A Study on Applying Genetic Algorithms to Games Search Trees

• Main Authors: Huang, Ke-Yuan, 黃科元
• Other Authors: Hong, Tzung-Pei
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/58395920091978456027

碩士 === 義守大學 === 資訊工程研究所 === 85 === 　　The computer chess game has become increasingly important to artificial intelligence researchers on showing the intelligence of machines. There are usually factors in improving the capabilities of a computer chess system-chess domain knowledge, search algorithm, and hardware used. The contents of chess domain knowledge depend on the king of chess played, and the speed of the hardware depends on the technical advance of computers. For traditional game search algorithms, the time spent and the space needed grow exponentially along with the search depth. The search is then usually not deep enough to have a good play due to the time and space constraints. 　　In this thesis. we attempt to apply genetic algorithms to game-tree search for raising the performance. Genetic algorithms are more and more used in recent years. They are mainly used to find solutions to a maximum or a minimum problem. They are based on Darwin's principle of the fittest surrival, and find an optimal of a nearly optimal solution in limited time. Traditional genetic algorithms cannot, however, solve minimax problems. 　　Three kinds of genetic game-tree algorithms are proposed, respectively for one-player, two-player, and multiple-player game trees. A genetic one-player game tree algorithm is first proposed, which uses a special encoding scheme to improve the search performance. Three genetic two-player scheme to improve the search performance. Three genetic two-player game tree algorithms, including the direct-choice, the minimax, and the tree-reservation approaches are then proposed. Among them, the tree-reservation genetic algorithm is the best. It uses the genetic evaluation values kept in the reservation tree to generate the offspring. After generations the best solution will be output as the next step to play. The tree-reservation genetic algorithm is then extended to solve the multiple-player game tree search problem. A genetic max-n algorithm is then proposed. 　　At last, experiments and analysis are made to show the effect of our proposed algorithms. Results show that the accuracy using our algorithms can be raised in a limited amount of time. Our proposed algorithms are thus practical and can be applied to real computer games.