Design and Optimization of a Chinese Chess engine Based on MCTS

Abstract

Chinese chess, as part of Chinese traditional culture, presents unique challenges and value in the field of artificial intelligence research. This study focuses on innovatively applying the Monte Carlo Tree Search (MCTS) algorithm to the design of a Chinese chess engine, breaking through the limitations of traditional engines that rely on Alpha-Beta pruning and manual evaluation functions. By implementing a mixed strategy generation framework, the engine achieves decision-making and optimization in complex situations. Experiments show that the MCTS-based engine demonstrates significant tactical combination discovery capabilities during the middle game, with win rates for red and black pieces reaching 62.5% and 50%, respectively (against intermediate traditional engines), with critical decision response times controlled within 500ms. Its innovation lies in dynamically allocating computational resources, integrating opening library rules with MCTS random search, ensuring flexibility and the ability to handle complex rule requirements. The study validates the transfer potential of MCTS in non-complete information games, providing a reusable framework for AI development in traditional board games such as Chinese chess and Korean chess. Future work can further enhance the precision of handling complex situations by introducing reinforcement learning, promoting deep integration and innovation between artificial intelligence and traditional chess skills.