Sensing-control integrated twin apparatus and method for fully-mechanized coal mining equipment group

Abstract

Abstract The present disclosure relates to the field of a digital twin technology for intelligent coalmining, and in particular, to a sensing-control integrated twin apparatus and method for a fully-mechanized coal mining equipment group comprising an electro-hydraulic controller, a hydraulic support, a sensing module and a control unit.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of a digital twin technology for intelligent coal mining, and in particular, to a sensing-control integrated twin apparatus and method for a fully-mechanized coal mining equipment group. BACKGROUND

[0002] As a core support device of a fully-mechanized coal mining face, a hydraulic support group is responsible for providing roof support and advancing the fully-mechanized coal mining face. Intelligent operation of the hydraulic support group is of great significance for safe, continuous, efficient, and green mining of the fully-mechanized coal mining face.

[0003] However, there are a large number of hydraulic support groups, which are characterized by a compact arrangement, frequent operations, a high failure rate, and a complex spatial pose relationship and collaborative control objective. This results in slow progress in intelligent research and application of the hydraulic support group. Consequently, most coal mines remain at an automation or semi-automation stage. Improving intelligent control over the hydraulic support group has become a key issue that urgently needs to be addressed in intelligent construction of the fully-mechanized coal mining face.

[0004] The application of the digital twin theory and the virtual reality technology in the field of fully-mechanized coal mining equipment is expected to provide new approaches for intelligent monitoring and operation of underground fully-mechanized coal mining equipment in a coal mine, and positively affects underground operations in the coal mine by improving production efficiency, reducing production costs, and reducing accident risks.

[0005] In the prior art, some researches have been carried out to adopt the digital twin theory and the virtual reality technology in the fully-mechanized coal mining equipment, but there are still following shortcomings:

[0006] 1) Due to a complex underground environment, a sensing information transmission path and a physical instruction transmission path are unstable. When there is only the sensing information transmission path or the physical instruction transmission path, it is impossible to guarantee that a movement process of a digital twin model will not be affected, and there is a significant difference between a movement of the digital twin model and an actual working condition.

[0007] 2) A virtual monitoring model is driven by data and can directly reach a target pose position 2025213699   11 Aug 2025 with a rapid response, but a virtual control model is driven by force and needs to take a long time to reach a target pose. This results in asynchronous movements between the two virtual models.

[0008] 3) In a multi-sensor fusion system, there is an urgent need to avoid redundant and conflicting sensing data.

[0009] 4) When sensing information and a physical instruction are input simultaneously, problems arise in conflict resolution and redundancy control, and the system cannot perform self-adaptive regulation to adapt to a current optimal working condition. SUMMARY

[0010] An objective of the present disclosure is to provide a sensing-control integrated twin apparatus and method for a fully-mechanized coal mining equipment group, such that a virtual control model and a virtual monitoring model can be driven by a physical entity to perform a high-fidelity movement, so as to virtually reconstruct a workspace in virtual space.

[0011] To achieve the above objective, the present disclosure provides following technical solutions.

[0012] According to a first aspect, the present disclosure provides a sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group, including an electro-hydraulic controller, a hydraulic support, a sensing module, and a control unit, where the control unit is embedded with a virtual monitoring model, a virtual control model, and a dual-drive agent model; and

[0013] when the virtual monitoring model receives only sensing information, the sensing module includes a plurality of pose sensors that are configured to collect pose data at different positions on the hydraulic support after the electro-hydraulic controller controls the hydraulic support to change a pose; the virtual monitoring model is configured to: determine a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, predict an optimal value of a driving variable based on the weight, drive a virtual support in the virtual monitoring model based on the optimal value of the driving variable, perform full pose calculation based on the optimal value of the driving variable, obtain a pose state of a virtual support whose appearance is consistent with a current appearance of the hydraulic support, and send the pose state to the virtual control model as a destination state; and the virtual control model is configured to determine a difference between the destination state of the virtual support and a current state of the virtual support, determine driving force based on the difference, and apply the driving force to drive a virtual support in the virtual control model to move; where the virtual support is a digital twin model of the hydraulic support;

[0014] when the virtual monitoring model receives only a physical instruction, the sensing module 2025213699   11 Aug 2025 includes a plurality of pressure sensors that are configured to monitor pressure at different positions on the hydraulic support; the virtual control model is configured to: receive a control instruction sent from the electro-hydraulic controller, predict optimal virtual force by using an optimal force agent model based on the pressure at the different positions after receiving the control instruction, drive, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously, and send a state of the virtual support after the synchronous movement to the virtual monitoring model; and the virtual monitoring model is configured to perform full pose calculation on the received state, and control a pose state of a virtual support in the virtual monitoring model based on data obtained after the full pose calculation; and

[0015] when the virtual monitoring model receives sensing information and a physical instruction at the same time, the sensing module includes a plurality of pose sensors and a plurality of pressure sensors; the dual-drive agent model is configured to: fuse a control instruction and pose data by using an adaptive control strategy, obtain fused pose data, determine a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the fused pose data, predict an optimal value of a driving variable based on the weight, predict a velocity parameter for a movement of an oil cylinder of the hydraulic support by using a velocity parameter training model, and interpolate the optimal value of the driving variable based on a predicted velocity parameter, such that a virtual support in the virtual monitoring model moves based on a velocity of the hydraulic support; and the dual-drive agent model is also configured to predict optimal virtual force by using an optimal force agent model based on pressure at different positions after the virtual control model receives the control instruction, and drive, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously.

[0016] According to a second aspect, the present disclosure provides a sensing-control integrated twin method for a fully-mechanized coal mining equipment group, including:

[0017] when detecting that only sensing information is input, obtaining, by a virtual monitoring module, pose data at different positions on a hydraulic support, determining a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, predicting an optimal value of a driving variable based on the weight, driving a virtual support in the virtual monitoring model based on the optimal value of the driving variable, performing full pose calculation based on the optimal value of the driving variable, obtaining a pose state of a virtual support whose appearance is consistent with a current appearance of the hydraulic support, and sending the pose state to a virtual control model as a destination state; and determining, by the virtual control module, a difference 2025213699   11 Aug 2025 between the destination state of the virtual support and a current state of the virtual support, determining driving force based on the difference, and applying the determined driving force to drive a virtual support in the virtual control model to move; where the virtual support is a digital twin model of the hydraulic support;

[0018] when detecting that only a physical instruction is input, receiving, by a virtual control model, a control instruction sent from an electro-hydraulic controller, predicting optimal virtual force by using an optimal force agent model based on pressure at different positions after receiving the control instruction, driving, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and a hydraulic support to move synchronously, and sending a state of the virtual support after the synchronous movement to a virtual monitoring model; and performing, by the virtual monitoring model, full pose calculation on the received state, and controlling a pose state of a virtual support in the virtual monitoring model based on data obtained after the full pose calculation; and

[0019] when detecting that sensing information and a physical instruction are input simultaneously, fusing, by a dual-drive agent model, a control instruction and pose data by using an adaptive control strategy, obtaining fused pose data, determining a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the fused pose data, predicting an optimal value of a driving variable based on the weight, predicting a velocity parameter for a movement of an oil cylinder of a hydraulic support by using a velocity parameter training model, and interpolating the optimal value of the driving variable based on a predicted velocity parameter, such that a virtual support in a virtual monitoring model moves based on a velocity of the hydraulic support; and also predicting, by the dual-drive agent model, optimal virtual force by using an optimal force agent model based on pressure at different positions after a virtual control model receives the control instruction, and driving, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously.

[0020] According to specific embodiments provided in the present disclosure, the present disclosure achieves following technical effects:

[0021] The present disclosure provides a sensing-control integrated twin apparatus and method for a fully-mechanized coal mining equipment group. Three working modes are provided: a sensing information-driven mode in which a virtual monitoring model receives only sensing information, a physical instruction-driven mode in which the virtual monitoring model receives only a physical instruction, and a dual-drive mode in which the virtual monitoring model receives the sensing information and the physical instruction at the same time, thereby ensuring that a movement process of a digital twin model is not affected and that the digital twin model moves based on an actual 2025213699   11 Aug 2025 working condition. In these three modes, information exchange is performed between the virtual monitoring model and a virtual control model, which enables the two virtual models to move synchronously. A weighted adaptive hybrid algorithm of a sensor is used to determine a weight of a corresponding driving variable of each piece of pose data, and an optimal value of a driving variable of the virtual monitoring model is predicted, thereby avoiding redundant and conflicting sensing data in a multi-sensor fusion system. When the sensing information and the physical instruction are input simultaneously, an adaptive control strategy is adopted to fuse a control instruction and pose data, which avoids conflicting and redundant information. In this way, adaptive regulation can be performed to adapt to a current optimal working condition. The present disclosure can virtually reconstruct a workspace in virtual space by using a physical entity to drive the virtual control model and the virtual monitoring model to perform a high-fidelity movement. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To describe the technical solutions in the embodiments of the present disclosure more clearly, the accompanying drawings required for describing the embodiments are briefly described below. Apparently, the accompanying drawings in the following description show merely some embodiments of the present disclosure, and those of ordinary skill in the art may still derive other accompanying drawings from these accompanying drawings without creative efforts.

[0023] FIG. 1 is a schematic diagram of a technical route of a sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to an embodiment of the present disclosure;

[0024] FIG. 2 is a schematic diagram of a technical route of a sensing information-driven mode according to an embodiment of the present disclosure;

[0025] FIG. 3 is a schematic diagram of a technical route of a physical instruction-driven mode according to an embodiment of the present disclosure;

[0026] FIG. 4 is a schematic diagram of a technical route of a dual-drive mode according to an embodiment of the present disclosure; and

[0027] FIG. 5 is a schematic structural diagram of a computer device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] The technical solutions in the embodiments of the present disclosure are clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are only some rather than all of the embodiments of the present disclosure. All other embodiments obtained by a person of ordinary 2025213699   11 Aug 2025 skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0029] To make the above objectives, features, and advantages of the present disclosure more obvious and easy to understand, the present disclosure will be further described in detail with reference to the accompanying drawings and specific implementations.

