Machine Learning-Based Performance Prediction of Photovoltaic-Thermal Energy Storage Systems: A Comparative Study

The intermittent nature of solar energy necessitates efficient storage systems, which can ensure grid stability and an uninterrupted supply. The present study presents a comprehensive comparative analysis of different machine learning (ML) algorithms employed in predicting the thermal performance of a hybrid PV/T system coupled with stratified water storage. The primary objective is to develop efficient forecasting models for two different thermal phenomena: the extremely volatile collector outlet temperature and the inertia-dominated temperature profile of the storage tank. A high-fidelity dataset, consisting of more than ۸,۷۰۰ data points, was generated using the TRNSYS dynamic simulation tool under Tehran’s meteorological conditions. Four different ML algorithms, i.e., Artificial Neural Networks (ANN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR), have been employed. The performance of each ML algorithm was assessed, and it was observed that the performance of each algorithm heavily relies on the physical properties of the variable. For predicting the extremely volatile collector outlet temperature, the gradient boosting framework (XGBoost) was observed to perform better, achieving an R^۲ value of ۰.۹۰۹ and an RMSE of ۳.۱۵ °C. Conversely, when employed in predicting the temperature profile of the storage tank, which is a continuous, inertia-dominated phenomenon, the Deep Neural Network (ANN) was observed to perform better, achieving an R^۲ value of ۰.۹۳۵ and the lowest value of RMSE, i.e., ۱.۴۱ °C.

کلیدواژه ها:

Photovoltaic-Thermal (PVT) ، Machine Learning ، Energy Storage ، Performance Prediction ، TRNSYS