Real-Time Water Quality Monitoring: Assessment of Multisensor Data Using Bayesian Belief Networks

Abstract

Real-time sensing in water distribution systems provides a potentially powerful analytical tool for providing water security. Through monitoring surrogate parameters (e.g., pH, turbidity, and residual chlorine) over time, the natural variations of a distribution system's parameters are established, allowing rapid detection of changes in water quality. However, the level of performance that water quality event detection algorithms have exhibited to date is insufficient for real-world utilization. Bayesian belief networks (BBNs) offer a formalized method of reasoning under uncertainty and are well suited to the analysis of multiple sources of information. The application of a BBN to water quality event detection is described. Surrogate parameters (pH, conductivity, and turbidity) were monitored during an experimental E. coli contamination. Difference filtration using a 60-s moving window of observations identified rapid rates of change present in the surrogate parameter signals, demonstrated as responsive to contamination as simulated in bench-scale studies. A BBN was constructed to assimilate the surrogate parameter variation and compute temporal probability distributions about the contamination of an experimental system. The BBN topology, probability distributions and data transformation techniques applied were validated through successful identification of contaminant injections.