Research Of Computer Go Based On Expert System And Monte Carlo Method

Computer go game is an important direction for research in artificial intelligence, and it bringsgreat challenges to the computer game researchers. At present, the best go program is still in theamateur level, go is a good environment for testing the level of the development of artificialintelligence, so computer go research has important theoretical significance and practical value.In the era of traditional computer go, go programs use expert system, the methods include expertknowledge-based pattern recognition and static evaluation of the influence function.Recently, MonteCarlo method and UCT algorithm has been widely used in computer go game systems, UCTalgorithm brings new era in the modern computer go. Setting in9×9go, this paper researched andanalyzed computer go game principles and methods, and gave valuable attempt to joseki codingmethod and the improvement of UCT algorithm optimization.In the go layout stage, the effective method is matching joseki by pattern recognition to guidehand.Based on the summary of the advantages and disadvantages of the pattern method widely usedin computer go, design and implement EHPEM method, the method is designed by the improvedHuffman coding. It uses four32-bit integers to encode the entire board, and the advantage is codedconflict-free, efficient and accurate pattern matching.After entering the go disk stage, the change is very complex, and it is difficult to conductaccurate assessment of the situation.The UCT algorithm based on the Monte Carlo evaluationspecializes in fighting in the disk, and also endgame move. In this paper, we introduce the UCTalgorithm to the go disk stage, and improve UCT algorithm optimization, one is about the sort andexpansion of optional point, the other is applying parallel algorithms for UCT through the multi-coreand MPI. Experimental results show that the parallel computing to speed up the game tree searchspeed, improve playing strength, and verify the feasibility of the algorithm.Finally, we achieve a nine-way computer go system Doubleminggo by using the C/C++language,The system, which effectively combines the traditional pattern recognition methods with modern UCTalgorithm, shows high level artificial intelligence to achieve the desired effect.