Desalination of brackish water and seawater are the main sources of many countries which afflict rainwater scarcity and are deprived of lakes and rivers. There are several methods for desalination of water such as membrane distillation, solar evaporation, electrodialysis reversal, multi-stage flash distillation, reverse osmosis, etc. Reverse osmosis is more popular because of its simple design and economic factors such as require low energy, minimal operating temperature, and low water production costs. In the current study, artificial intelligence (AI)-based reverse osmosis water desalination models have been developed using AI techniques viz. artificial neural networks (ANN) and support vector regression (SVR). The input parameters of the models include sodium chloride concentration, feed temperature and pressure. The permeate rate is used as the output parameter. The developed AI-based models are evaluated and validated against the reported experimental data present in the literature. These AI-based models are then further compared with the widely used multiple linear regression (MLR) over the virgin test (unseen) dataset based on statistical measures like average absolute relative error (AARE), coefficient of determination (R2), etc. The SVR-based model exhibits a low value of AARE of 1.95% and a value of high R2 of 0.9963 while the corresponding values for ANN and MLR models are 9.35%, 52.04%, 0.9899 and 0.9157, respectively. Thus, the structural risk minimization (SRM) principle based SVR model is found to be the best, more accurate and generalized in comparison to the empirical risk minimization (ERM) based MLR and ANN models for the permeate rate prediction. Furthermore, through these AI techniques, excellent predictions can be made for the unseen data which not only reduces the number of experiments to be done but also helps the more effective design and fabrication of membrane-based desalination unit.
Reverse osmosis, SVR, ANN, R2, AARE
