A method for predicting periodic pressure of large inclination working face by region

Abstract

The application provides a large-dip-angle working face periodic weighting sub-region prediction method, and belongs to the technical field of coal mining. The method creates a twin model corresponding to a physical entity by using digital twinning technology, analyzes and predicts the collected data in combination with various algorithms in machine learning. The working face is divided into regions by using a MeanShift clustering algorithm, and an algorithm model is constructed for the hydraulic support resistance data of different regions to improve the prediction accuracy and reliability. Meanwhile, consistency testing is performed on the prediction results of the twin model and the algorithm model to ensure the accuracy and consistency of the prediction results. The application adopts the above-mentioned large-dip-angle working face periodic weighting sub-region prediction method, cooperates digital twinning and multiple algorithms, realizes dynamic monitoring and high-precision prediction of periodic weighting under complex geological conditions, and provides a reliable basis for intelligent management, disaster warning and safety decision of coal mines.

Description

Technical Field

[0001] This invention relates to the field of coal mining technology, and in particular to a regional prediction method for periodic pressure on steeply inclined working faces. Background Technology

[0002] During coal mining, as mining progresses deeper into more difficult-to-mine coal seams, ground pressure manifestations and various geological disasters occur more frequently. This is especially true in steeply inclined working faces, where the frequency of ground pressure manifestations increases significantly due to complex geological conditions and unusual geological structures. Currently, the coal mining industry faces the significant challenge of accurately predicting cyclical ground pressure manifestations for prevention and control. In recent years, with the rapid development of digital twin technology and machine learning, these emerging technologies have provided new ideas and methods for predicting cyclical ground pressure in coal mines. Digital twin technology primarily creates a virtual mirror of the physical system, enabling real-time monitoring and updating of information such as the mine's condition, environment, and status, forming a dynamic predictive analysis platform. Machine learning, leveraging large amounts of historical data, uses algorithmic models for pattern recognition and prediction to analyze and assess the potential risks of ground pressure manifestations. Combining digital twin technology with machine learning can more quickly achieve ground pressure prediction in complex working faces.

[0003] In steeply dipping working faces, due to their unique geological structure and ore-rock properties, pressure inrush phenomena not only exhibit randomness but also show significant regional characteristics. Differences in geological parameters such as pressure and stress lead to significant variations in the frequency and intensity of periodic pressure inrushes across different regions. Therefore, traditional prediction methods often fail to effectively capture the complex characteristics of these regions, resulting in predictions that do not match the actual situation. To address this issue, it is necessary to propose a novel pressure inrush prediction method. Summary of the Invention

[0004] The purpose of this invention is to provide a regional prediction method for periodic pressure in steeply inclined working faces. This method aims to solve a series of problems encountered in steeply inclined working face mining through digital twin technology and machine learning algorithms. These problems include, but are not limited to, traditional prediction methods that rely primarily on empirical rules and simple mathematical models, lacking effective modeling of complex geological environments and dynamic working conditions, and the inability of mines to fully utilize sensor data and advanced data analysis techniques, thus reducing the timeliness and reliability of prediction results. This invention can accurately capture the pressure variation patterns of complex working faces, significantly improve the accuracy of pressure prediction, identify potential risks in advance, enhance operational safety, and thereby optimize resource allocation and improve operational efficiency.

[0005] To achieve the above objectives, the present invention provides a regional prediction method for periodic pressure on a large-angle working surface, comprising the following steps:

[0006] Step S1: The perception layer collects data through sensors deployed on the mine working face and hydraulic supports, and transmits the data to the information layer through the communication network via the interaction layer.

[0007] Step S2: After cleaning and performing correlation analysis on the collected data, the information layer obtains the processed data and transmits it to the model layer.

[0008] Step S3: The model layer is set with a digital twin to suppress the prediction model. The digital twin to suppress the prediction model includes a twin model and an algorithm model, and the constructed twin model and algorithm model are trained and optimized.

[0009] Step S4: Perform a consistency test on the prediction results of the trained and optimized twin model and the algorithm model.

[0010] Preferably, in step S1, the sensors include a pressure sensor arranged on a hydraulic support and an underground camera arranged in the tunnel.

[0011] Preferably, in step S2, data cleaning includes deleting duplicate data, filling in missing values, and handling outliers.

