate, resulting in ineffective support bearing (underpressure). A reading above the upper limit indicates severe pressure from the top plate and excessive load on the support (overpressure). Based on the predetermined normal working range, all support working resistance values ​​in the hydraulic support working resistance data are divided into low-pressure, normal, and high-pressure over-limit ranges.

[0043] The arithmetic mean of all sampled values ​​of each hydraulic support in the working resistance data of the hydraulic support is taken to obtain the representative pressure value of the hydraulic support in this period. This eliminates the instantaneous spike interference caused by random noise such as coal mining machine cutting vibration and hydraulic system pulsation, so that the statistical results reflect the average load state of the support in this period.

[0044] The representative pressure value of each hydraulic support is compared with the predetermined normal operating range. The number of supports with representative pressure values ​​lower than the lower limit of the predetermined normal operating range is counted. The ratio of the number of supports with representative pressure values ​​lower than the lower limit of the predetermined normal operating range to the total number of hydraulic supports is the indicator of insufficient support. The number of supports with representative pressure values ​​higher than the upper limit of the predetermined normal operating range is counted. The ratio of the number of supports with representative pressure values ​​higher than the upper limit of the predetermined normal operating range to the total number of hydraulic supports is the indicator of pressure strength.

[0045] The system comprehensively assesses both the inadequate support indicators and the pressure intensity indicators, calculating the overall risk percentage and taking the larger value. The rationale is that regardless of whether it's low or high pressure, a large-scale anomaly in either direction significantly increases the risk of coal wall spalling. Based on the range of the overall risk percentage, a corresponding pressure anomaly risk factor is output using a four-level mapping: 0 for 0%; 0.1 for 0 < overall risk percentage ≤ 10%; 0.4 for 10% < overall risk percentage ≤ 25%; 0.7 for 25% < overall risk percentage ≤ 40%; and 1 for overall risk percentage > 40%. This four-level range is based on statistical analysis of mine pressure monitoring data and spalling events from multiple working faces. Statistics show that when the abnormal support percentage is below 10%, the probability of spalling events is low; when the percentage exceeds 25%, the probability of spalling increases significantly; and when it exceeds 40%, large-scale spalling is almost always present. The pressure anomaly risk factor is a discrete risk characteristic value output based on the combination of the support insufficiency indicator and the pressure intensity indicator, comprehensively reflecting the abnormal state of the working face pressure field. For example, if the support insufficiency indicator is 25.3% and the pressure intensity indicator is 14.7%, then the overall risk ratio is 25.3%, and therefore the pressure anomaly risk factor is 0.7.

[0046] By statistically analyzing the overall pressure distribution of all supports rather than monitoring the pressure values ​​of one or two supports in isolation, misjudgments caused by single sensor failures or localized instantaneous disturbances are effectively avoided. Simultaneous monitoring of anomalies in two opposite directions—low pressure and high pressure exceeding limits—achieves comprehensive coverage of the two main mechanical causes of coal wall spalling. Both can induce spalling, but their mechanisms differ; the parallel monitoring of these two indicators enables complete coverage of the mechanical causes.

[0047] The statistical analysis module 14 is used to perform statistical analysis on the actual mining height data and output mining height risk factors based on the mining height deviation rate and the degree of mining height fluctuation exceeding the limit.

[0048] Furthermore, the statistical analysis module 14 in the adaptive control system based on coal face spalling risk assessment is also used for: a deviation rate calculation unit, used to calculate the deviation rate of the average mining height of the working face relative to the designed mining height and the degree of exceedance of the mining height fluctuation amplitude relative to the preset fluctuation threshold within the current monitoring period; and a mining height risk factor determination unit, used to output the mining height risk factor based on the combination of the deviation rate and the degree of exceedance.

[0049] Specifically, the actual mining height data is the actual vertical height of the coal face after the coal mining machine drum has cut it. It is calculated by the tilt sensor installed on the rocker arm of the coal mining machine in combination with the machine body attitude, or directly measured by the ultrasonic ranging sensor installed under the top beam of the hydraulic support. It reflects the actual execution of the current mining process parameters and is an important geometric parameter for evaluating the exposed height and stress state of the coal face.

[0050] The current monitoring period is a set fixed time window, which is used for statistical analysis of the mining height data within the time period. For example, it is set to 10 minutes to ensure the alignment and comparability of multi-source data in the time dimension, while ensuring that there are sufficient sampling points. Calculated at a sampling rate of 0.5Hz, there are 300 data points in 10 minutes in total, making the statistical results representative.

[0051] The arithmetic average of the actual mining height data in the current monitoring period is taken to obtain the actual average mining height. The designed mining height is read, and the mining height deviation rate is calculated, that is, the mining height deviation rate D=(H-H1) / H1 ×100%, where H is the actual average mining height, and H1 is the designed mining height. The larger the mining height deviation rate, the farther the overall deviation of the actual mining height from the designed value. When the actual average mining height is greater than the designed mining height, the exposed height of the coal wall increases, and the cantilever length and loaded area of the coal wall increase; when the actual average mining height is less than the designed mining height, although the exposed height decreases, it may mean that the cutting is not in place, the cutting state of the shearer is abnormal, or the fluctuation of the roof and floor is too large, which is also not conducive to surrounding rock control.

