基于差分进化算法的模糊神经网络控制器

A fuzzy neural network controller based on differential evolution algorithm

针对模糊神经网络控制器在应用误差反向传播算法训练时,易于陷入局部最优的问题,提出了一种将差分进化算法与BP算法相结合的学习法,首先利用差分进化算法的全局寻优能力,给BP算法一个好的寻优初始点;然后再以一定的概率进行BP算法的寻优。对一个带有滞后环节的二阶系统进仿真表明,控制性能优于基于BP的模糊神经网络控制器。

In order to avoid the local optimum resulted from back propagation algorithm, we present a novel training approach for fuzzy neural network controller, which combines differential evolution algorithm and BP algorithm. First, global optimization ability of DE is used to get good initial points for BP algorithm. Then BP algorithm is required at a certain probability, using its good local optimization ability. The approach is applied to a second order system with time delay. Simulation results indicate the efficiency of this approach.