东南大学学报（自然科学版）2013，Vol.43Issue(2)：322-327,6.DOI:10.3969/j.issn.1001-0505.2013.02.018
基于混合型鲁棒输入训练神经网络的非线性数据校正方法及其应用
Nonlinear data correction method and its application based on hybrid robust input-training neural network
摘要
Abstract
A nonlinear data correction model, hybrid robust modified input-training neural network model, is presented by integrating process mechanism constraint equations into the neutral network model. In the network structure of proposed model, constraint equations are included in the objective function of network training by penalty function method. The network learning method is derived based on BP(back propagation) algorithm, and the algorithm steps are also given in this paper. The case study is conducted to detect and validate some data sets measured from and 1# high pressure heater in a 1 000 MW unit and a five-dimension nonlinear system, and the result indicates the validity and robustness of the proposed model, which can ensure the accuracy and reliability of data correction in case of multi-point fault under system mechanism constraints.关键词
混合型鲁棒输入训练神经网络/故障诊断/机理约束/罚函数/数据校正Key words
hybrid robust input-training neural network/ fault diagnosis/ mechanism constraints/penalty function/ data correction分类
信息技术与安全科学引用本文复制引用
任少君,司风琪,李欢欢,徐治皋..基于混合型鲁棒输入训练神经网络的非线性数据校正方法及其应用[J].东南大学学报（自然科学版）,2013,43(2):322-327,6.基金项目
国家自然科学基金资助项目(51176030). （51176030）