[0052] Taking the designed mining height as the benchmark, the preset fluctuation threshold is set at a ratio of ±7%: the upper limit is 1.07 times the designed mining height, and the lower limit is 0.93 times the designed mining height. The relationship between each mining height sampling value in the current period and the upper and lower limits is compared one by one. A point is defined as an over-limit point if it is greater than the upper limit value or less than the lower limit value, the number of over-limit points is counted, and the ratio of the number of over-limit points to the total number of sampling points is calculated to obtain the proportion of over-limit points. The over-limit degree is the ratio of the amplitude by which the instantaneous value of the actual mining height exceeds the preset fluctuation threshold to the threshold itself, which quantifies the severity of mining height fluctuation. The greater the over-limit degree, the more severe the mining height fluctuation, the farther it exceeds the normal range, and the stronger the dynamic load disturbance to the coal wall.

[0053] A comprehensive judgment is made on the deviation rate and the over-limit degree: first, the basic risk level is determined according to the value of the deviation rate, then the risk is corrected according to the over-limit degree, and finally the larger one of the corresponding risk values of the two is taken as the mining height risk factor. The four-level intervals of the deviation rate D: D≤5% corresponds to 0.1, 5%<D≤12% corresponds to 0.4, 12%<D≤20% corresponds to 0.7, and D>20% corresponds to 1.0; the four-level intervals of the average over-limit degree E: E≤0.1 corresponds to 0.1, 0.1<E≤0.35 corresponds to 0.4, 0.35<E≤0.6 corresponds to 0.7, and E>0.6 corresponds to 1.0. The final mining height risk factor = max (the risk value corresponding to D, the risk value corresponding to E). The logic of taking the larger value of the two lies in that the deviation rate reflects the overall deviation risk of the absolute value of the mining height, and the over-limit degree reflects the disturbance risk of the instantaneous fluctuation of the mining height. The two have independent and superimposed effects on the stability of the coal wall, and taking the maximum value represents the most unfavorable risk scenario.

[0054] Risk assessment module 15 is used to input the image recognition risk factor, the pressure anomaly risk factor and the mining height risk factor into the risk assessment model, and output the coal wall spalling risk level of the fully mechanized mining face after reasoning by the risk assessment model.

[0055] Furthermore, the risk assessment module 15 in the adaptive control system based on coal wall spalling risk assessment is also used for: the risk assessment model is a machine learning classification model trained based on historical monitoring data and corresponding actual spalling results, trained using cross-validation and configured with an incremental learning interface; wherein, the risk assessment model is also configured with a physical constraint verification mechanism, the physical constraint verification mechanism includes: while executing the machine learning classification model to infer and output the risk level, calculating the coal wall stability coefficient based on the current working face's geomechanical parameters, the coal wall stability coefficient being positively correlated with roof pressure and mining height; comparing the change trend of the coal wall stability coefficient with the change trend of the risk level output by the machine learning classification model; if the two change trends are inconsistent, it is determined that the current model inference result is abnormal, triggering the model to re-infer and output the risk level corrected by physical constraints; if the two change trends are consistent, the model inference result is directly output.

[0056] Specifically, the risk assessment model is a decision-making model that integrates image recognition risk factors, pressure anomaly risk factors, and mining height risk factors to output the risk level of coal wall spalling. The risk assessment model is a supervised learning-based machine learning classification model; the input is a three-dimensional feature vector, and the output is a discrete risk level, namely, level one, level two, and level three.

[0057] Historical monitoring data for at least three months were collected from fully mechanized mining faces or adjacent working faces with similar geological conditions. Each sampling point included three risk factor values: image factor, pressure anomaly factor, and mining height risk factor, along with a corresponding severity label for rockfall. Rockfall severity labels were determined through on-site manual inspection records or point cloud change analysis from a 3D laser scanner, and were categorized into three levels. After cleaning and standardization, the dataset was trained using a 5-fold cross-validation method: all samples were randomly divided into five equal parts, with one part used as the validation set and four parts as the training set. An XGBoost multi-class classification model was trained, and after five rounds of this rotation, the average classification accuracy and F1-score were used as model performance evaluation metrics. After model training, the data was exported in ONNX format and deployed to the inference server in the control center. An incremental learning interface was also configured, allowing the system to incrementally update the model monthly using newly accumulated labeled data. Specifically, the risk assessment model uses the XGBoost algorithm as the basic classifier. Its internal structure consists of three parts: the input layer receives three-dimensional feature vectors, the hidden layer consists of 200 classification decision trees with a maximum depth of 6 layers, and the output layer uses the Softmax function to summarize the prediction results of each decision tree into a probability distribution of three risk levels. During the offline training phase, sample data containing three dimensions of features—image recognition risk factors, pressure anomaly risk factors, and mining height risk factors—as well as the corresponding on-site measured slope level labels are extracted from the historical database of the working face. After data cleaning to remove samples from sensor failure periods and those inconsistent with manual annotations, all samples are randomly shuffled and divided into 5 subsets. A 5-fold cross-validation strategy is used for training. One subset is used as the validation set, and the other four subsets are used as the training set for model training and performance evaluation. During each fold training process, the model drives 200 decision trees to grow one by one by minimizing the log loss function. Each tree randomly selects a subset of features from the three-dimensional features to find the optimal split node. The maximum depth of the trees is limited to 6 layers to prevent overfitting. The learning rate is set to 0.1 to control the contribution weight of each tree. The training of that fold is terminated early when the loss of the validation set no longer decreases after 10 consecutive iterations. The final output is a weighted average of the combined 5-fold models, or the single-fold model with the best performance on the validation set is selected as the final deployment model. After training, the data is exported in an open neural network exchange format and deployed to the inference server of the central control center.

