A Closed-Loop Cut-Off Planning Method Driven by the Fusion of VR Navigation and Coal Rock Identification Online Planning

Abstract

This invention discloses a closed-loop cutting planning method driven by the fusion of VR navigation and online coal and rock identification. By introducing virtual reality (VR) navigation dynamic pre-planning technology, the cutting path of the coal mining machine is planned in advance at a macroscopic level. Simultaneously, real-time coal and rock identification technology is combined with online sensors and algorithms to identify the hardness, structure, and location characteristics of coal and rock, obtaining precise data at the microscopic level. The fusion of these two technologies, through a closed-loop feedback control mechanism, combines the real-time perceived coal and rock information with the virtual pre-planning results to dynamically adjust the cutting strategy of the coal mining machine. The core of this invention lies in the deep integration of the virtual and physical layers. Virtual pre-planning guides the real-time operation of physical equipment, while physical feedback optimizes the virtual planning, thereby achieving closed-loop control and dynamically adjusting the cutting plan.

Description

Technical Field

[0001] This invention relates to the field of intelligent coal mining technology, and in particular to a closed-loop cutting planning method driven by the fusion of VR navigation dynamic pre-planning and online planning based on coal and rock identification. Background Technology

[0002] With the continuous advancement of coal mining technology, the level of intelligence in fully mechanized mining equipment (such as fully mechanized mining machines and coal mining machines) is constantly improving. However, existing cutting planning methods still face some challenges.

[0003] First, the coal mining environment is complex and variable, including factors such as the distribution and thickness of coal seams and the hardness of rocks, which significantly affect the cutting effect. In existing technologies, many devices rely on fixed cutting schemes, which often cannot adapt to changes in coal and rock conditions during actual operations, leading to low cutting efficiency and resource waste.

[0004] Furthermore, existing coal and rock identification technologies largely rely on fixed sensor data and manual inspection. These technologies have a slow response time in real-time dynamic coal mine environments and struggle to accurately capture changes in coal and rock properties. The real-time state of the coal seam has a significant impact on the cutting process, but traditional systems often cannot adjust the cutting path and parameters in a timely manner. This can lead to problems such as equipment overload, reduced cutting quality, or resource waste, thereby affecting the safety and efficiency of the operation.

[0005] The introduction of virtual reality (VR) technology has provided new ideas for mining operations. By creating a three-dimensional virtual model of the mine, preliminary cutting planning can be carried out in a simulated environment. However, current VR pre-planning systems are usually limited to static models and cannot reflect the actual situation inside the mine in real time. This leads to a gap between the pre-planning results and the changes in actual operation, reducing the practical application effectiveness of the system.

[0006] To improve the intelligence level of fully mechanized mining equipment in complex mining environments, there is an urgent need for a method that combines VR navigation technology with a real-time coal and rock identification system to form a dynamically adjustable cutting plan. This method should have dynamic adjustment capabilities, responding in real time to changes in coal seams and rocks, optimizing the cutting path, and ensuring efficient equipment operation and maximum resource utilization. By achieving a seamless integration of dynamic pre-planning and online adjustment, the overall efficiency and safety of fully mechanized coal mining operations can be significantly improved.

[0007] The invention patent with publication number CN117744604A provides a coal mine fully mechanized mining planning and cutting method and system based on the Internet of Things and data visualization. The system involves an upper-level computer editing the coal mining machine's process documents and sending them to a lower-level computer. The lower-level computer parses the received process documents and controls the coal mining machine to perform its first trial run. The upper-level computer collects the operational data from the first trial run, calculates the operational data using a planning and cutting method algorithm, and generates planning and cutting method data. This data is then edited according to actual conditions. The upper-level computer sends the edited data to the lower-level computer, which parses the edited data and controls the coal mining machine based on the parsing results. The beneficial effects of this invention are: it achieves intelligent planning and optimization of coal mining machine operation, realizes closed-loop control of the coal mining machine, improves the automation level of coal mining machine operation, enables remote monitoring through a virtual interface, realizes data collection, analysis, and utilization, improves the efficiency and output of coal mining machine operation, and reduces safety hazards.

[0008] The invention patent with publication number CN116398134A provides a method and equipment for coal mining machine cutting planning based on visual perception during inspection. This method combines the spatial mapping relationship between the coordinate system of the moving platform in the mining area and the geological coordinate system to digitally construct the mining area space and automatically detect and identify the coal mining machine. It performs perception analysis on the movement state of the coal mining machine to obtain its movement state information and perform coal mining machine capture and tracking. It also performs state perception on target objects of the coal mining machine components in the cutting space and determines the optimal safe cutting curve based on the state perception data. This invention enables a real-time cutting planning method that can visually and intelligently perceive the equipment and environmental conditions in the cutting space, possessing strong data timeliness and engineering implementation continuity. It provides real-time visualization and interaction of the perception results of related equipment and the surrounding environment in the cutting space, allowing staff to provide real-time cutting assistance and obtain more accurate cutting planning data.

[0009] The invention patent with publication number CN113722979A provides a digital twin-based virtual-physical interaction system for coal mining machines and its construction method. This system enables real-time data analysis of the coal mining machine's operating status and assessment and prediction of its health status. Based on a twin model and iterative algorithms, it achieves advanced simulation of the coal mining machine's cutting actions and optimizes the cutting trajectory. By utilizing real-time status data collected during the coal mining machine's operation and the real-time prediction simulation results from the twin model, it provides real-time dynamic feedback control of the coal mining machine, enabling efficient and autonomous cutting. Using real-time data from the coal mining machine's operation, it can drive the dynamic simulation of the twin model in real time. The twin model can perceive state signals that are not easily directly measured by the physical entity of the coal mining machine, thus facilitating the comparison between measured data from the coal mining machine and perceived data from the twin model. Algorithms are used to assess and predict the real-time health status of the coal mining machine, and a database of coal mining machine health status assessment results is established for the management and predictive maintenance of the coal mining machine's health status throughout its entire life cycle.

[0010] The aforementioned patent documents utilize IoT and data visualization, inspection visual perception, and digital twin technology to achieve coal mining machine cutting planning, respectively. These methods have achieved good results in improving the intelligence and efficiency of coal mining operations. However, they still have some shortcomings. First, while the cutting planning method based on IoT and data visualization can achieve remote data monitoring and optimized operation, it has poor adaptability to real-time changing environments and lacks dynamic adjustment capabilities. Although the inspection visual perception method achieves real-time perception of the coal mining machine and environmental status through visual technology, the visual system is easily affected by factors such as light and coal dust in complex coal mining environments, resulting in limited perception accuracy. While digital twin technology has highly accurate prediction and simulation capabilities, it is heavily dependent on real-time data, and lag in model updates can affect decision accuracy. Overall, these methods still suffer from poor dynamic adaptability, perception limitations, and lag in model feedback. Summary of the Invention

[0011] The technical problem to be solved by this invention is to provide a closed-loop cutting planning method driven by the integration of VR navigation and online planning for coal and rock identification. It combines virtual reality technology for dynamic pre-planning of the environment, and at the same time obtains accurate coal and rock feature information through real-time coal and rock identification technology, so as to realize dynamic cutting planning of the closed-loop control system, thereby improving the system's adaptability and real-time performance.

[0012] To solve the above technical problems, the technical solution adopted by this invention is: a closed-loop cutting planning method driven by the fusion of VR navigation and coal and rock identification online planning, comprising the following modules:

[0013] (1) VR navigation pre-planning module;

[0014] First, historical coal mine data is acquired and preprocessed to build a Transformer model. Real-time data or new input data are then input into the Transformer model. The Transformer model extrapolates based on these inputs to predict future coal and rock conditions and changes in the morphology of the coal seam roof and floor.

