Research And Implementation Of Agent Decision Making In Short Track Speed Skating Simulation System

Decision-making of agent in simulation system is a very important problem in the fields of Artificial Intelligence. A lot of research findings of Artificial Intelligence can be used in the agent's controlling of the different simulation systems, which has received an excellent performance. But, at the same time, as the different simulation systems have the different characteristics, some of which demand real-time and some cares about the predicting, so, we choose the different artificial intelligence methods according to the different simulation systems, and do some improvements.Firstly, according to the features that the short track simulation system is more commending real-time, we choose the reactive-agent system as the framework of the agent's decision-making process. And we also carried out the mathematical description model through the abstracting of Markov decision process, which greatly make the chosen of the method in agent decision-making process more convenient.According to the design of decision-making model, decision tree is firstly chosen as the method of logical reasoning in the agent's decision-making process. Meantime we introduce tactics to require the different performance of agents.Concerning that the simulation system demand the agent can simulation some special athletes, we choose artificial neural networks to carry out the difference of decision-making. The program firstly obtains the samples by the way of artificial controlling, and secondly obtains the weight of neural networks through the training of samples. Finally the agents can reflect the difference of decision-making through the difference of the weight.The method of decision tree and artificial neural networks are applied to the decision-making process of the agent, and the result is very good. Firstly the agent can finish the match under the demand of no-out and no-fault, secondly the decision of the agent has obviously different, and finally the agent has much more challenged when contesting with the users.