[0012] Preferably, in step S3, the steps for constructing the twin model are as follows:

[0013] The motion posture of the hydraulic support in a steep working face was analyzed using SolidWorks software, and a three-dimensional solid model including the physical and mechanical structure and hydraulic system was constructed. The motion posture of the hydraulic support in actual application was set, and the corresponding motion parameters and constraints were defined. At the same time, the accuracy and precision of the model were verified and compared, and adjustments were made to address any errors and discrepancies, so as to simulate the actual working process of the hydraulic support.

[0014] Convert the constructed 3D solid model into an STL interface file, import it into Maya software to adjust the model's materials, textures, lighting and shadows, and perform animation design to optimize the rendering effect;

[0015] In Unity3D, a virtual scene of a large-angle working surface is created, and the background, lighting, and sound effects are configured. C# scripts are used to program the motion of the hydraulic support and control the parameters of the hydraulic system. Sensors and communication modules are set up to realize data exchange and information transmission, thereby completing the construction of the twin model.

[0016] Preferably, in step S3, the steps for constructing the algorithm model are as follows:

[0017] The raw time-series data of hydraulic supports in the upper, middle and lower regions of the hydraulic support online monitoring system for a specific time period are extracted as a dataset. Data preprocessing is performed, including data cleaning, transformation, feature selection and extraction, and smoothing and noise reduction.

[0018] The MeanShift clustering algorithm was used to perform regional feature analysis on the steep-angle working face and divide the working face into regions.

[0019] For hydraulic support resistance data in different working areas, various machine algorithm models are constructed.

[0020] The accuracy of the algorithm model prediction results for each working face area is compared. Based on the mean square error, root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination, the algorithm model with the highest prediction accuracy in each working face area is selected for the final prediction.

[0021] Preferably, in step S4, when the prediction results of the twin model and the algorithm model are consistent, the results are visualized and relevant control decisions are made based on the prediction results. When the prediction results of the twin model and the algorithm model are inconsistent, the digital twin is updated to suppress the prediction model, and the test is repeated.

[0022] Therefore, the present invention adopts the above-mentioned regional prediction method for periodic pressure on steep-angle working faces, and the beneficial technical effects are as follows: it can realize accurate monitoring of pressure on steep-angle working faces, which is conducive to the further realization of intelligent management in coal mines. At the same time, the accurate pressure prediction results help decision-makers make correct decisions before environmental changes, thereby improving the safety and intelligence level of mine management. Attached Figure Description

[0023] Figure 1 Architecture diagram of a regional prediction system for periodic pressure on a steep-angle working surface;

[0024] Figure 2 A flowchart of a regional prediction method for periodic pressure on a steeply inclined working face;

[0025] Figure 3 Build a graph for the twin model;

[0026] Figure 4 This is a diagram for analyzing the mechanical model;

[0027] Figure 5 Simulation curves for the twin model;

[0028] Figure 6 Flowchart for region-based prediction of the algorithm model;

[0029] Figure 7 Clustering diagram of the steep-angle working surface region;

[0030] Figure 8 This is a comparison chart of load predictions for multiple regions of the working face; among them, Figure 8 (a) in the text refers to hydraulic support No. 3; Figure 8 (b) in the text refers to hydraulic support No. 27; Figure 8 (c) in the text refers to hydraulic support No. 43; Figure 8 (d) in the text refers to hydraulic support No. 66;

[0031] Figure 9 Comparison chart of multi-regional pressure prediction at the working face; Figure 9 (a) in the text refers to hydraulic support No. 3; Figure 9 (b) in the text refers to hydraulic support No. 27; Figure 9 (c) in the text refers to hydraulic support No. 43; Figure 9 (d) in the figure represents hydraulic support No. 66. Detailed Implementation

[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0034] Example 1

[0035] I. Architecture of the regional prediction system for periodic pressure on steep working surfaces.

[0036] To address the challenges posed by the complex geological environment and outdated, inaccurate prediction methods in current steep-angle working faces, this invention proposes a regional prediction method for periodic pressure in steep-angle working faces. This method integrates various sensor data features, industrial internet information interaction technology, and a fusion module combining digital twins and machine learning to construct a perception layer, an interaction layer, an information layer, and a model layer.

