Optimization of Jamming Type Selection for Countering Multifunction Radar Based on Generative Adversarial Imitation Learning

In recent years, deep reinforcement learning (DRL) has made some progress in jamming type selection (JTS). However, during the training process of the agent, exploration of the action space is necessary, which leads to poor jamming effects in the early stages of training, posing a significant threat to the aircraft’s survivability. Therefore, this paper proposes a method of JTS based on generative adversarial imitation learning (GAIL). Firstly, the agent learns from expert strategies to achieve high reward returns, avoiding wasted time from unguided exploration, thereby ensuring that the agent maintains a good jamming effect throughout its application process. Secondly, based on generative adversarial theory, the discriminator measures the difference between the generated and expert strategies. This difference is used as an internal reward to assist in updating the neural network parameters, effectively reducing the complexity of reward function design. Finally, through case analysis, it can be observed that using the GAIL algorithm can achieve rewards close to those of the expert strategy. When applied online, it does not rely on accurate predictions or precise modeling of the external environment, allowing for quick real-time decision-making. Additionally, its performance surpasses that of traditional JTS strategies.