Development and Comparison of ANN-LM and ANN-BR Models for Predicting the Performance of Membrane Based Novel Liquid Desiccant Drying/Air Conditioning cum Desalination System

Abstract

The major issue in extracting pure water from humid air is design of energy-efficient dehumidifier and regenerator as well as optimization of the working processes. In the current study, a novel liquid desiccant air conditioning/drying cum desalination system is analyzed by incorporating M-cycle based dehumidification and flat plate type polyvinylidene difluoride (PVDF) membrane-based indirect contact regenerator has been proposed which is investigated by using two artificial neural network (ANN) models i.e., Levenberg-Marquardt (ANN-LM) and Bayesian Regularization (ANN-BR). To investigate the model three different activation functions i.e., Tangent Sigmoid in both layers, Tangent Sigmoid & Linear, and Logarithmic Sigmoid & Linear for hidden and output layers are applied using experimental data from the literature. To perform the numerical analysis four inlet parameters (mass flow rate of water and liquid desiccant, liquid desiccant concentration at inlet, regenerator inlet temperature of water, and liquid desiccant) is considered to predict three-outlet parameters which are pure water extraction rate, liquid desiccant regenerator outlet concentration, and liquid desiccant regenerator outlet temperature. These combinations are explored and it has been established that ANN-LM with activation function Tan-Sigmoid + Tan-Sigmoid is the best performing combination having R2 value of 0.98 for pure water extraction rate, 0.97 for liquid desiccant solution concentration, and 0.99 for liquid desiccant outlet temperature. It has also been observed that a combination of ANN-BR with activation function Log-Sigmoid + Linear was the least accurate for predicting the exit parameters. The predicted values have been found to be in excellent concurrence with the experimental data.