[0015] (2) Parameter construction and pose calculation module;

[0016] The morphological and physical property models of the coal seam are constructed using the Unity 3D platform, and the pose and cutting trajectory of the coal mining machine are calculated. Based on the above parameters, the pose of the coal mining machine is calculated by solving the motion changes of the roof and floor, and the real-time pose of the coal mining machine is obtained. Then, combined with the pose data of the coal mining machine, the cutting trajectory of the coal mining machine is further calculated.

[0017] (3) Coal and rock identification module;

[0018] By combining ultrasonic sensors and laser-induced breakdown spectroscopy (LIBS) technology, coal seams and rock strata are identified in real time to obtain coal and rock characteristic parameters. After coal and rock identification is completed, a fine model of the coal seam roof is constructed in Unity 3D to realize the reconstruction of the coal seam roof. Based on the reconstructed roof, the virtual coal mining machine is cut and simulated in Unity 3D to correct the attitude of the coal mining machine.

[0019] (4) Drum adjustment and dynamic speed regulation module;

[0020] The motion trajectory of the roller is optimized using coal and rock feature data provided by the coal and rock identification module and pose data provided by the pose calculation module.

[0021] (5) Path execution and feedback module;

[0022] Based on the prediction and optimization results jointly generated by the VR navigation pre-planning module, parameter construction and pose calculation module, coal and rock identification module and drum adjustment and dynamic speed regulation module, the actual operation path of the coal mining machine is executed, and the operation data and operation feedback information of the coal mining machine are transmitted back to the system.

[0023] (6) Online prediction and decision-making module;

[0024] Based on the coal and rock data obtained from the coal and rock identification module, the coal and rock status is updated in real time through data processing and the predictive capabilities of the Transformer model, dynamic prediction is made, and the results are fed back into the working path planning of the coal mining machine. In addition, the parameters of the coal mining machine are dynamically adjusted according to the real-time coal and rock data to optimize the mining strategy.

[0025] Furthermore, the steps for constructing a VR navigation pre-planning module include:

[0026] Step 101: Historical Data Acquisition and Preprocessing: First, acquire historical coal mine data, record the data according to time series, then perform data preprocessing, and divide the data into training set, validation set and test set according to time order;

[0027] Step 102, Data Segmentation and Feature Extraction: After data preprocessing, the historical coal mine data is divided into segments according to the mining location intervals. For each segment, feature extraction is performed to construct a feature matrix, and the feature matrix is ​​arranged in chronological order.

[0028] Step 103, Transformer Model Construction: The Transformer model is constructed using the deep learning framework TensorFlow;

[0029] Step 104, Model Training and Validation: In the Transformer model, load the training set and validation set to train the model. After training, perform a comprehensive evaluation on the test set. The model that has completed training and validation is used for coal and rock conditions and coal seam roof and floor prediction.

[0030] Step 105, Model Validation and Feedback: Apply the model to field operation data independent of the training process, evaluate whether the model runs stably in actual operation, analyze the robustness of the model under extreme conditions, and feed the deviation back to the system to improve the model's algorithm or add new input features.

[0031] Furthermore, the steps for constructing the parameter construction and pose calculation module include:

[0032] Step 201, Morphological Parameter Construction: Using the Unity 3D platform, simulate the actual geometric structure of the coal mining machine, and digitally reconstruct the morphology of the top and bottom plates of the coal seam based on the changes in the morphology of the top and bottom plates of the coal seam.

[0033] Step 202, Material Property Parameter Construction: Obtain historical coal mine data including stress, hardness, density, and chemical element composition. Based on this historical data, establish a material property parameter library in Unity 3D. Integrate the material property parameters with the geometric model constructed in Step 201, and add C# scripts to the coal mining machine to enable its operation and control. Use Unity3D's Transform component to dynamically change the X, Y, and Z coordinates of the coal mining machine model, as well as the pitch (RX), yaw (RY), and roll (RZ) angles, to precisely control the virtual coal mining machine's position and orientation in three-dimensional space, thereby simulating its actual operation in the coal mine working environment.

[0034] Step 203: Coal mining machine pose calculation;

[0035] Step 204: Calculate the cutting trajectory.

[0036] Furthermore, in step 202, during the construction process, virtual sensors are introduced to simulate the function of actual sensors, collect data and feed it back to the physical property parameter model to acquire and monitor the state of coal seams and rock strata in real time. These virtual sensors automatically update data according to changes in physical property parameters and provide real-time feedback.

[0037] Furthermore, step 203, the coal mining machine pose calculation steps, include:

[0038] To construct a virtual scene of the coal mining face, it is first necessary to establish an absolute coordinate system with the initial position of the coal mining machine as the origin. , where X n The axis points in the direction of the coal mining machine's movement, Y n Vertically upward, Zn The axis points in the direction of the working face advance. The pose calculation of the coal mining machine is performed by analyzing the axis encoder data to calculate its position coordinates and travel angle in the virtual workspace, so as to accurately describe the dynamic trajectory of the coal mining machine;

[0039] According to the coal mining machine at the i The position and the first i The position information of the coal mining machine is deduced from the encoder data at +1 position. The specific position coordinates are calculated using formula (2-1):

[0040] (2-1)

[0041] Among them, X i With Y i These represent the number of coal mining machines. i The x and y coordinates at each position s i For the first coal mining machine i The axis encoder reading at each position θ i For the first coal mining machine i Pitch angle at each position;

[0042] To accurately describe the rotational attitude of the coal mining machine, quaternions are used to represent its attitude in space. Let the coal mining machine be at position... i The attitude quaternion is q i Its form is:

[0043]

[0044] Where w is the real part of the quaternion, and x, y, and z are the imaginary components used to represent the current rotation state of the coal mining machine;

[0045] Coal mining machine in position i arrive i The rotation increment between +1 is Δq, which represents the increments of pitch, yaw, and roll angles. Convert them to their corresponding quaternion increments:

[0046]

[0047] in:

[0048]

[0049] The increment is applied to the original pose using the quaternion update formula to obtain the next pose. i +1 posture quaternion q i+1 :

[0050]

[0051] The position information and attitude parameters obtained through pose calculation provide necessary initial pose references for the roller adjustment and dynamic speed regulation module, and also provide real-time pose data for the path execution and feedback module.

[0052] Furthermore, step 204, the calculation of the cutting trajectory, includes:

[0053] Based on the mining height information and the structure of the coal mining machine (such as...) Figure 4 The rocker arm tilt angle is calculated as shown in equation (2-2).

[0054] (2-2)

[0055] in and The coal mining machine was in the first i The cutting height of the front and rear rollers at each position; h is the distance from the rocker arm pin to the bottom plate of the scraper conveyor; D is the roller diameter; L1 is the rocker arm length;

[0056] Then, based on the corrected trajectory of the coal mining machine, the cutting trajectory is calculated using the calculated rocker arm inclination angle. The coordinates of the upper drum cutting point and the lower drum cutting point are calculated according to equations (2-3) and (2-4), respectively: (2-3)

[0057] (2-4)

[0058] Where L is the distance from the front rocker arm connecting pin to the rear rocker arm connecting pin. θ i For the coal mining machine at the first i Pitch angle at each position.

[0059] Furthermore, the feature is that the steps for constructing the coal and rock identification module include:

[0060] Step 301, Equipment Initialization and Calibration: Before starting the coal and rock identification module, the ultrasonic sensor and LIBS sensor are initialized and calibrated.

[0061] Step 302, Ultrasonic sensor detection: The ultrasonic sensor detects the coal seam and rock strata. The ultrasonic sensor determines the physical properties by emitting high-frequency sound waves and receiving the echo signals reflected back from the coal and rock.

[0062] Step 303, LIBS Detection: After the ultrasonic sensor completes the preliminary determination of coal and rock density, the system activates the LIBS sensor to perform a more accurate chemical composition analysis of the coal and rock.

