摘要
针对液压伺服系统动态流量软测量模型中神经网络训练精度和训练速度难以同时提升的问题,引入减聚类(SCM)算法将原训练样本集映射成初始径向基函数(RBF)中心集,并确定基函数宽度;利用敏感性分析算法(SenV)对基函数的中心进行优化,从而减少神经网络隐层节点数目;在根本上为同时提升神经网络训练精度和训练速度提供保障。实验表明,神经网络的隐层节点数可降低至少30%。
It is a complex problem for neural network to obtain a quick and exact training result synchronously in the measurement of dynamic flow in hydraulic pressure servo systems. To solve this problem, the subtractive clustering algorithm is introduced. In this method, the initial centers of their radial basis functions are selected and then the widths of RBFs are determined. On the basis of subtractive clustering, sensitivity analysis is applied to optimize the centers of RBFs, thus the number of hidden neurons is decreased. Using the optimization algorithm, the training speed and precision can be improved synchronously. Compared with the sensitivity analysis algorithm, experiment results show that the new one can decrease the hidden nodes at least by 30%.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2009年第12期2619-2623,共5页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(60970073)
河北省自然科学基金(F2006000267)资助项目
关键词
动态流量
软测量
径向基函数神经网络
减聚类算法
敏感性分析
dynamic flow
soft measurement
radial basis function neural network
subtractive clustering method
sensitivity analysis