Short-term self consumption PV plant power production forecasts based on hybrid CNN-LSTM, ConvLSTM models

Global electricity consumption has raised in the last century due to many reasons such as the increase in human population and technological development. To keep up with this increasing trend, the use of fossil resources has increased. But these resources are not environmentally friendly, and for this reason, many countries and governments are encouraging the use of green sources. Among these sources, PV technology is widely promoted and used due to its improved efficiency and lower prices for photovoltaic panels. Therefore, the importance of forecasting power production for these plants is necessary. In this work, two hybrid models were proposed (CNN-LSTM and ConvLSTM) to effectively predict the power production of a self-consumption PV plant. To confirm the efficiency of the proposed models, the LSTM model was used as a baseline for comparison. The three models were trained on two datasets, a univariate dataset containing only the power output of the previous days, while the multivariate dataset contains more features (weather features) that affect the production of the PV plant. The time frames for the forecast ranged from one day to one week ahead of time. The results show that the proposed methods are more accurate than a normal LSTM model. • CNN-LSTM and ConvLSTM models are proposed for forecasting PV output power. • The proposed models' results are compared to the LSTM model. • The models were tested on real data, acquired from a PV plant in Rabat, Morocco. • The univariate dataset performs better based on the proposed models than the multivariate dataset. • The accuracy of the designed hybrid models' increases as the forecasting horizon, shortens.