[0037] like Figure 1 The diagram shown is a regional prediction system architecture for periodic pressure on the dip working surface, including:

[0038] The sensing layer is primarily used to detect the status of hydraulic supports, the mine's working environment, and the stress state of the surrounding rock. Sensors are deployed at the actual coal mine working face to collect experimental data. Pressure sensors are placed at key components of the hydraulic supports, such as hydraulic cylinders, columns, and crossbeams, to monitor real-time pressure changes from the roof during mining. Simultaneously, underground cameras are deployed in the roadways to monitor equipment operation and integrate with the monitoring system for remote control. The placement of these sensors comprehensively considers the different working environments and characteristics of each production stage in the mine, ensuring timely feedback on actual pressure conditions under various loads.

[0039] As the key connection between the perception layer and the information layer, the interaction layer plays a vital role in information transmission, data processing, and control execution. The interaction layer mainly involves two aspects: communication connectivity and edge control, to ensure the efficient and intelligent operation of the system.

[0040] In communication connections, communication equipment is fundamental for data transmission, including various sensors, data acquisition devices, and communication modules. These devices are responsible for collecting key parameters from the steep-angle working face, such as hydraulic support pressure and its corresponding time, and converting the data into a transmittable format through appropriate signal processing. To achieve efficient data transmission, the interaction layer employs multiple communication networks, including wired networks (such as Ethernet) and wireless networks (such as Wi-Fi and NB-IoT). This flexible network architecture enables rapid and efficient information transmission in various complex working environments. The effectiveness of the communication layer also depends on the appropriate application of data transmission protocols. Commonly used protocols such as MQTT, CoAP, and HTTP enable standardized communication between different devices. Protocol parsing ensures the correctness and accuracy of transmitted data, allowing received data to be quickly understood and processed.

[0041] Edge control runs on edge computing nodes close to the data source, processing and analyzing data from the perception layer. It enables real-time on-site analysis, utilizing machine learning algorithms for status assessment and trend prediction, saving bandwidth and reducing response time. Edge gateways connect edge devices to the cloud, handling data forwarding, caching, and protocol conversion, managing real-time and historical data to ensure efficient system operation. They also adapt to data protocols between different devices. Edge control refers to control strategies executed in real-time through edge computing nodes. Based on the processing and analysis of real-time data, edge control can quickly make decisions and execute corresponding operations. For example, when a sensor in the perception layer detects abnormal pressure, edge control will immediately adjust the working face or activate an alarm to reduce danger. With the help of advanced communication equipment and networks, the interaction layer ensures real-time and efficient information transmission; through edge services, edge gateways, and edge control, the system can achieve rapid response and intelligent decision-making, thereby improving the safety and stability of steep-angle working faces.

[0042] The information layer acts as a bridge for real-time mapping and interaction between physical entities and virtual twins, primarily responsible for data collection, cleaning, correlation analysis, and algorithm model optimization. The construction of the information layer is crucial for ensuring data quality and the accuracy of analytical results. Data cleaning is the core of the information layer, including identifying and removing duplicate data, filling in missing values, and handling outliers, aiming to improve effectiveness by removing redundant, erroneous, and irrelevant data. These processes ensure the accuracy and reliability of the data used in subsequent analyses, making model training and prediction results more credible. After cleaning, the data undergoes further operations such as classification, clustering, and regression to extract effective information and eliminate outliers, providing a high-quality data foundation for subsequent analysis and applications, facilitating the application and processing of related twin models and algorithm models later on.

[0043] The model layer is equipped with a digital twin to support the prediction model, which consists of a twin model and an algorithm model, and the two interact closely through twin data.

[0044] The twin model integrates two specific models: a load model and a simulation model. The load model is a mathematical representation based on the coupling dynamics and load variation characteristics between the surrounding rock and the support system, and it also provides relevant data support for the subsequent simulation model. The simulation model is mainly constructed using software such as SolidWorks, incorporating actual motion parameters, and is used to represent the actual working process.

[0045] The algorithm model is divided into data algorithm, model algorithm, and application algorithm. The algorithm model is mainly trained using historical load data or simulated data. The specific process is as follows: the data algorithm preprocesses and smooths the hydraulic support data, and then uses it for training and testing the model algorithm; the trained algorithm is optimized for hyperparameters to achieve accurate prediction of pressure conditions; the application algorithm analyzes the prediction results and formulates intelligent decision-making schemes.

