Gene expression models for prediction of dam breach parameters

Abstract

Data from a large database of 140 dam failure cases are used with gene expression programming (GEP) to develop new empirical formulae of physical meaning for prediction of non-dimensional key dam breach parameters. The GEP models are trained on 75% of the data set and validated on the remaining 25%. Parametric and error analyses are conducted to confirm the robustness of the developed relations. Moreover, uncertainty analyses using the Monte Carlo technique is performed to check for the output uncertainty of key dam breach parameters and the contribution of various input parameters to the overall output uncertainty. It is found that uncertainties of 20 to 40% are calculated for the developed GEP models with reservoir shape factor and dam erodibility being main influential predictors.