In the face of the high requirements of vocational education for practical skills training, as well as the poor results and high costs of traditional training, this study aims to build an efficient and intelligent virtual training and simulation operation platform by introducing generative artificial intelligence technology to improve teaching quality and learning efficiency. This paper uses the adversarial generation ability of GAN and combines it with the knowledge graph in the field of vocational education to generate a high-fidelity and scalable virtual training environment. The variational autoencoder (VAE) is used to dynamically adjust the scene parameters to adapt to the needs of different majors and skill levels. Based on the Transformer multimodal dialogue model, a real-time operation guidance system integrating virtual and real is constructed. The physical engine is combined to simulate the feedback of equipment operation, and the long short-term memory network (LSTM) is used to capture the characteristics of the trainee’s operation sequence, dynamically predict potential errors and trigger correction prompts. This paper also uses the deep reinforcement learning (DRL) framework to build a personalized evaluation agent, using the trainee’s operation data as the state space to generate targeted training suggestions. The average operation accuracy of the experimental group students reaches 89.715%, the highest skill assessment pass rate reaches 100%, and the overall distribution is more concentrated, reducing the cost of practical training. In addition, teachers and students highly evaluated the teaching adaptability, operation guidance and error correction functions of the platform. The teaching platform proposed in this study provides a new path and strategy for the modernization of vocational education. © 2025 Copyright held by the owner/author(s).
Author Keywords
Index Keywords