[0046] II. Cycle-based pressure prediction steps.

[0047] Figure 2 The flowchart of the regional prediction method for periodic pressure on steep working faces provides a more detailed explanation of the specific prediction process.

[0048] This invention proposes to combine digital twin technology with machine learning theory for predicting periodic pressure in mine working faces, so as to achieve accurate, dynamic and efficient support measures.

[0049] The specific steps are as follows:

[0050] Step 1: The relevant equipment in the working face perception layer collects data on the physical hydraulic support in real time, and transmits the data to the information layer through the interaction layer via communication connection.

[0051] Step 2: After preprocessing the collected data, the information layer transmits the data to the twin model and the algorithm model respectively for simulation and training.

[0052] Step 3: Conduct simulation experiments on the twin model using hydraulic support data to achieve visualization.

[0053] Step 4: Divide the preprocessed hydraulic support data into training set and test set to train, evaluate and test the algorithm model, and then generate algorithm prediction results.

[0054] Step 5: The digital twin system for predicting load on the steep working face calls the algorithm model through the application program interface (API) to predict the load and transmits the results to the Unity3D interface. By integrating the digital twin system with the actual hydraulic support through the API, real-time data interaction and feedback of prediction results are realized.

[0055] Step 6: Perform a consistency test on the results of the twin model and the algorithm model. If the results are consistent, display the prediction results in real time and make relevant control decisions based on the prediction results. If the results are inconsistent, update the prediction model to improve the prediction accuracy, improve the system, and repeat the consistency test until the test results are consistent.

[0056] Step 7: The reasonable prediction data displayed in real time by the twin model will be re-imported into the database to optimize the database.

[0057] Step 8: The prediction results of the twin model interact directly with the physical hydraulic support entity to achieve automated control, decision optimization and feedback.

[0058] III. Construction and mechanical analysis of the twin model.

[0059] Figure 3 The diagram illustrates the construction process of the twin model in digital twins, specifically describing the construction of the twin model in digital twins.

[0060] In the construction of the hydraulic support twin model, software such as SolidWorks and 3ds Max were used to create the twin model. These models were then imported into Unity3D software to achieve docking and real-time interaction between the physical entity and the virtual model. The specific steps are as follows: First, using SolidWorks software, the motion posture of the hydraulic support in the steep working face was analyzed. Simultaneously, relevant structural parameters of the hydraulic support collected from the information layer were imported, such as column angle, column length, top beam length, base length, and balance bar length, to construct a three-dimensional solid model containing the physical and mechanical structure and hydraulic system. During this process, the spatial pose state, hydraulic cylinder motion state, and relative motion of key components of the hydraulic support during actual working face mining were set. Geometric motion parameters, dynamic response parameters, mechanical parameters, and constraints such as stroke limits, angle limits, load-bearing capacity limits, stability constraints, and kinematic constraints were defined. The accuracy and precision of the model were verified and compared, and adjustments were made to address any errors and discrepancies to accurately simulate the actual working state of the hydraulic support. After adjustment, the model was converted into an STL interface file.

[0061] Next, the obtained STL interface file was imported into Maya software. In Maya, the model's materials, textures, lighting, and shadows were adjusted to realistically display the surface features of the hydraulic support. Simultaneously, related animation design was performed to achieve smooth movement and state changes, further optimizing the rendering effect and enhancing the model's realism and visual appeal. Finally, a virtual scene with a large-angle working surface was created in Unity3D to simulate the actual working environment, and background, lighting, and sound effects were configured. Meanwhile, C# scripts were used to program the hydraulic support's motion and control the hydraulic system parameters. Sensors and communication modules were set up to achieve data exchange and information transmission, ensuring real-time interaction between the model and the physical hydraulic support.

[0062] The generated large-angle working face pressure prediction digital twin system calls relevant algorithm models for prediction via an application programming interface (API) and transmits the results to the Unity3 system. The digital twin system is then integrated with the actual hydraulic support system via the API, enabling real-time data interaction and prediction result feedback for the hydraulic support group. Specifically, the data portion of the digital twin system receives real-time data, performs predictions using the prediction model within the digital twin system, and then feeds the prediction results back to the physical portion of the digital twin system to update model parameters and status, thereby achieving real-time prediction and management of hydraulic support pressure.

