Intelligent hybridized modeling approach to predict the bedload sediments in gravel-bed rivers

Bedload transport due to approved complexity and challenges has been the subject of different modeling approaches. Due to imprecise of the empirical equations, the potencies of the intelligent techniques in developing more accurate bedload predictive models have been highlighted. In this paper, an optimum hybridized artificial neural network (ANN) with firefly metaheuristic algorithm (FA) through a dynamic setting parameter approach was developed and introduced. The model was applied on 879 datasets including 5 dominant parameters of bedload transport (discharge, flow velocity, slope, depth, mean grain size) from 19 gravel-bed streams of Idaho- USA. Detailed analyses using different analytical error metrics as well as comparison with several empirical equations showed an improved R-2-value from 0.1 in empirical equation to 0.95 in hybrid model. The assessed performances of applied model demonstrated for 6.03% progress in ANN and at least 63.08% in empirical equations. According to observed results, the hybrid model with 84.65% accuracy was outperformed than others in providing closer and more compatible outputs to measurements. Referring to carried out sensitivity analysis, the discharge and velocity were identified as the most effective factors on predicted bedload.

Automated summary

Bedload transport has been studied using different modeling approaches, with the potential of intelligent techniques highlighted for developing more accurate predictive models. This study introduces an optimum hybridized artificial neural network with firefly metaheuristic algorithm, showing significant improvement in accuracy compared to empirical equations. The hybrid model demonstrated better performance in providing closer and more compatible outputs to measurements, with discharge and velocity identified as the most effective factors on predicted bedload.