980 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 61, NO. 4, APRIL 2012 Abrupt Event Monitoring for Water Environment System Based on KPCA and SVM Jianjun Ni, Member, IEEE, Chuanbiao Zhang, Li Ren, and Simon X. Yang, Senior Member, IEEE Abstract—The abrupt event monitoring is a challenging and critical issue in water environment systems. There are two main different abrupt events in the monitoring system, namely, the emergency water pollution accident and the abrupt sensor fault. The two different abrupt events have similar data characteristics, and few methods can be used to recognize the events. In this paper, a novel abrupt event monitoring approach based on kernel prin- cipal component analysis (KPCA) and support vector machines is proposed, which is combined with the physical redundancy method. The trust mechanism is introduced into the proposed approach to reduce the interference of external noise and improve the performance of quick response for the abrupt events. A spare data area is set up to store the data for the KPCA modeling. The data in the spare data area are updated continuously, and the KPCA model is updated subsequently to improve the adaptivity of the KPCA model for the abrupt event monitoring. The experimen- tal results show that the proposed approach is capable of detecting and recognizing the two different abrupt events efficiently. Index Terms—Abrupt event, kernel principal component (PC) analysis (KPCA), support vector machines (SVMs), water environ- ment system. I. I NTRODUCTION T HE FREQUENT occurrences of water pollution accidents have attracted more and more attention in the whole human society, so it is very important to monitor the water environment automatically by sensor systems [1], [2]. Because of the complexity of water environment, there are two main different abrupt events in the water environment sensor system, namely, the emergency water pollution accident and the abrupt sensor fault. The emergency water pollution is mostly caused by discharging the polluted water into the water environment directly without any treatments. However, the abrupt sensor fault is one type of sensor fault, which is mostly caused by mon- Manuscript received May 17, 2011; revised August 27, 2011; accepted September 30, 2011. Date of publication November 22, 2011; date of current version March 9, 2012. This work was supported in part by the National Natural Science Foundation of China under Grant 61074056, by the Open Fund of Jiangsu Key Laboratory of Power Transmission and Distribution Equipment Technology under Grant 2010JSSPD02, by the Hohai University Innovation Foundation under Grant XZX/09B002-02, and by the Fundamental Research Funds for the Central Universities under Grant 2011B04614. The Associate Editor coordinating the review process for this paper was Dr. V. R. Singh. J. Ni and C. Zhang are with the College of Computer and Informa- tion, Hohai University, Changzhou 213022, China (e-mail: nijj@hhuc.edu.cn; zhangcbhhuc@163.com). L. Ren is with the College of Hydrology and Water Resources, Hohai University, Nanjing 210098, China (e-mail: renli@hhu.edu.cn). S. X. Yang is with the Advanced Robotics and Intelligent Systems Labo- ratory, School of Engineering, University of Guelph, Guelph, ON N1G 2W1, Canada (e-mail: syang@uoguelph.ca). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIM.2011.2173000 itoring objects (such as the heavy corrosive pollutants in the polluted water) or the design flaws. If it is an emergency water pollution accident, it should be detected and dealt with timely. Otherwise, it will lead to serious consequences, such as an increase of fish death, shortage of drinking water, and economic losses. If it is an abrupt sensor fault, the fault sensor should be detected and replaced timely to keep the sensor system working normally. The treatments of these events are different, so how to monitor the abrupt event timely and accurately is a challenging and critical issue in water environment sensor systems. Various methods have been proposed to deal with the sensor fault problem [3], [4]. The sensor faults can be classified into drift faults and abrupt faults. Because the abrupt faults may cause much more serious results than drift faults, research has focused on the abrupt sensor fault detection and isolation recently. For example, Samara et al. [5] proposed a statistical method for sensor abrupt faults. Oosterom et al. [6] devel- oped a sensor management system based on soft computing techniques, and the abrupt fault of sensor was analyzed by using the majority voting concept. Zhang and Yan [7] proposed a wavelet-based approach to the abrupt fault detection and diagnosis of sensors. The problem of monitoring water pollution accidents is a focus recently. Reddy et al. [8] proposed a mathematical model to continuously monitor the impurities that are present in water. Zhao et al. [9] proposed a novel optical fiber sensor for remote monitoring of salinity in water. Zhu et al. [10] discussed some roles for emergency pollution accident in drink- ing water sources, including monitoring parameters, methods, and procedures. Kunkel et al. [11] used the interdisciplinary model network REGFLUD to predict the actual mean nitrate concentration in percolation water at the scale of the Weser river basin (Germany). There is much research on monitoring the abrupt sensor fault or the emergency water pollution accident, but few consid- ered the two problems together. The change characteristics of the measured data during the two different abrupt events are similar, so conventional methods cannot recognize the abrupt event efficiently. There are two main tasks in the abrupt event monitoring. One task is that the abnormality in the sensor system should be detected quickly and accurately. The other one is that the type of abrupt events should be recognized correctly as soon as the abnormality is detected. Although those methods introduced earlier could deal with some problems in fault diagnoses, few of them could be used directly for the abrupt event monitoring of water environment systems. To accomplish the abrupt event monitoring task efficiently, a novel adaptive approach based on kernel principal component (PC) 0018-9456/$26.00 © 2011 IEEE