[0058] Regarding the integration of the risk assessment model with specific mining scenarios, the three input features have a clear physical causal relationship with the risk of coal face spalling. The image recognition risk factor directly characterizes the visible damage scale that has occurred in the coal face. The pressure anomaly risk factor comprehensively reflects the active control capability of the support over the roof and the degree of passive compression of the coal face by the roof pressure. The mining height risk factor quantifies the disturbance effect of the deviation of the mining height from the design value and the cutting fluctuation on the stability of the coal face. The three factors jointly depict the full picture of coal face spalling risk from three independent dimensions: morphology, mechanics, and technology, making the input features of the model have clear physical interpretability. When the risk assessment model is inferring, it receives the three-dimensional feature vector generated by the front-end perception system within the current monitoring period. After step-by-step judgment by 200 decision trees and Softmax normalization, it outputs the probability values ​​of the three risk levels. The level with the highest probability is taken as the coal face spalling risk level of the current working face and output to the execution control module. Regarding the online operation mechanism, the risk assessment model has a built-in incremental learning interface. At the end of each month, it automatically uses newly collected sample data that has been manually verified and labeled on-site for incremental model updates. The newly added samples account for about 5% to 10% of the total sample size. During incremental learning, the weight parameters of the original 200 decision trees are retained, and 20 to 30 small-weight decision trees are added on top of them to fit the new data distribution. At the same time, a sliding window mechanism is used to automatically eliminate historical samples older than 6 months to maintain the timeliness of the model in the current mining pressure pattern. The incremental update process is completed offline in the background without interrupting the online inference service. After the update is completed, the new model parameters are automatically loaded and take effect in the next inference.

[0059] Two computational chains are executed in parallel within each monitoring cycle. Chain 1 is the model inference chain, which receives image recognition risk factors, pressure anomaly risk factors, and mining height risk factors, concatenates them into a three-dimensional feature vector, inputs it into the XGBoost model for forward inference, and outputs probability values ​​corresponding to the three risk levels. The level corresponding to the highest probability is taken as the preliminary risk level output. Chain 2 is the physical constraint verification chain, which is used to compare the inference results with the output of the machine learning model to identify whether the risk assessment model has produced AI illusion-like erroneous outputs when encountering new and unseen working conditions. At the same time, it reads the geomechanical parameters of the current working face from the database, which are key parameters characterizing the mechanical properties of the coal seam and the roof and floor rock mass, including experimental parameters such as coal cohesion c, internal friction angle φ, and unit weight γ, as well as the average roof pressure and average mining height of the current cycle. These are substituted into the coal wall stability coefficient calculation model to obtain the specific value of the coal wall stability coefficient for that cycle.

[0060] The historical values ​​of coal wall stability coefficient and risk level output from the risk assessment model for the most recent three consecutive monitoring periods are cached in memory. After each new period is calculated, the trend of stability coefficient over the most recent three periods is calculated. If the three values ​​increase successively, it is marked as an upward trend; if they decrease successively, it is marked as a downward trend; if there is no obvious unidirectional change, it is marked as a flat trend. The trend of risk level over the most recent three periods is also determined and marked as upward / downward / flat. The consistency of the two trend markings is compared: if both are in the same direction (both are upward or both are downward) or both are flat, it is determined that the trend is consistent, and the inference result of the risk assessment model is directly output as the final risk level; if one is upward and the other is downward, in opposite directions, it is determined that the trend is inconsistent, and the anomaly handling mechanism is triggered.

[0061] When a trend inconsistency is triggered, the physical constraint correction procedure is initiated. It checks whether the average roof pressure and mining height of the current cycle exceed the normal range of historical statistical distribution. If they exceed ±2 standard deviations of the historical average, the current operating condition is determined to have entered an unknown region not covered by the model training data. In this case, the risk assessment model output is not adopted; instead, a preset conservative strategy is used: if the risk assessment model outputs level one and the physical coefficients show an upward trend, the output level is upgraded to level two; if the risk assessment model outputs level two and the physical coefficients show an upward trend, the output level is upgraded to level three; if the risk assessment model has already output level three, it remains unchanged. The corrected risk level is used as the final output, and a physical constraint intervention log is generated for subsequent risk assessment model optimization and manual verification. If the current operating condition is within the historical data coverage but the model output and physical coefficient trends are inconsistent, the risk assessment model is triggered to re-infer. The current feature vector is input into the model multiple times, and a dropout strategy (randomly discarding parts of the decision tree) is used to observe the output stability. If the multiple inference results are inconsistent, the output is corrected according to the majority vote result.

[0062] Purely data-driven models may output unreasonable or even erroneous results when encountering extreme geological conditions not present in the training data, i.e. AI illusion. By introducing an independent physical constraint verification mechanism, using the stability coefficient of classical rock mechanics as a second opinion, the model output is reverse-verified, fundamentally improving the reliability and safety under unknown working conditions.

[0063] The control strategy execution module 16 is used to execute a graded adaptive control strategy according to the coal wall spalling risk level. The graded adaptive control strategy includes: executing a first control measure when the risk level is the first level, executing a second control measure when the risk level is the second level, and executing a third control measure when the risk level is the third level.

