Abstract: Aiming at the problems of missing information of the pose data of the equipment group in the longwall mining face, the low reliability of the sensing data, and the difficulty of solving the pose and bottom plate morphology of the floating connection mechanism, a technology for accurate virtual perception and reconstruction of the state of the equipment group and the bottom plate system in the longwall mining face is proposed.Firstly, 3D LiDAR and 2D LiDAR are installed in the shearer body and hydraulic support column in the physical space to obtain the point cloud data of the equipment group, and the overall equipment pose information is enhanced.Then, the virtual point cloud is obtained in the virtual scene with the integration of the physics engine and joint constraints, and the translation rotation matrix is obtained by registering the virtual and real point clouds with the physical feature point cloud with deep learning algorithms, which is used to drive the position reconstruction of the hydraulic support;the shearer trajectory is converted into driving data to drive the synchronous movement of the shearer virtual model;at the same time, based on the two-dimensional point cloud and the shearer sliding shoe trajectory fitting plane, the position of the scraper conveyor is inverted, and then combined with the hydraulic support to calculate the position and pose of the floating connection mechanism in both directions, the overall pose reconstruction of the equipment group is completed.Finally, based on the coupling relationship between the floor plate and the equipment, the coal seam surface is constructed by combining the shearer trajectory and rocker drum data, and the surface reconstruction and reverse verification of the working face are realized with the help of equipment pose information and physics engine.The results show that the position error of the reconstructed equipment is within 0.05 m, the attitude error is within 1°, and the attitude error of the scraper conveyor after reverse verification of the floor plate is within 0.8°, and the overall accuracy is high, which provides technical support for the accurate virtual reconstruction and normalized autonomous cutting of the working face.