Deep Learning Architectures for Solar PV Forecasting

Abstract

Integration of solar photovoltaic (PV) based energy systems into the electric grid has increased during the last decades because solar energy is one of the most promising and immediate energy resources to convert sunlight energy to electric energy. However, solar PV energy systems rely on atmospheric weather conditions. Due to the nature of solar PV energy outputs, intermittent, volatile, and nonlinear characteristics create uncertainty in solar PV energy production for the electricity grid. Accurate solar energy forecasting requires mitigating uncertainties from solar energy outputs. In this study, we generated residential-level solar energy outputs from the physical model to train the model. We focused on a method, combining Convolution Neural Network (CNN) and Long Short-Term Memory (LSTM), to investigate solar energy forecasting a 4.9 kW residential rooftop PV while considering weather conditions for the State of Connecticut. We also implemented multivariable solar forecasting techniques. Various Deep Learning method has been analyzed, and Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) was computed to evaluate the performance metrics of the DL methods and compare their results for multivariable solar energy forecasting.