AWWA ACE56265

The design of drinking water treatment plants must consider several objectives and satisfy multiple constraints. The use of mathematical programming techniques can assist in determining the optimal treatment plant design. Unfortunately, common practice assumes that raw water characteristics and model parameters are known (perfect information) when, in fact, they include either natural variation or experimental uncertainty. Including variability and uncertainty in the design framework allows for a robust design. A framework is presented for including variability and uncertainty into the design formulation for particulate removal under conventional treatment (rapid mix, flocculation, sedimentation, and filtration). As an example, a deterministic design that assumes perfect information is performed and shown not to be robust with respect to influent variability and model parameter uncertainty. Individually incorporating one of four variable influent parameters or three uncertain model parameters in the design process increased design costs up to 21.2%. The resulting designs were, however, robust with respect to the individual variabilities/uncertainties. Including multiple variable/uncertain parameters resulted in even greater design costs than the sum of the individual variability/uncertainty values. Includes 13 references, tables, figures.