Research and Implementation of Chinese Chess Game Algorithm Based on Reinforcement Learning

Chinese chess is a two-person zero-sum game with complete knowledge like the widely used game Go and Chess. This paper designs a Chinese chess game algorithm based on reinforcement learning. It adopts a self-play learning model constructed by combining deep convolutional neural networks and Monte Carlo search tree algorithm, and uses Monte Carlo search tree algorithm to simulate chess moves that constantly change roles. The deep neural network improves the chess strategy, and trains the deep neural network in the reverse direction after winning or losing, so as to try to obtain a certain level of chess skill in Chinese chess through self-learning without initial training data. After a period of training, through the design and the traditional alpha-beta pruning algorithm and Monte Carlo algorithm comparison and its own competition analysis and other experiments, it is found that the model has strong self-learning ability, and has a good performance in Chinese chess. There is more room for improvement in chess power.