Chance-Constrained Watershed Management Using Evolutionary Algorithms

A modeling approach has been developed that identifies cost-effective detention pond configurations and land management plans for a watershed such that pollutant removal is achieved with a specified reliability level under conditions of uncertainty in pollutant loading factors. The design procedure incorporates reliability estimation into a genetic algorithm-based search procedure. This search procedure identifies cost-effective detention pond allocations (i.e., locations and sizes) that meet the pollutant removal targets at a specified level of likelihood and also allocates the land use development to meet system-wide land development plans. However, the implementation previously used was very computationally intensive and required repeated runs with different starting random seeds to find a good solution. In this paper, various constraint-handling methods, including penalty functions, multiobjective constraint handling techniques, and a new stochastic selection method, are applied to the reliability constraints and compared with respect to constraint satisfaction, quality of solution, and consistency. The results are used to provide guidelines for constraint handling techniques that perform relatively efficiently and consistently in evolutionary algorithms applied to watershed management under conditions of uncertainty.