Multiscale water quality contamination events detection based on sensitive time scales reconstruction

With the help of statistical technology and artificial intelligence algorithms, online water quality monitoring and detecting have significant importance to national water security. This paper proposed amulti-scale and multivariate water quality event detection approach for detecting accidental or intentional water contamination events. The approach is based on the ensemble empirical mode decomposition (EEMD), which is a novel algorithm for the analysis of nonstationary and nonlinear data of the type used in this paper. With EEMD as a dyadic filter bank, original water quality time series are decomposed into a sequence of intrinsic mode functions (IMFs). The local time scale is an important feature for statistical analysis and multi-scale representation. In this paper, the fluctuation characteristic for newly available measurements is estimated dynamically, and the corresponding membership degree to the constructed time scale reference which depends on offline long-term normal data analysis is calculated with Gaussian fuzzy logic. Taking the various membership as weight values, the anomalous signal can be enhanced and sifted out by the selection and reconstruction of sensitive time scales. Compared with traditional water quality detection methods with receiver operating characteristic (ROC) curves, the proposed multi-scale method can improve the detection accuracy and reduce the false rate.