基于改进BP算法的RoboCup传球策略研究

Study of passing strategy in RoboCup 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中传球策略,但是存在最速下降法收敛速度慢和易陷入局部极小的缺点。提出一种改进的BP算法,增加了附加动量项的方法对BP算法进行了改进,将之应用于离线的传球学习。最后在RoboCup环境中与传统的BP算法进行了比较,结果表明该改进算法有效地提高了收敛成功率。

Traditional BP neural network algorithm has been effectively applied to deal with passing strategy in RoboCup, but it has the shortcomings of low speed with the convergence of the steepest descent method and easy local minimization. The paper proposes an improved BP algorithm, with the additional momentum term method, BP algorithm is improved, and is applied to offline learning. Finally, in the Robo- Cup environment the traditional BP algorithm are comoared. It Droves the convergence rate of success.