Data driven identification of industrial reverse osmosis membrane process

Abstract

A dynamic artificial neural network (ANN) is developed for the identification of an industrial reverse osmosis membrane process under fouling effect. 4-year historical data on feed properties and measured process variables in the plant were used for the ANN training and validation. The ANN considers the current and previous week's online measurements as inputs and provides one-week and two-week ahead permeate flow predictions. A sensitivity analysis is provided at various periods to determine the variables with high impact on the permeate flow, and thus the plant performance. Based on the sensitivity analysis, cartridge filter pressures and pH have the highest impact on the output. Plant operating window is also calculated under such complex multivariable and nonlinear nature. The results show quantitative and intuitive conclusions, parallel to existing literature, and provide significant insight on the identification of industrial and large-scale reverse osmosis processes. (c) 2022 Elsevier Ltd. All rights reserved.