[0064] Furthermore, the control strategy execution module 16 in the adaptive control system based on coal wall spalling risk assessment is also used for: the first control measures include: increasing the support force of the hydraulic support side plate and maintaining the existing cutting parameters of the coal mining machine.

[0065] Furthermore, the control strategy execution module 16 in the adaptive control system based on coal wall spalling risk assessment is also used for: the second control measures include: increasing the support force of the side plate and controlling the side plate to move closer to the coal wall side to increase the coal wall confinement pressure, while increasing the initial support force replenishment frequency and replenishment target value of the hydraulic support, and reducing the cutting speed of the coal mining machine.

[0066] Furthermore, the control strategy execution module 16 in the adaptive control system based on coal wall spalling risk assessment is also used for: the third control measure includes: further reducing the cutting height of the coal mining machine to reduce the exposed height of the coal wall on the basis of the second control measure, and activating the underground high-flow horizontal long-hole regional fracturing device pre-embedded in the hard roof to inject high-pressure liquid into the directional long borehole to form a three-dimensional fracture network.

[0067] Specifically, based on the different risk levels output by the risk assessment model, an intelligent decision-making mechanism automatically triggers sets of control measures of varying intensities and types. After the risk assessment is completed in each monitoring cycle, the control strategy execution module receives the risk level determination results. The control strategy execution module has three pre-set independent and progressively hierarchical sets of control instructions, which are automatically matched and issued for execution according to the risk level.

[0068] If the risk level is Level 1, the control strategy execution module sends a backing force adjustment command to the hydraulic support electro-hydraulic control system via industrial Ethernet. This increases the pressure compensation target value of each support backing plate jack from the daily maintenance value to 85% of the rated working pressure (e.g., from 30MPa to 25.5MPa if the rated pressure is 30MPa). The pressure compensation frequency remains at the existing setting, such as automatic pressure compensation every 15 minutes. This first control measure only involves adjusting the pressure compensation parameters of the backing plate jacks and does not send any speed or height adjustment commands to the coal mining machine control system. The coal mining machine maintains its current cutting speed and cutting height, continuing to advance with conventional parameters to ensure that production efficiency is not affected.

[0069] If the risk level is Level 2, the control strategy execution module sends commands to the hydraulic support electro-hydraulic control system and the coal mining machine traction control system in parallel. Regarding support control: the target pressure replenishment value of the sidewall jacks is further increased to 95% of the rated working pressure, such as to 28.5 MPa; the sidewall jacks are extended via the electro-hydraulic directional valve, causing the front end of the sidewall to actively move towards the coal wall until the sidewall and coal wall surface form a tight contact, with a target contact confining pressure set not less than 0.3 MPa; the pressure replenishment frequency is increased from once every 15 minutes to once every 5 minutes, using more frequent pressure replenishment to cope with the pressure attenuation caused by roof subsidence. Regarding coal mining machine control: a speed reduction command is sent to the coal mining machine traction inverter, linearly reducing the cutting speed from the current value (e.g., 8 m / min) to the target value (e.g., 5 m / min) within 5 seconds, to reduce the transient impact energy of the cutting teeth on the coal wall. All commands adopt a ramp transition method, that is, the control quantity changes smoothly from the current value to the target value within 3 to 5 seconds, avoiding the impact of abrupt changes on the hydraulic system and the coal mining machine traction system.

[0070] If the risk level is level three, the control strategy execution module adds two more powerful control actions to the second control measure. An additional cutting height adjustment command is sent to the coal mining machine control system. By adjusting the angle of the coal mining machine's rocker arm, the drum cutting height is gradually reduced to the designed mining height, such as from the current 6.8m to 6.2m. The reduction process is completed uniformly within 15 seconds to reduce the vertical height of the exposed coal face and the length of the lever arm. A start command is sent to the underground high-flow-rate horizontal long-hole fracturing unit pre-embedded in the hard roof of the working face. This command is transmitted to the control cabinet of the fracturing pump station, activating the high-flow-rate high-pressure pump. High-pressure fluid with a pressure of no less than 35 MPa is injected at a rate of no less than 500 L / min into the directional long borehole already drilled in the roof strata. Fracturing continues for 20 to 30 minutes. The high-pressure fluid expands fractures within the natural joints and bedding of the roof strata, forming a three-dimensional fracture network. This reduces the overall stiffness and energy storage capacity of the hard roof, and the pressure energy from the roof is gradually dissipated along the fracture network, thereby reducing the peak mining stress acting on the coal face. During the fracturing operation, the system continuously monitors the injection pressure and flow rate parameters of the fracturing pump station. Fracturing automatically stops when the pressure steadily decreases or the injection volume reaches the design value. Three-dimensional fracture network is a three-dimensional network of fractures formed by the expansion of high-pressure fluid along natural joints and newly formed fractures in the roof strata. The formation of three-dimensional fracture network significantly reduces the integrity and strength of the roof strata, allowing the pressure energy from the roof to be dissipated through the opening and sliding of the fracture network, thereby reducing the peak stress transmitted to the coal face during mining.

