基于蚁群BP神经网络的动态测量误差预测

针对BP算法在神经网络参数学习中局部性能好但易陷入局部极小值而蚁群算法拥有良好的全局性能的特点,提出了神经网络的综合训练方法(蚁群BP算法,ACBP Algorithm)即首先利用蚁群算法对神经网络参数进行全局搜索训练,然后再利用BP算法对参数进行局部学习。设计了一种自适应蚁群算法,一定程度上解决了传统蚁群算法收敛速度和拓宽搜索空间之间的矛盾。最后利用仿真实例验证了ACBP神经网络较之BP神经网络和AC神经网络在动态误差预测方面的优越性,得出了相关结论。

Forecast of dynamic measurement error based on ant colony BP neural network

While training neural network,the BP algorithm has good local performance but it is easy to fall into local minimum,and the ant colony algorithm has good global performance,so the following combinatorial method (ACBP algorithm) is put forward.First of all,the neural network is trained based on ant colony algorithm in global space,then,the parameters of neural network is trained using BP algorithm in local space.A kind of self-adjusting ant colony algorithm is designed.The new method eases up the contradiction between convergence speed and expanding searching space in traditional ant colony algorithm.At last,that the ACBP neural network outperforms BP neural network and AC neural network in the aspect of dynamic error forecast is verified by computer emulation example,and related conclusions are given.