Calibration of Flow and Water Quality Modeling Using Genetic Algorithm

In mathematical simulation for flow prediction and water quality management, the inappropriate use of any model parameters, which cannot be directly acquired from measurements, may introduce large errors or result in numerical instability. In this paper, the use of a genetic algorithm for determining an appropriate combination of parameter values in flow and water quality modeling is presented. The percentage error of peak value, peak time, and total volume of flow and water quality constituents are important performance measures for model prediction. The parameter calibration is based on field data of tidal as well as water quality constituents collected over five year span from 1991 to 1995 in Pearl River. Another two-year records from 1996 to 1997 are utilized to verify these parameters. Sensitivity analysis on crossover probability, mutation probability, population size, and maximum number of generations is also performed to determine the most befitting algorithm parameters. The results demonstrate that the application of genetic algorithm is able to mimic the key features of the flow and water quality process and that the calibration of models is efficient and robust.