[0071] Level 1 control can eliminate minor hidden dangers with zero production loss; Level 2 control effectively controls the development of medium risks through a combination of speed reduction, pressure replenishment, and side support; Level 3 control demonstrates extremely strong emergency response capabilities, especially with the introduction of downhole fracturing technology, which reduces roof pressure from the source and transforms passive support into active pressure relief; significantly improving the working face's disaster resistance and intelligent mining level.

[0072] Furthermore, the adaptive control system based on coal wall spalling risk assessment is also used for: a fully mechanized mining face division module, used to divide the fully mechanized mining face into a first section, a second section, and a third section along the dip direction when the fully mechanized mining face is an inclined coal seam; a control strategy execution module 16, used to assess the coal wall spalling risk level of each section of the first section, the second section, and the third section respectively and execute an appropriate adaptive control strategy, wherein a higher control intensity level is configured for the first section and the second section where the roof pressure is severe; and a joint verification module, used to perform joint verification by combining the pressure sensor data and height sensor data of the hydraulic support corresponding to the monitoring area when a spalling event is determined to exist in the monitoring area of ​​any image acquisition device, so as to reduce the false alarm rate of image recognition.

[0073] Specifically, when the fully mechanized longwall face is an inclined coal seam, the design parameters of the longwall face are read, including the dip length and the dip angle of the coal seam. An inclined coal seam is defined as a coal seam with a dip angle greater than 8° and less than 45°. Unlike near-horizontal coal seams, during the mining of inclined coal seams, the roof pressure undergoes significant component decomposition along the dip direction of the longwall face. The component of gravity along the dip direction leads to uneven stress distribution in different sections of the longwall face. Typically, the lower and middle parts of the longwall face are more complexly stressed due to the effects of roof sliding force and gangue filling in the goaf, resulting in significant zonal differences in the risk of coal wall spalling.

[0074] Based on on-site mine pressure monitoring data at the working face, such as the periodic pressure step distance and pressure intensity distribution data of each support along the dip direction, the working face is divided into three consecutive sections along the dip direction. The first section corresponds to the lower part of the working face (0 to L / 3 range), the second section corresponds to the middle part of the working face (L / 3 to 2L / 3 range), and the third section corresponds to the upper part of the working face (2L / 3 to L range). Each section contains approximately one-third of the total number of hydraulic supports. Each section is equipped with a corresponding group of image acquisition devices, a group of pressure sensors, and a group of height sensors. The monitoring data of each group are independently statistically analyzed and calculated in subsequent processing.

[0075] Within each monitoring cycle, independent feature extraction and risk factor calculation were performed for the three sections. For image recognition, each section calculated its image recognition risk factor based solely on images from cameras within its own monitoring area; specifically, the maximum percentage of the area covered by all cameras within a section was used as the image factor for that section. For pressure statistics, the percentage of low-pressure and high-pressure exceeding limits for the supports within each section was statistically analyzed, and the pressure anomaly factor for each section was calculated. For mining height statistics, since the mining height data was continuously sampled along the entire length of the working face, the data was allocated to corresponding sections based on the spatial location of the sampling points, and the mining height deviation rate and fluctuation exceeding limit degree for each section were calculated. Ultimately, each section obtained a set of three-dimensional feature vectors: image recognition risk factor, pressure anomaly risk factor, and mining height risk factor.

[0076] The three-dimensional feature vectors of each of the three sections are input into the risk assessment model for independent reasoning to obtain the risk levels of the first, second, and third sections. Corresponding control measures are matched according to the risk level of each section, with the enforcement intensity of the control measures weighted by the section location. Specifically, based on historical mine pressure observation data and mechanical analysis of inclined coal seams, areas of intense roof pressure are typically located in the lower part of the working face (first section) and the lower-middle part of the working face (second section). Therefore, higher control intensity levels are configured for the first and second sections. That is, when the same risk level occurs in the first or second section, the actual control parameters are more biased towards the upper limit of that level or directly increased by half a level; for example, if the original risk level was second level, parameters close to the third level are implemented. For the third section (upper part of the working face), standard level parameters are implemented or slightly reduced. Control commands for each section are issued independently to the hydraulic support group and coal mining machine of the corresponding section. When the coal mining machine advances along the dip, its cutting parameters are adjusted uniformly across the entire working face according to the requirements of the high-risk section. However, local control measures such as side protection plates are executed independently by section.

[0077] When a coal face spalling event is detected within the monitoring area of ​​any image acquisition device, the event is not immediately confirmed. Instead, a joint verification procedure is initiated, automatically retrieving pressure and height sensor data from the hydraulic supports within the monitoring area corresponding to that image acquisition device. Each camera covers the area of ​​2 to 3 supports, retrieving pressure and mining height sampling values ​​for all supports within that area within 10 seconds before and after the specified moment. The pressure data of the supports in that area is checked for abnormal drops. If the average pressure value at the current moment drops by more than 8% compared to the average value within the previous 5 minutes, it indicates that the roof load transfer after coal face spalling has caused unloading of the supports, and the pressure verification passes. The height sensor data in that area is checked for abnormal fluctuations. If the mining height value at the current moment increases by more than 0.1m compared to the average value within the previous minute, it indicates that the cutting height measurement value of the coal mining machine after coal face spalling has temporarily increased or suddenly decreased due to coal collapse, indicating abnormal changes in distance measurement caused by coal pile cover, and the height verification passes. If the image recognition judgment matches the logic of pressure verification and height verification, that is, if the image judgment shows a rockfall event with a decrease in pressure and an abnormal height, then the rockfall event is confirmed as a real event, and the corresponding image recognition risk factor is output. If the image judgment shows a rockfall event but the pressure data does not show an abnormal decrease and the height data does not show an abnormal fluctuation, then the image recognition is determined to be a false alarm, such as the model misdetection caused by shadows on the coal wall surface, changes in lighting, or dust obstruction. No risk factor is output, and the false alarm sample is automatically recorded.

