Research On The Enhancement Of UCT Algorithm Based On Global Static Evaluation In Computer Go

Computer game is an important area of artificial intelligence research, chess game is an important branch of computer game. For now, other chess games have been satisfactorily resolved, only the highest level of computer Go is worse than the primary human players.Go can test the level of development of artificial intelligence, developing a computer Go program which can reach a considerable degree of professional players will help to explore human cognitive science. Making an efficient static evaluation for the current game can not only assist players to judge their situation but also decide the point of attack and defense. Furthermore, it will lead the next direction and help UCT to find a balance between explore and exploit. This paper builds a quantitative model based on a new influence function to solve the static evaluation problem, after that,the result and UCT can be combined to construct a pruning method for UCT. This article’s algorithm can promote the efficiency of the program and improve the Go level, it plays an important role in theory and practical use.The main achievements are as follows:A new influence function called Influ is designed. Most of the past influence function didn’t consider the other pieces’ obstruction, which results in a low accuracy.In this case, linear distance and diagonal distance can be used to calculate between the pieces and spots. Besides, A* algorithm is used to find the optimal path. So finally the value of a single piece to one spot can be obtained.A quantitative model called Value is built based on the influence function. First of All, a linear superposition about the value of singer piece to one spot is made. Then some edge angle adjustments, Rise Drop optimization algorithm and Bisection optimization algorithm are used to gain the value of all the spots above the chessboard.Finally, Multi species compete die out method can be applied to optimize the parameters, so can be determined.The experiments show that the quantitative model called Value is very close to the judgment of professional players. Its accuracy can be applied for the opening game and middle game. It also lies a foundation for the following pruning strategy.The result of static evaluation and UCT can be combined to construct a pruningmethod for the UCT algorithm. To reduce the search depth and increase the number of Monte Carlo simulations, the value can be used as an evaluation standard for the nodes, so finally the search can be leaded to a meaningful direction.The experimental results show that the pruning strategy can improve the search efficiency, it can also contributes to improve the program’s level.