Selection of Water Treatment Processes Using Bayesian Decision Network Analyses
Zhu, Zoe J. Y., McBean, Edward A., Journal of Environmental Engineering and Science
Abstract: Decisions on selection between alternative water treatment sequences require numerous considerations, including treatment effectiveness and costs of the individual technologies, water quality guidelines, and the characteristics of the influent source water. Bayesian decision networks (BDNs) are demonstrated to provide a normative framework to assist in decision making for problem domains where tradeoffs are needed between uncertainty in the domain and the preferences of decision makers. A methodology is developed to apply BDNs and the methodology, and is illustrated with experimental results.
Key words: Bayesian decision network, uncertainty, risk assessment, water quality, water treatment processes.
Resume: Les decisions concernant la selection des sequences de traitement des eaux demandent de tenir compte de plusieurs points dont l'efficacite du traitement et les couts des techniques individuelles, les lignes directrices sur la qualite de l'eau et les caracteristiques de l'eau de l'influent source. Les reseaux de decision bayesienne (BDN) sont demontres afin de fournir un cadre normatif pour aider a la prise de decision quant aux domaines problematiques lorsque des compromis sont requis entre l'incertitude dans le domaine et les preferences des preneurs de decisions. Un processus est mis sur pied pour appliquer les BDN et leur methodologie a l'aide de resultats experimentaux.
Mots cles: reseau de decision bayesienne, incertitude, evaluation des risques, qualite de l'eau, procedes de traitement des eaux.
[Traduit par la Redaction]
Chlorination has been used to treat drinking water for more than a century, and it remains the predominant disinfection method because of its low cost and well established operations. However, in the 1970s chlorine disinfection in water treatment plants was identified as causative of undesirable disinfection byproducts. These include trihalomethanes and total haloacetic acids (HAAs), where their subspecies have been classified as probable and possible carcinogens (USEPA 1999x, 1999b; Sadiq and Rodriguez 2004). As a consequence, while chlorine has continued to be widely employed for its disinfection potential, there is extensive interest in minimizing the formation of disinfection byproducts to decrease health impacts on human recipients of treated water.
More generally, questions continue about how water processing may be viewed as a decision problem under uncertainty. Raw water quality may be characterized by multiple factors that involve significant temporal variability, where these factors include, for example, organic matter concentrations, turbidity, pH, and alkalinity. In response, a sequence of treatment processes, or multiple barriers, is employed to ensure appropriate and cost-effective treatment of the raw water. Different combinations of treatment processes produce varying qualities of treated water and involve different costs. Due to the numerous factors that cannot be modeled precisely, including raw water quality variability and the variable functionality of alternative processing methods, modeling cannot escape uncertainty in both inputs and mathematical structure, which in turn results in uncertainty in predictions.
The uncertainty demands a normative framework to guide decision-making on how to treat the water. The search for normative frameworks has been conducted in numerous other areas of science and engineering, including decision analysis, statistics, economics, operations research, and artificial intelligence, with fruitful results (Solberg 1987; Cheeseman 1993; Horvitz and Barry 1995). Studies of probability have shown that a rational decision maker's subjective belief about events in an uncertain domain can be represented by Bayesian probability (de Finetti 1937; Pearl 1991). Studies on multivariable utility theory and decision theory show that a rational decision maker represents his preference by a utility function, and trades off uncertainty and desirability according to the principle of maximum expected utility (Von Neumann and Morgenstern 1944; Pearl 1991). …