Design of real-time hybrid nanofiltration/reverse osmosis seawater desalination plant performance based on deep learning application

Abstract

The advancements brought by Artificial Intelligence (AI) have revolutionized various research domains in solving highly dynamic and complex problems such as desalination. Recently, there has been a growing trend toward modeling the effectiveness of the hybrid nanofiltration (NF) and reverse osmosis (RO) of desalination. In this study, we develop a deep learning (DL)-based approach to model the performance of hybrid NF/RO desalination plants based on permeate conductivity (PC), permeate flow rate (PF), and permeate recovery (PR). For this purpose, three configurations of a convolutional neural network (CNN), recurrent neural network (RNN), and relevance vector machine (RVM) were designed to achieve the modeling task. Before the modeling process, data preprocessing and feature selection were conducted based on the raw input-output parameters. The outcomes were evaluated based on several statistical variables. The results demonstrated that CNN-M3 achieved the best performance in all the five statistical performance criteria employed for PC$\left(\mathrm{\mu S}/\mathrm{cm}\right)$, PF$\left(m^3/h\right)$, and PR (%) modeling during the calibration and verification phase. The quantitative results proved that CNN-M3 achieved an accuracy of (MAE = 0.0780), (MAE = 0.0657), and (MAE = 0.0491) for PC$\left(\mathrm{\mu S}/\mathrm{cm}\right)$, PF $\left(m^3/h\right)$, and PR (%), respectively. The results were also drawn in a 2D-dimensional Taylor diagram to show the probability cumulative distribution function (CDF) in a scatter plot. Results reveal that DL-based models like CNN perform superiorly against RNN and RVM. Therefore, they can be deployed as a reliable and efficient tool for simulating the performance of a hybrid NF/RO seawater desalination system.