Optimization of Multi-Platform Dynamic Weapon-Target Assignment Based on Multi-Agent Reinforcement Learning

This paper focuses on the 'Weapon Platform-Weapon-Target Assignment' (W-WTA) Problem, aiming to minimize the expected threat values of incoming targets during combat while considering the minimal consumption of resources for effective resource allocation. To solve this problem, we propose a multi-agent reinforcement learning method based on the attention mechanism. We have established threat value model for targets and damage probability model. Each weapon platform is characterized by distinct attributes, such as range, weapon type and the number of missiles, etc. The multi-head attention mechanism integrates target information and weapon platform information to adjust the output of each agent, enabling each agent to consider the overall combat effectiveness. Experimental results indicate the in the most cases, the proposed algorithm demonstrates superior performance with a lower time cost. Compared with the Genetic Algorithm (GA) and Greedy Algorithm, it shows outstanding effectiveness in solving the W-WTA problem. This research takes into account the actual combat process and provides an advanced and reliable method for military operation decision-making.