Data-driven models of a solar field used to power membrane distillation systems: A comparison study

Abstract

The pressing issue of water scarcity has led to increased research focussing on enhancing access to fresh water, with sustainable desalination emerging as a prominent solution. The use of solar energy is often proposed because of the geographical coincidence of high solar irradiance and water scarcity. However, the variability of the energy source in a stationary-designed process such as desalination must be addressed, and modelling solar desalination systems is crucial to understanding the dynamics and optimising the performance. Solar thermal energy is cheaper to store than photovoltaic energy, and powers advanced desalination technologies such as membrane distillation (MD) that can reach higher water recovery. This study investigates the application of data-driven modelling techniques to an innovative solar collector field providing heat for a MD system. The novelty of using mirrors in the solar field to boost the thermal power yielded renders the classical first-principles-based models presented in the literature invalid, as they cannot account for the nonlinear impact of mirrors in the solar field performance. This justifies the use of new data-driven techniques, and four modelling methodologies are compared, with the NARX artificial neural network that proves the most effective, with an R${}^2$ value of 0.9741 and an RMSE value of 6.3151. The best model is validated by simulation of a solar MD plant.