A simulation-optimization approach to determine optimum features of detention rockfill dams under flood condition

Abstract

Detention rockfill dams are one of the structural methods that can control flood risks with no impermeable core or membrane. These dams decrease the maximum discharge of flood hydrograph and postpone the time of its occurrence reducing life and financial losses in downstream. In this research study, maybe for the first time, a new methodology based on Multi-Layer Perceptron (MLP) artificial neural network meta-model and genetic algorithm optimization model is proposed for optimum design of detention rockfill dam. Hence, results of experiments done on the rockfill dam model, from Khorramshokouh (1391), are used in order to train and validate the MLP neural network model. The MLP neural network model is in fact a meta-model of simulating the hydraulic performance of detention rockfill dam which is linked to genetic algorithm optimization model to determine the optimal features of detention rockfill dam regarding the relationship between design variables and flood hydrograph passed through the porous media. Results of the proposed methodology depict that for the total discharges of probable floods, the optimum thickness of detention rockfill dam and the mean diameter of rockfills in the porous media are 17.6 cm and 2 cm, respectively. Also, in the optimum state, the peak discharge of flood hydrograph has 47.18% reduction and flood duration increases 39.94%.

Keywords

• detention rockfill dam
• multilayer perceptron neural network
• meta-model
• genetic algorithm optimization model
• cohesive and non-cohesive mixed sediment