Artificial neural network model for optimizing operation of a seawater reverse osmosis desalination plant

An artificial neural network (ANN) was developed to predict the performance of a seawater reverse osmosis (SWRO) desalination plant, and was then applied to the simulation of feed water temperature. The model consists of five input parameters (i.e., feed temperature, feed total dissolved solids (TDS), trans-membrane pressure (TMP), feed flow rate, and time) and two output parameters (i.e., permeate TDS and flow rate). Then, the one-year operation data (n = 200) from the Fujairah SWRO plant was divided into three data sets (i.e., training, validation, and test data set) to develop the ANN model. The trained ANN model was subsequently found to produce good agreement between the observed and simulated data (TDS: R2 = 0.96; flow rate: R2 = 0.75) in the test data set. The results of this study show that the variation of the feed water temperature and TMP was found to significantly affect both the permeate TDS and flow rate. From subsequent simulations with various temperature controls, it is further suggested that the permeate TDS can be reduced using a linear increase control (from 27.5 to 29.5°C) for the feed temperature in an SWRO hybrid system with multi-stage flash (MSF) distillation, such as the Fujairah plant.

In this study, we proposed an applicable framework for optimizing the operation of the Fujairah SWRO plant using an artificial neural network (ANN), based on the one-year operational data for 2005. We also investigated effects of feed temperature and TMP on permeate TDS and flow rate. As a result of the input temperature control based on the optimal temperature points extracted from this study, it was suggested that permeate TDS can be reduced through optimized operation using a linear increase control (from 27.5 to 29.5°C) for feed temperature, determined from the SWRO hybrid system with MSF in the Fujairah plant. Additionally, this study suggests that the hybrid system contain a heat exchanger such that the feed temperature can be adjusted to the optimized temperature, and thereby achieve a higher water recovery rate based on the maximum allowable TDS criteria (<500 ppm). Finally, a framework for developing an ANN model, applicable to SWRO process operating data, was suggested for predicting the performance and optimizing the operation of SWRO desalination plants. Based on the framework proposed here, furthermore, artificial neural network can be combined with deterministic models that include physical laws as a hybrid model for analyzing and/ or predicting complex SWRO desalination processes: fouling/scaling prediction models, process optimization models, etc.