[0030] In the prior art, the Chinese patent application with a publication number CN115685240A discloses a hybrid data and knowledge-driven virtual reconstruction method for a relative pose of a hydraulic support. A digital twin model system is constructed for a physical prototype of the hydraulic support, and sensor arrangement in a physical monitoring system is guided through a sensor position planning system. Sensor data in the physical monitoring system is collected through a data collection module and transmitted to a mechanism model for data processing. Processed data is corrected through a parameter preprocessing module, and then corrected data is transmitted to a relative position calculation module as a known quantity for calculation. A calculation result is mapped into a pose deduction system in real time through dynamic linking. A pose correction module iterates a result of the pose deduction system to construct a hybrid knowledge and data-driven reconstruction method. Finally, pose information of the hydraulic support is displayed in a human-computer interaction system, and error evaluation is performed on deduced pose information of the hydraulic support in an accuracy evaluation system. The human-computer interaction system is used to perform state monitoring on all systems.

[0031] The Chinese patent with a publication number CN112945160A discloses a physical-virtual integrated testing platform and method for a relative pose between hydraulic supports. A physical-virtual integration approach is used to simulate working processes and support scenarios of adjacent hydraulic supports in a real coal seam environment, including a physical support testing system, a physical-virtual data exchange system, and a multi-support virtual testing scenario. The multi-support virtual testing scenario simulates movement processes, real-time relative pose states, and coal seam inclinations of underground adjacent supports. Adjacent virtual supports and an actual tested hydraulic support move alternately, so as to display real-time pose images and relative pose data of the adjacent hydraulic supports, and real-time pressure data of the tested hydraulic support. In this way, when relative pose states of the adjacent hydraulic supports are studied, conditions of inclination angles of a coal seam roof and floor and a change in roof pressure can be taken into account, which solves problems of low experimental efficiency, difficulty in real-time observation of a full pose of the hydraulic support, and a huge manpower and material resource consumption in a pose monitoring and determination algorithm and a pose adjustment method for a hydraulic support group.

[0032] The public paper "Digital Twin-Driven Key Technology and System for Virtual Production 2025213699   11 Aug 2025 of Fully-Mechanized Coal Mining Equipment" uses various sensors on fully-mechanized coal mining equipment of a physical system to obtain real-time pose data of the fully-mechanized coaling mining equipment during operation. Based on a physical-virtual bidirectional data exchange technology and a virtual monitoring technology, processed pose data of the fully-mechanized coal mining equipment during the operation is assigned to a corresponding variable of a virtual model to drive virtual fully-mechanized coal mining equipment to move in real time. In a physical fully-mechanized coal mining face, a data collection module monitors a potential change of a control-end controller, converts the potential change into a control signal, and sends the control signal to a database in a message. At a virtual end, data in the database is received and converted into a control signal to control the virtual fully-mechanized coal mining equipment to move.

[0033] The public paper "Research on Digital Twin-Driven Intelligent Pose Regulation Method for Hydraulic Support Group" combines sensing and monitoring data with a pose reconstruction method to obtain a digital description for a pose and an operational state of a support. The description corresponds to a driving interface of a virtual model, thereby achieving sensing data-driven operation of the virtual model and restoring a pose state of a physical support, that is, achieving virtual monitoring of the physical support. Communication between virtual space and a physical object control system is established, enabling write-in of a control instruction. A single hydraulic support group in the virtual space completes decision-making on its movement behavior based on a distributed control strategy, and converts a result into a control instruction to reversely control a corresponding physical entity to move. Through three key steps: virtual monitoring, virtual decision-making, and reverse control, working condition sensing data is converted into a control instruction that can directly guide operation of a physical device. The virtual space serves physical space and provides an analysis and calculation platform for the physical space. With the help of closed-loop linkage between the virtual space and the physical space, intelligent operation of a real device is empowered.

[0034] The prior art cannot drive a virtual control model and a virtual monitoring model to perform a high-fidelity movement to reconstruct a workspace in the virtual space. In order to address problems in the prior art, in an exemplary embodiment, as shown in FIG. 1, a sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group is provided, including an electro-hydraulic controller, a hydraulic support, a sensing module, and a control unit. The control unit is embedded with a virtual monitoring model, a virtual control model, and a dual-drive agent model.

[0035] When the virtual monitoring model receives only sensing information, the sensing module includes a plurality of pose sensors that are configured to collect pose data at different positions on the hydraulic support after the electro-hydraulic controller controls the hydraulic support to change 2025213699   11 Aug 2025 a pose; the virtual monitoring model is configured to: determine a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, predict an optimal value of a driving variable based on the weight, drive a virtual support in the virtual monitoring model based on the optimal value of the driving variable, perform full pose calculation based on the optimal value of the driving variable, obtain a pose state of a virtual support whose appearance is consistent with a current appearance of the hydraulic support, and send the pose state to the virtual control model as a destination state; and the virtual control model is configured to determine a difference between the destination state of the virtual support and a current state of the virtual support, determine driving force based on the difference, and apply the driving force to drive a virtual support in the virtual control model to move. The virtual support is a digital twin model of the hydraulic support.

[0036] When the virtual monitoring model receives only a physical instruction, the sensing module includes a plurality of pressure sensors that are configured to monitor pressure at different positions on the hydraulic support; the virtual control model is configured to: receive a control instruction sent from the electro-hydraulic controller, predict optimal virtual force by using an optimal force agent model based on the pressure at the different positions after receiving the control instruction, drive, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously, and send a state of the virtual support after the synchronous movement to the virtual monitoring model; and the virtual monitoring model is configured to perform full pose calculation on the received state, and control a pose state of a virtual support in the virtual monitoring model based on data obtained after the full pose calculation.

[0037] When the virtual monitoring model receives sensing information and a physical instruction at the same time, the sensing module includes a plurality of pose sensors and a plurality of pressure sensors; the dual-drive agent model is configured to: fuse a control instruction and pose data by using an adaptive control strategy, obtain fused pose data, determine a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the fused pose data, predict an optimal value of a driving variable based on the weight, predict a velocity parameter for a movement of an oil cylinder of the hydraulic support by using a velocity parameter training model, and interpolate the optimal value of the driving variable based on a predicted velocity parameter, such that a virtual support in the virtual monitoring model moves based on a velocity of the hydraulic support; and the dual-drive agent model is also configured to predict optimal virtual force by using an optimal force agent model based on pressure at different positions after the virtual control model receives the control instruction, and drive, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model 2025213699   11 Aug 2025 and the hydraulic support to move synchronously.

[0038] For example, the virtual monitoring model, the virtual control model, and the dual-drive agent model are constructed in simulation software Unity3D.

[0039] As an optional implementation, the pose sensor is an inclination angle sensor. A plurality of inclination angle sensors are respectively disposed at a rear connecting rod, a roof beam, and a shield beam of the hydraulic support. The plurality of inclination angle sensors are configured to collect inclination angle data of the rear connecting rod, the roof beam, and the shield beam.

[0040] As an optional implementation, the sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group further includes an actuator. The electro-hydraulic controller controls the actuator to move according to the control instruction, to change the pose of the hydraulic support.

[0041] As an optional implementation, the sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group further includes a data collection module. The data collection module is embedded in the control unit. The data collection module is configured to collect pose data of the plurality of pose sensors and pressure monitored by the plurality of pressure sensors, correct and filter the collected pose data and pressure, and transmit corrected and filtered pose data and pressure to the virtual monitoring model.

[0042] A sensing and monitoring data variable in physical space and an equipment operation control variable can be integrated into a programmable logic controller (PLC). Data from a data block (DB) in the PLC is read and written, such that an upper computer can remotely obtain sensing data from a hydraulic support group. A real-time communication interface is constructed to import data obtained by a sensor such as the inclination angle sensor into the simulation software Unity3D as a source of subsequent simulation data. Main programming software used is Visual Studio or a third-party library such as pcl. The inclination angle sensor communicates with the upper computer through a serial port, and a C# program is compiled to read sensor data. The sensor data is transmitted to the Unity3D by using the Transmission Control Protocol (TCP). After receiving the sensor data, the virtual monitoring model modifies a rotation angle or a position of a corresponding component of the virtual support to simulate a motion support.

[0043] Data processing of the inclination angle sensor mainly involves data correction and filtering, and a sensor fusion algorithm (such as Kalman filtering or particle filtering) is used to reduce conflicts. The inclination angle sensor obtains data by measuring a change in an inclination angle of a device relative to the Earth's gravity field. Most inclination angle sensors operate based on an accelerometer principle to measure a change in an acceleration direction, which is closely related to the inclination angle of the device. Collected raw data may be affected by noise, and therefore, usually needs to be filtered. A low-pass filter is used to remove high-frequency noise. Optimal 2025213699   11 Aug 2025 estimation is performed on noise-contaminated data through the Kalman filtering. To ensure accuracy of a measurement result of a sensor, the sensor usually needs to be calibrated. An error of the sensor may include a bias error, a nonlinear error, and the like. The calibration usually involves: zero calibration in which an output value of the device in a horizontal state is measured and corrected for a deviation; and temperature compensation in which an impact of a temperature change on inclination angle measurement is adjusted based on a temperature characteristic of the sensor.

[0044] The data collection module communicates with the Unity3D in a plurality of communication modes, depending on its application scenario, environmental condition, data transmission requirement, and real-time performance requirement.

[0045] A wireless communication network Wi-Fi uses a Wi-Fi enabled sensor or microcontroller (such as ESP8266 or ESP32). The microcontroller is connected to a network through the Wi-Fi and runs a web server. The microcontroller can listen to a specified port through the Hyper Text Transfer Protocol (HTTP), receive an HTTP request from the Unity3D, and return a response. The Unity3D uses a Unity3DWebRequest class to send the HTTP request and receive data from the microcontroller. Firmware is compiled to send data to a server via the HTTP, and the Unity3D receives the data through a network library. The wireless communication network Wi-Fi has high flexibility, can be easily deployed, and supports multi-device connection. In a mine, wireless communication can avoid a limitation of a physical connection. In some places where wiring is difficult, the wireless communication can be used to reduce construction difficulty and costs. However, the wireless communication is greatly affected by environmental factors (such as signal interference and an obstacle).

[0046] TCP / Internet Protocol (IP) communication uses a microcontroller with an Ethernet interface (such as Raspberry Pi), and a TCP / IP server is built on a device to receive sensor data. The Unity3D uses a TCPClient class to establish a connection, and processes and visualizes the received data. The TCP / IP communication is suitable for a wired network environment. When there is a stable wired network, the TCP / IP can be used to achieve high-velocity and reliable data transmission. The TCP / IP can ensure integrity and an order of data, making it suitable for an application that has a high reliability requirement for data transmission. However, the TCP / IP communication requires wiring and network infrastructure. In a complex network environment, more configuration and management are required.