[0063] Step 304, Coal Seam Roof Reconstruction: After completing coal and rock identification, the system will reconstruct a detailed model of the coal seam roof in Unity 3D using voxelization and Marching Cubes algorithms;

[0064] Step 305, Coal Mining Machine Attitude Correction: Based on the reconstructed roof obtained in Step 304, the virtual coal mining machine is subjected to cutting simulation in Unity 3D. The front roller of the virtual coal mining machine is attached to the coal seam roof, and a new roller cutting control strategy is obtained, which can correct the attitude of the coal mining machine.

[0065] Furthermore, the steps for constructing the drum adjustment and dynamic speed regulation module include:

[0066] Step 401, Coal and Rock Analysis and Preliminary Judgment: The system receives coal and rock feature data from the coal and rock identification module and pose data from the parameter construction and pose calculation module via the MQTT protocol. First, the coal and rock features are analyzed, and the hardness, density, and other physical properties of the current cutting area are calculated as a preliminary basis for judging the coal and rock structure. At the same time, the system judges the shape of the top and bottom plates of the coal seam based on the pose data, identifies whether there is a height difference or tilt change, and preliminarily determines whether the height and angle of the roller need to be adjusted so as to further optimize the roller movement trajectory in subsequent steps.

[0067] Step 402, Drum Height Adjustment: After the coal and rock analysis and preliminary judgment are completed, the system calculates the ideal cutting height based on the height difference between the top and bottom plates of the coal seam, the hardness and density of the coal and rock, and uses it as the target height for drum adjustment; the drum height is adjusted by changing the extension of the hydraulic cylinder of the coal mining machine to adapt to the needs of different cutting layers and coal and rock characteristics.

[0068] Step 403, Coal mining machine speed adjustment: The system uses the MQTT protocol to receive sensor data and obtain real-time drum load and speed information, and dynamically adjusts the coal mining machine's travel speed by analyzing this data;

[0069] Step 404, Feedback and Optimization: The operation data of the coal mining machine is fed back in real time to adapt to the changing characteristics of the coal and rock strata; the optimization module compares the actual state with the target state and further optimizes the speed and cutting depth through reinforcement learning algorithms.

[0070] Furthermore, the steps for building the path execution and feedback module include:

[0071] Step 501, Path Execution: Based on the optimized cutting strategy and travel path, the system sends instructions to the coal mining machine, which then begins actual operation according to the instructions. During the operation, the system monitors the key data of the coal mining machine in real time through sensors.

[0072] Step 502, Feedback and Data Analysis: During the operation of the coal mining machine, the system will continuously collect and provide real-time feedback of key data from the sensors. The system will compare this feedback data with the optimized instructions, detect any deviations or anomalies in the operation, and update the operation status immediately.

[0073] Step 503, Memory Cutting and Path Reproduction: The system establishes a "demonstration cutter" as a benchmark during the demonstration operation. When actually performing memory cutting, the system calls the data recorded by the demonstration cutter, compares the current operation with the parameters of the demonstration cutter in real time, minimizes the deviation, and reproduces the motion trajectory and operation mode of the demonstration cutter. The reference data is transmitted to the actuator through the MQTT protocol. The system can dynamically adjust the height and angle of the roller and the propulsion speed to ensure consistency with the demonstration cutter.

[0074] Furthermore, the steps for constructing an online prediction and decision-making module include:

[0075] Step 601, Data Processing and Status Update: After the system acquires real-time coal and rock data, it first performs data preprocessing, such as noise filtering and outlier handling; the processed data is then input into the Transformer model trained in Step 104 to update the coal mining status and ensure that the model predictions can reflect the latest coal and rock conditions.

[0076] Step 602, Dynamic prediction using Transformer model: After data processing and status update, the system calls the Transformer model for real-time prediction. By continuously inputting real-time sensor data and prediction results into the Transformer model, the system makes detailed predictions of future coal and rock changes. Through continuous iterative prediction, the system can gradually optimize the path and cutting strategy.

[0077] Step 603, Cutting Strategy Optimization: Based on the dynamic prediction results provided by the Transformer model, the cutting strategy of the coal mining machine is optimized by combining real-time data and prediction information. According to the changing trend of coal and rock structure, the cutting depth, cutting path and travel speed of the coal mining machine are adjusted to ensure that the equipment always maintains the best working state. By analyzing key factors including the hardness and thickness of the coal seam and the coal-rock boundary, the optimal cutting parameters are determined.

[0078] Step 604, Dynamic Adjustment and Feedback: After the cutting strategy is optimized, the system dynamically adjusts the operating parameters of the coal mining machine based on real-time feedback. The system continuously monitors the operation of the equipment and changes in coal and rock through sensor data. The system adjusts the cutting depth, path and speed in real time to ensure that the coal mining machine operates in the best condition. The feedback loop is used to analyze the difference between the actual effect and the optimized strategy and make adjustments quickly.

[0079] This invention introduces virtual reality (VR) navigation dynamic pre-planning technology to pre-plan the cutting path of the coal mining machine on a macroscopic level. Simultaneously, it combines real-time coal and rock identification technology, using online sensors and algorithms to identify the hardness, structure, and location of coal and rock, obtaining precise data at the microscopic level. The fusion of these two technologies, through a closed-loop feedback control mechanism, combines real-time perceived coal and rock information with the virtual pre-planning results to dynamically adjust the coal mining machine's cutting strategy. The core of this invention lies in the deep integration of the virtual and physical layers. Virtual pre-planning guides the real-time operation of physical equipment, while physical feedback optimizes the virtual planning, thereby achieving closed-loop control and dynamically adjusting the cutting plan. Attached Figure Description

[0080] Figure 1 This is a general framework diagram of the closed-loop cut-off planning method described in this invention;

[0081] Figure 2 This is a flowchart of the coal and rock identification process of the present invention;

[0082] Figure 3 This is a flowchart of the navigation map process of the present invention;

[0083] Figure 4 This is a structural diagram of the coal mining machine of the present invention. Detailed Implementation

[0084] This invention provides a closed-loop cut-off planning method driven by the fusion of VR navigation and coal and rock identification online planning, such as... Figure 1 As shown, it includes the following 6 modules: (1) VR navigation pre-planning module; (2) parameter construction and pose calculation module; (3) coal and rock identification module; (4) drum adjustment and dynamic speed regulation module; (5) path execution and feedback module; (6) online prediction and decision-making module.

[0085] Based on the above 6 modules, such as Figure 1 , Figure 3As shown, the overall concept of this invention is as follows: The VR navigation pre-planning module provides the parameter construction and pose calculation module with the future state of coal and rock and the morphological changes of the coal seam roof and floor; the parameter construction and pose calculation module provides the drum adjustment and dynamic speed regulation module with pose parameters and cutting trajectory parameters, and also provides the coal and rock identification module with physical property parameters through virtual sensors; the coal and rock identification module detects the distribution of coal and rock based on ultrasonic and LIBS sensors, corrects the coal mining machine pose, and feeds it back to the drum adjustment and dynamic speed regulation module; the drum adjustment and dynamic speed regulation module receives the pose and trajectory parameters, and optimizes the drum operation in real time by combining the feedback information from the coal and rock identification module. The first four modules jointly generate a pre-planning scheme and transmit it to the path execution and feedback module, which transmits the feedback data of the equipment operation back to each module to optimize the entire system; the online prediction and decision-making module, in conjunction with the drum adjustment and dynamic speed regulation module, the coal and rock identification module, and the path execution and feedback module, performs online prediction and dynamic decision-making adjustments on the feedback data to achieve optimization of the working state and path.

[0086] To enable those skilled in the art to better understand the present invention, the present invention will be further described clearly and completely below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the examples of this application can be combined with each other.

[0087] 1. VR Navigation Pre-planning Module

[0088] VR navigation pre-planning is a planning method based on virtual reality technology. By integrating historical and real-time data, it uses the Transformer model to pre-determine the changes in coal and rock conditions and the morphology of the roof and floor of the coal seam, providing dynamic navigation maps and predictive data support for subsequent modules.

