摘要
针对BP算法在测向定位中收敛速度慢、易陷入局部极小等缺点,将模拟退火方法应用到BP神经网络中,同时结合变步长方法,利用隐层节点的动态合并与删除策略,在满足定位精度的同时使网络结构最小化,使用三层前馈网络建立了三站测向定位模型。通过仿真实验,新方法在收敛速度和有效性方面都远高于BP算法。
According to the BP algorithm for direction finding of slow convergence speed,easy to fall into local minimum and other shortcomings,simulated annealing algorithm is applied to BP neural network,combined with the variable step size method,by using the hidden layer nodes dynamically merged with deletion strategy,to meet the positioning accuracy and make the network structure minimized,using three layer feed forward neural network establish three station location model.Through the simulation,the new method in convergence speed and effectiveness are far higher than those of BP algorithm.
出处
《安阳工学院学报》
2012年第2期35-38,共4页
Journal of Anyang Institute of Technology
基金
安徽科技学院青年科研基金(No.ZRC2011270)
关键词
神经网络
测向定位
模拟退火
neural network
DOA location
simulated annealing
