Real-Time Water Quality Assessment with Bayesian Belief Networks

Abstract

Real-time sensing in water distribution systems provides a new and potentially powerful analytical tool with which water security and quality may be characterized. However, current real-time sensing technology is relegated to what are generally considered indirect indicators of quality or 'surrogate parameters' (e.g. pH, turbidity, residual chlorine, etc.). Through monitoring the quality of the water in a distribution process over time, the natural variation of the system's parameters may be established. Subsequently, operation of a real-time sensing system would rapidly detect quality changes within a distribution system. This process would allow response actions to take place much quicker and more reliably than conventional 'grab sample' analyses. However, the level of performance that water quality event detection methods have exhibited to date is insufficient for real world utilization. In response, Bayesian Belief Networks (BBNs) offer a formalized method of reasoning under uncertainty. BBN-based analysis allows the assimilation of multiple sources of sensor information over time and the generation of temporal probability distributions. The development/application of a BBN is described. Surrogate parameters monitored for the development of the BBN include pH, dissolved oxygen, conductivity, oxidation-reduction potential and turbidity. Difference filtration using a 60 second moving window of observations identified the rapid rate of change present in the signals for the surrogate parameters pH, conductivity and turbidity proved responsive to contamination as simulated in bench-scale studies.