Artificial intelligence based forecasting and optimization of hybrid solar-powered desalination systems combining reverse osmosis and humidification-dehumidification

Abstract

This study develops, models, and evaluates a fully solar-powered hybrid desalination system integrating Reverse Osmosis (RO) and Humidification–Dehumidification (HDH) technologies driven by photovoltaic (PV) energy. While many previous studies have examined RO–HDH hybridization under fixed or steady-state assumptions, this work presents a unified MATLAB/Simulink framework that integrates PV power input with both membrane-based and thermal desalination processes within a single simulation environment under idealized operating conditions. Individual subsystem models were validated against experimental and literature data, showing low average deviations of 1.83% for PV, 3.37% for RO, and 2.86% for HDH. Parametric analysis revealed that RO freshwater production is strongly governed by membrane surface area and pump pressure, while HDH performance is dominated by inlet air temperature and condenser cooling conditions. A key contribution of this study is the integration of RO brine reuse as feed for the HDH unit within a single simulation environment, resulting in up to 220% enhancement in total freshwater recovery and a 35% reduction in energy consumption compared to standalone operation, which has not been quantitatively demonstrated in prior RO–HDH studies. To address the lack of predictive control tools in existing hybrid desalination models, an Adaptive Neuro-Fuzzy Inference System (ANFIS) was embedded within the framework to forecast system performance under fluctuating solar conditions. The ANFIS models were trained primarily using data generated from the validated Simulink framework, while the experimental datasets used for subsystem validation were also employed as external references in evaluating the AI outputs; however, independent real-time experimental validation of the integrated ANFIS-based system remains necessary. The ANFIS models achieved high predictive accuracy, with R² values ranging from 0.983 to 0.992 and low RMSE and MAPE values. The key innovation of this study is the creation of a unified simulation framework for a PV-powered RO–HDH hybrid system, coupled with ANFIS-based prediction and optimization informed by a validated simulation framework and supported by subsystem-level experimental references, which enables brine reuse, improves freshwater recovery, and lowers energy consumption.