Very Short-Term Power Forecasting for Photovoltaic Power Plants Using a Simple LSTM Model Based on Short-Term Historical Datasets: Case Study

Prediction of small-scale and even large-scale solar energy is a hot research field. In fact, the intermittent nature of solar energy can disrupt the stability and management of electrical energy. Therefore, photovoltaic output power forecasts are recommended to solve this problem. In this paper, we will present a neural network-based photovoltaic (PV) power prediction method, in particular the long-short-term memory (LSTM) method. We are interested in this work in very short-term predictions because of its lack of presence in the literature. The proposed LSTM method is trained for this prediction category ranging from 5 min to 60 min (1 h) ahead. Then, the LSTM approach is trained and tested on real data measured from the PV power plant installed in northern Morocco. Based on the evaluation indices of the prediction models, i.e. R2, MAE, RMSE and MAE, the results found show that the optimized LSTM model showed its robustness and prediction efficiency on all validated prediction horizons. Moreover, the proposed method exhibits more accuracy in terms of RMSE errors compared to other well-known methods in the literature, including RNN, CNN and ELM.