Optimized Calibration for Unsteady Flow Modeling using a Genetic Algorithm

Abstract

The calibration of an unsteady flow model generally relies on the estimation of roughness coefficients, which act as descriptors of friction resistance along the channel. Here an optimized procedure is implemented by incorporating a genetic algorithm (GA) to automate the process. This approach is suitable because discrete parameters are considered. The target is a one-dimensional unsteady flow model in which the roughness coefficient is allowed to vary not only along the river axis, but also dynamically with discharge. The objective function is the root mean square of the difference between measured and computed water levels at a given cross section. This function is minimized using several strategies for reproduction and crossover. Several fitness functions are tested as well. The application results are encouraging and the implementation is less complicated than in methods where the objective function has to be related explicitly to the constraints of the application.