[0089] Step 101: Historical Data Acquisition and Preprocessing: First, acquire historical coal mine data, including coal seam hardness, density, structure, and other characteristics. Record the data according to time series, then perform data denoising and normalization. Divide the data into training, validation, and test sets according to time order. The training set is used to train the model, the validation set is used to verify the model's performance, and the test set is used for final testing. Finally, compile the data into an XML file and use Unity 3D code to read the data, enabling the reading of specific data types for subsequent model processing and pre-planning analysis.

[0090] Step 102: Data Segmentation and Feature Extraction: After data preprocessing, the historical coal mine data is divided into segments according to mining location intervals, with each segment consisting of a 10-meter square. This allows for the segmented capture of local feature changes in the coal seam, enhancing the consistency of the time-series data. For each segment, features are extracted based on physical properties (e.g., hardness, density, coal seam thickness), chemical composition (e.g., sulfur content, carbon content), and structural characteristics (e.g., coal bedding, interbedded rock thickness). During feature extraction, signal denoising is used to optimize the accuracy of sensor data such as hardness and density. Then, the physical property data is standardized to construct a feature matrix for unified analysis of each coal seam feature. Subsequently, a feature selection algorithm is used to filter out features important for prediction, reducing interference from redundant information. Finally, these feature matrices are arranged in chronological order to provide clear, time-consistent input features for subsequent predictions of coal rock conditions and the morphology of the coal seam roof and floor.

[0091] Step 103: Transformer Model Construction: The deep learning framework TensorFlow is used to construct the Transformer model to achieve accurate prediction of coal and rock conditions and the roof and floor of the coal and rock.

[0092] First, the TensorFlow and Keras libraries are imported to provide the foundation for model building. When constructing the initial structure, the core parameters of the model are defined, including the embedding dimension, the number of attention heads, and the number of hidden layer units in the feedforward network. The input layer uses an embedding layer to embed the raw data, transforming it into a high-dimensional feature representation, which facilitates the model's capture of complex sequence information. Then, a multi-head self-attention mechanism is introduced in the encoder part. This mechanism, through multi-head parallel processing, allows the model to focus on features at different positions in the input sequence. Each layer of the encoder consists of a multi-head self-attention layer, a feedforward network, and residual connections. Within each encoder layer, self-attention operations are first used to make the model focus on the dependencies between features; then the output is passed through the feedforward network to further extract and abstract important features. To maintain the model's stability and efficiency, layer normalization is introduced after each layer operation, combined with residual connections to ensure gradient stability and accelerate model convergence. The final layer of the model is the output layer, which generates the prediction results.

[0093] During model compilation, the Adam optimizer and the MSE loss function are used to continuously optimize the model. The Adam optimizer is responsible for updating the model's weights during training. By continuously adjusting the weights, the optimizer gradually brings the model closer to the optimal solution, determining how the model adjusts its parameters in each training iteration. The MSE loss function measures the difference between the model's predictions and the true values; the smaller the loss function, the higher the model's prediction accuracy. After the model is built, the model weights and structure are saved for subsequent training, validation, and evaluation.

[0094] Step 104, Model Training and Validation: First, by loading the collected and preprocessed coal and rock data, including training and validation sets, ensure that the input data is aligned with the predicted labels, and import the data into the model in batches according to the specified batch size for efficient training. The core of model training lies in iterative optimization through forward and backpropagation. After training begins, the input data is passed to the model, and the input data will be processed sequentially through the embedding layer, the multi-head self-attention layer, the encoder's self-attention module, and the feedforward network to extract important features from the sequence data layer by layer, and finally generate predicted values ​​through the output layer. Then, the error between the model's predicted values ​​and the true labels is measured by the MSE (mean squared error) loss function, and the error is propagated back layer by layer through the backpropagation algorithm.

[0095] Model optimization is performed by the Adam optimizer, which adjusts the weights of each layer to gradually approach the minimum of the loss function, thereby improving the model's prediction accuracy. At the end of each training cycle, validation set data is input into the model to evaluate its performance on unseen data, and the validation set error is used to detect overfitting. If the validation error continues to decrease, it indicates that the model has good generalization ability. If the model's validation error begins to rise, indicating overfitting, the EarlyStopping callback function can be used to implement an early stopping strategy, halting training to prevent further overfitting.

[0096] Simultaneously, a learning rate scheduler is used to dynamically adjust the learning rate, ensuring smooth convergence of the model at different training stages. After training, a comprehensive evaluation is conducted on the test set, taking into account metrics such as MSE and MAE to confirm the model's predictive accuracy and stability. The model that has completed training and validation will be used for predicting coal and rock conditions and the roof and floor of coal seams. At this point, real-time data or new input data will be fed into the model, which will then extrapolate based on these inputs to predict future coal and rock conditions and changes in the morphology of the roof and floor of coal seams, thus providing a basis for subsequent parameter construction and pose calculation modules.

[0097] Step 105, Model Validation and Feedback: After model training and validation, model validation and feedback are mainly used to evaluate the model's performance in real-world application environments and provide optimization suggestions.

[0098] In this step, the model is applied to field operational data independent of the training process, and its accuracy and adaptability in predicting coal and rock properties are observed through field testing. The focus of validation is to evaluate whether the model operates stably during actual operations and to analyze the model's robustness under extreme conditions.

[0099] Furthermore, deviation data from actual operations is collected and fed back to the system to improve the model's algorithm or add new input features to better match on-site requirements. Ultimately, the verification and feedback results are used to confirm the model's production feasibility, providing more reliable data support for closed-loop path execution and feedback, thus forming a complete dynamic optimization closed-loop system.

[0100] 2. Parameter Construction and Pose Calculation Module

[0101] The parameter construction and pose calculation module is responsible for constructing the morphological and physical property parameter models of the coal seam and calculating the pose and cutting trajectory of the coal mining machine. First, the morphological parameters (such as the height and structure of the roof and floor) and physical property parameters (including hardness, density, and chemical element distribution) of the coal seam are generated using the Unity 3D platform to accurately reflect the physical characteristics of the coal seam. Then, based on these parameters, the pose of the coal mining machine is calculated by solving the motion changes of the roof and floor, obtaining the real-time pose of the coal mining machine. Finally, combined with the pose data, the cutting trajectory of the coal mining machine is further calculated to support subsequent path planning and dynamic adjustments.

[0102] Step 201, Morphological Parameter Construction: This part mainly involves the accurate modeling of the geometry of the coal mining machine and the roof and floor of the coal seam. Using the Unity 3D platform, the actual geometry of the coal mining machine is simulated, and the roof and floor morphology of the coal seam is digitally reconstructed based on the changes in the roof and floor morphology, including information such as the height and undulation characteristics of the roof and the structure of the floor.

[0103] The system uses real-time coal and rock data collected, along with predictions of future coal and rock conditions and changes in the roof and floor morphology of the coal seam, from the Transformer model as basic data inputs. To ensure data accuracy and consistency, this basic data undergoes processing steps such as data smoothing, noise reduction, and interpolation to eliminate noise and inconsistencies during data acquisition. Based on this data, the coal mining machine and a 3D navigation map (i.e., the roof and floor of the coal seam) containing the geometry of the coal face and the distribution of coal and rock are modeled in UG, and the necessary pins are repaired. The model is then saved in STL format and imported into 3D Max. In 3D Max, the model is exported in FBX format and imported into Unity 3D. Preliminary parameter settings are made for the model in the virtual scene, and parent-child hierarchical relationships are established to ensure the collaborative interaction of different objects in the virtual scene.

[0104] Step 202, Material Property Parameter Construction: First, acquire historical coal mine data, including stress, hardness, density, and chemical elemental composition. Smooth, denoise, and interpolate this data to ensure accuracy and consistency. Next, create a new project in Unity 3D and establish a material property parameter library. Each material property parameter will be managed as an independent script or component. During this process, C# scripts are used to define the properties of each parameter, including its value, unit, and range of variation, facilitating dynamic adjustment and precise control during simulation.

