摘要
神经网络能够以任意精度逼近任意复杂的非线性关系,具有高度的自适应和自组织性,在解决高度非线性和严重不确定系统的控制方面具有巨大的潜力。但一般神经网络训练算法如BP算法训练速度慢,受初值影响大且易陷入局部极小点,该文提出了一种基于模糊神经网络的间接自校正控制系统,控制器以高斯隶属度函数的径向基函数(RBF)神经网络结构,利用改进的遗传算法(GA)对结构和参数进行同步优化,改进适应度函数指导搜索过程,在保证稳定情况下大大加快了收敛的速度。神经网络正向模型(NNP)利用弹性BP算法进行离线辨识,使得到的模型泛化性能好。
The artificial neural networks can approximate any non -linear function with any given precision. It has a high self - adaptability and self - organization, which endows the neural networks with large potential to solve the control of the system with high nonlinearity and serious uncertainty. Meanwhile, the currently used training algorithm for the neural networks such as BP is often inclined to local minimum, affected greatly by the initialization of the weights and has a low convergence velocity. An indirect self - adaptive fuzzy - neural network controller （FNNC） has been presented with its parameters and the structure tuned simultaneously by GA. The structure of the controller is based on the radical basis function （RBF） neural network with Gaussian membership functions. Dynamic crossover, mutation probabilistic rates as well as modified fitness function have been used for faster convergence. Flexible BP algorithm has been applied for neural network identification off - line of system forward model. Simulation results show that the FNNC presents encouraging advantages.
出处
《计算机仿真》
CSCD
2005年第9期136-139,共4页
Computer Simulation
关键词
非线性
不确定性
遗传算法
模糊神经网络控制器
径向基函数
神经网络辨识
Non -linear
Uncertainty
Genetic algorithm
Fuzzy neural network controller
Radical basis function
Neural network identiiication
