Forecasting of Permeate Conductivity using MLR and ANN Methods of Boujdour Seawater Reverse Osmosis Desalination Plant

Abstract

Background: For many years, seawater desalination technique has been operational to deal with water scarcity. In Boujdour region, located near the Atlantic Ocean southwest of Morocco, most water drinking is produced by a reverse osmosis seawater desalination plant. The permeate conductivity prediction is used to evaluate the performance of desalination plants. Objective: The present paper focuses on the modeling and comparison of the Artificial Neural Network (ANN) and Multiple Linear Regression (MLR) for the prediction of permeate conductivity for a one-year period. Methods: Six input variables are considered, including turbidity, temperature, pH, feed conductivity, feed flow, and transmembrane pressure (TMP). Firstly, the MLR identifies the most important variables influencing the permeate conductivity with the aim of developing a regression model for the dependent and independent parameters. Secondly, the ANN method is examined to analyze the performance of desalination plant. A study of the effect of the number of neurons and the number of hidden layers on the efficiency of the neural network has been made. Results and Conclusion: Results confirm that the MLR and ANN models forecast the permeate conductivity with a suitable coordination coefficient of the real and predicted values. ANN model has been successfully tested for reliability with a correlation coefficient R2 of 99.097% and a mean square error (MSE) of 0.002607.