[0105] During the construction process, virtual sensors are introduced to acquire and monitor the state of coal seams and rock strata in real time. These virtual sensors can simulate the function of actual sensors, collecting data and feeding it back to the physical property parameter model, thereby enhancing the model's accuracy and real-time performance. These virtual sensors can automatically update data based on changes in physical property parameters, providing real-time feedback.

[0106] In Unity 3D, a deep black, high-gloss material is used for the coal seam to reflect its high carbon content and smooth properties. The rock strata, on the other hand, use a light gray, rough material to showcase their hardness and porosity, realistically simulating the physical structure of rock. Dynamic deformation effects are used to demonstrate the deformation of the coal seam under stress, visually reflecting its response during mining.

[0107] Furthermore, interactive functionality was added to the physical property parameter model, allowing users to adjust parameters in real time by clicking or swiping, and observe the impact of these changes on coal seam characteristics and simulation results. Next, the physical property parameters were integrated with the geometric models (such as the coal mining machine and the roof and floor of the coal seam) constructed in step 201, and a C# script was added to the coal mining machine to enable its operation and control. By using Unity3D's Transform component, the X, Y, and Z coordinates of the coal mining machine model, as well as the pitch (RX), yaw (RY), and roll (RZ) angles, were dynamically changed to precisely control the virtual coal mining machine's position and orientation in three-dimensional space, thereby simulating its actual operation in a coal mine working environment.

[0108] Step 203, Coal Mining Machine Pose Calculation: To construct a virtual scene of the coal mining face, it is first necessary to establish an absolute coordinate system with the initial position of the coal mining machine as the origin. Among them, X n The axis points in the direction of the coal mining machine's movement, Y n Vertically upward, Z n The axis points in the direction of the working face advance. The pose calculation of the coal mining machine is performed by analyzing the axis encoder data to calculate its position coordinates and travel angle in the virtual workspace, so as to accurately describe the dynamic trajectory of the coal mining machine.

[0109] According to the coal mining machine at the i The position and the first i The position information of the coal mining machine can be deduced from the encoder data at +1 position. The specific position coordinates are calculated using formula (2-1):

[0110] (2-1)

[0111] Among them, X i With Y i These represent the number of coal mining machines. i The x and y coordinates at each position s i For the first coal mining machine i The axis encoder reading at each position θ i For the first coal mining machine i Pitch angle at each position.

[0112] To accurately describe the rotational attitude of the coal mining machine, quaternions are used to represent its attitude in space. Let the coal mining machine be at position... i The attitude quaternion is q i Its form is:

[0113]

[0114] Where w is the real part of the quaternion, and x, y, and z are the imaginary components, used to represent the current rotation state of the coal mining machine.

[0115] Coal mining machine in position i arrive i The rotation increment between +1 is Δq, which represents the increments of pitch, yaw, and roll angles. Convert them to their corresponding quaternion increments:

[0116]

[0117] in:

[0118]

[0119] The increment is applied to the original pose using the quaternion update formula to obtain the next pose. i +1 posture quaternion q i+1 :

[0120]

[0121] The position information and attitude parameters obtained through pose calculation provide necessary initial pose references for the roller adjustment and dynamic speed regulation module, and also provide real-time pose data for the path execution and feedback module.

[0122] Step 204, Cutting Trajectory Calculation: The coal mining machine travels along the scraper conveyor, and the drum cuts the coal face to form the top and bottom plates of the coal seam. The cutting trajectory of the coal mining machine reflects the changing trend of the coal seam bottom plate. Based on the mining height information and the structure of the coal mining machine (such as...), the cutting trajectory is calculated. Figure 4 The rocker arm tilt angle is calculated as shown in equation (2-2).

[0123] (2-2)

[0124] in and The coal mining machine was in the first i The cutting height of the front and rear rollers is at position 1; h is the distance from the rocker arm pin to the bottom plate of the scraper conveyor. Since the coal seam is relatively flat, the influence of the rotation angle of the support slipper relative to the machine body is ignored, so this value is constant; D is the roller diameter; L1 is the rocker arm length.

[0125] Then, based on the corrected trajectory of the coal mining machine, the cutting trajectory is calculated using the calculated rocker arm inclination angle. The coordinates of the upper drum cutting point and the lower drum cutting point are calculated according to equations (2-3) and (2-4), respectively: (2-3)

[0126] (2-4)

[0127] Where L is the distance from the front rocker arm connecting pin to the rear rocker arm connecting pin. θ i For the coal mining machine at the first i The pitch angle at each position. These coordinate data provide a precise trajectory reference for the roller adjustment and dynamic speed control modules, and also provide position information support for the path execution and feedback modules.

[0128] 3. Coal and Rock Identification Module

[0129] like Figure 2 As shown, the coal and rock identification module is responsible for identifying coal seams and rock strata in real time using sensor technology, helping the coal mining machine accurately determine the geological conditions of the working face. In this module, ultrasonic sensors are used to detect the physical properties of coal and rock, such as hardness and density, helping to distinguish between coal seams and rock strata; while laser-induced breakdown spectroscopy (LIBS) technology identifies the chemical composition of coal and rock by analyzing the spectral information of the coal and rock surface. These two technologies work together, enabling the system to perform a comprehensive analysis from both physical and chemical perspectives, ensuring the accuracy of the identification.

[0130] Step 301, Equipment Initialization and Calibration: Before starting the coal and rock identification module, the ultrasonic sensor and LIBS sensor are initialized and calibrated to ensure the equipment can accurately detect the physical and chemical properties of coal and rock strata. The calibration of the ultrasonic sensor mainly involves adjusting the wave velocity, transmission frequency, and sensitivity to ensure accurate measurement of propagation characteristics in different media. Simultaneously, the LIBS sensor needs to have its laser power and spectral detection range set to ensure accurate identification of the elemental composition of coal and rock strata. During calibration, the system reads the known characteristics of a reference sample and adjusts the sensor parameters by comparing the deviation between the detection results and standard values ​​until the expected accuracy is achieved.

[0131] Step 302, Ultrasonic Sensor Detection: After equipment calibration, the ultrasonic sensor begins detecting the coal seam and rock strata. The ultrasonic sensor determines the physical properties of materials by emitting high-frequency sound waves and receiving echo signals reflected from different media (such as coal and rock). Specifically, ultrasonic waves propagate at different speeds in materials with different densities and hardnesses. The time delay, attenuation, and waveform changes of the echoes can reveal information such as the material's density, thickness, and internal structure. The system acquires these signals in real time and improves data accuracy through data filtering, noise reduction, and signal enhancement. Next, the system performs density assessment based on the measured sound wave propagation speed and density model, comparing it with the system's built-in coal and rock characteristic database to determine if the density is below a preset density threshold. If the measured density is low, the system preliminarily identifies the current detection area as a coal seam; if the density is high, it identifies it as a rock stratum. This preliminary judgment provides the basis for subsequent LIBS detection.

[0132] Step 303, LIBS Detection: After the ultrasonic sensor completes the preliminary density assessment of the coal and rock, the system activates the Laser-Induced Breakdown Spectroscopy (LIBS) sensor for a more precise chemical composition analysis of the coal and rock. LIBS uses high-energy laser pulses to irradiate the surface of the coal and rock, exciting plasma. The spectrum emitted by the plasma is then analyzed by a spectrometer to extract elemental composition information. For areas initially identified as coal seams, if the LIBS detection shows a high carbon content, the area will be definitively identified as a coal seam. For areas initially identified as rock strata, if a low carbon content and a high content of typical rock strata elements such as silicon or aluminum are detected, the area is confirmed as a rock strata.

