The drinking water industry has an increased focus on the effectiveness of drinking water treatment plants in removing waterborne pathogens such as Giardia and Cryptosporidium. The most notable Cryptosporidium incident is the Milwaukee outbreak in April 1993 (Craun 1998), which affected nearly 400,000 people and led to national awareness for the need to protect public water supplies from this pathogen. A major cause of the Milwaukee outbreak was changes in coagulation that resulted in filter failures.; Aerobic spores and particle counts can be used to predict Giardia and Cryptosporidium removal (Nieminski and Bellamy 2000), but particle counts are a more useful surrogate, as they can be monitored in real-time (Gelder et al., 1999; O'Shaughnessy et al., 1997).; Theoretical models utilize parameters that are sometimes difficult to measure and change over time as the filter operates (Ohja and Graham 1994, Burganos et al., 1995, Conlin et al., 1997). Therefore, empirical models for headloss development, filter-effluent turbidity, and effluent particles in the 2- to 5-μm range were predicted using full-scale data. The models were developed using multivariate polynomial regression (MPR).; MPR is based on a class of nonlinear models developed by Chen and Billings (1989). NOR produces fits to data comparable to those of artificial neural networks, but the resulting models are parsimonious (have few coefficients) and mathematically simple. They can be analyzed by standard graphical and statistical methods, including computation of confidence intervals (Wang and Vaccari 2003). These models are capable of describing complex relationships including multivariable chaotic systems and arbitrary truth table relationships.; The total number of particles entering a filter directly impacts headloss. Filter influent that is too clean can result in increased particle breakthrough in the 2- to 5-μm range and extended filter-ripening times. These models can provide a real-time prediction tool for operations staff to anticipate the impact of changes in filter operation on water quality. The benefit of these models is that the drinking water industry can now begin to assess the impact of changes to filter influent water quality on the filter effluent water quality in such areas as turbidity and particle counts.
展开▼