[0047] Serial communication connects a serial interface of the data collection module of the hydraulic support to a universal serial bus (USB) interface of a computer. An RS232 serial protocol is used for data transmission, which is suitable for short-range communication and also incorporates a Wi-Fi module for a scenario that requires remote monitoring. In a C# script of the Unity3D, a 2025213699   11 Aug 2025 serialPort class is used to create a serial port object and configure a serial port parameter. It is ensured that serial port settings (a COM port and a baud rate) in the Unity3D are consistent with those of the data collection module. A user interface (UI) component is utilized to visualize a monitored object. The serial communication is suitable for the short-range transmission, typically data transmission within a few meters, and is suitable for a connection to a simple device such as a single-chip microcontroller and a sensor. The serial communication is characterized by simple implementation and low costs. The serial communication has good real-time performance, and is suitable for transmitting a small amount of data. However, the serial communication has a limited transmission distance, and is not suitable for a large-scale or complex network connection.

[0048] A communication mode between the data collection module and the Unity3D needs to be selected by comprehensively considering following factors: an environmental condition such as whether there is interference from a wireless signal, and wiring difficulty; a data amount, including whether there is a need to transmit a large amount of data, and a requirement for real-time performance; and a device type, including compatibility between the sensor and a main control device, and an interface type. By taking into account the above factors, a most suitable communication mode is flexibly selected, or a plurality of communication modes are combined to ensure that the sensor data is smoothly transmitted to the Unity3D. The inclination angle data collected by the inclination angle sensor on the rear connecting rod, the roof beam, and the shield beam is transmitted to the constructed virtual monitoring model in the Unity3D in one or more of the above communication modes, thereby achieving the communication between the collection module and the Unity3D.

[0049] The sensing-control integrated twin apparatus in the present disclosure has three working modes: a sensing information-driven mode, a physical instruction-driven mode, and a dual-drive mode. The sensing information-driven mode is used for a working condition in which a support is provided with the pressure sensor and the pose sensor. Real-time sensing data of the hydraulic support is collected through the data collection module to achieve "sensing-control" integration and appearance consistency when there is no physical instruction information. The physical instruction-driven mode is used for a working condition in which no sensing information is available. A current pose is estimated through the velocity parameter training model and the optimal force agent model, thereby achieving physical-virtual synchronization. The dual-drive mode is used for a working condition in which both a physical instruction transmission path and a sensing information path are intact, and based on the adaptive control strategy, achieves real-time consistent movement and regulation for the virtual monitoring model and the virtual control model.

[0050] Each of the above working modes is described in detail below.

[0051] (1) Sensing information-driven mode 2025213699   11 Aug 2025

[0052] As shown in FIG. 2, in the sensing information-driven mode, the electro-hydraulic controller of the hydraulic support sends an instruction to control the actuator to move correspondingly, thereby changing the pose of the hydraulic support, and a plurality of sensors installed on the hydraulic support collect information in real time. A plurality of sensors of a plurality of hydraulic supports transmit data to the data collection module in real time, and collected data is transmitted to the virtual monitoring model through a serial communication port compiled by the Unity3D. Based on the weighted adaptive hybrid algorithm of the sensor, a weight of a corresponding driving variable of each piece of sensing information is calculated, and an optimal value of a driving variable of the virtual monitoring model is predicted. The full pose calculation is performed in the virtual monitoring model based on the optimal value, and a pose state of a virtual hydraulic support that has a consistent appearance with a current physical support, namely, a destination state Sd. A current state Sc is transmitted to the virtual control model through a Message Queuing Telemetry Transport (MQTT) protocol, and a difference between each pose value of the current state Sc and each pose value of a current state of the virtual control model is obtained. Based on a magnitude of the difference, different levels of force are assigned, and a larger difference leads to greater force. In this way, the virtual control model rapidly responds to the virtual monitoring model, such that both the virtual control model and the virtual monitoring model are synchronized with a physical entity.

[0053] A virtual monitoring agent model includes an MQTT protocol communication module and the weighted adaptive hybrid algorithm of the sensor. The MQTT protocol communication module first needs to download a MQTTnet.DLL file and import the MQTTnet.DLL file into a project in the Unity3D. An MQTT protocol proxy (server) is compiled for a client using a Factory class in an MQTTnet library. A client script is compiled, a client theme and topic are set, and the script is configured into the project in the Unity3D. The project is run in the virtual control model, and a server IP address is input and connected to an MQTT protocol communication server. The project is run in the virtual monitoring model, and an IP address is set and connected to the MQTT protocol communication server. A communication script is compiled in another client, a server IP address and port are set, a theme published by a client of the virtual control model is subscribed to, and a message sent by the virtual control model is received. A theme and a topic are set in the virtual monitoring model, and a message is sent. In this way, mutual communication between the two models is achieved.

[0054] After establishing a state equation and an observation equation of a system through the Kalman filtering, the weighted adaptive hybrid algorithm of the sensor predicts a state at a next time point and its uncertainty (covariance matrix) by using a current state estimate and control input based on a dynamic model of the system. The state estimate and the covariance matrix are updated 2025213699   11 Aug 2025 based on a new observed value obtained from the sensor. A key is to adjust weights of different sensors through an error covariance matrix of the sensor. In multi-sensor fusion, the Kalman filtering achieves dynamic weight adjustment by recursively minimizing an error covariance matrix of each sensor in combination with the dynamic model and the observation model of the system. In this way, in each step, a state can be optimized based on a current predicted value and an actual observed value, thereby achieving adaptive weighting.

[0055] The optimal value of the driving variable is predicted based on respective weight proportions of the inclination angles that are of the rear connecting rod, the roof beam, and the shield beam and calculated by the weighted adaptive hybrid algorithm of the sensor. The optimal value of the driving variable is transmitted to the monitoring model and analyzed by using a four-bar linkage structure of the hydraulic support, to perform the full pose calculation. A calculated pose is used as a destination support state Sd, which is sent to the virtual control model through the MQTT protocol communication module for comparison to obtain a difference, such that the control model makes a corresponding response subsequently.

[0056] Specifically, an execution process of the weighted adaptive hybrid algorithm of the sensor includes a plurality of stages. At an initialization stage, the system sets an initial state for each sensor, including initialization of a state estimate and the error covariance matrix. The state estimate is usually a measurement result of the sensor at a time point, and the error covariance matrix reflects a magnitude of an estimation error of the sensor. In addition, a mutual relationship between the dynamic model of the system and the sensor is set, and weights of all sensors are initialized to be equal. At a sensor data obtaining and real-time observation stage, a plurality of sensors obtain data in real time, including pose information of different parts of the hydraulic support. Because the sensor may have a different timestamp, time synchronization is required to ensure reliability of all sensor data at a same time point. Next, the weighted adaptive hybrid algorithm of the sensor an algorithm dynamically calculates a weight of the sensor based on performance, reliability, and precision of the sensor. The weight is calculated based on the error covariance matrix of the sensor. A smaller error covariance leads to a higher weight, indicating that observation data of the sensor has a significant impact on a final result. The dynamic weight adjustment ensures preferential reliance on a sensor with higher reliability under different conditions.

[0057] At a system state prediction stage, the dynamic model of the system is used to predict the current state, taking into account factors such as an acceleration and external interference. Based on a state estimate and control input at a previous time point, a predicted value and a predicted error covariance matrix are calculated, providing reference for a subsequent fusion stage. At a weighted fusion stage, by using a weighted average algorithm, a Kalman filtering algorithm, or the like, weighted fusion is performed on weights dynamically calculated based on observation data of the 2025213699   11 Aug 2025 plurality of sensors, thereby generating a more accurate state estimate. The weighted fusion is intended to reduce impacts of observation noise and a system error. At a state estimate update stage, a state estimate of the system is updated based on new observed data and predicted state. By combining fused data, a predicted state value, and the error covariance matrix, the state estimate is optimized based on a Kalman gain, such that an optimized state estimate is closer to an actual situation, thereby reducing a bias caused by inaccurate sensor data. As new data is input, the system dynamically adjusts the weight of the sensor. For each data update, the algorithm recalculates the weight based on an error and reliability of the sensor, to ensure sustained high precision and reliability of a fusion result.

[0058] The entire process iterates continuously, and when new sensor data arrives, the weight of the sensor is continuously adjusted in real time. Through this continuous optimization process, the system can track a state in real time, optimize the state estimate, ensure that an estimation result is consistent with the sensor data, and perform adaptive adjustment based on an environmental change to cope with an external environment, a sensor failure, or a precision change.

[0059] Equations involved in the weighted adaptive hybrid algorithm of the sensor are as follows:

[0060] A. System definition and core architecture

[0061] 1. State variables are shown in Table 1. Table 1 State variables Variable symbol Physical meaning Unit Obtaining method / Calculation basis e True inclination angle of the rear connecting rod (relative to a base) rad Calculated according to a kinetic equation w Angular velocity of the rear connecting rod rad / s Differential term of the state equation Pcyl Equivalent pressure of the oil cylinder (equivalent pressure for driving the hydraulic cylinder) Pa Calculated by a pressure model of a hydraulic system

[0062] 2). Input variables are shown in Table 2. Table 2 Input variables 2025213699   11 Aug 2025 Variable symbol Physical meaning Unit Obtaining method / Calculation basis Prock Surrounding rock pressure (distribution pressure of a roof on the support) Pa Measured by a distributed pressure sensor array Tvalve Delay of a valve control instruction (a response delay of an electromagnetic valve) s Calibrated through a characteristic experiment of the electromagnetic valve) V Vibration intensity (amplitude of a three-axis vibration acceleration of the support) m / s2 Measured by a three-axis accelerometer

[0063] 3. Observation variables are shown in Table 3. Table 3 Observation variables Variable symbol Physical meaning Unit Sensor type e i Measured value of an inclination angle sensor of the rear connecting rod rad Inclination angle sensor 92 Measured value of an inclination angle sensor of the roof beam rad Inclination angle sensor e3 Measured value of an inclination angle sensor of the shield beam rad Inclination angle sensor Pmeas Measured value of a pressure sensor of the oil cylinder Pa Pressure sensor

