Abstract
Multiple Linear Regression (MLR) is utilized to correlate the process parameters to the output permeate flow rate for a Seawater Reverse Osmosis (SWRO) desalination plant. Thorough residual analysis robust regression provided alternative methods to predict the response with higher accuracy compared to basic MLR analysis. Robust regression methods are developed in this study to predict the productivity of a seawater reverse osmosis plant based on key empirical correlations. The robust regression methodology considers important input operating parameters such as feed flow rate, feed pressure, outlet pressure of the multi-media filter, inlet and outlet cartridge filter pressures, outlet pressure of the high-pressure pump, and inlet seawater flow rate to pressure exchanger and correlates them to the permeate flow rate. The robust regression models are capable of accurately predicting response for any input operating parameters for the reverse osmosis plant. The regression models demonstrated strong statistical goodness-of-fit measures using Huber’s method in terms of three-way interactions with a high R2 of approximately 0.99 and a mean absolute percentage error of 2.4%. Furthermore, the robust regression results have been validated experimentally and the results showed very good agreement with measured values, with an error of approximately 0.8%.