Incomplete Information Machine Game Based On CNN And MCTS

Since the concept of artificial intelligence was put forward,machine game has been one of the most challenging research directions.Machine game is divided into complete information machine game and incomplete information machine game.The characteristic of incomplete information machine game is that the agent cannot obtain all situation information in the game process.Many decision-making problems in the real world can be abstracted into incomplete information game problems,such as airport planning,network security,financial and energy issues.Therefore,the study of machine games with incomplete information is of great practical significance.The traditional methods to solve the problem of incomplete information machine game are partial observable markov decision process model and reinforcement learning algorithm.However,reinforcement learning cannot converge under the condition of incomplete information and high-dimensional state space.It is impossible to traverse all states only through limited data and repeated tests.In this paper,a deep learning network model is proposed to solve the problem of large state space in some incomplete machine game problems.Due to the introduction of human experience,the model based on neural network can simulate human cooperation in game.In this paper,an incomplete information game method based on monte carlo tree search and simple risk model is proposed.In this paper,the deep learning network is used to replace the state-action value function in reinforcement learning to solve the problem that reinforcement learning cannot converge in the high-dimensional state space.In this paper,the deep neural network and the improved deep reinforcement learning algorithm are applied to the machine game with incomplete information,and a two-play onepoker machine game system is realized.Compared with traditional learning algorithms,learning strategies from perception to action end to end reach a higher game level.The improved deep reinforcement learning provides a feasible method for the realization of largescale machine game system and also provides a possibility for extending it to real life.