模糊神经网络的混沌优化算法设计

摘要

提出了一种基于混沌变量的多层模糊神经网络优化算法设计.离线优化部分采用混沌算法,将混沌变量引入到模糊神经网络结构和参数的优化搜索中,使整个网络处于动态混沌状态,根据性能指标在动态模糊神经网络中寻找较优的网络结构和参数.在线优化部分采用梯度下降法,把混沌搜索后得到的参数全局次优值作为梯度下降搜索的初始值,进一步调整模糊神经网络的参数,实现混沌粗搜索和梯度下降细搜索相结合的优化目的,能较快地找到全局最优解.最后对二阶延迟系统进行仿真,结果表明混沌优化方法控制精度高、超调小、响应快和鲁棒性强.%An optimization algorithm design based on chaotic variable is proposed for multilayer fuzzy neural network. Offline optimization uses chaos algorithm and chaos variables are applied to search for network structure and parameters, in which the network is in dynamic chaos state. An approximate optimal network structure and parameters are found from dynamic network according to performance index. On-line optimization uses gradient descent algorithm and the initial values of gradient descent searching are parameters approximately global optimal values from chaos searching, the parameters of fuzzy neural network are further adjusted. The global optimal values of network are searched quickly by means of combination of chaos global searching and gradient descent local searching. Finally, second order delay system is simulated, and the results show that the chaos optimal control is of high precision, small overshoot, fast response and good robustness.