Artificial neural networks approach for estimation of sediment removal efficiency of vortex settling basins

The mechanism of flow in vortex settling basins is so complicated that it is difficult to establish a general regression model to accurately estimate the sediment removal efficiency. Hence, in this study an alternative approach using artificial neural networks (ANNs) is proposed to determine the sediment removal efficiency of the vortex settling basins. Laboratory and field data collected from the literature having a wide range of hydraulic and geometrical variables are used to train, test and validate the network. A network architecture complete with trained values of connection weight and bias and requiring input of ungrouped parameters pertaining to Qi, Qu, Zh, hp, DT, B, du, d50, and ωo is recommended in order to predict the removal efficiency of vortex settling basin. Predictions based on original raw data (Qi, Qu, Zh, hp, DT, B, du, d50, ωo) were better than those based on grouped dimensionless forms of the data. Findings of the sensitivity analysis showed that DT was the most significant parameter for the prediction of removal efficiency. The variables in order of decreasing level of sensitivity were DT, Qi, hp, du, ωo, B, d50, Zh, and Qu. But in view of the variability in the outcome resulting from sensitivity analysis, it is felt that the network, which requires all input quantities, may be followed for generality. The results of this modeling were also compared with the available regression models and it was found that the ANN results are highly satisfactory.