Artificial Neural Networks for Defining the Operational Parameters of Sand Filters for Irrigation

ABSTRACT The difficulty in defining the working configuration during the operation of sand filters arises from the complexity of the relationships among the variables in the filtration and backwashing processes. The objective was to use artificial neural networks to model the pressure loss and removal efficiency of filtration in a commercial sand filter. The experimental tests covered three grain sizes of silica sand (fine, medium and coarse) and four filtration rates (20, 40, 60 and 75 m h −1 ). Three artificial neural networks (ANN1, ANN2 and ANN3) were built to define the duration of the filtration cycle, pressure loss and filtered volume. ANN1 was based on modelling pressure loss through the sand filter, whereas ANN2 and ANN3 were based on the effluent suspended solids. The variables used were the filtration rate, filtered volume, grain size, suspended solids and pressure loss. ANN1 presented good performance with the 5‐4‐1 architecture (five neurons in the input layer, four neurons in the hidden layer, and one neuron in the output layer). An increase in the filtration rate reduced the duration of the filtration cycle, whereas an increase in the grain size extended the duration. The ANN2 and ANN3 networks performed poorly and, therefore, were disregarded.