A Tactical Decision-making Method for Short Track Speed Skating based on Deep Reinforcement Learning

Inhalt:
Short track speed skating is a sport that has attracted much attention at the Winter Olympics. In addition to athlete's physical level and physiological state, competition performance is more easily affected by athlete's tactical level. An athlete with excellent tactical decision-making ability tends to have better skating performance. Traditional methods mainly use decision trees and genetic algorithms to learn tactical strategies. However, these methods often require complex rule design, and the learned tactical strategies are limited by training data and are not always effective. Therefore, we propose a novel tactical decision-making method for short track speed skating based on deep reinforcement learning. Specifically, we build a short track speed skating simulation competition environment through detailed analysis and research. Further, we adopt the deep reinforcement learning algorithm to learn and explore the agent's tactical strategy. In addition, we design a state representation network to effectively model the state features of the agent. Based on real short-track speed skating competition data, we used agents instead of athletes to conduct competitions in the simulation environment. Tactical strategies learned by the agent can significantly improve competition performance. The experimental results verify the effectiveness of our method.