Research On Water Quality Abnormal Detection Based On Time And Spatial Correlation Analysis In Distribution System

As water quality poses significant impact on people’s health, it is of great importance to build up an online monitoring system for the urban water supply pipeline system. The key part of this system is to detect the abnormal fluctuation in the monitored water quality parameters and distinguish water quality events caused by accidentally leaking or deliberately dosing from routine hydraulic changes. With the help of the accurate detection of the abnormal changes, the contamination events can be found and controlled in time, providing a safe guard to the daily water supply.With the development of detection technology and distributed sensor networks, it appears to be a trend of diversification and spatialization in the research with respect to water quality monitoring. Traditional abnormal detection methods usually fail to explore the relationships between multi-dimensional data. In this paper, we proposed a new abnormal detection method based on correlation analysis on multivariate observed data collected from multiple monitor sites. The laboratory experiments and simulation results have indicated the effectiveness of the proposed algorithms.The main contents and innovative points are summarized as follows:1. Most of the existing water quality abnormal detection methods depend on the volatility analysis of every observed time series. The performance of detection will be easily affected by routine hydraulic fluctuation and signal transmission absence. To deal with this problem, a research on the detrended fluctuation analysis with a moving slide window has been implemented to the water quality monitoring system. Self-correlation analysis is applied to every detrended time series and then fused on D-S theory to obtain the multivariate abnormal probability. The proposed method is tested with experiment contaminant intrusion data collected from laboratory water supply system. In addition, the influence of slide window size on the proposed method is considered and will be further discussed afterwards.2. More than one observed parameters will show response to the injection of contaminant, and corresponding variations indicate strong connection among all the parameters. Taking this into consideration, we bring in longest common subsequence as a measurement of the correlation between each pair of compared time series and build up the mutual correlation matrix. The accuracy of abnormal detection could be enhanced with the fused probability of both detrended self-correlation and mutual correlation. Meanwhile, to minimize the influence of slide window size, an adaptive adjustment method has been brought in to find a better balance between failure reports and false alarms. The benefits of introducing the mutual correlation have been proved through laboratory contamination experiments.3. Urban water supply system usually covers a large spatial span. Time and spatial correlations can be found among monitored parameters from different sites along the topological structure of the pipeline networks. An abnormal detection method based on multi-site time and spatial correlation analysis is proposed to improve the detection performance of the whole network. Topology diagram and Bayesian networks has been introduced to procure a prior probability for the downstream sites based on the monitoring results obtained from upstream sites. Both laboratory experiments and software simulation have been applied to prove the feasibility and validity of the proposed method.To put all into a nutshell, in this thesis, a water quality abnormal detection method based on time and spatial correlation is proposed for better detection performance with the analysis of multi-variate and multi-site observation time series. Moreover, the effectiveness and advantages of the proposed correlation analysis method towards abnormal event detection are demonstrated with experimental and simulated data.