基于核主元分析和动态递归模糊神经网络的软测量建模  

Soft Sensor Modeling Based on KPCA and DRFNN

作　　者：张瑶[1] 

机构地区：[1]南京工程学院工业中心,江苏南京211167

出　　处：《南京工程学院学报（自然科学版）》2013年第1期19-24,共6页Journal of Nanjing Institute of Technology(Natural Science Edition)

摘　　要：针对软测量建模过程中的误差数据剔除、特征提取,及模型的动态辨识问题,提出基于核主元分析和动态递归模糊神经网络软测量建模方法.首先,利用样本间马氏距离进行样本相似程度分析,去除样本中错误数据以确保数据质量;然后利用核主元分析提取系统的非线性主元,作为动态递归模糊神经网络的输入;最后利用新样本数据训练动态递归模糊神经网络.将该方法应用于赖氨酸发酵过程的产物浓度预测,仿真结果表明该方法具有较高的预测精度,满足现场测量要求.To address the problems associated with eliminating gross error data, feature extraction and dynamic identification of models, in software measurement modeling, this paper proposes a modeling method using kernel principal component analysis （KPCA） and dynamic recursive fuzzy neural network （DRFNN）. Firstly, to ensure the data quality, incorrect data in the samples is eliminated by means of sample similarity analysis using Mahalanobis distance between samples. Kernel principal component analysis is then carried out to extract the nonlinear principal components of the system, which is used as a dynamic recursive fuzzy neural network input. Finally, new sample data is presented to train dynamic recursive fuzzy neural network. This method is then applied to predict the production concentration of lysine fermentation process. The simulation results show that this method can meet the demand for spot measurement because of its high predicting accuracy.

关 键 词：马氏距离 核主元分析法 动态递归模糊神经网络(DRFNN) 软测量 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]