Optimization of a solar-based PEM methanol/water electrolyzer using machine learning and animal-inspired algorithms

Abstract

This contribution scrutinizes the performance of the integrated photovoltaic thermal (PVT) system with the proton exchange membrane methanol electrolyzer (PEMME) and water electrolyzer (PEMWE) as sustainable ways to produce hydrogen. Artificial neural networks (ANNs), are adopted to evaluate the effect of various operating parameters on the performance of the systems. The adopted ANNs are radial basis function (RBF), extreme learning machine (ELM), long short-term memory (LSTM), and gated recurrent unit (GRU). Moreover, to obtain the optimum performances of the systems, the multi-objective whale optimization algorithm (MOWOA) and multi-objective bat algorithm (MOBA) are implemented. It is found that solar radiation is the most influential parameter on the hydrogen production rate of the systems, followed by ambient temperature, inlet temperature, working fluid mass flow rate, and wind speed. According to the machine learning outputs, the highest hydrogen production rate for the PVT-PEMWE and PVT-PEMME systems are around 2.45 mol h−1 m−2 and 5.44 mol h−1 m−2, respectively. Moreover, the MOBA shows that the hydrogen production rate and electrical efficiency of the PVT-PEMWE and PVT-PEMME systems are 2.29 mol h−1 m−2 and 18.74%, and 5.06 mol h−1 m−2, and 18.77%, respectively, at their global optimum point by considering hydrogen production rate and electrical efficiency as the objective functions. This study reveals that although at the same working conditions, the electrical and thermal output of the water-based PVT system is superior to that of the water/methanol-based PVT system, the PVT-PEMME system outperforms the PVT-PEMWE system from the hydrogen production viewpoint, due to the higher efficiency of the PEM methanol electrolyzer.