摘要
针对现有径向基函数(RBF)神经网络训练算法在非线性动态系统大范围辨识中的不足,借鉴免疫原理,提出了一种新颖的RBF神经网络在线学习算法,通过分析RBF神经网络学习过程和免疫系统的相似性,采用免疫记忆、克隆选择、扩增和细胞凋亡机制在线动态调节网络隐层节点,并确定相应的数据中心和宽度,从而使网络具有在线学习和记忆新样本的功能,并将该网络应用于某300MW火电机组主汽压的多工况辨识.实验结果表明该算法不仅能精简网络的结构,而且能很好地适应对象的时变特性.
A novel online training algorithm＇ for RBF neural network based on immune principles is presented to overcome the weaknesses of the traditional neural network training algorithm in the long time-period identification of non-linear dynamical system. With the analysis of the comparability between the learning of RBF neural network and immune system, this algorithm uses immune memory, clone selection and cell languish mechanisms to adjust the nodes in hidden layer dynamically, and determine their core function＇s center and width. As a result, the RBF neural network could have the function of on-line learning and memory of new samples. The improved neural network is then used for modeling the main steam pressure of 300MW thermal power unit with varying operating region, and the results show that the new algorithm can not only simplify the structure of neural network but also adapt the time-varying character of dynamical system effectively.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2006年第9期14-19,共6页
Proceedings of the CSEE
基金
国家自然科学基金项目(50576011)
教育部高等学校博士学科点专项科研基金(20020286001)~~
关键词
热工过程
径向基函数神经网络
在线学习
免疫原理
系统辨识
thermal process
radial basis function neuralnetwork
on-line learning
immunity principle
systemidentification
