Very Short-Term Photovoltaic (PV) Power Forecasting Using Deep Learning (LSTMs)

Transition from carbon-intensive sources to renewable energy sources is of paramount importance and will curb effects of greenhouse gas emissions. India has witnessed rapid expansion in the solar power sector. Prime utilization of solar power calls for a reliable prediction model. Precise real time data of PV systems is the need of the hour. Market today requires the use of very short-term forecasts. This study aims to present a multivariate approach to forecast solar power generation of PV systems on a very short-term basis using LSTM, a deep-learning algorithm. System performance and capabilities are validated on the basis of a standard metric, Mean Absolute Error (MAE). The model incorporates other features of Machine Learning like hyper parameter tuning and optimum feature selection which further refines the accuracy.