A generalized artificial neural network approach to model multiple-extraction humidification-dehumidification systems

This study constructs a sophisticated Artificial Neural Network (ANN) to predict the efficacy of a highly nonlinear heat and mass transfer humidification-dehumidification (HDH) desalination system for different arrangements. The incorporation of ANN allows for superior handling of the non-linearities inherent in the heat and mass transfer processes of the HDH desalination system, resulting in markedly improved predictive accuracy. This is particularly beneficial for optimizing system operational parameters, which directly influence the efficiency and effectiveness of water production. The ANN is extensively trained, validated, and tested using a wide-ranging dataset covering numerous operational scenarios. The evaluation focuses on critical metrics such as the gained output ratio, water-to-air mass flow ratio, recovery ratio, seawater and air temperatures, energy efficiency, and the optimal number of air extractions. The ANN receives four primary inputs: the temperature of incoming seawater, the peak water temperature, the enthalpy pinch, and the number of air extractions, and is optimized for inlet temperatures of 10 to 40 °C and heating temperatures from 60 to 90 °C. The model accuracy impressively reaches a minimum of 96.1 %. The findings reveal that the maximum feasible number of air extractions is 22 with an enthalpy pinch of 0.1 kJ kg_d⁻¹, suggesting the method's potential, efficiency, and durability. Moreover, at an enthalpy pinch of 3 kJ kg_d⁻¹, the optimal operational parameters near the theoretical maximum (99 % of infinite extractions) are achieved with seven extractions.