[0064] B. Integrated mathematical model

[0065] 1. The state equation (including the surrounding rock pressure, a valve delay, and oil cylinder pressure) is as follows: 2025213699   11 Aug 2025 A Q - a + (aFrock sin Q + a2V 1 '® - J [KpPcyl (t - Tvalve ) - Cd® - mgl sin Q] . ^ Pcyl - ^( U (t - Tvalve) - Pcyi )+ 7V

[0066] Parameter description of the state equation is shown in Table 4. Table 4 Parameter description of the state equation Parameter symbol Physical meaning Calculation / Calibration method Typical value range a i Coefficient for coupling between the surrounding rock pressure and the inclination angle (characterizing an amplification effect of the surrounding rock pressure on an inclination angle change) Calibrated through a rock mechanics experiment 0.05-0.2 N^m / Pa a i Vibration disturbance coefficient (equivalent torque coefficient per an inclination angle change induced by vibration) Fitted through vibration spectrum analysis 0.01-0.1 rad / (m / s2) J Moment of inertia of the rear connecting rod Calculated through a computer-aided design (CAD) model of a mechanical structure 50-200 kg^ Kp Coefficient for conversion between the oil cylinder pressure and a torque (efficiency of a hydraulic driving torque) Calibrated based on a characteristic of the hydraulic system 0.5-2.0 m2 2025213699   11 Aug 2025 Cd Damping coefficient (including friction and hydraulic damping) Identified through a dynamic response experiment 10-50 N^m^s / rad p Response efficient of the oil cylinder pressure (reflecting a dynamic characteristic of a hydraulic pipeline) Obtained through frequency response testing on the hydraulic pipeline 0.1-1.0 1 / s

[0067] Dynamic coupling of the surrounding rock pressure: Nonlinear interference of the surrounding rock pressure Prock on an inclination angle of the support is incorporated into the state equation based on the a 1. A value of the a 1 is calibrated through the rock mechanics experiment.

[0068] Explicit modeling of the valve delay: The oil cylinder pressure Pcyl (t -tvalve) directly characterizes an impact of the transmission delay tvalve of the electromagnetic valve instruction.

[0069] Vibration disturbance compensation: The vibration intensity V is used to respectively correct dynamic inclination angle and oil pressure errors based on the a2 (calibrated through frequency spectrum analysis) and the y (calibrated through a transmission experiment of the hydraulic pipeline).

[0070] 2. The observation equation is as follows: r 0 i 1  00 r    n ' V1 ' 1 0 1 o2 2 cos 6 sin 6 0 V2 = ty + 0 0      /   0 p, V3 P meas _ 0      0    V _ L cyl J _ vP _

[0071] Parameter description of the observation equation is shown in Table 5. Table 5 Parameter description of the observation equation Parameter symbol Physical meaning Calculation / Calibration method Typical value range 0 Articulation angle of the shield beam (geometrically constrained angle) Design parameter of the mechanical structure Fixed value (for example, 30°) n Pressure-displacement conversion coefficient (linear Calibrated based on a characteristic of the oil 0.01-0.1 m / Pa 2025213699   11 Aug 2025 Parameter symbol Physical meaning Calculation / Calibration method Typical value range relationship between oil pressure and a displacement of the roof beam) cylinder

[0072] Geometric constraint matrix: a mechanical kinematic constraint relationship constructed based on the articulation angle ^ of the shield beam (the ^ is a structure design parameter of the support), which eliminates a measurement coupling error between a sensor of the roof beam and a sensor of the rear connecting rod.

[0073] Observation channel of the oil cylinder pressure: The pressure sensor data Pmeas is integrated into an observation system based on the n (the pressure-displacement conversion coefficient is calibrated by a size of the oil cylinder) to achieve cross validation of a state of the hydraulic system and a mechanical pose.

[0074] C. Noise covariance model

[0075] 1. A process noise covariance is as follows: 00 0 0  Op O - f (Prock, V).

[0076] Parameter description of the process noise covariance is shown in Table 6. Table 6 Parameter description of the process noise covariance Parameter symbol Physical meaning Dynamic relationship equation O 2 Inclination process noise variance Constant or related to the vibration intensity °. 2 Noise variance for angular velocity estimation Constant or related to the vibration intensity OP2 Noise variance for oil pressure estimation f (Prock, V) - k1 Prock + k2V

[0077] A process noise variance OP2  is dynamically adjusted based on the surrounding rock P pressure rock and the vibration intensity V, and robustness of a filter is automatically enhanced 2025213699   11 Aug 2025 when there is pressure from the roof.

[0078] 2. A measurement noise covariance is as follows: ’^1-1 0 0 0 R = 0 V 0 0 k 0 0 23’1 0 _ 0 0 0 A

[0079] Parameter description of the measurement noise covariance is shown in Table 7. Table 7 Parameter description of the measurement noise covariance Parameter symbol Physical meaning Weight coefficient formula 2i Sensor measurement weight (i=1, 2, 3, corresponding to an inclination angle 1        A                1 . sensor, and p in p corresponds to pressure) A =-----------1----------- €i+IProck IVi +8i +KITvalve I €i Basic sensor reliability coefficient (factory calibration error) Determined by sensor precision (for example, 0.001 rad2) Vi Coefficient of interference of the surrounding rock pressure to the sensor Calibrated through an experiment (for example, 0.005 Pa-1) 8i Residual feedback term (a bias between a predicted value and an actually measured value) 8t =| -   | -1| (a predicted residual of incl K Weight coefficient for valve delay compensation Determined through stability analysis ofa control system

[0080] The diagonal element 2 / 1 is dynamically updated through a weighting algorithm to achieve real-time evaluation of sensor reliability.

[0081] D. Core algorithm module

[0082] 1. Weighted adaptive hybrid algorithm

[0083] A dynamic weight is calculated as follows: _____________________1_____________________ €i + I Prock I Vi + 8 / + K | T valve | . 2025213699   11 Aug 2025 ||w

[0084] In the above formula, rock i represents an interference compensation term of ...            .                    a ,        ., ,.,. .h surrounding rock pressure of an ith pose sensor; i represents pose data of the ith pose sensor; ^. ak1 k 1 represents estimated pose data of the ith pose sensor; and K |7valve | presents a compensation term of the valve delay.

[0085] 2. Time lag compensation observer Pc* (t) = P (t - 7„,„) + K, J^^_ (u (f ) - Py, (f))d? .

[0086] A pressure accumu,ation error during the va,ve de,ay is compensated for through an integrator. Parameter description of the time ,ag compensation observer is shown in Tab,e 8. Table 8 Parameter description of the time lag compensation observer Variable symbol Physical meaning Mathematical expression u(t) Input control instruction (expected oil pressure) Output from a controller e Integral variable (dummy variable for a time integral) Integral interval ranging from t—Tvalve to t

[0087] Embodiment 1: Core parameter ca,ibration method

[0088] (1) Calibrating the coefficient a1 for the coupling between the surrounding rock pressure and the inc,ination ang,e

[0089] Experimental equipment: an MTS 815 rock mechanics testing machine (loading precision: ± 0.5 MPa), an inclination angle sensor array (resolution: 0.01°), and a hydraulic support prototype.

[0090] Experimental steps are as follows:

[0091] 1. Install the support on the testing machine, with the roof beam attached to a simulated surrounding rock contact surface.

[0092] 2. Apply gradient surrounding rock pressure (5 MPa ^ 20 MPa , with a step size of 5 MPa ), and maintain the vibration intensity V at 0.

[0093] 3. Record data reflecting that the inclination angle a of the rear connecting rod varies P with the rock under a steady state. ^e a1 =---------- .       .-   .      ....        ... P , • sin a

[0094] 4. Perform formula fitting by using a least squares method:        rock , where a typical value is as follows: a 1=0.12 N^m / Pa. 2025213699   11 Aug 2025

[0095] Data processing: Remove vibration disturbance data and take an average value of three experiments.

[0096] (2) Calibrating the vibration disturbance coefficient a2

[0097] Experimental equipment: a three-axis vibration table (frequency range: 0 Hz to 100 Hz), a laser displacement sensor, and an accelerometer.

[0098] Experimental steps are as follows:

[0099] 1. Fix the support on the vibration table, and set the vibration intensity V to 1 m / s2 to 5 m / s2 (simulating an underground working condition).

[0100] 2. Turn off the hydraulic system (Pyl =0), and record a curve reflecting that the 0 varies with the V. ^e / At ^2 =   rz

[0101] 3. Calculate an inclination angle offset rate caused by the vibration:         V , where a typical value is as follows: a2=0.03 rad / (m / s2).

[0102] Innovative point: Actual underground random vibration is simulated by using a white noise vibration spectrum.

[0103] (3) Calibrating the response efficient P of the oil cylinder pressure

[0104] Reference standard: Section 5.2 "Step Response Method" in "Dynamic Characteristic Testing of Hydraulic System".

[0105] Brief description: Apply a step control signal u(t) to the oil cylinder, record an exponential P (t) decay curve of the cyl through the pressure sensor, and perform calculation according to a V - — tt following formula:      63% , where 63% represents time required for the pressure to reach 63% of a steady-state value, and a typical value of the P is 0.81 / s.

[0106]

[0107]

[0108]

[0109]

[0110] (4) Calibrating a vibration-oil pressure transfer coefficient y Experimental steps are as follows: 1. Apply fixed-frequency vibration (f=10 Hz, V=2 m / s2) on the vibration table. AP 2. Measure a pressure fluctuation amplitude cyl of the oil cylinder. AP , cyl / = -^ 3. Calculate the transfer coefficient according to a following formula:       V , where a typical value is as follows: ^ °’05Pa / (m / s ).

[0111] Data validation: Compare values of the y at different frequencies (5 Hz, 20 Hz, and 50 Hz) and take an average value.

[0112] Embodiment 2: Dynamic parameter calibration 2025213699   11 Aug 2025

[0113] 1. Calibrating a dynamic relationship of the process noise variance &P2 P

[0114] Experimental conditions: The surrounding rock pressure ( rock =5 Mpa to 15 MPa) and the vibration (V=1 m / s to 4 m / s2) are synchronously applied, and there are a total of 25 working conditions.

[0115] Data processing: P| mes

[0116] .     .            AP ,= I P , 1. Record an oil pressure estimation error cyl cyl

[0117] 2. Establish a model through multiple linear regression: ^p 2 = 0.021 Pmck + 0.15k

[0118] 3. Verify that it is considered valid when R2>0.85.

