This study addresses the issues of high risk and low efficiency in safety education and training for new energy power generation by innovatively developing an immersive training system based on “Internet+VR.” By improving the DH parameter method for kinematic modeling of the HTC VIVE controller, combined with Monte Carlo method analysis of the workspace, the system ensures the physical accuracy of virtual operations. A trajectory optimization algorithm based on equidistant interval sampling and statistical outlier removal (SOR) is proposed, combined with quadratic resampling to fix the trajectory points to 25. The system was developed using the Unity3D engine, compatible with HoloLens devices, and validated using the CMU-Hand dataset to assess the performance of the hand pose estimation algorithm, achieving an average accuracy of PCK=0.981, AUC=0.875, and E_mean=4.739mm. This significantly outperforms the other four algorithms. The experimental group using this system demonstrated significantly higher mastery of knowledge points and exam scores compared to traditional teaching methods, particularly in practical knowledge points such as “high-altitude work safety” and “personal protective equipment usage.” The system constructed in this paper addresses the core requirements of high environmental simulation, precise operation mapping, and zero-contact safety risk in new energy power training.