[0078] The stress state and pressure characteristics of different sections of the working face in inclined coal seams vary significantly. The lower and middle sections typically experience higher stress and more intense roof pressure, while the upper section is relatively less affected. Zonal control provides stronger support and gentler cutting parameters to high-stress areas, while maintaining normal advance speed in low-stress areas. This maximizes production efficiency while ensuring safety, avoiding large-scale efficiency losses caused by conservative control of the most dangerous sections across the entire working face. Image recognition has a certain probability of false detection in complex underground environments. By introducing pressure and height sensor data for joint verification of image recognition results, a roof collapse event is confirmed only when three conditions are simultaneously met: visually observed roof collapse, mechanically observed support unloading, and geometrically observed sudden changes in mining height. This forms a visual-mechanical-geometric three-modal interlocking verification mechanism, preventing ineffective control actions and production interruptions caused by frequent false alarms.

[0079] In summary, the adaptive control system based on coal face spalling risk assessment provided in this application has the following technical effects: A data acquisition module is used to acquire multi-source monitoring data from the fully mechanized mining face, including coal face image data, hydraulic support working resistance data, and actual mining height data; a target recognition module is used to preprocess and identify targets in the coal face image data, and outputs an image recognition risk factor based on the area ratio of the spalling region when a spalling event is detected; an interval statistics module is used to perform interval statistics on the hydraulic support working resistance data, and outputs a pressure anomaly risk factor based on the proportion of supports in the low-pressure interval and the proportion of supports in the high-pressure over-limit interval; statistical analysis... The system comprises three modules: a first module for statistical analysis of actual mining height data, outputting mining height risk factors based on mining height deviation rate and mining height fluctuation exceeding limits; a second module for inputting image recognition risk factors, pressure anomaly risk factors, and mining height risk factors into a risk assessment model, which then outputs the coal wall spalling risk level of the fully mechanized mining face after inference by the model; and a third module for executing a tiered adaptive control strategy based on the coal wall spalling risk level. This tiered adaptive control strategy includes: executing a first control measure when the risk level is first level, executing a second control measure when the risk level is second level, and executing a third control measure when the risk level is third level. In other words, by constructing a multi-source data fusion assessment model integrating coal wall image recognition, hydraulic support pressure monitoring, and mining process data, and executing a tiered adaptive progressive control strategy, advanced and accurate early warning and differentiated flexible intelligent control of coal wall spalling risk are achieved, improving the safety and production efficiency of coal mining operations.

[0080] Example 2: Based on the same inventive concept as the adaptive control system based on coal wall spalling risk assessment in Example 1, this application also provides an adaptive control method based on coal wall spalling risk assessment. Please refer to the appendix. Figure 2 The adaptive control system based on coal wall spalling risk assessment includes: Multi-source monitoring data of the fully mechanized mining face is acquired, including coal wall image data, hydraulic support working resistance data, and actual mining height data. The coal wall image data is preprocessed and target identification is performed; when a spalling event is detected, an image recognition risk factor is output based on the area ratio of the spalling region. Interval statistics are performed on the hydraulic support working resistance data, and a pressure anomaly risk factor is output based on the proportion of supports in the low-pressure range and the proportion of supports in the high-pressure over-limit range. The actual mining height data is statistically analyzed, and a mining height risk factor is output based on the mining height deviation rate and the degree of mining height fluctuation exceeding limits. The image recognition risk factor, the pressure anomaly risk factor, and the mining height risk factor are input into a risk assessment model, and the model outputs the coal wall spalling risk level of the fully mechanized mining face after inference. A graded adaptive control strategy is executed based on the coal wall spalling risk level, including: executing a first control measure when the risk level is level one, executing a second control measure when the risk level is level two, and executing a third control measure when the risk level is level three.

[0081] Further, the preprocessing and target recognition of the coal wall image data includes: performing filtering preprocessing on the coal wall image data to obtain filtered image data; performing frame-by-frame inference on the filtered image data to obtain target coal, wherein the target coal includes flared coal and piled coal; when the number of detection frames for flared coal reaches a preset threshold and the detection frame for piled coal is detected simultaneously, a flared event is determined to exist, and the bounding rectangle of all detection frames is taken as the flared region; the ratio of the area of ​​the flared region to the area of ​​the entire image is calculated as the flared area ratio; and the image recognition risk factor is output by matching the preset interval to which the flared area ratio belongs.

[0082] Furthermore, the step of performing interval statistics on the hydraulic support working resistance data and outputting a pressure anomaly risk factor based on the proportion of supports in the low-pressure range and the proportion of supports in the high-pressure over-limit range includes: obtaining a predetermined normal working range, and dividing the support working resistance value based on the predetermined normal working range to obtain a low-pressure range and a high-pressure over-limit range; statistically analyzing the proportion of supports in the low-pressure range during the current monitoring period as an indicator of insufficient support; statistically analyzing the proportion of supports in the high-pressure over-limit range as an indicator of pressure intensity; and outputting the pressure anomaly risk factor based on the combination of the insufficient support indicator and the pressure intensity indicator.

