The Research And Application Of Bayesian Networks In Fencing Training Decision Support

Fencing is an important item in Olympic Games, and China Fencing Team achieved great compliments in 2008 Olympic Games. In fencing training, the training plan should adjust according to athletes'physical condition. The data of physical state is measured and collected periodically, and feed back to the making of training plan by coaches. Although, it is revealed that the training course is dominated by coaches'experience instead of science of sport. So, efficient decision making assistance is imperative.In information age, the fields of expert system, data mining, and decision making is an efficient tool to manage and analyze information. Bayesian networks, a popular kind of them, have been used in different application including diagnosis, medical expert system, and software debugging. Bayesian networks are based on directed graphs. It has connections with the statistical, neural networks, and uncertainty communities.Consider Bayesian networks on discrete variables; they consist of a network structure and its associated conditional probability tables. The network structure is represented by a Directed Acyclic Graph (DAG), which simplifies the full joint probability distribution for a set of variables and show independencies between the variables. Conditional probability tables are needed to specify a probability distribution based on the network. In this paper, basic concepts for learning and Bayesian networks are introduced and methods are then reviewed. Methods are discussed for learning the structure and parameters of a probability network.In order to improve the quality of training, the system of fencing training decision making is developed. Unlike the previous research, which focused on the relation between load and physiological parameters, in this project, the relation between training courses and physiological parameters are emphasized. An inference model based on Bayesian network to discover the relation in optimization of fencing-training was proposed. By combining experts'experience, this paper presented how to construct network structures and configure parameters, applied the sample data concerning women's epee in the experiments, and illustrated the performance-comparisons with BP artificial neural network. Besides, the BP algorithm we used is improved by add a momentum item. The results show that this model is helpful in coaches'decision-making.