[0133] For detection results that differ from the initial assessment—for example, if the initial assessment identifies it as a coal seam but the carbon content is low, while the rock strata have high elemental content, or vice versa—the system will mark these results as anomalous data and store them in the coal and rock characteristics database. This anomalous data will then be analyzed using machine learning algorithms to optimize the coal and rock identification model, ensuring detection accuracy and robustness of the identification process, and gradually improving the system's adaptability to complex changes in coal and rock characteristics.

[0134] Step 304, Coal Seam Roof Reconstruction: After completing coal and rock identification, the system will construct a detailed model of the coal seam roof in Unity 3D using voxelization and Marching Cubes algorithms.

[0135] The coal and rock data are voxelized to enhance the model's spatial resolution, and the Marching Cubes algorithm is used to triangulate the voxel mesh to generate a smooth, natural mesh structure, ensuring that the coal seam surface highly replicates the true geological features. Based on this, a height map and Perlin Noise function are further applied to introduce detailed unevenness into the coal seam surface, simulating the irregularity and thickness differences of the rock strata. The system also reconstructs the rock strata portion of the mesh based on the rock strata thickness information determined in the preliminary analysis, appropriately reducing the height of the rock strata to achieve separation between the coal seam and the rock strata, ultimately obtaining a complete coal-rock boundary.

[0136] During the modeling process, the system will also utilize Unity 3D's physics engine to enhance the physical interaction between the coal seam roof and the fully mechanized mining equipment. Specifically, a Rigidbody component will be added to the fully mechanized mining equipment model to make it subject to physical factors such as gravity and inertia. At the same time, a Mesh Collider component will be added to both the equipment and the coal seam to simulate the weight and motion feedback of the equipment on the coal seam roof.

[0137] Step 305, Coal Mining Machine Attitude Correction: Based on the reconstructed roof obtained in Step 304, the virtual coal mining machine is subjected to cutting simulation in Unity 3D. The front roller of the virtual coal mining machine is attached to the coal seam roof, and a new roller cutting control strategy is obtained, which can correct the attitude of the coal mining machine.

[0138] 4. Drum adjustment and dynamic speed control module

[0139] The drum adjustment and dynamic speed control module is primarily responsible for real-time adjustment of the drum angle and height of the coal mining machine, and dynamic adjustment of the machine's speed to adapt to changes in the coal and rock strata. This module optimizes the drum's trajectory using coal and rock feature data provided by the coal and rock identification module and pose parameters provided by the pose calculation module, achieving more precise cutting control. Simultaneously, the system generates drum adjustment strategies and implements dynamic speed control based on changes in the state and physical properties of the coal seam's roof and floor.

[0140] Step 401, Coal and Rock Analysis and Preliminary Judgment: The system receives coal and rock feature data from the coal and rock identification module and pose data from the parameter construction and pose calculation module via the MQTT protocol, including hardness, density, and real-time morphological parameters of the coal seam roof and floor. The system first analyzes these coal and rock features, calculating the hardness, density, and other physical properties of the current cutting area as a preliminary basis for judging the coal and rock structure. Simultaneously, based on the pose data, the system judges the morphology of the coal seam roof and floor, identifying any height differences or tilt changes, and preliminarily determines whether the height and angle of the roller need to be adjusted for further optimization of the roller's movement trajectory in subsequent steps.

[0141] Step 402, Drum Height Adjustment: After coal and rock analysis and preliminary judgment, the system calculates the ideal cutting height based on the height difference between the top and bottom of the coal seam, coal and rock hardness, density, and other characteristic information, and uses this as the target height for drum adjustment. Drum height adjustment is achieved by changing the extension of the hydraulic cylinder of the coal mining machine to adapt to the needs of different cutting layers and coal and rock characteristics. In specific execution, the system compares the target height with the real-time drum height, receiving the drum height data obtained from pose calculation at a frequency of 10Hz per second via the MQTT protocol to ensure the continuity and real-time nature of drum height feedback, and transmitting it to the PID controller. The PID controller performs real-time calculation and feedback adjustment of the error value between the drum and the target height during the adjustment process. The control process consists of three parts: proportional (P), integral (I), and derivative (D): proportional control adjusts the drum position according to the magnitude of the instantaneous error to quickly approach the target height; integral control is used to eliminate long-term deviations and ensure that the drum remains stable near the target height; derivative control smooths the adjustment based on the error change rate, avoiding excessive or insufficient displacement. Under this control framework, the PID controller makes feedback adjustments based on the magnitude of the error and sends control signals to the hydraulic cylinder actuator via the MQTT protocol to dynamically adjust the extension of the hydraulic cylinder, thereby precisely adjusting the height of the drum.

[0142] Step 403, Coal Mining Machine Speed ​​Adjustment: The system uses the MQTT protocol to receive sensor data at a frequency of 10 Hz and acquire real-time drum load and rotational speed information. By analyzing this data, the system dynamically adjusts the coal mining machine's travel speed. To this end, the system employs an adaptive speed control method, determining whether to accelerate or decelerate based on real-time changes in drum load and physical properties such as coal and rock hardness. Specifically, when the load increases or a high-hardness coal or rock layer is encountered, the speed is appropriately reduced to avoid equipment overload; conversely, the speed is increased to improve work efficiency. The control signal is then transmitted to the coal mining machine's hydraulic drive system via the MQTT protocol to achieve precise speed control, ensuring system stability and cutting efficiency.

[0143] Step 404, Feedback and Optimization: This step dynamically optimizes control parameters by providing real-time feedback on the coal mining machine's operating data, such as speed, drum height, and angle, to adapt to the changing characteristics of the coal and rock strata. The optimization module compares the actual state with the target state and further optimizes the speed and cutting depth through reinforcement learning algorithms, thereby improving coal mining efficiency and equipment stability. Combining the drum adjustment, speed control, and coal and rock identification analysis from the previous three modules, the system generates an optimized pre-planned scheme and transmits it to the path execution and feedback module for real-time path control and decision-making.

[0144] 5. Path execution and feedback module

[0145] The path execution and feedback module is mainly responsible for executing the actual operation path of the coal mining machine based on the previous prediction and optimization results, and at the same time transmitting the coal mining machine's operation data and operation feedback information back to the system for further optimization and adjustment.

[0146] Step 501, Path Execution: Based on the optimized cutting strategy and travel path, the system sends instructions to the coal mining machine, which then begins actual operation. During execution, the coal mining machine strictly follows the system's optimized path, cutting depth, and speed parameters. Simultaneously, the system monitors key data of the coal mining machine in real time through sensors, such as travel speed, position, cutting depth, and equipment load.

[0147] Step 502, Feedback and Data Analysis: During the operation of the coal mining machine, the system continuously collects and provides real-time feedback of key data from sensors, such as travel speed, equipment position, and cutting depth. The system compares this feedback data with optimized instructions, detects any deviations or anomalies in the operation, and immediately updates the operation status to ensure that the system can quickly adjust the path, speed, or cutting depth.

[0148] Step 503, Memorizing Cutting and Path Reproduction: The system will establish a "demonstration cutter" as a benchmark during the demonstration operation. The purpose of the demonstration cutter is to record the key cutting parameters when the coal mining machine passes through the middle trough of each scraper conveyor and store these parameters in XML format. These key parameters include the height and angle of the drum, the advance speed and position of the coal mining machine, etc.

[0149] During actual memory cutting, the system retrieves data recorded by the demonstration cutter, compares the current operation with the parameters of the demonstration cutter in real time, minimizes deviations, and reproduces the motion trajectory and operation mode of the demonstration cutter. These reference data are transmitted to the actuator via the MQTT protocol, allowing the system to dynamically adjust the height and angle of the rollers and the advance speed to ensure consistency with the demonstration cutter.