[0119] (2) Rule for calculating the sensor measurement weight coefficient Ai

[0120] Calibrating the residual feedback oi: In a dynamic movement process of the support, take statistics on a residual of each sensor: N 0, = A Z M k) - “' k I k -1)1 N k=1

[0121] In the above formula, N=1000 sampling points, and it is considered reliable when oi <0.1 rad.

[0122] Calibrating the coefficient k for the valve delay compensation: Perform measurement k =-----1----- 1 + T     / T„          T through a step response experiment, and perform calculation:          value 0 , where 0 =0.1 s, indicating a reference relay, and k=0.2 to 0.8.

[0123] Embodiment 3: Explanation of reference-type parameters

[0124] (1) Calculating the moment of inertia J

[0125] The moment of inertia J of the rear connecting rod is directly obtained by a quality attribute analysis module of three-dimensional (3D) modeling software (SolidWorks 2023), namely J = 128.5kg • m2 C

[0126] (2) Identifying the damping coefficient d

[0127] An envelope of an angular velocity ^(t) is measured through a free attenuation experiment C = 32.4N •m •s / rad of hydraulic support, and d                       is fitted.

[0128] Therefore, the determining a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, and predicting an optimal value of a driving variable based on the weight can be replaced by following steps 101 to 103:

[0129] Step 101: Estimate a state of the hydraulic support by means of the Kalman filtering based 2025213699   11 Aug 2025 on the pose data at the different positions, and obtain estimated pose data.

[0130] Step 102: Calculate the weight of the corresponding driving variable of each piece of pose data based on the pose data at the different positions and the estimated pose data by using a formula A- _____________________1_____________________ £i+ I Prock I Vi+ $i+ * I ? valve I .

[0131] Step 103: Perform the weighted fusion on the pose data at the different positions based on the weight of the corresponding driving variable of each piece of pose data, and generate the optimal value of the driving variable.

[0132] The full pose calculation utilizes a seamless linkage method. A four-bar linkage mechanism of the hydraulic support is analyzed, the four-bar linkage mechanism and the roof beam are collaboratively analyzed, and the four-bar linkage mechanism, the roof beam, and front and rear columns are collaboratively analyzed. In this way, all pose data of the support is analyzed based on the inclination angles that are of the rear connecting rod, the roof beam, and the shield beam and are monitored by the inclination angle sensors. Given structural parameters such as L1, L2, L3, and L4, as well as the parameters 0 and 9, following conditions are met: L2 sinP + L4sin9 = L1siny + L3sin0. L2cosP + L1cosy = L4cosp + L3cos0.

[0133] As described above, L1 represents a distance between a pin shaft connecting a front connecting rod and the shield beam and a pin shaft connecting the rear connecting rod; L2 represents a length of the front connecting rod; L3 represents a length of the rear connecting rod; and L4 represents a distance between a pin shaft connecting the front connecting rod and the base and a pin shaft connecting the rear connecting rod and the base. 0 represents the inclination angle of the rear connecting rod; 9 represents a horizontal inclination angle between a connection line of the pin shafts that respectively connecting the front and rear connecting rods and the base; P represents an inclination angle of the front connecting rod, y represents the inclination angle of the shield beam; and n represents an included angle between the front column of the base and the base.

[0134] It is obtained that: Y = arcsine)+ ;'..

[0135]

[0136] P = arccos( L4cos^+L3Cos0-L3CosY L2 ). Intermediate variables a, b, and c are as follows: a = 2 x L1 x (L3sin0 — L4sm^); b =— 2 x L1 x (L3cos6 + L4cos^); and c = L22 L2 — (L3cos0 + L4cosp)2 — (L3sin0 — L4sinp)2. Structures of the roof beam and the base are separately analyzed based on the inclination 2025213699   11 Aug 2025 angle 5 of the roof beam. In this way, coordinates of a top pin shaft point C of the front column, a pin shaft point D of the rear column, a bottom pin shaft point A of the front column, and a pin shaft point B of the rear column in a base coordinate system (with a pin shaft point of the rear connecting rod as an origin) can be determined. Y n =— arcsin—. XAC

[0137] As described above, XAC represents a horizontal distance between the bottom pin shaft point A and the top pin shaft point C of the front column, and YAC represents a vertical distance between the bottom pin shaft point A and the top pin shaft point C of the front column. Lelongated = XAC + YAC LAC original •

[0138] As described above, LAC original represents an original length before a column-type oil cylinder extends, and Lelongated represents a difference before and after the column-type oil cylinder extends, namely an elongation.

[0139] In this way, the included angle n between the front column and the base, as well as an extension length or a contraction length of the column in this process can be obtained. A height H of the hydraulic support can also be calculated. H = sinn J XAc + YAc.

[0140] A full pose of the hydraulic support can be obtained by using the above full pose calculation method, including the inclination angle of the shield beam, the inclination angle of the roof beam, the inclination angles of the front and rear connecting rods, and angles and elongations of the front and rear columns.

[0141] A rapid response of the virtual control model is to drive the monitoring model to move to a target position in a data-driven manner through a script mounted on the monitoring model after the full pose calculation of the virtual monitoring model. A current pose of each component in the hydraulic support of the monitoring model is defined as the destination state Sd, and a current state that is of the support of the control model and presented by a virtual sensor in the control model is defined as the Sc. The Sd is sent to the virtual control model through the MQTT protocol, and a difference between the Sd and the Sc is obtained. Driving force of the virtual control model is set based on the difference. A larger difference leads to larger driving force, and a smaller difference leads to smaller driving force, thus achieving the fast response of the virtual control model.

[0142] 2. Physical instruction-driven mode

[0143] As shown in FIG. 3, in the physical instruction-driven mode, the electro-hydraulic controller of the physical hydraulic support issues a digital code of the control instruction, which is transmitted to the virtual control model in the Unity3D through the MQTT protocol. Each digital 2025213699   11 Aug 2025 code represents a movement instruction. Based on the oil cylinder pressure and the surrounding rock pressure that are monitored by the pressure sensor, as well as factors such as a delay of a switch valve, the optimal force agent model predicts driving force that is of a virtual model and is most suitable for a current working condition, and instructs the virtual control model and the physical entity to move synchronously. Virtual sensor information in the control model is sent to the virtual monitoring model through the MQTT protocol. The full pose calculation is performed on received data, and a Transform component of a corresponding component is controlled to control the virtual monitoring model to move synchronously. Therefore, when there is no sensor, sensor information can also be simulated only based on a physical control instruction. Pose data of each component in the virtual monitoring model is simulated sensing information. In this case, simulated operations include: issuing and transmitting the control instruction, constructing a virtual control agent model, sending the virtual sensor information, performing the full pose calculation, and obtaining the simulated sensor information.

[0144] The issuing and transmitting the control instruction is to transmit an instruction of the electro-hydraulic controller of the hydraulic support to the Unity3D. Firstly, it is necessary to connect the electro-hydraulic controller to the computer through an appropriate interface (such as a serial communication interface or a TCP / IP). Secondly, a program for collecting a signal from the electro-hydraulic controller on the computer is compiled by using a C# language, to achieve data collection and analysis. The received data is transmitted to the Unity3D by using the C# script in the Unity3D through the TCP / IP, the User Datagram Protocol (UDP), or the serial communication. The Unity3D supports a plurality of network communication modes, and a built-in System.IO.Ports.SerialPort class or custom socket communication in the Unity3D may be used. After the instruction is received, in the Unity3D, the transmitted data is analyzed according to a written script, and according to the instruction, a virtual hydraulic support model is controlled to move.

[0145] The virtual control agent model includes the MQTT protocol communication module, the velocity parameter training model, and the optimal force agent model. The optimal force agent model is to jointly drive a virtual geometric model of the physical hydraulic support in the Unity3D based on knowledge such as hydraulic support positioning and a single-support kinematic model, as well as real-time inclination angle, displacement, and other data, to achieve appearance consistency between the physical hydraulic support and the virtual support. A reserved pose driving variable is associated with a real-time pose and other data to drive the virtual hydraulic support model to move synchronously, thereby achieving spatio-temporal consistency between the physical hydraulic support and the hydraulic support model. The hydraulic support model established by Unigraphics NX (UG) is imported into the Unity3D as a rendering part of the virtual control model, and is added 2025213699   11 Aug 2025 with a collision body. Driving force is added to the virtual control model, a force range is set, a force value within the force range is traversed, and a movement velocity of the oil cylinder is obtained through GetComponent<Rigidbody>().velocity and compared with a velocity predicted by the above velocity parameter training model. If a difference between a value of a velocity. magnitude parameter in the obtained movement velocity and a value of the velocity. magnitude parameter in the predicted velocity is within 1%, the force traversal is stopped, and an optimal force value is output. The optimal force value is sent to the virtual control model through the MQTT protocol communication server. After receiving the movement instruction, the virtual control model in the Unity3D calls a corresponding movement method, change force applied to the virtual control model to the above optimal driving force, so as to achieve maximum fidelity. Therefore, the predicting optimal virtual force by using an optimal force agent model based on the pressure at the different positions can be replaced by following steps 201 to 206:

[0146] Step 201: Set a simulated driving force range, and select any piece of driving force within the simulated driving force range.

[0147] Step 202: Add the selected driving force to the virtual support in the virtual control model, and simulate a movement velocity of a corresponding oil cylinder, where the oil cylinder includes a left column-type oil cylinder, a right column-type oil cylinder, a balance jack oil cylinder, and a push-pull oil cylinder.

[0148] Step 203: Predict the movement velocity of the oil cylinder by using the velocity parameter training model based on pressure from an underground oil cylinder, where the velocity parameter training model is a multi-layer perceptron (MLP).

[0149] The velocity parameter training model is configured to perform velocity parameter interpolation for a data-driven process of the monitoring model and provide a reference velocity for the optimal force agent model. The velocity parameter training model is a key technical means for implementing the dual-drive mode.

[0150] The velocity parameter training model first needs to process existing pressure and surrounding rock pressure from an underground real column-type oil cylinder and corresponding movement velocity parameters, existing pressure and surrounding rock pressure from an underground real push-pull oil cylinder and corresponding movement velocity parameters, and existing pressure and surrounding rock pressure from an underground real balance jack oil cylinder and corresponding movement velocity parameters. For a missing value, an approximate value is selected to make up for a corresponding data group. If there is no approximate value, the corresponding data group is removed. Then processed data is stored in a matrix. In order to ensure randomness of the data, a random number generator can be used to shuffle an order of the data. A dataset is divided into a training set (containing 70% of data in the dataset), a validation set 2025213699   11 Aug 2025 (containing 15% of the data in the dataset), and a test set (containing 15% of the data in the dataset).

