Deep Learning Techniques for Very Short-Term Forecasting of PV Power

Precisely forecasting the power output during solar power generation can significantly mitigate the effects of unpredictability and instability on the power grid system, leading to improved stability, optimal management, and reduced operational expenses. Solar photovoltaic (PV) power generation is influenced by various environmental factors, including temperature, relative humidity, global sun radiation, and wind speed. It is susceptible to significant variations in production under different weather conditions. Hence, it is crucial to prioritize the development of accurate models that can effectively forecast very short-term solar PV generation. This article explores an approach for forecasting the very short-term power output of photovoltaic power plants using deep learning methods. A deep learning strategy using the Long Short Term Memory (LSTM) algorithm is assessed for its ability to predict solar power data. We compared the performance of the LSTM network to the Gated Recurrent Unit (GRU), Multi-layer Perceptron (MLP) and Convolutional Neural Network (CNN) using MAE, RMSE and R2.