Artificial neural networks applied to manage the variable operation of a simple seawater reverse osmosis plant

Abstract

For the purpose of managing the operation of a small-scale prototype of a sea water reverse osmosis desalination plant installed on the island of Gran Canaria (Spain) and enabling it to function with fluctuating power input, artificial neural network (ANN) models were incorporated into its control system. The ANN models were developed to generate feed flow and operating pressure setpoints (with the restriction of having to maintain the permeate recovery rate within a certain range) after taking into account not only the available electrical power but also the temperature and conductivity of the feedwater. It is concluded that the ANN models that were used after training and validation were able to successfully manage the random and widely varying available electrical power. The statistical hypothesis testing that was also performed showed no significant statistical differences (at 5% level) between the errors (both MAE and MAPE) conunitted when adapting power consumption of the plant to the available electrical power in the various operational tests using different feedwater characteristics.