[0063] To ensure the accuracy of the numerical representation of the load model in the twin model, a planar mechanical analysis was performed to comprehensively understand the stress conditions of the hydraulic support in the mine roadway under steep-angle working face loads. This helps to more accurately simulate the actual working state of the hydraulic support under cyclic pressure conditions. Figure 4 The diagram shown is an analysis diagram of the mechanical model of the hydraulic support.

[0064] During the mining of steeply inclined working faces, the hydraulic supports tend to move relative to the overlying roof due to its movement and cyclic pressure. This is influenced by the roof pressure and the subsidence of the floor. As the contact pattern and load characteristics between the roof and the hydraulic supports change, the behavior of the hydraulic supports also changes, leading to subsidence, slippage, and rotation. In-depth analysis of the load and instability phenomena of the hydraulic support mechanical model under cyclic pressure helps to accurately construct and effectively utilize the twin model in areas such as structural parameter optimization, dynamic characteristic analysis, data interaction optimization, intelligent decision support, and visualization.

[0065] like Figure 4 During the movement of the roof slab, the position of the hydraulic support changes as the overlying roof collapses and applies pressure. The amount of subsidence of point O along the z-axis of the hydraulic support is first denoted as z. o Meanwhile, the rotation angle around point O is denoted as Based on the theory of elastic foundations, the distributed load q at the upper and lower edges of the hydraulic support base along the dip direction can be obtained. A and q B The calculation formula is:

[0066]

[0067] Where, k o The foundation soil coefficient (kN·m) -3 ); c represents the length of the hydraulic support base; α represents the coal seam dip angle.

[0068] From equations (1) and (2) above, we can obtain the resultant force F of the normal load on the hydraulic support from the base plate under the current hydraulic support position. N The position of action x2 is:

[0069] F N =ack o z o (3);

[0070]

[0071] Where 'a' represents the width of the hydraulic support.

[0072] According to such Figure 4 The mechanical model of the hydraulic support shown can be used to derive the equilibrium conditions of the hydraulic support under the condition of a large inclined working surface:

[0073]

[0074] Among them, F R F represents the frictional force between the hydraulic support and the roof rock strata. F The frictional force between the hydraulic support and the underlying rock strata is represented by: G; the weight of the hydraulic support; P; the working resistance of the hydraulic support; x1; the position of action between the hydraulic support and the load on the roof; b; and h. o This indicates the height of the center of gravity of the hydraulic support.

[0075] Since the working resistance of the hydraulic support is much greater than its own weight, the influence of the hydraulic support's rotation angle on its own weight will be ignored in the subsequent analysis for the sake of theoretical solution. Based on equations (5) to (7), the subsidence z of point O under the large tilt angle position shown in the figure can be calculated. O Rotation angle And the frictional force F between the hydraulic support and the base plate F As shown below:

[0076]

[0077] F F=Gsinα-F R (10);

[0078] The analysis of the aforementioned mechanical model helps identify key parameters and their impact on the performance of the hydraulic support. Furthermore, the twin model, through optimization and adjustment of these parameters, makes the model more closely resemble actual working conditions, improving its predictive accuracy and adaptability. It also facilitates understanding the overall dynamic characteristics of the hydraulic support, providing accurate data for the dynamic simulation of the twin model, enabling it to realistically reproduce the actual motion state of the hydraulic support.

[0079] IV. Validation of the twin model.

[0080] Since the accuracy of the hydraulic support's prediction of incoming pressure has a significant impact on the entire twin model, verification experiments are needed to validate the model's rationality and ensure correct modeling. After setting the parameters of the twin system, simulation experiments are conducted to determine whether the time-load curve and the incoming pressure display are consistent. The twin system simulation curves are shown below. Figure 5 As shown.

