Artificial Neural Network Modelling of Reverse Electrodialysis

Reverse electrodialysis (RED) is an emerging technology with the potential to generate energy from salinity power gradients with limited environmental pollution. In a salinity-gradient power generator (SGP-RED), cation and anion exchange membranes are stacked alternately and due to the difference in concentrations of river and seawater, the diffusion of ions generates an electrochemical potential. In this work, a data-driven approach has been adopted for predicting the performance (Power Density) of SGP-RED power generators over different operating conditions and membrane types, using Artificial Neural Network (ANN) models. Experimental data was gathered and mined from 130 research publications from the last ten years to predict the output Power density of the system against a set of nine input parameters, e.g. membrane types, thickness, resistance, current density, perm selectivity, temperature, etc. A simple mathematical model to estimate the Power density is proposed and used to validate the experimental data before using them with ANN. We used ANN models using Bayesian Regularisation as the backpropagation algorithm (ANN-BR) along with a combination of six sets of pair-wise activation functions amongst Tan-sigmoid, Log-Sigmoid and Linear, in hidden and output layers. The ANN-BR model with Tan-Sigmoid activation functions in both hidden and output layers is shown to predict most accurately the Power density of the SGP-RED system.