Prediction by neural networks of filtered volume and outlet parameters in micro-irrigation sand filters using effluents

Abstract

Filtration is an essential operation in irrigation that removes particles carried by water that could clog drip emitters. Prediction models for filtered volume and effluent parameters are not available for the sand filters used in micro-irrigation systems. The objective was to develop, and assess the performance of, artificial neural networks (ANNs) able to predict filtered volume and sand filter outlet values of dissolved oxygen (DO) and turbidity; both are related to emitter clogging risks. Data from 770 experimental filtration cycles of a sand filter operating with effluent were used for training, cross-validation and testing the ANNs. The ANN with a best performance was compared with prediction equations obtained with the multi-linear regression stepwise method. The best ANN developed, which needed as input parameters effective sand media size, head loss across the filter and filter inlet values of DO, turbidity, electrical conductivity (EC), pH and temperature, had a higher accuracy predicting the target parameters than the multi-linear equations adjusted to the experimental data. ►No models are available to predict sand filter outlet parameters using effluents. ►ANNs can predict volume and outlet parameters in a sand filter. ►ANNs predict these outputs better than multiple linear regressions.