A Deep Learning Approach to Short-term PV Power Generation Forecasting

This paper presents PV power generation forecasting. The proposed forecasting model is based on long short-term memory (LSTM) model. PV power plant datasets, and cloud covers are used as input for training the model. Cloud cover was estimated from the all-sky images. Images taken from all-sky camera feature details of cloud movement and the sun’s position over the sky hemisphere. To evaluate the performance of the developed model, the mean absolute error (MAE) and the skill score (SS) are measured. The comparison results between the persistent, MLP, CNN, and LSTM (without cloud cover) illustrate that the LSTM model has the most accurate. However, the difference in performance between LSTM (without cloud cover) and LSTM (with cloud cover) is ambiguous.