Optimal Load Allocations by Linkage of Evolutionary Optimization Algorithms with Inductive Models of Watershed Response

Two separate optimization strategies (genetic algorithm and box complex method) are compared in the development of optimal nutrient load allocations for the water quality-impaired Beargrass Creek Watershed in Louisville, Jefferson County, Kentucky. The optimal load allocations are determined by linking the optimization algorithms with receiving water inductive models developed for the lower reaches of the watershed. The inductive models are developed from (1) a synthesis of both input and output response variables as derived from a continuous simulation of the watershed using a calibrated HSPF model, and (2) a synthesis of the continuous and discrete water quality data sampled over the last 2 years in the watershed. Inductive model construction is performed by use of artificial neural networks and the use of functional fixed-set genetic programming. The use of inductive models provides a more computational efficient framework for linkage with an optimization model for use in developing an optimal loading strategy.