Operational Patterns Recognition Using Multi-class Cost-sensitive Learning for Alumina Evaporation Process  

作　　者：TANG Mingzhu YANG Chunhua QING Jingjing GUI Weihua 

机构地区：[1]School of Information Science and Engineering, Central South University, Changsha 410083, China [2]School of Energy and Power Engineering, Changsha University of Science and Engineering, Changsha 410007, China

出　　处：《Chinese Journal of Electronics》2013年第2期282-286,共5页电子学报（英文版）

基　　金：This work is supported in part by the National Natural Science Foundation of China （No.60874069）, the National Science Fund for Distinguished Young Scholars of China （No.61025015）, and the Research Foundation of Education Bureau of Hunan Province, China （No.12A007）.

摘　　要：The characteristics of Alumina evaporat- ing process （AEP） are analyzed firstly. Operational pat- tern is defined to describe the AEP. A new framework is proposed, which formulates the operational pattern recog- nition problem as a multi-class class-imbalanced problem of unequal misclassifieation costs. Aim to the multi- class class-imbalanced problem of unequal misclassifica- tion costs in the operational pattern set of AEP, a multl- class cost-sensitlve Probabilistic neural network （mc-PNN） method is proposed. A spectral clustering based on Bal- anced iterative reducing and clustering using hierarchies （BIRCH） is used to optimize the number of pattern layer nodes of me-PNN. Experimental results show that the pro- posed method reduces effectively average miselassification costs and increases recognition rate of excellent and faulty classes.

关 键 词：Cost-sensitive probabilistic neural net- work Operational pattern  Alumina evaporating process  Class-imbalanced problem Unequal misclassification costs. 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TQ028.61[自动化与计算机技术—计算机科学与技术]