[0150] 6. Online Prediction and Decision-Making Module

[0151] The online prediction and decision-making module is responsible for ensuring that the coal mining machine's operating path can adapt to the complex and ever-changing coal and rock environment through real-time monitoring and dynamic adjustments. Based on accurate coal and rock data obtained from the coal and rock identification module, this module uses data processing and the predictive capabilities of the Transformer model to update the coal and rock status in real time, perform dynamic predictions, and feed this information back into the coal mining machine's working path planning. The module also dynamically adjusts parameters such as the coal mining machine's working path, speed, and cutting depth based on real-time coal and rock data to optimize the mining strategy. This module is closely related to the coal mining machine's VR navigation pre-planning: VR navigation pre-planning sets the initial path and operation plan for the coal mining machine, while the online prediction and decision-making module continuously calibrates and optimizes the path based on real-time feedback during actual operation, thereby achieving more accurate and efficient coal mining operations.

[0152] Step 601, Data Processing and Status Update: After acquiring real-time coal and rock data, the system first performs data preprocessing, such as noise filtering and outlier handling. Then, the processed data is input into the Transformer model trained in Step 104 to update the coal mining status, ensuring that the model's predictions reflect the latest coal and rock conditions.

[0153] Step 602, Dynamic Prediction using the Transformer Model: After data processing and status updates, the system invokes the Transformer model for real-time prediction. Unlike the Transformer model in the VR navigation pre-planning stage, which focuses on global path planning, this Transformer model is dynamically updated and makes short-term predictions based on real-time data from the actual operation process (such as coal and rock conditions, equipment operating parameters, etc.). By continuously inputting real-time sensor data and prediction results into the Transformer model, the system can make detailed predictions about future coal and rock changes. Especially during operation, the Transformer model predicts the changing trends of coal and rock structure, potential obstacles, and the working status of the coal mining machine. Through continuous iterative prediction, the system can gradually optimize the path and cutting strategy.

[0154] Step 603, Cutting Strategy Optimization: Based on the dynamic prediction results provided by the Transformer model, the system optimizes the cutting strategy of the coal mining machine by combining real-time data and prediction information. The main goal of optimization is to adjust the cutting depth, cutting path, and travel speed of the coal mining machine according to the changing trends of the coal and rock structure, ensuring that the equipment always remains in optimal working condition. The system analyzes key factors such as the hardness and thickness of the coal seam, as well as the coal-rock boundary, to determine the optimal cutting parameters, thereby improving coal mining efficiency and reducing equipment wear.

[0155] Step 604, Dynamic Adjustment and Feedback: After optimizing the cutting strategy, the system dynamically adjusts the operating parameters of the coal mining machine based on real-time feedback. By continuously monitoring equipment operation and coal / rock changes through sensor data, the system adjusts the cutting depth, path, and speed in real time to ensure the coal mining machine operates under optimal conditions. The feedback loop is used to analyze the difference between the actual results and the optimized strategy, enabling rapid adjustments.

Claims

1. A closed-loop cut-off planning method driven by the fusion of VR navigation and coal and rock identification online planning, characterized in that, Includes the following modules: (1) VR navigation pre-planning module; First, historical coal mine data is acquired and preprocessed to build a Transformer model. Real-time data or new input data are then input into the Transformer model. The Transformer model extrapolates based on these inputs to predict future coal and rock conditions and changes in the morphology of the coal seam roof and floor. The VR navigation pre-planning module provides the parameter construction and pose calculation module with the future state of coal and rock and the morphological changes of the top and bottom plates of the coal seam. (2) Parameter construction and pose calculation module; The morphological and physical property models of the coal seam were constructed using the Unity 3D platform, and the pose and cutting trajectory of the coal mining machine were calculated. Based on the morphological and physical property parameters of the coal seam, the pose of the coal mining machine was calculated by solving the motion changes of the roof and floor, and the real-time pose of the coal mining machine was obtained. Then, combined with the pose data of the coal mining machine, the cutting trajectory of the coal mining machine was further calculated. The parameter construction and pose calculation module provides pose parameters and cutting trajectory parameters for the drum adjustment and dynamic speed regulation module, and also provides physical property parameters for the coal and rock identification module through virtual sensors; (3) Coal and rock identification module; By combining ultrasonic sensors and laser-induced breakdown spectroscopy (LIBS) technology, coal seams and rock strata are identified in real time to obtain coal and rock characteristic parameters. After coal and rock identification is completed, a fine model of the coal seam roof is constructed in Unity 3D to realize the reconstruction of the coal seam roof. Based on the reconstructed roof, the virtual coal mining machine is cut and simulated in Unity 3D to correct the attitude of the coal mining machine. (4) Drum adjustment and dynamic speed regulation module; The motion trajectory of the roller is optimized by using the coal and rock feature parameters provided by the coal and rock identification module and the pose parameters provided by the pose calculation module. (5) Path execution and feedback module; Based on the prediction and optimization results jointly generated by the VR navigation pre-planning module, parameter construction and pose calculation module, coal and rock identification module and drum adjustment and dynamic speed regulation module, the actual operation path of the coal mining machine is executed, and the operation data and operation feedback information of the coal mining machine are transmitted back to the system. (6) Online prediction and decision-making module; Based on the coal and rock data obtained from the coal and rock identification module, the coal and rock status is updated in real time through data processing and the predictive capabilities of the Transformer model, dynamic prediction is made, and the results are fed back into the working path planning of the coal mining machine. In addition, the parameters of the coal mining machine are dynamically adjusted according to the real-time coal and rock data to optimize the mining strategy.

2. The method according to claim 1, characterized in that: The steps to build a VR navigation pre-planning module include: Step 101: Historical Data Acquisition and Preprocessing: First, acquire historical coal mine data, record the data according to time series, then perform data preprocessing, and divide the data into training set, validation set and test set according to time order; Step 102, Data Segmentation and Feature Extraction: After data preprocessing, the historical coal mine data is divided into segments according to the mining location intervals. For each segment, feature extraction is performed to construct a feature matrix, and the feature matrix is ​​arranged in chronological order. Step 103, Transformer Model Construction: The Transformer model is constructed using the deep learning framework TensorFlow; Step 104, Model Training and Validation: In the Transformer model, load the training set and validation set to train the model. After training, perform a comprehensive evaluation on the test set. The model that has completed training and validation is used for coal and rock conditions and coal seam roof and floor prediction. Step 105, Model Validation and Feedback: Apply the model to field operation data independent of the training process, evaluate whether the model runs stably in actual operation, analyze the robustness of the model under extreme conditions, and feed the deviation back to the system to improve the model's algorithm or add new input data.

3. The method according to claim 1 or 2, characterized in that: The steps for constructing the parameter construction and pose calculation module include: Step 201, Morphological Parameter Construction: Using the Unity 3D platform, simulate the actual geometric structure of the coal mining machine, and digitally reconstruct the morphology of the top and bottom plates of the coal seam based on the changes in the morphology of the top and bottom plates of the coal seam. Step 202, Material Property Parameter Construction: Obtain historical coal mine data including stress, hardness, density, and chemical element composition. Based on this historical data, establish a material property parameter library in Unity 3D. Integrate the material property parameters with the geometric model constructed in Step 201, and add C# scripts to the coal mining machine to enable its operation and control. Use Unity3D's Transform component to dynamically change the X, Y, and Z coordinates of the coal mining machine model, as well as the pitch, yaw, and roll angles, to precisely control the virtual coal mining machine's position and orientation in three-dimensional space, thereby simulating its actual operation in the coal mine working environment. Step 203: Coal mining machine pose calculation; Step 204: Calculate the cutting trajectory.

4. The method according to claim 3, characterized in that: In step 202, during the construction process, virtual sensors are introduced to simulate the function of actual sensors, collect data and feed it back to the physical property parameter model to acquire and monitor the state of coal seams and rock strata in real time. These virtual sensors automatically update data according to changes in physical property parameters and provide real-time feedback.

