Online fault detection methods for chillers combining extended kalman filter and recursive one-class SVM

详细信息    查看全文

• 作者：Ke Yan^aAuthor Vitae ; Zhiwei Ji^b ; ^{jzw18@hotmail.com}Author Vitae ; Wen Shen^cAuthor Vitae
• 关键词：Chiller ; Fault detection ; One-class support vector machine ; Extended kalman filter
• 刊名：Neurocomputing
• 出版年：2017
• 出版时间：8 March 2017
• 年：2017
• 卷：228
• 期：Complete
• 页码：205-212
• 全文大小：608 K
• 卷排序：228

文摘

Automatic, accurate and online fault detection of heating ventilation air conditioning (HVAC) subsystems, such as chillers, is highly demanded in building management system (BMS) to prevent energy waste and high maintenance cost. However, most fault detection techniques require rich faulty training data which is usually unavailable. In this study, a novel hybrid method is proposed to detect faults for chiller subsystems without any faulty training data available, i.e. by training the normal data only. A hybrid feature selection algorithm is applied to the chiller dataset collected by ASHRAE project 1043-RP to select the most significant feature variables. An online classification framework is introduced by combining an Extended Kalman Filter (EKF) model and a recursive one-class support vector machine (ROSVM). Experiment results show that the proposing algorithm detects typical chiller faults with high accuracy rates and requires less feature variables compared to existing works.