Research On Water Quality Anomaly Detection Using Multi-sensor Data Fusion

Water quality events detection has been an active research topic in the field of drinking water insurance for many years; however, it hasn’t been widely applied in practice due to some issues. The issues include high false alarm rate, limited types of contamination events the approach can detect, and that such approach can’t perform real-time anomaly detection. Rapid detection of water quality events within a water quality early warning system is desirable for the protection of drinking water against both accidental and malevolent contamination events. The key to this advanced detection lies in anomalous water quality detection algorithms.Base on data fusion, two methods which combine online water-quality data and other information have been proposed to improve the detection of water-contamination events in this thesis. A large number of simulation studies and experimental analysis have been implemented to test the performance of the proposed algorithms, and the initial development of the matched water quality events detection software has been completed.The main contents and innovative points are summarized as follows:1. Using existing data from in-situ water quality sensors, a multi-parameters fusion algorithm including AR model and Fuzzy C-Means has been presented. Prediction deviations are obtained within AR and then classified into two clusters within FCM. To test this algorithm, the distance between a new prediction deviation and the anomalous cluster center are calculated and compared to a constant threshold. Results show that this fusion algorithm produces the lowest false alarm rate and highest detection rate for all cases of simulated event strength, which has better performance than single-parameter algorithm, the fusion algorithm base on Euclidean distance and algorithms fused water quality data directly.2. A large number of real water-quality contamination events data is needed for FCM model training which is difficult to obtain. Therefore, a water-contamination event detector based on an extension and improvement of the Dempster-Shafer (D-S) evidence theory has been proposed. The extended D-S theory for detection of water-contamination events relies on the times series of residuals of water-quality parameters predictions and the use of weighted-averaging and time-dimension information to resolve conflicts or ambiguities that arise when attempting to detect water-contamination events. First, an objective probability distributive function is given to change the AR prediction deviations to anomaly probability series, and incorporated weighted average method and time dimension features for evidence conflict resolution. And then the detector fuses multiple anomaly probability series and compares to a constant threshold to decide whether the current water quality is normal. Also, some detailed performance comparison such as advantages and applicability between the extended D-S evidence fusion method and FCM clustering method has been analyzed.3. Contaminant injection in water, transport, monitoring, and detection testing were conducted in an experimental pipe network to test the performance of the two data fusion methods in detecting water-contamination events. Two toxic compounds with potassium ferricyanide and ferric ammonium sulfate were tested under a series of different concentrations. Experimental tests demonstrated that the fusion methods achieve a low false alarm rate and high probabilities in detecting water contamination events.4. Basic functional modules of the matched water quality events detection software have been developed base on VC++and MATLAB, including detailed description of software framework design, data flow diagrams, main function algorithms, and main software interface.In general, the proposed two fusion methods in this study can detect most of the water quality events, which are suitable for routine monitoring of water quality early warning system.