[0081] exist Figure 5 The twin model clearly demonstrates that as the longwall mining face advances, the basic roof undergoes periodic fractures, causing a significant increase in the load on the hydraulic supports, sometimes even reaching peak values. Simultaneously, the "masonry beam" structure formed by the fractured basic roof blocks undergoes periodic motion and change, resulting in periodic pressure phenomena. The load curves of the twin model accurately depict each stage of the periodic pressure process, as follows:

[0082] In the initial stage, the hydraulic support is not yet in full contact with the roof or only has initial support force, resulting in relatively low working resistance. The load pressure of the hydraulic support is generally around 15 MPa. As the working face gradually advances, the geological conditions and stress state around the hydraulic support change. The exposed roof may fracture and exert pressure on the hydraulic support, creating an additional load. Simultaneously, the concentrated pressure from the roof on the hydraulic support during periods of strong localized pressure causes the working resistance of the hydraulic support to gradually increase, reaching a high peak value. This peak load is the resistance at the end of the cycle. After the peak load, due to large-scale fractures or collapses of the roof, or passive / active unloading of the hydraulic support, the working resistance of the hydraulic support rapidly decreases, indicating the end of one cycle of pressure. As the working face continues to advance to the next position, the roof will fracture and subside again, and the working resistance will gradually increase, initiating a new cycle of pressure.

[0083] The simulation results from the twin model show that the twin model's variation curves are basically consistent with the actual dynamic change trends, and the correlation error is within an acceptable range. The dynamic characteristics at each stage meet the experimental requirements. This demonstrates the reliability and effectiveness of the twin model in predicting pressure phenomena in a simulated inclined mining environment.

[0084] V. Construction of the algorithm model.

[0085] Figure 6 The flowchart illustrates the regional pressure prediction process of the algorithm model, primarily explaining the model construction process and related analysis results. By applying the MeanShift clustering algorithm to analyze the hydraulic support resistance and using the algorithm model to predict the hydraulic support resistance, the relevant pressure prediction results are obtained. These results are then applied to the prediction and analysis of pressure conditions in steep working faces. The specific steps are as follows:

[0086] (1) Define the sample data. Extract the original time series data of hydraulic supports No. 1 to No. 67 in the upper, middle and lower areas of a certain area from January 2, 2024 to February 6, 2024 from the hydraulic support online monitoring system as the dataset, with a sampling interval of 10 minutes.

[0087] (2) Data Preprocessing. During coal mining, sudden events such as periodic pressure and coal wall spalling can cause drastic changes in the load on hydraulic supports. Wavelet decomposition can reveal local features in the signal, making subsequent signal analysis easier. Based on this, the extracted hydraulic support data is cleaned, transformed, and its features are selected and extracted. Smoothing and noise reduction are then performed to optimize the training and prediction performance of the model.

[0088] (3) Regional characteristic analysis of the steep-angle working face shows that the periodic pressure and fracture of the top plate of the steep-angle working face have significant zoning and temporal characteristics. In order to effectively deal with these complex patterns, various algorithm models can be used to predict the performance of the hydraulic support during the pressure process.

[0089] (4) Data category classification employs a MeanShift-based clustering algorithm. First, bandwidth parameters are appropriately set to determine the neighborhood search range. Kernel density estimation is used to accurately calculate the weighted average of each data point within its neighborhood. Next, the mean shift vector of each data point is precisely calculated, pointing precisely in the direction of increasing density. Subsequently, based on the mean shift vector, the positions of data points are iteratively updated until the position changes of all data points are below a preset threshold, at which point convergence terminates. After convergence, neighboring data points are accurately merged into cluster centers, thus completing the initial determination of the clustering results. Afterward, the rationality of the clustering results is evaluated, and any abnormal or unreasonable classification results are deeply optimized until the expected requirements are met. Finally, the clustering results are visually visualized and comprehensively evaluated to fully support subsequent in-depth analysis and scientific decision-making.

[0090] Based on the monitoring of time-load sequence data of the hydraulic supports at the working face, the MeanShift clustering algorithm was used to autonomously divide the sample set density, mainly based on the differences in the loads borne by the hydraulic supports and their different locations, thereby achieving accurate differentiation of the working face area. The selection of monitoring points for the hydraulic support column loads followed the principle of equidistant distribution. In a working face with a total of 67 hydraulic supports, the time-load sequences of hydraulic supports No. 5, 11, 17, 23, 29, 35, 41, 47, 53, 59, and 65 were monitored. Simultaneously, to ensure the continuity of the working face area division, the hydraulic supports with the above-numbered numbers were used as the central frame, and the central frame, along with the three preceding and two following hydraulic supports, were divided into the same area. The division results are as follows: Figure 7 As shown, using the hydraulic support load parameters and hydraulic support number as input parameters for the clustering algorithm, the hydraulic support group on the steep-angle working face can be divided into four regions, with relatively clear boundaries between each region. The resulting regions on the steep-angle working face are hydraulic supports 1-20, 21-37, 38-50, and 51-67. Data from hydraulic supports 3, 27, 43, and 66 were selected as the research objects to train and predict the algorithm model.