[0083] Furthermore, the statistical analysis of the actual mining height data, and the output of the mining height risk factor based on the mining height deviation rate and the degree of mining height fluctuation exceeding the limit, includes: calculating the deviation rate of the average mining height of the working face relative to the designed mining height and the degree of mining height fluctuation exceeding the preset fluctuation threshold within the current monitoring period; and outputting the mining height risk factor based on the combination of the deviation rate and the degree of exceeding the limit.

[0084] Furthermore, the risk assessment model is a machine learning classification model trained based on historical monitoring data and corresponding actual measurement results of the face slope. It is trained using cross-validation and configured with an incremental learning interface. The risk assessment model also includes a physical constraint verification mechanism, which comprises: simultaneously executing the machine learning classification model to infer and output the risk level, calculating the coal wall stability coefficient based on the current working face's geomechanical parameters. The coal wall stability coefficient is positively correlated with roof pressure and mining height; comparing the trend of the coal wall stability coefficient with the trend of the risk level output by the machine learning classification model; if the trends are inconsistent, the current model inference result is determined to be abnormal, triggering the model to re-infer and output the risk level corrected by physical constraints; if the trends are consistent, the model inference result is directly output.

[0085] Furthermore, the first control measure includes: increasing the support force of the hydraulic support side plate and maintaining the existing cutting parameters of the coal mining machine.

[0086] Furthermore, the second control measure includes: increasing the support force of the side guard plate and controlling the side guard plate to move closer to the coal wall side to increase the coal wall confinement pressure, while increasing the initial support force replenishment frequency and replenishment target value of the hydraulic support, and reducing the cutting speed of the coal mining machine.

[0087] Furthermore, the third control measure includes: further reducing the cutting height of the coal mining machine to reduce the exposed height of the coal wall based on the second control measure, and activating the underground high-flow-rate horizontal long-hole regional fracturing device pre-embedded in the hard roof to inject high-pressure liquid into the directional long borehole to form a three-dimensional fracture network.

[0088] Furthermore, the adaptive control method based on coal wall spalling risk assessment further includes: when the fully mechanized mining face is an inclined coal seam, dividing the fully mechanized mining face into a first section, a second section, and a third section along the dip direction of the working face; assessing the coal wall spalling risk level of each of the first, second, and third sections respectively and implementing an appropriate adaptive control strategy, wherein a higher control intensity level is configured for the first and second sections where roof pressure is severe; simultaneously, when a spalling event is determined to exist in the monitoring area of ​​any image acquisition device, the pressure sensor data and height sensor data of the corresponding hydraulic support in the monitoring area are jointly verified to reduce the false alarm rate of image recognition.

[0089] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The adaptive control system and specific examples based on coal wall spalling risk assessment in the aforementioned embodiment one are also applicable to the adaptive control method based on coal wall spalling risk assessment in this embodiment. Through the foregoing detailed description of the adaptive control system based on coal wall spalling risk assessment, those skilled in the art can clearly understand the adaptive control method based on coal wall spalling risk assessment in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0090] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0091] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. An adaptive control system based on coal wall spalling risk assessment, characterized in that, include: The data acquisition module is used to acquire multi-source monitoring data of the fully mechanized mining face, including coal face image data, hydraulic support working resistance data, and actual mining height data. The target recognition module is used to preprocess and recognize targets in the coal wall image data. When a coal wall collapse event is detected, the module outputs an image recognition risk factor based on the area ratio of the collapse region. The interval statistics module is used to perform interval statistics on the working resistance data of the hydraulic support, and output the pressure anomaly risk factor based on the proportion of supports in the low pressure range and the proportion of supports in the high pressure over-limit range. The statistical analysis module is used to perform statistical analysis on the actual mining height data and output mining height risk factors based on the mining height deviation rate and the degree of mining height fluctuation exceeding the limit. The risk assessment module is used to input the image recognition risk factor, the pressure anomaly risk factor, and the mining height risk factor into the risk assessment model, and output the coal wall spalling risk level of the fully mechanized mining face after reasoning by the risk assessment model. The control strategy execution module is used to execute a graded adaptive control strategy according to the coal wall spalling risk level. The graded adaptive control strategy includes: executing a first control measure when the risk level is the first level, executing a second control measure when the risk level is the second level, and executing a third control measure when the risk level is the third level.

2. The adaptive control system based on coal wall spalling risk assessment according to claim 1, characterized in that, The target recognition module includes: A filtering preprocessing unit is used to perform filtering preprocessing on the coal wall image data to obtain filtered image data; A frame-by-frame inference unit is used to perform frame-by-frame inference on the filtered image data to obtain the target coal, wherein the target coal includes flared coal and piled coal; The coal spalling event determination unit is used to determine that a coal spalling event exists when the number of detection frames of the detected coal spalling reaches a preset number threshold and the detection frames of the coal pile are detected at the same time, and to take the outer rectangle of all detection frames as the coal spalling area. The area ratio calculation unit is used to calculate the ratio of the area of ​​the baffle region to the area of ​​the entire image as the baffle area ratio. The image recognition risk factor determination unit is used to match the preset interval to which the area ratio of the sheet belongs and output the image recognition risk factor.