5. The method according to claim 4, characterized in that: Step 203, the steps for calculating the coal mining machine's pose include: To construct a virtual scene of the coal mining face, it is first necessary to establish an absolute coordinate system with the initial position of the coal mining machine as the origin. , where X n The axis points in the direction of the coal mining machine's movement, Y n Vertically upward, Z n The shaft points in the direction of the working face advance; the pose calculation of the coal mining machine is to calculate its position coordinates and travel angle in the virtual workspace by parsing the shaft encoder data, so as to accurately describe the dynamic trajectory of the coal mining machine. According to the coal mining machine at the i The position and the first i The position information of the coal mining machine is deduced from the encoder data at +1 position. The specific position coordinates are calculated using formula (2-1): （2-1） Among them, X i With Y i These represent the number of coal mining machines. i The x and y coordinates at each position s i For the first coal mining machine i The axis encoder reading at each position θ i For the first coal mining machine i Pitch angle at each position; To accurately describe the rotational attitude of the coal mining machine, quaternions are used to represent its attitude in space. Let the coal mining machine be at position... i The attitude quaternion is q i Its form is: ； Where w is the real part of the quaternion, and x, y, and z are the imaginary components used to represent the current rotation state of the coal mining machine; Coal mining machine in position i arrive i The rotation increment between +1 and Δq represents the pitch angle increment. Increment of yaw angle and the increment of roll angle Convert them to their corresponding quaternion increments: ； in: ； The increment is applied to the original pose using the quaternion update formula to obtain the next pose. i +1 posture quaternion q i+1 : ； The position information and attitude parameters obtained through pose calculation provide an initial pose reference for the roller adjustment and dynamic speed regulation module, and also provide real-time pose data for the path execution and feedback module.

6. The method according to claim 5, characterized in that: Step 204, the steps for calculating the cutting trajectory include: Based on the mining height information and the structure of the coal mining machine, the rocker arm inclination angle is calculated as shown in equation (2-2). （2-2） in and The coal mining machine was in the first i The cutting height of the front and rear rollers at each position; h is the distance from the rocker arm pin to the bottom plate of the scraper conveyor; D is the roller diameter; L1 is the rocker arm length; Then, based on the corrected trajectory of the coal mining machine, the cutting trajectory is calculated using the calculated rocker arm inclination angle. The coordinates of the upper drum cutting point and the lower drum cutting point are calculated according to equations (2-3) and (2-4), respectively: （2-3） （2-4） Where L is the distance from the front rocker arm connecting pin to the rear rocker arm connecting pin. θ i For the coal mining machine at the first i Pitch angle at each position.

7. The method according to claim 6, characterized in that: The steps to construct a coal and rock identification module include: Step 301, Equipment Initialization and Calibration: Before starting the coal and rock identification module, the ultrasonic sensor and LIBS sensor are initialized and calibrated. Step 302, Ultrasonic sensor detection: The ultrasonic sensor detects the coal seam and rock strata. The ultrasonic sensor determines the physical properties by emitting high-frequency sound waves and receiving the echo signals reflected back from the coal and rock. Step 303, LIBS Detection: After the ultrasonic sensor completes the preliminary determination of coal and rock density, the system activates the LIBS sensor to perform a more accurate chemical composition analysis of the coal and rock. Step 304, Coal Seam Roof Reconstruction: After completing coal and rock identification, the system will reconstruct a detailed model of the coal seam roof in Unity 3D using voxelization and Marching Cubes algorithms; Step 305, Coal Mining Machine Attitude Correction: Based on the reconstructed roof obtained in Step 304, the virtual coal mining machine is subjected to cutting simulation in Unity 3D. The front roller of the virtual coal mining machine is attached to the coal seam roof, and a new roller cutting control strategy is obtained, which can correct the attitude of the coal mining machine.

8. The method according to claim 7, characterized in that: The steps for constructing the drum adjustment and dynamic speed control module include: Step 401, Coal and Rock Analysis and Preliminary Judgment: The system receives coal and rock feature parameters from the coal and rock identification module and pose parameters from the parameter construction and pose calculation module via the MQTT protocol. First, the coal and rock features are analyzed, and the hardness and density of the current cutting area are calculated as a preliminary judgment basis for the coal and rock structure. At the same time, the system judges the shape of the top and bottom plates of the coal seam based on the pose data, identifies whether there is a height difference or tilt change, and preliminarily determines whether the height and angle of the roller need to be adjusted so as to further optimize the roller movement trajectory in subsequent steps. Step 402, Drum Height Adjustment: After the coal and rock analysis and preliminary judgment are completed, the system calculates the ideal cutting height based on the height difference between the top and bottom plates of the coal seam, the hardness and density of the coal and rock, and uses the ideal cutting height as the target height for drum adjustment; the drum height is adjusted by changing the extension of the hydraulic cylinder of the coal mining machine to adapt to the needs of different cutting layers and coal and rock characteristics. Step 403, Coal mining machine speed adjustment: The system uses the MQTT protocol to receive sensor data and obtain real-time drum load and speed information, and dynamically adjusts the coal mining machine's travel speed through analysis; Step 404, Feedback and Optimization: The operation data of the coal mining machine is fed back in real time to adapt to the changing characteristics of the coal and rock strata; the optimization module compares the actual state with the target state and further optimizes the speed and cutting depth through reinforcement learning algorithms.

9. The method according to claim 8, characterized in that: The steps for building the path execution and feedback module include: Step 501, Path Execution: Based on the optimized cutting strategy and travel path, the system sends instructions to the coal mining machine, which then begins actual operation according to the instructions. During the operation, the system monitors the key data of the coal mining machine in real time through sensors. Step 502, Feedback and Data Analysis: During the operation of the coal mining machine, the system will continuously collect and provide real-time feedback of key data from the sensors. The system will compare this feedback data with the optimized instructions, detect any deviations or anomalies in the operation, and update the operation status immediately. Step 503, Memory Cutting and Path Reproduction: The system establishes a "demonstration cutter" as a benchmark during the demonstration operation. When actually performing memory cutting, the system calls the data recorded by the demonstration cutter and compares the parameters of the current operation with those of the demonstration cutter in real time to minimize the deviation and reproduce the motion trajectory and operation mode of the demonstration cutter. The reference data is transmitted to the actuator through the MQTT protocol. The system can dynamically adjust the height and angle of the roller and the propulsion speed to ensure consistency with the demonstration cutter.

10. The method according to claim 9, characterized in that: The steps to build an online prediction and decision-making module include: Step 601, Data Processing and Status Update: After the system acquires real-time coal and rock data, it first performs data preprocessing, including noise filtering and outlier handling; the processed data is then input into the trained Transformer model to update the coal mining status and ensure that the model predictions can reflect the latest coal and rock conditions. Step 602, Dynamic prediction using Transformer model: After data processing and status update, the system calls the Transformer model for real-time prediction. By continuously inputting real-time sensor data and prediction results into the Transformer model, the system makes detailed predictions of future coal and rock changes. Through continuous iterative prediction, the system can gradually optimize the path and cutting strategy. Step 603, Cutting Strategy Optimization: Based on the dynamic prediction results provided by the Transformer model, the cutting strategy of the coal mining machine is optimized by combining real-time data and prediction information. According to the changing trend of coal and rock structure, the cutting depth, cutting path and travel speed of the coal mining machine are adjusted to ensure that the equipment always maintains the best working state. By analyzing key factors including the hardness and thickness of the coal seam and the coal-rock boundary, the optimal cutting parameters are determined. Step 604, Dynamic Adjustment and Feedback: After the cutting strategy is optimized, the system dynamically adjusts the operating parameters of the coal mining machine based on real-time feedback. The system continuously monitors the operation of the equipment and changes in coal and rock through sensor data. The system adjusts the cutting depth, path and speed in real time to ensure that the coal mining machine operates in the best condition. The feedback loop is used to analyze the difference between the actual effect and the optimized strategy and make adjustments quickly.

Images