[0091] (5) Constructing an algorithm model to predict the pressure situation of the working face cycle. First, the data of different categories are classified, and the end resistance of the measured curve at the end of the cycle is regarded as a periodic variation feature. In order to achieve effective training of the model, 80% of the data is selected as the training set, and the remaining 20% ​​is used as the test set. In the process of model construction, the regional characteristics of the actual production process of the steep angle working face are taken into account, and the prediction accuracy of the algorithm in different regions is compared.

[0092] Due to the unique geological structure and the influence of steeply inclined working faces, hydraulic support load data exhibits significant nonlinear regionalization and temporal dependence. In this context, Random Forest (RF) and Support Vector Regression (SVR) perform well in modeling nonlinear relationships, while Long Short-Term Memory (LSTM) excels at capturing long-term dependencies. Therefore, it is necessary to compare the performance of these three algorithms and select the most appropriate model for predictive analysis. Finally, the prediction results of the algorithm model are visualized and consistency tested using a twin model to verify the model's rationality, which is beneficial for the model's continuous learning and optimization.

[0093] To comprehensively evaluate the accuracy of the three prediction models, the evaluation metrics used include: mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²). 2 The comparison results of the prediction accuracy of multiple algorithms are shown in Table 1. After comparing and analyzing the prediction accuracy of the three models, the algorithm model with higher accuracy was selected for prediction. Specifically, the X1-X20 hydraulic support was predicted using the RF model, the X21-X35 hydraulic support was predicted using the RF model, the X36-X50 hydraulic support was predicted using the SVR model, and the X51-X67 hydraulic support was predicted using the SVR model. A detailed comparison of the prediction accuracy of the specific algorithm models is shown in Table 1. Figure 8 .

[0094] (6) To further improve the model's prediction performance, ISSA was used to optimize the hyperparameters of the aforementioned algorithms. Through initial hyperparameter setting, iterative updates, performance evaluation, and optimal solution selection, the efficiency and accuracy of the algorithm model were improved, the search strategy and local information utilization were enhanced, and the optimal hyperparameter combination was identified. Simultaneously, the optimized model was evaluated. If the predicted values ​​met the requirements, the model was exported and its performance was tested and evaluated. If the requirements were not met, the parameters were adjusted to optimize the model again. Finally, to demonstrate the advantages of the optimized model, its parameters were compared with the initial model, as shown in Table 2. This indicates that the overall prediction accuracy of the model was further improved after ISSA optimization. ISSA referenced "An Improved Sparrow Algorithm Integrating Cauchy Mutation and Backward Learning".

[0095] Table 1 Comparison of prediction accuracy among multiple algorithms

[0096]

[0097] Table 2 Comparison of prediction accuracy after optimization of multiple algorithms

[0098]

[0099]

[0100] (7) Output and analyze the prediction results. Output the hydraulic support resistance predicted by the algorithm model and analyze the pressure situation of each hydraulic support during the working face advancement process. The time-weighted average working resistance is used as the pressure criterion. The sum of the average weighted working resistance of each hydraulic support cycle and its variance is used as the criterion for judging the pressure from the top plate. The calculation formula is as follows:

[0101]

[0102] Where p′t represents the pressure criterion; This represents the average value of the initial support force; This represents the mean squared error.

[0103] VI. Analysis of Simulation Results.

[0104] The load conditions of the hydraulic support during the advancement process in different areas are visualized, and the load prediction comparison results for multiple areas of the working face are as follows: Figure 8 As shown. A comparison of actual and predicted pressure results reveals that this invention can accurately predict the working resistance of the hydraulic support, achieving a prediction accuracy of over 95% for areas with significant periodic pressure during the working face advancement process, thus enabling accurate prediction and early warning.

