基于改进的BP算法的RoboCup防守策略研究

RoboCup defensive strategy based on improved BP algorithm

传统的BP神经网络算法已被有效地应用于处理RoboCup中防守策略,但是它具有最速下降法收敛速度慢和易陷入局部极小的缺点。针对该问题提出了一种改进的BP算法,通过增加附加动量项对BP算法进行了改进,并将之应用于离线的防守学习。随后,在RoboCup环境中与传统的BP算法进行了比较,结果表明:该方法可有效提高收敛成功率。

Traditional BP neural network algorithm has been effectively applied to deal with the defensive strategy in RoboCup,but it has the defects of the steepest descent method such as being slow in convergence and being easy falling into the local minimum.Thus,a new BP algorithm was proposed by improving the traditional BP algorithm with the additional momentum term method,and then it was applied to the offline learning of defense.Finally,in the RoboCup environment,the improved BP algorithm was compared with the traditional one.The result shows that the algorithm can effectively improve the success rate of convergence.