[0151] Divided data is input. The MLP can be easily implemented and trained, and has a powerful representation capability. The MLP can approximate any continuous function when there are sufficient hidden layers and neurons, making it suitable for problem prediction. Therefore, a neural network model architecture with the MLP is selected. Four neurons are set in an input layer, respectively corresponding to pressure of the left column-type oil cylinder, pressure of the right column-type oil cylinder, pressure of the balance jack oil cylinder, and pressure of the push-pull oil cylinder. A hidden layer consists of 15 neurons. An output layer consists of four neurons, respectively corresponding to a velocity parameter of the left column-type oil cylinder, a velocity parameter of the right column-type oil cylinder, a velocity parameter of the balance jack oil cylinder, and a velocity parameter of the push-pull oil cylinder.

[0152] The most commonly used Rectified Linear Unit (ReLU) activation function is selected. A formula for the ReLU activation function is as follows: f (x)=max (0, x), which is characterized by simple calculation and can accelerate forward and backward propagation processes. In the formula for the ReLU activation function, x represents an input, and f() represents the ReLU activation function. In addition, the ReLU activation function can also alleviate a problem of gradient vanishing. In a positive interval, a derivative of the ReLU activation function is always 1, which makes a network less prone to gradient vanishing during training, especially in a deep network. A Mean Squared Error (MSE) loss function is sensitive to an outlier because a squared term amplifies a large error. Therefore, the MSE loss function is selected.

[0153] Layer 2 (L2) regularization is achieved by adding a penalty term for an L2 norm of a weight parameter to a loss function of a model. In the L2 regularization, the penalty term is typically defined as a square of the L2 norm of the weight parameter. Specifically, a loss function of the L2 regularization can be expressed as follows: LL2 = Ldata + ^||wH2 .

[0154] As described above, Ldata represents a data loss of the virtual control model, which is an error between a predicted value of the virtual control model and a true label; A represents a regularization parameter, which is used to control regularization intensity; and ||«||2 represents a square of an L2 norm of a weight vector, which is expressed as a sum of squares of various parameters in the weight vector.

[0155] The loss function of the L2 regularization is used. An optimization algorithm considers both a data loss and a regularization term in an optimization process, in order to minimize a size of a model parameter and reduce model complexity while maintaining a capability of fitting training data.

[0156] An Adam optimizer is selected, the model parameter is transferred to the Adam optimizer, 2025213699   11 Aug 2025 and a learning rate and other hyperparameters are set, specifically, epochs is set to 100, and a batch size is set to 32. A plurality of iterations (epochs) are performed, and each batch of data is trained in each epoch. Model performance is evaluated on the test set to monitor overfitting and a model generalization capability. The Adam optimizer optimizes a hyperparameter based on the model performance, that is, adjusts the learning rate, the batch size, and other hyperparameters based on performance of the validation set to improve the model performance. After training under most suitable model performance is completed, the virtual control model is saved.

[0157] A trained neural network model is exported in an ONNX format by using an exportONNXNetwork function of a matrix library (MATLAB). The Unity3D supports an ONNX model and uses a Barracuda library in the Unity3D to load and run the ONNX model, thereby achieving communication between the Unity3D and a neural network training model.

[0158] Step 204: Determine a difference between a simulated movement velocity of the oil cylinder and a predicted movement velocity of the oil cylinder.

[0159] Step 205: If the difference between the simulated movement velocity of the oil cylinder and the predicted movement velocity of the oil cylinder is greater than a preset difference threshold, select another piece of driving force within the simulated driving force range, and return to the step 202.

[0160] Step 206: If the difference between the simulated movement velocity of the oil cylinder and the predicted movement velocity of the oil cylinder is less than or equal to a preset difference threshold, determine the selected driving force as the optimal virtual force.

[0161] The sending the virtual sensor information is to construct the virtual sensor at a control model end by using transform.localPosition() and transform.localRotation() methods, and display a key pose of a current support. A driving variable that is required by the monitoring model are sent to the monitoring model through the MQTT protocol communication module, so as to make the monitoring model respond.

[0162] The obtaining the simulated sensor information is to obtain all poses of the hydraulic support after the full pose calculation, such that when there is no sensing information, the simulated sensing information can be obtained to provide an initial pose reference for a next movement instruction. In this way, a movement process of the hydraulic support can be monitored, and a fully-mechanized coal mining process is transparent.

[0163] 3. Dual-drive mode

[0164] As shown in FIG. 4, the dual-drive mode integrates pressure and pose sensing information on the hydraulic support and a movement instruction on the physical entity, and transmits the pressure and pose sensing information and the movement instruction to the virtual monitoring model and virtual control model respectively. The sensing information is processed through the 2025213699   11 Aug 2025 weighted adaptive hybrid algorithm of the sensor to calculate a weight of each piece of sensing data, as well as the velocity parameter for the movement of the oil cylinder under current oil cylinder pressure and surrounding rock pressure. An optimal value of a key driving variable for the virtual monitoring model is predicted based on the weight. Driving data is interpolated based on the velocity parameter to make a movement characteristic of the monitoring model closer to an actual support movement velocity. After receiving the control instruction, a virtual control end drives, based on optimal virtual force corresponding to actual oil cylinder pressure obtained through training in the optimal force agent model, the virtual control model to move. The adaptive control strategy is used to perform conflict resolution and redundancy control on the sensing information and the input physical instruction, so as to enable the virtual monitoring model and the virtual control model to move consistently and synchronously.

[0165] The dual-drive mode performs interpolation based on a velocity parameter predicted by the dual-drive agent model to make a movement velocity of the monitoring model approximate a movement velocity of a physical support. A magnitude of optimal virtual force predicted by the dual-drive agent model enables the virtual control model and the physical support to move synchronously. The adaptive control strategy is used to perform the conflict resolution and the redundancy control on the sensing information and the input physical instruction, so as to enable the virtual monitoring model and the virtual control model to move consistently and synchronously, which is a necessary condition for achieving "sensing-control" integration. Issuance of the control instruction, collection of the sensing information, the dual-drive agent model, the velocity parameter training model, a velocity parameter interpolation method, and the adaptive control strategy are included.

[0166] The issuance of the control instruction and the collection of the sensing information include sensing information collection and processing in the sensing information-driven mode, as well as control instruction issuance and transmission in the physical instruction-driven mode. This is applicable to a working condition under which both the sensing information path and the physical instruction path exist.

[0167] The dual-drive agent model includes the MQTT protocol communication module, the weighted adaptive hybrid algorithm of the sensor, and the optimal force agent model. The weighted adaptive hybrid algorithm of the sensor is also used to predict an optimal driving variable of the monitoring model, while the optimal force agent model is used to predict the optimal virtual force and drive the control model to move. The MQTT protocol communication module is used to achieve communication between a virtual detection model and the virtual control model, facilitating feedback adjustment.

[0168] The velocity parameter interpolation method needs to obtain a current velocity parameter 2025213699   11 Aug 2025 CurrentSpeed from the velocity parameter training model in each frame, and multiplies a velocity specified in the current velocity parameter CurrentSpeed by Time.deltTime to obtain a distance that needs to be moved for each frame. A Vector3.MoveTowards method is used to move an object from its current position to a target position, and also monitor the current position and the target position of the object. After the target position is reached, the movement is stopped, such that the virtual monitoring model performs a high-fidelity movement at an appropriate velocity. Therefore, the interpolating the optimal value of the driving variable based on a predicted velocity parameter can be replaced by following steps 301 and 302:

[0169] Step 301: Multiply a predicted velocity parameter in each frame by time, and obtain a distance that needs to be moved for each frame.

[0170] Step 302: Interpolate the optimal value of the driving variable based on the distance that needs to be moved for each frame.

[0171] The adaptive control strategy is designed for the dual-drive mode in which both the sensing information and the control instruction exist. If credibility of inclination angle sensor data is high, a high weight is allocated to the inclination angle sensor data, and a weight of the control instruction is correspondingly reduced. If credibility of the control instruction is high, a high weight is allocated to the control instruction, and a weight of the sensor data is correspondingly reduced. If the credibility of the inclination angle sensor data and the credibility of the control instruction are equivalent, the weights of the inclination angle sensor data and the control instruction are dynamically adjusted based on an adaptive algorithm.

[0172] To implement the above strategy, it is necessary to first clarify a credibility evaluation indicator. For the sensing information, the credibility evaluation indicator includes: data integrity, indicating whether data is complete and whether there is a missing value or an abnormal value; data consistency, indicating whether the data is consistent with historical data, other sensor data, or a prediction result of a physical model; a sensor state, indicating whether a sensor is working properly and whether there is a fault or drift; and environmental interference, indicating whether an environmental factor (such as a temperature, a humidity, or electromagnetic interference) has an impact on the sensor data. For the control instruction, the credibility evaluation indicator includes: instruction rationality, indicating whether the instruction conforms to a current state and working logic of a system; instruction source reliability, indicating whether an instruction source is reliable and whether there is a risk of tampering with or sending by mistake the instruction; and historical instruction consistency, indicating whether a current instruction is consistent with a historical instruction sequence and whether there is any contradiction.

[0173] Secondly, based on the above indicators, a credibility evaluation model is designed. In terms of a qualitative aspect, a rule-based model is designed, that is, a series of rules are formulated 2025213699   11 Aug 2025 based on expert experience and domain knowledge to determine credibility of information. For example, if the sensor data exceeds a normal range, its credibility decreases. If the control instruction conflicts with the historical instruction sequence, credibility of the control instruction decreases. In terms of a quantitative aspect, a statistics-based model needs to be designed, that is, distribution characteristics of the sensor data and the control instruction are statistically analyzed based on historical data, and a probability model is established to perform credibility evaluation. For example, a Bayesian network is used to calculate a posterior probability as credibility based on prior and conditional probabilities of the sensor data and the control instruction. A system state and observation noise are estimated through the Kalman filtering based on a system model and observation data, in order to evaluate the credibility of the information. Specifically, it is necessary to first construct the Bayesian network, where each node represents a different variable such as the sensor data or the control instruction, and an edge represents a conditional dependency relationship between variables. In a practical application, when the system obtains real-time sensor data and a real-time control instruction, a posterior probability of each variable in the network can be updated according to the Bayes' theorem, thereby evaluating credibility of the system state. A higher posterior probability leads to higher credibility of the information. This method effectively quantifies the credibility of the information through reasoning and probability calculation. Next, the Kalman filtering recursively estimates the system state in combination with the system model and the observation data, thereby further evaluating the credibility of the information. The Kalman filtering consists of two steps: prediction and updating. At the prediction stage, a system state at a current time point is calculated based on a state transition model of the system and a state estimate at a previous time point, and an error covariance is calculated. At the updating stage, a state estimate is corrected based on new observation data, and a prediction error is updated based on the Kalman gain. An updated error covariance matrix can reflect an uncertainty of the information, and a smaller error leads to higher credibility of the information. By combining the Bayesian networks and the Kalman filtering, the system can comprehensively evaluate the credibility of the information, and enhance estimation accuracy and reliability.

