The Research Of Sports Game Decision-making System Based On BP Neural Network

Game AI mainly researches on how to apply the creatures in game with intelligent behavior, conceptual work and other characters as human. Sustainable learning game AI is one of present research hotpots. In this thesis, the main contents of this research are as follows:1) A design method of sports game decision-making system based on BP neural network has been proposed. According to the characteristics of the sports game, the ability of evolution learning is provided in the decision-making system by using BP neural network. At the same time, the training data of BP neural network has been gotten by processing the game data using clustering analysis, then achieve the target that let the decision-making system have the ability of sustainable learning.2) The structure of BP neural network has been determined aiming at the characteristics of sports game. The initial weight value, hidden layer and the selection of excitation function have been determined according to the characteristics of sports game. A 3v3 basketball game prototype has been designed as the test platform and the data source. The training data have been gotten from the data of player in game, and the experiment has been designed to compare the training speed and the memory consumption of different training algorithm, then the Fletcher Reeves conjugate gradient method has been determined as the training algorithm in this thesis.3) The training data processing method based on double cluster analysis has been proposed. Through the process of the statistic data by using cluster analysis, the inconsistency data of different tactic tendency in sports game has been separated and statistic data sets of consistently tactic tendency have been gotten.Then the real-time data that in statistic data has been clustered to optimize the distribution of training data. Final the experiment that using the data after double clustering to train the BP neural network has been designed to verify the effectiveness of the double clustering method through comparing with the original data in the convergence speed and prediction accuracy.4) The 3v3 basketball game decision-making system prototype has been realized. To verify the effectiveness of the method in this thesis, the sports game decision-making system has been used in 3v3 basketball game. This thesis realize the decision module based on BP neural network and the data processing module based on cluster analysis. The original data, the data after once clustering and the data after double clustering has been used to train the BP neural network,and the training result has been compared in the intelligent and identification of game AI. The conclusion of the research is that the decision-making system can separate and learn from the player data of different tactic tendencies, and the intelligent and identification of the sports game decision-making system has been improved.