Research And Implementation Of Human Computer Game Algorithm Optimization Of Gobang

Computer game,also known as machine game,is the product of the combination of game theory and computer technology,and is an important research direction in the field of artificial intelligence.At the same time,it is also the research foundation and experimental field of many artificial intelligence application fields,such as machine intelligence,military chess deduction,intelligent decision-making system,etc,In recent years,with the deepening of the research of artificial intelligence technology represented by various deep learning algorithms,especially the great success of the machine game system represented by alphago,further promote the rapid development of machine game theory and related technologies.Modern machine game research is mainly about go,chess,Gobang and so on.The computer game research of chess has the longest history,and has experienced a magnificent"fight",This paper summarizes a set of core technical points about the process modeling,state representation,movement generation,chess game evaluation,game tree search,opening and remaining database development,system testing and parameter optimization of chess machine game,and determines the research direction for the follow-up related research.As one of the most popular chess categories in our life,Gobang has all the characteristics of simple rules and typical zero sum complete information game.It is convenient to carry out in-depth research and research can reflect the advantages and disadvantages of the game algorithm to a certain extent,In this paper,the game tree search and system self-learning ability training are studied for the zero sum complete info rmation game.The main research contents and achievements are as follows:(1)In the research of game tree search algorithm,through the combination of UCB algorithm and Monte Carlo algorithm to solve the multi arm bandit problem,the applicability of uct algorithm in the game problem is analyzed,as well as some shortcomings of traditional uct algorithm,and the corresponding improved algorithm is,proposed.On the one hand,combining the pruning method to determine the pruning conditions,to a certain extent,reduce the search space of the game.On the other hand,the replacement table algorithm is introduced,which combines the replacement table with the uct algorithm to store the historical node states,(2)In the aspect of the realization of self-learning ability,firstly,the whole framework of traditional artificial neural network is studied and analyzed,and the convolutional neural network used in this project is determined.Then,by comparing the characteristics and performance of several architectures,combined with the basic model architecture,the internal structure of strategic value network model required by the project is designed.Based on the characteristics of Gobang,the method of generating and expanding data set is planned,and the back propagation algorithm in training is optimized with the additional momentum algorithm.(3)Combined with the above research results,the prototype design and preliminary implementation of the Gobang game system are carried out.The experimental results show that the proposed method is effective in improving the search speed of the machine game and enhancing the self-learning ability of the system.