Research on Missile Cooperative Adversarial Decision Making Based on Deep Reinforcement Learning

Collaborative missile strikes play a crucial strategic role in beyond-visual-range aerial combat, yet with the advancement of technology, air combat scenarios have become increasingly complex. However, traditional methods such as Genetic algorithms and Matrix countermeasure method have the problems of insufficient information utilization and lack flexibility in handling complex combat situations. To better utilize environmental information, enhance the flexibility of missile maneuver decision-making and increase the lethality of dual missile strikes, this paper proposes a collaborative decision-making approach based on the Dueling DQN algorithm. We discretize the missile maneuver action space in three-dimensional space and design a smooth reward function based on factors such as the timing of attacks and approach angles, which guides intelligent agents to collaborate with non-intelligent agents to achieve maximum offensive advantage. The experimental results demonstrate that even when the enemy employs a fully random escape strategy, the Dueling DQN algorithm can still converge rapidly and outperforms the DQN and DDQN algorithms in addressing aerial combat problems.