Application of Deep Reinforcement Learning in Team Competitive Strategy Prediction

초록

This paper proposes a team competition strategy prediction method based on deep reinforcement learning. By constructing a multi-agent environment model and a hierarchical state representation mechanism, a dual-channel deep reinforcement learning architecture was designed, and the ResNet-50 strategy prediction network and value evaluation network were integrated. The improved Actor-Critic framework and priority experience playback are used for training optimization. Experiments show that the average prediction accuracy of the proposed method on the five benchmark sets is 85.3%, which is 15.6% higher than that of the existing methods. The prediction delay is controlled within 8.5ms, and the policy coverage rate is 92.7%, which is better than the traditional method in high-intensity confrontation scenarios. © 2024 IEEE.

키워드