[0174] The dynamic weight adjustment mainly involves the following three aspects: time decaying: the credibility of the information decreases over time, so it is necessary to introduce a time decaying factor for the dynamic weight adjustment; new information updating: when new sensor data or a new control instruction is obtained, it is necessary to update a credibility evaluation result in a timely manner and perform weight adjustment accordingly; and a feedback mechanism: based on an actual operational state of the system, the credibility evaluation model and a weight adjustment strategy are fed back and optimized. The feedback mechanism optimizes the credibility evaluation model and the weight adjustment strategy in real time based on the actual operational state of the 2025213699   11 Aug 2025 system. During operation of the system, a weight and an evaluation model are dynamically adjusted by monitoring a deviation between actual and expected results. For example, if a sensor or a control instruction repeatedly has a significant error, the system will automatically identify and reduce a weight of the information source. Conversely, if an information source performs well, the system will appropriately increase its weight. In addition, the feedback mechanism can also iteratively optimize the evaluation model based on the historical data and the actual operational state to gradually improve the weight adjustment strategy to more accurately predict and evaluate the credibility of the information, thereby ensuring reliability and accuracy of the system in a dynamic environment.

[0175] Therefore, the fusing a control instruction and pose data by using an adaptive control strategy, and obtaining fused pose data can be replaced by following steps 401 to 405:

[0176] Step 401: Determine the credibility evaluation indicator of the control instruction and a credibility evaluation indicator of the pose data.

[0177] Step 402: Calculate the credibility of the control instruction and credibility of the pose data by using the rule-based model and the statistics-based model based on the credibility evaluation indicator of the control instruction and the credibility evaluation indicator of the pose data.

[0178] Step 403: If the credibility of the control instruction is not equal to the credibility of the pose data, set the weight of the control instruction and a weight of the pose data based on a ratio of the credibility of the control instruction to the credibility of the pose data.

[0179] Step 404: If the credibility of the control instruction is equal to the credibility of the pose data, determine the weight of the control instruction and a weight of the pose data by using the weighted adaptive hybrid algorithm of the sensor.

[0180] Step 405: Perform the weighted fusion on the control instruction and the pose data based on the weight of the control instruction and the weight of the pose data, and obtain the fused pose data.

[0181] Table 9 compares the three modes. Table 10 shows switching among the three modes. Table 9 Model comparison Mode Sensing information-driven mode Physical instruction-driven mode Dual-drive mode Applied technical module The weighted adaptive hybrid algorithm of the sensor and the MQTT protocol communication The optimal force agent model, the velocity parameter training model, and the MQTT protocol communication The adaptive control strategy, the velocity parameter interpolation method, the velocity parameter training model, the weighted adaptive hybrid algorithm of the sensor, 2025213699   11 Aug 2025 module module the optimal force agent model, and the MQTT protocol communication module Function © Collecting and processing the sensing information; @ transmitting the sensing information to the Unity3D; © implementing the communication between the monitoring and control models; © predicting the optimal driving variable; @ performing the full pose calculation; @ rapid response of the virtual control model; and © processing redundant or conflicting sensing information ©Issuing and transmitting the control instruction; © predicting the optimal force applied to the control model; © implementing the communication between the monitoring and control models to subscribe to and send the virtual sensor information; © performing the full pose calculation; and © simulating the sensor information ©Issuing the control instruction and collecting the sensing information; © predicting the optimal driving variable for the monitoring model and performing the full pose calculation; © predicting the velocity parameter for the movement of the oil cylinder and interpolating the velocity parameter by the monitoring model; © predicting optimal simulated force and driving the control model to move; @ performing feedback regulation to ensure that the monitoring and control models always move synchronously; and © performing the conflict resolution and redundancy control on the physical instruction and the sensing information Applicable working condition No physical instruction is available. No sensing information is available. Both the physical instruction and the sensing information exist. Table 10 Mode switching methods Working condition for triggering mode switching Parameter change © to © The transmission of the sensing Physical instruction parameters are 2025213699   11 Aug 2025 information is interrupted, and the transmission of the physical instruction is restored. prioritized. © to © The sensing information path is maintained, and the transmission of the physical instruction is restored. The sensing information and the physical instruction are adaptively regulated, and the conflict resolution and the redundancy control are achieved. © to © The transmission of the physical instruction is interrupted, and the transmission of the sensing information is restored. The sensing data is prioritized. © to © The physical instruction path is maintained, and the transmission of the sensing information is restored. The sensing information and the physical instruction are adaptively regulated, and the conflict resolution and the redundancy control are achieved. ©to© The transmission of the physical instruction is interrupted. The control model responds based on the sensing data. ©to© The transmission of the sensing information is interrupted. The sensing data is simulated based on physical instruction data.

[0182] Note: In Table 10, © represents the sensing information-driven mode, © represents the physical instruction-driven mode, and © represents the dual-drive mode.

[0183] The present disclosure uses the physical entity to drive the virtual control model and the virtual monitoring model to perform the high-fidelity movement to virtually reconstruct the workspace in the virtual space, and enables the two models to move consistently through a neural network training algorithm and an agent model.

[0184] Based on the same inventive concept, the embodiments of the present disclosure also provide a sensing-control integrated twin method for a fully-mechanized coal mining equipment group, which is applied to the sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group. A problem-solving implementation solution provided by the sensing-control integrated twin method is similar to the implementation solution described in the above sensing-control integrated twin apparatus. Therefore, for following specific limitations on one or more embodiments of the sensing-control integrated twin method for a fully-mechanized coal mining equipment group, reference may be made to the above limitations on the 2025213699   11 Aug 2025 sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group, and details are not described herein again.

[0185] In an exemplary embodiment, a sensing-control integrated twin method for a fully-mechanized coal mining group includes following steps 501 to 503:

[0186] Step 501: When detecting that only sensing information is input, a virtual monitoring module obtains pose data at different positions on a hydraulic support, determines a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, predicts an optimal value of a driving variable based on the weight, drives a virtual support in the virtual monitoring model based on the optimal value of the driving variable, performs full pose calculation based on the optimal value of the driving variable, obtains a pose state of a virtual support whose appearance is consistent with a current appearance of the hydraulic support, and sends the pose state to a virtual control model as a destination state; and a virtual control module determines a difference between the destination state of the virtual support and a current state of the virtual support, determines driving force based on the difference, and applies the determined driving force to drive a virtual support in the virtual control model to move; where the virtual support is a digital twin model of the hydraulic support.

[0187] Step 502: When detecting that only a physical instruction is input, a virtual control model receives a control instruction sent from an electro-hydraulic controller, predicts optimal virtual force by using an optimal force agent model based on pressure at different positions after receiving the control instruction, drives, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and a hydraulic support to move synchronously, and sends a state of the virtual support after the synchronous movement to a virtual monitoring model; and the virtual monitoring model performs full pose calculation on the received state, and controls a pose state of a virtual support in the virtual monitoring model based on data obtained after the full pose calculation.

[0188] Step 503: When detecting that sensing information and a physical instruction are input simultaneously, a dual-drive agent model fuses a control instruction and pose data by using an adaptive control strategy, obtains fused pose data, determines a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the fused pose data, predicts an optimal value of a driving variable based on the weight, predicts a velocity parameter for a movement of an oil cylinder of a hydraulic support by using a velocity parameter training model, and interpolates the optimal value of the driving variable based on a predicted velocity parameter, such that a virtual support in a virtual monitoring model moves based on a velocity of the hydraulic support; and the dual-drive agent model also predicts optimal virtual 2025213699   11 Aug 2025 force by using an optimal force agent model based on pressure at different positions after a virtual control model receives the control instruction, and drives, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously.

[0189] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal, and an internal structure thereof may be as shown in FIG. 5. The computer device includes a processor, a memory, an input / output (I / O) interface, and a communication interface. The processor, the memory, and the I / O interface are connected through a system bus. The communication interface is connected to the system bus through the I / O interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store sensing information and a state of a virtual support. The I / O interface of the computer device is configured to exchange information between the processor and an external device. The communication interface of the computer device is configured to connect to and communicate with an external terminal through a network. The computer program is executed by the processor to implement a sensing-control integrated twin method for a fully-mechanized coal mining equipment group.

[0190] Those skilled in the art may understand that the structure shown in FIG. 5 is only a block diagram of a part of the structure related to the solutions of the present disclosure and does not constitute a limitation on a computer device to which the solutions of the present disclosure are applied. Specifically, the computer device may include more or less components than those shown in the figure, or combine some components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the computer program is executed by the processor to implement the steps in the above method embodiments.

[0191] In an exemplary embodiment, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in the above method embodiments.

[0192] In an exemplary embodiment, a computer program product is provided, including a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments. 2025213699   11 Aug 2025

[0193] It should be noted that information of a user (including but not limited to device information of the user, personal information of the user, and the like) and data (including but not limited to data for analysis, stored data, displayed data, and the like) in the present disclosure are information and data authorized by the user or fully authorized by each party.

[0194] Those of ordinary skill in the art may understand that all or some of the procedures in the method of the foregoing embodiments may be implemented by a computer program instructing related hardware. The computer program may be stored in a non-volatile computer-readable storage medium. When the computer program is executed, the procedures in the embodiments of the above method may be performed. Any reference to a memory, a database, or other media used in the embodiments of the present disclosure may include at least one of a non-volatile memory and a volatile memory. The non-volatile memory may include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical memory, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, and the like. The volatile memory may include a random access memory (RAM) or an external cache memory. As an illustration rather than a limitation, the RAM may be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM).

