An Intelligent Design Method of Kill-web Based on Knowledge-enhanced Pre-trained Language Model

The existing rule-based reasoning design method of kill-web has insufficient reasoning depth, poor flexibility and limited adaptability. This paper proposes an intelligent design method of kill-web based on knowledge-enhanced pre-trained language model. A multi-agent workflow of "scenario extraction-element completion-scheme reasoning-effectiveness assessment" is established based on the OODA loop theory and human command decision logic, and an intelligent design framework of the kill-web is constructed by using domain knowledge. A knowledge system composed of equipment entity knowledge base, tactical indicator library and design rule library is constructed to provide support for the knowledge enhancement model. Based on the retrieval-augmented generation technology and Crew-AI framework, the scenario extraction agent, element completion agent, scheme reasoning agent and effectiveness assessment agent are established to form a deep reasoning agent of multi-level pre-trained language model. The rapid intelligent design of kill-web in dynamic battlefield is realized through multiple interactions with the knowledge enhanced pre-trained language model in stages. A typical air defense and anti missile combat scenario is taken as a case to design the kill-web, and the feasibility and effectiveness of the proposed method are verified through comparative experiments.