Machine learning-based performance prediction and multi-objective optimization of hybrid photovoltaic/thermal systems

Abstract

Hybrid Photovoltaic/Thermal (PVT) systems merge photovoltaic and thermal technologies to generate electricity and heat from solar energy at the same time. These systems boost energy efficiency by producing electricity from sunlight and using surplus heat for heating applications, which makes them an eco-friendly and multi-purpose solution for power and heat generation. This study mathematically models a nanofluid-based PVT energy system under varied environmental conditions and utilize different machine learning algorithms to discover complex relationships between system parameters. Therefore, a predictive Elastic Net model optimized through hyperparameter tuning by the Dynamic Differential Annealed Optimization Algorithm (DDAOA) is developed for accurate projection of the system's electrical and thermal energy output. This model demonstrates outstanding performance in predicting the electrical and thermal outputs of the nanofluid-based PVT system with high R 2 of 0.9999 and 0.9749, respectively. Multi-objective optimization using Non-dominated Sorting Genetic Algorithm II is employed in this work to identify the desired execution condition of a hybrid PVT system for both thermal and electrical maximum outputs, which delivers an electricity output of 135.670 W/m 2 and a heat output of 205.620 W/m 2 .