Dynamic model of an alkaline electrolyzer based an artificial neural networks

Abstract

This paper presents an alkaline electrolyzer (AE) modelling based on artificial neural networks (ANN). Artificial neural networks can be applied to develop models for predicting the performance of complex and nonlinear systems. An alkaline electrolyzer behavior was modeled with success using a Multilayer Perceptron Network (MLP). The dynamic model which is used has been trained by using a Levenberg-Marquardt back propagation algorithm to learn the relationships that govern the electrolyzer and then predict its behavior without any physical equations. The absorbed electric current and the operating temperature were used as input vector of the neural networks which allows to predict the cell voltage behavior. The performance of this predictive neural network model is carried out using Matlab/Simulink software. Simulation results show that this predictive model estimated accurately the electrolyzer's cell voltage with the tracking errors within ± 0.01 V, which is less than ± 0.44 %.