Abstract—The steam turbine generator faults not only damage the generator itself, but also cause outages and loss of profits. The traditional fault diagnosis systems care only about high diagnosis accuracy. But different misdiagnoses may lead to quite different losses and it is unreliable if misdiagnoses were accepted. In order to reduce the total loss caused by misdiagnoses and improve the diagnosis reliability, in this paper, cost integrated multiclass SVM with reject option (CIMCR-SVM) is proposed. Firstly, we present a very simple and effective method to make the multi-class classifiers cost-sensitive. Secondly, diagnosis reliabilities were evaluated by a reliability evaluator, and reject option is introduced for rejecting classified samples with lower diagnosis reliabilities. Experimental results demonstrate that CIMCR-SVM is able to minimize the average cost and improve the diagnosis reliability.
Index Terms—SVM, multi-class, cost-sensitive, fault diagnosis, reject option.
Chao Zou is with the China Jiliang University, Hangzhou, CO 310018 P. R. China (e-mail: zouc@cjlu.edu.cn).
En-hui Zheng, corresponding author of this paper, is with the China Jiliang University, Hangzhou, CO 310018 P. R. China (phone: +86-571-86914549 5; e-mail: ehzheng@cjlu.edu.cn).
Hong-wei Xu is with the China Jiliang University, Hangzhou, CO 310018 P. R. China (e-mail: xhw@cjlu.edu.cn).
Le Chen is with the China Jiliang University, Hangzhou, CO 310018 P. R. China (e-mail: cl7788@126.com).
Cite: Chao Zou, En-hui Zheng, Hong-wei Xu, Le Chen, "Cost-sensitive Multi-class SVM with Reject Option: A Method for Steam Turbine Generator Fault Diagnosis," International Journal of Computer Theory and Engineering vol. 3, no. 1, pp. 77-83, 2011.
Copyright © 2008-2024. International Association of Computer Science and Information Technology. All rights reserved.
