Weapon Targets Assignment for Electro-optical System Countermeasures based on Multi-objective Reinforcement Learning

A reasonable electro-optical weapon assignment scheme can improve the combat efficiency and reduce the cost. Considering the spectrum type, field angle, and effective interference distance of electro-optical weapon, an optimization model is established with the objectives of maximum interference effectiveness and minimum weapon cost. A multi-objective deep reinforcement learning algorithm based on convex envelope Q learning (CEQL-EWTA)is designed to solve the electro-optical weapon-target assignment model. The states are designed by the number and characteristics of weapons and incoming targets, the actions were designed by heuristic rules, and the reward functions were designed based on the objective functions. The importance of objective functions is converted into preferences, and Q-value learning under multiple preferences is performed by deep Q-networks. Simulation experiments show that CEQL-EWTA can effectively solve the weapon assignment problem for electro-optical countermeasures system, which has higher efficiency compared with genetic algorithms and deep Q networks.