[0195] The database in the embodiments of the present disclosure may include at least one of a relational database and a non-relational database. The non-relational database may include a blockchain-based distributed database, but is not limited thereto. The processor in the embodiments of the present disclosure may be a general processor, a central processing unit (CPU), a graphics processor, a digital signal processor (DSP), a programmable logic device, and a data processing logic device based on quantum computing, but is not limited thereto.

[0196] The technical characteristics of the above embodiments can be employed in arbitrary combinations. To provide a concise description of these embodiments, all possible combinations of all the technical characteristics of the above embodiments may not be described; however, these combinations of the technical characteristics should be construed as falling within the scope defined by the specification as long as no contradiction occurs.

[0197] Several examples are used herein for illustration of the principles and implementations of the present disclosure. The description of the above embodiments is used to help illustrate the method of the present disclosure and the core principles thereof. In addition, those of ordinary skill in the art can make various modifications in terms of specific implementations and the scope of application in accordance with the teachings of the present disclosure. In conclusion, the content of the present specification shall not be construed as a limitation to the present disclosure.

Claims

1. A sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group, comprising: an electro-hydraulic controller, a hydraulic support, a sensing module, and a control unit, whereinthe control unit is embedded with a virtual monitoring model, a virtual control model, and a dual-drive agent model; andwhen the virtual monitoring model receives only sensing information, the sensing module comprises a plurality of pose sensors that are configured to collect pose data at different positions on the hydraulic support after the electro-hydraulic controller controls the hydraulic support to change a pose; the virtual monitoring model is configured to: determine a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, predict an optimal value of a driving variable based on the weight, drive a virtual support in the virtual monitoring model based on the optimal value of the driving variable, perform full pose calculation based on the optimal value of the driving variable, obtain a pose state of a virtual support whose appearance is consistent with a current appearance of the hydraulic support, and send the pose state to the virtual control model as a destination state; and the virtual control model is configured to determine a difference between the destination state of the virtual support and a current state of the virtual support, determine driving force based on the difference, and apply the driving force to drive a virtual support in the virtual control model to move; wherein the virtual support is a digital twin model of the hydraulic support;when the virtual monitoring model receives only a physical instruction, the sensing module comprises a plurality of pressure sensors that are configured to monitor pressure at different positions on the hydraulic support; the virtual control model is configured to: receive a control instruction sent from the electro-hydraulic controller, predict optimal virtual force by using an optimal force agent model based on the pressure at the different positions after receiving the control instruction, drive, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously, and send a state of the virtual support after the synchronous movement to the virtual monitoring model; and the virtual monitoring model is configured to perform full pose calculation on the received state, and control a pose state of a virtual support in the virtual monitoring model based on data obtained after the full pose calculation; andwhen the virtual monitoring model receives sensing information and a physical instruction at the same time, the sensing module comprises a plurality of pose sensors and a plurality of pressure sensors; the dual-drive agent model is configured to: fuse a control instruction and pose data by2025213699   11 Aug 2025using an adaptive control strategy, obtain fused pose data, determine a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the fused pose data, predict an optimal value of a driving variable based on the weight, predict a velocity parameter for a movement of an oil cylinder of the hydraulic support by using a velocity parameter training model, and interpolate the optimal value of the driving variable based on a predicted velocity parameter, such that a virtual support in the virtual monitoring model moves based on a velocity of the hydraulic support; and the dual-drive agent model is also configured to predict optimal virtual force by using an optimal force agent model based on pressure at different positions after the virtual control model receives the control instruction, and drive, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously.

2. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, further comprising: an actuator, whereinthe electro-hydraulic controller controls the actuator to move according to the control instruction, to change the pose of the hydraulic support.

3. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, further comprising: a data collection module, whereinthe data collection module is embedded in the control unit; andthe data collection module is configured to collect pose data of the plurality of pose sensors and pressure monitored by the plurality of pressure sensors, correct and filter the collected pose data and pressure, and transmit corrected and filtered pose data and pressure to the virtual monitoring model.

4. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, wherein the pose sensor is an inclination angle sensor;a plurality of inclination angle sensors are respectively disposed at a rear connecting rod, a roof beam, and a shield beam of the hydraulic support; andthe plurality of inclination angle sensors are configured to collect inclination angle data of the rear connecting rod, the roof beam, and the shield beam.

5. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, wherein the determining a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, and predicting an optimal value of a driving variable based on the2025213699   11 Aug 2025weight comprises:estimating a state of the hydraulic support by means of Kalman filtering based on the pose data at the different positions, and obtaining estimated pose data;calculating the weight of the corresponding driving variable of each piece of pose data based on the pose data at the different positions and the estimated pose data by using a formula 2,. =-------------------------------, wherein in the formula, e, represents basic reliability of ane, + I ^rock I 7 / +    + I ^valve Izlh pose sensor, / / represents a pressure-noise transfer coefficient of the 7th pose sensor, / jock represents surrounding rock pressure, | Prod, | 77, represents an interference compensation term of surrounding rock pressure of the 7th pose sensor, 5. represents a feedback term of a predicted residual of the z1h pose sensor, 5t =| 0j -      |, represents pose data of the z1h pose sensor,represents estimated pose data of the zlh pose sensor, K represents a delay sensitivity coefficient, rvalve represents a delay of a valve control instruction, and k | rvalve | represents a compensation term of a valve delay; andperforming weighted fusion on the pose data at the different positions based on the weight of the corresponding driving variable of each piece of pose data, and generating the optimal value of the driving variable.

6. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, wherein the destination state comprises: an inclination angle of a shield beam, an inclination angle of a roof beam, an inclination angle of a front connecting rod, an inclination angle of a rear connecting rod, an angle of a front column, an angle of a rear column, and an elongation of a column-type oil cylinder.

7. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, wherein the predicting optimal virtual force by using an optimal force agent model based on the pressure at the different positions comprises:setting a simulated driving force range, and selecting any piece of driving force within the simulated driving force range;adding the selected driving force to the virtual support in the virtual control model, and simulating a movement velocity of a corresponding oil cylinder, wherein the oil cylinder comprises a left column-type oil cylinder, a right column-type oil cylinder, a balance jack oil cylinder, and a push-pull oil cylinder;2025213699   11 Aug 2025predicting the movement velocity of the oil cylinder by using the velocity parameter training model based on pressure from an underground oil cylinder, wherein the velocity parameter training model is a multi-layer perceptron (MLP);determining a difference between a simulated movement velocity of the oil cylinder and a predicted movement velocity of the oil cylinder; andif the difference between the simulated movement velocity of the oil cylinder and the predicted movement velocity of the oil cylinder is greater than a preset difference threshold, selecting another piece of driving force within the simulated driving force range, and returning to the step of "adding the selected driving force to the virtual support in the virtual control model, and simulating a movement velocity of a corresponding oil cylinder"; orif the difference between the simulated movement velocity of the oil cylinder and the predicted movement velocity of the oil cylinder is less than or equal to a preset difference threshold, determining the selected driving force as the optimal virtual force.

8. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, wherein the interpolating the optimal value of the driving variable based on a predicted velocity parameter specifically comprises:multiplying a predicted velocity parameter in each frame by time, and obtaining a distance that needs to be moved for each frame; andinterpolating the optimal value of the driving variable based on the distance that needs to be moved for each frame.

9. The sensing-control integrated twin apparatus for a fully-mechanized coal mining equipment group according to claim 1, wherein the fusing a control instruction and pose data by using an adaptive control strategy, and obtaining fused pose data specifically comprises:determining a credibility evaluation indicator of the control instruction and a credibility evaluation indicator of the pose data;calculating credibility of the control instruction and credibility of the pose data by using a rule-based model and a statistics-based model based on the credibility evaluation indicator of the control instruction and the credibility evaluation indicator of the pose data;if the credibility of the control instruction is not equal to the credibility of the pose data, setting a weight of the control instruction and a weight of the pose data based on a ratio of the credibility of the control instruction to the credibility of the pose data; orif the credibility of the control instruction is equal to the credibility of the pose data, determining a weight of the control instruction and a weight of the pose data by using the weighted2025213699   11 Aug 2025adaptive hybrid algorithm of the sensor; andperforming weighted fusion on the control instruction and the pose data based on the weight of the control instruction and the weight of the pose data, and obtaining the fused pose data.

10. A sensing-control integrated twin method for a fully-mechanized coal mining equipment group, comprising:when detecting that only sensing information is input, obtaining, by a virtual monitoring module, pose data at different positions on a hydraulic support, determining a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the pose data at the different positions, predicting an optimal value of a driving variable based on the weight, driving a virtual support in the virtual monitoring model based on the optimal value of the driving variable, performing full pose calculation based on the optimal value of the driving variable, obtaining a pose state of a virtual support whose appearance is consistent with a current appearance of the hydraulic support, and sending the pose state to a virtual control model as a destination state; and determining, by the virtual control module, a difference between the destination state of the virtual support and a current state of the virtual support, determining driving force based on the difference, and applying the determined driving force to drive a virtual support in the virtual control model to move; wherein the virtual support is a digital twin model of the hydraulic support;when detecting that only a physical instruction is input, receiving, by a virtual control model, a control instruction sent from an electro-hydraulic controller, predicting optimal virtual force by using an optimal force agent model based on pressure at different positions after receiving the control instruction, driving, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and a hydraulic support to move synchronously, and sending a state of the virtual support after the synchronous movement to a virtual monitoring model; and performing, by the virtual monitoring model, full pose calculation on the received state, and controlling a pose state of a virtual support in the virtual monitoring model based on data obtained after the full pose calculation; andwhen detecting that sensing information and a physical instruction are input simultaneously, fusing, by a dual-drive agent model, a control instruction and pose data by using an adaptive control strategy, obtaining fused pose data, determining a weight of a corresponding driving variable of each piece of pose data by using a weighted adaptive hybrid algorithm of a sensor based on the fused pose data, predicting an optimal value of a driving variable based on the weight, predicting a velocity parameter for a movement of an oil cylinder of a hydraulic support by using a velocity parameter training model, and interpolating the optimal value of the driving variable based on a2025213699   11 Aug 2025predicted velocity parameter, such that a virtual support in a virtual monitoring model moves based on a velocity of the hydraulic support; and also predicting, by the dual-drive agent model, optimal virtual force by using an optimal force agent model based on pressure at different positions after a virtual control model receives the control instruction, and driving, based on the control instruction and the optimal virtual force, a virtual support in the virtual control model and the hydraulic support to move synchronously.

Images