面向火炮模拟器人机交互的大模型长时记忆机制OA

Aiming at the problems of low training data utilization rate in the traditional human-computer interaction(HCI)module of artillery simulation training systems and insufficient interactivity during the training process,we propose to construct a HCI application system based on large language models(LLMs).We build a vector database that integrates professional knowledge in the artillery field and historical Q&A data,and enable the connection between LLMs and the vector database via LangChain.Additionally,we optimize the text processing algorithm to enhance the system's ability to recognize professional terms and improve word segmentation performance.Experimental results demonstrate that the system exhibits long-term memory capabilities,can provide professional responses to training-related questions,and effectively improves both interaction efficiency and training data utilization.