Comparative analysis of photovoltaic system technologies using advanced Machine Learning algorithms

Abstract

Abstract The aim of this study is the comparison between three photovoltaic (PV) technologies using Machine Learning (ML) algorithms such as the Convolutional Neural Network (CNN), Artificial Neural Network (ANN), Bidirectional Long-Term Memory (BiLSTM), Long-Term Memory (LSTM), and CNN-LSTM. This comparison is based on different evaluation metrics including the energy yield( Y f ), Performance Ratio (PR), and the Capacity factor( C f ). Our results indicate that the ANN model offers the highest accuracy for polycrystalline and amorphous technologies, while the BiLSTM model performs best for monocrystalline technology. Specifically, the ANN model achieves an RMSE of 35.71 W at 5 min intervals for polycrystalline technology, and daily RMSE values do not exceed 1.53 kWh for monocrystalline units and 0.9 kWh for amorphous units. The average error in the daily performance ratio (PR) is less than 0.7% for monocrystalline units and does not exceed 0.6% for polycrystalline and amorphous units. In addition, the study reveals an error of 0.1% for the capacity factor and a very low error for the energy efficiency, on the order of 10 −2 kWh kWp −1 for all three measurements. The results demonstrate the effectiveness of using ML models to predict photovoltaic energy production, thus contributing to the efficient management and optimization of photovoltaic systems.