Research on the Generalization Ability of Reinforcement Transfer Learning Based on Self Game

In order to improve the generalization ability of reinforcement learning algorithm, a transfer learning training method integrating self game is proposed. This method can enable both sides to jointly optimize a strategy, share the experience pool, and use the transfer learning algorithm to continuously migrate the trained model to a more complex air combat environment. Finally, the trained model can be applied to a variety of air combat environments. This method not only accelerates the training speed of the strategy, improves decision performance, but also makes the samples more diverse, allowing the decision model to learn more air combat knowledge and improve generalization ability.