[0105] Comparison of multi-zone pressure prediction for steep-angle working faces (see below) Figure 9 As can be clearly seen from the figure, the method proposed in this invention can effectively predict the pressure in different regions of the working face, and the predicted phenomena are consistent with the actual situation. The experimental prediction results show that in steeply inclined working faces, the roof fractures and falling gangue in the middle and upper regions will slide, thus supporting the roof in the lower region, causing the working face to tilt. The mine pressure in the middle and upper regions is significantly more intense than in the lower region, and the pressure step distance is greater in the middle than on both sides, which conforms to the basic law of mine pressure during steeply inclined coal seam mining. Therefore, the prediction method proposed in this invention can achieve high-precision dynamic prediction of periodic pressure in different regions of steeply inclined working faces.

[0106] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0107] Therefore, this invention adopts the above-mentioned regional prediction method for periodic pressure on steep working faces. Through the synergy of digital twins and multiple algorithms, it realizes dynamic monitoring and high-precision prediction of periodic pressure under complex geological conditions, providing a reliable basis for intelligent coal mine management, disaster early warning and safety decision-making.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A regional prediction method for periodic pressure on a steeply inclined working face, characterized in that, Includes the following steps: Step S1: The perception layer collects data through sensors deployed on the mine working face and hydraulic supports. The sensors include pressure sensors deployed on key parts of the hydraulic supports and underground cameras deployed in the roadway. The key parts of the hydraulic supports are hydraulic cylinders, columns, and cross beams. The collected data includes the attitude parameters and load parameters of the hydraulic supports. The attitude parameters are subsidence and rotation angle, and the load parameters are working resistance and normal load. The data is transmitted to the information layer through the communication network via the interaction layer. Step S2: After the information layer cleans and performs correlation analysis on the collected data, it obtains the processed data and transmits it to the model layer. Data cleaning includes deleting duplicate data, filling missing values, and handling outliers. Correlation analysis is to identify key parameters and their impact on the performance of hydraulic supports through the analysis of the mechanical model. At the same time, the twin model optimizes and adjusts these parameters to make the model closer to the actual working conditions and improve the prediction accuracy and adaptability of the model. Step S3: The model layer is equipped with a digital twin pressure prediction model, which includes a twin model and an algorithm model. The constructed twin model and algorithm model are trained and optimized to achieve dynamic monitoring and high-precision prediction of periodic pressure under complex geological conditions. The twin model needs to be based on the gravity-tilt effect of the steep working face, and construct a three-dimensional solid model including the physical and mechanical structure of the hydraulic support and the hydraulic system. The motion parameters and constraints adapted to the steep working conditions are set. The motion parameters are the column angle and the angle between the top beam and the side plate. The constraints are the stroke limit, the bearing capacity limit, and the stability constraint. The twin model integrates two specific models: the load model and the simulation model. The load model is a mathematical representation based on the coupling dynamics and load change characteristics between the surrounding rock and the support system. It can also provide relevant data support for the subsequent simulation model. The simulation model is a model built by SolidWorks software in combination with the actual motion parameters and is used to represent the actual working process. The algorithm model takes the posture-load coupling characteristics of the hydraulic support as input. First, it uses the MeanShift clustering algorithm to realize the dynamic region division of the large-angle working face. Then, it builds a dedicated machine learning model for different regions. By comparing the accuracy, it selects the dedicated model with the highest prediction accuracy for each region and optimizes the hyperparameters using the ISSA algorithm. Step S4: Perform a consistency test on the prediction results of the trained and optimized twin model and the algorithm model. Use the time-weighted average working resistance as the criterion for determining pressure on the top plate. The sum of the average weighted working resistance of the hydraulic support for each cycle and its variance (1x) is used as the criterion for judging pressure on the top plate. The calculation formula is as follows: ； in, Indicates the criterion for applying pressure; This represents the average value of the initial support force; Indicates the mean squared error; In step S4, when the prediction results of the twin model and the algorithm model are consistent, the results are visualized and relevant control decisions are made based on the prediction results. When the prediction results of the twin model and the algorithm model are inconsistent, the digital twin is updated to suppress the prediction model, and the test is repeated. The prediction accuracy for areas with significant periodic pressure during the working face advancement reaches over 95%, with the accuracy calculated using the root mean square error.

Images