LSTM Network and Box and Jenkins Methodology for Time Series Forecasting: Solar Energy Production

Solar power generation experiences a significant surge, the forecasting of incoming solar energy is becoming increasingly crucial. The aim of this study is twofold. First, it aims to show how to overcome some of the shortcomings of standard methods of forecasting solar energy production using time series models and to propose an alternative empirical approach based on Recurrent neural network (LSTM algorithm). Secondly, it aims to fill the research gaps in forecasting in the Sfax region. In this regard, for the production of solar energy from June 2019 to March 2020, we use the ARIMA model for solar energy forecasting. Based on the metrics, we evaluate the effectiveness of time series models. By implementing the LSTM method, we note that, empirically, the LSTM models give better results in forecasting solar energy production for the Sfax region.