Research On Water Quality Abnormal Detection Based On Dynamic Incidence Matrix In Distributed System

Water quality abnormal detection is the key of water quality detection system in urban distributed system. The purpose is detecting possible water abnormal fluctuation in distributed system quickly and actually. Traditional water quality abnormal detection methods usually analyzed water monitoring data in time domain or frequency domain. With the development of water monitoring technology. water monitoring data turned to be diversification. Traditional methods were difficult to take advantage of the inner relationships among multidimensional data to enhance accuracy of water quality abnormal detection methods. Concerning this issue, a new water quality abnormal detection method is proposed based on multiple water quality parameters with correlation analysis of multidimensional data. Furthermore, the method is expanded to solve multiple sites monitoring data. Lots of experiments and simulations are brought in to test the method, and possible problems of method are discussed.The main contents and innovative points are summarized as follows.(1)Most of present water quality abnormal detection methods could not deal with the relationships among water quality parameters, therefore, expected result is difficult to obtain. In this paper, a method of water quality abnormal detection based on dynamic incidence matrix is presented. On-line detection of water quality in distributed system is realized through relationship analysis among time series of water quality parameters with dynamic time warping algorithm and fluctuation of water quality parameters. Also, water quality monitoring data with different sample intervals is used to discuss the influences which are brought in by background change and baseline drift.(2)With the help of feature extraction based on dynamic incidence matrix, water quality abnormal fluctuation is analyzed further. In virtue of random forest algorithm, classification of water quality abnormal fluctuations which are caused by different contaminants is realized. In the cases where contaminant is known, accuracy of water quality abnormal detection could be enhanced with the classification of water quality abnormal fluctuations. Moreover, with the help of probability distributions on weak classifiers of abnormal fluctuation samples, the influences of the method which are caused by operating under working conditions are partly weakened.(3)Patterning of the method with multiple water quality parameters on a monitoring site, similar method is presented to solve water quality abnormal detection on multiple monitoring sites. Accuracy of the method is improved with redundant data from multiple monitoring sites. In addition. the method is tested through simulations on software and measured data from distributed system in Hangzhou, and result is compared with other traditional algorithms.The method presented in this paper could analyze the relationships among multidimensional water quality data better, and more accurate result can be obtained with feature extraction of correlated behaviors.