3. The adaptive control system based on coal wall spalling risk assessment according to claim 1, characterized in that, The working resistance data of the hydraulic supports are statistically analyzed across intervals. Based on the proportion of supports in the low-pressure range and the proportion of supports in the high-pressure over-limit range, pressure anomaly risk factors are output, including: The working range division unit is used to obtain a predetermined normal working range and divide the working resistance value of the support based on the predetermined normal working range to obtain a low-pressure range and a high-pressure over-limit range. The low-pressure percentage statistics unit is used to count the percentage of stents in the low-pressure range during the current monitoring period as an indicator of insufficient support. The high-pressure percentage statistics unit is used to count the percentage of supports in the high-pressure over-limit range as a characterization index of the pressure intensity. The pressure anomaly risk factor determination unit is used to output the pressure anomaly risk factor based on the combination of the support insufficiency characterization index and the pressure intensity characterization index.

4. The adaptive control system based on coal wall spalling risk assessment according to claim 1, characterized in that, Statistical analysis is performed on the actual mining height data, and mining height risk factors are output based on the mining height deviation rate and the degree of exceeding the mining height fluctuation limit, including: The deviation rate calculation unit is used to calculate the deviation rate of the average mining height of the working face relative to the designed mining height and the degree of exceedance of the mining height fluctuation amplitude relative to the preset fluctuation threshold within the current monitoring period. A high-risk factor determination unit is used to output the high-risk factor based on a combination of the deviation rate and the degree of exceeding the limit.

5. The adaptive control system based on coal wall spalling risk assessment according to claim 1, characterized in that, The risk assessment model is a machine learning classification model trained based on historical monitoring data and corresponding actual test results of the project. It is trained using cross-validation and configured with an incremental learning interface. The risk assessment model is also equipped with a physical constraint verification mechanism, which includes: While executing the machine learning classification model to infer and output the risk level, the coal wall stability coefficient is calculated based on the geomechanical parameters of the current working face. The coal wall stability coefficient is positively correlated with the roof pressure and mining height. The trend of change of the coal wall stability coefficient is compared with the trend of change of risk level output by the machine learning classification model. If the two trends are inconsistent, it is determined that the current model inference result is abnormal, the model is triggered to re-infer and output the risk level after physical constraint correction; if the two trends are consistent, the model inference result is directly output.

6. The adaptive control system based on coal wall spalling risk assessment according to claim 1, characterized in that, The first control measures include: increasing the support force of the hydraulic support side plate and maintaining the existing cutting parameters of the coal mining machine.

7. The adaptive control system based on coal wall spalling risk assessment according to claim 6, characterized in that, The second control measures include: increasing the support force of the side guard plate and controlling the side guard plate to move closer to the coal wall side to increase the coal wall confinement pressure, while increasing the initial support force replenishment frequency and replenishment target value of the hydraulic support, and reducing the cutting speed of the coal mining machine.

8. The adaptive control system based on coal wall spalling risk assessment according to claim 7, characterized in that, The third control measure includes: further reducing the cutting height of the coal mining machine to reduce the exposed height of the coal wall based on the second control measure, and activating the underground high-flow-rate horizontal long-hole regional fracturing device pre-embedded in the hard roof to inject high-pressure liquid into the directional long borehole to form a three-dimensional fracture network.

9. The adaptive control system based on coal wall spalling risk assessment according to claim 1, characterized in that, Also includes: The fully mechanized mining face division module is used to divide the fully mechanized mining face into a first section, a second section, and a third section along the dip direction of the mining face when the fully mechanized mining face is an inclined coal seam. The control strategy execution module is used to assess the risk level of coal wall spalling in the first section, the second section and the third section respectively and execute the appropriate adaptive control strategy. Among them, the first section and the second section, which are subject to severe roof pressure, are configured with a higher control intensity level. The joint verification module is used to perform joint verification by combining the pressure sensor data and height sensor data of the corresponding hydraulic support in the monitored area when a panel event is determined to exist in the monitoring area of ​​any image acquisition device, so as to reduce the false alarm rate of image recognition.

10. An adaptive control method based on coal wall spalling risk assessment, characterized in that, Executed by the adaptive control system based on coal wall spalling risk assessment as described in any one of claims 1 to 9, wherein the adaptive control system based on coal wall spalling risk assessment comprises: Acquire multi-source monitoring data of the fully mechanized mining face, including coal face image data, hydraulic support working resistance data, and actual mining height data; The coal wall image data is preprocessed and target recognition is performed. When a rockfall event is detected, an image recognition risk factor is output based on the area ratio of the rockfall region. The working resistance data of the hydraulic support is statistically analyzed in intervals, and pressure anomaly risk factors are output based on the proportion of supports in the low-pressure range and the proportion of supports in the high-pressure over-limit range. Statistical analysis is performed on the actual mining height data, and mining height risk factors are output based on the mining height deviation rate and the degree of mining height fluctuation exceeding the limit. The image recognition risk factor, the pressure anomaly risk factor, and the mining height risk factor are input into the risk assessment model, and the risk level of coal wall spalling of the fully mechanized mining face is output after reasoning by the risk assessment model. A graded adaptive control strategy is implemented based on the risk level of coal wall spalling. The graded adaptive control strategy includes: implementing a first control measure when the risk level is the first level, implementing a second control measure when the risk level is the second level, and implementing a third control measure when the risk level is the third level.

Images