Prediction of wind energy by using long short-term memory neural networks

• Main Authors: Bouafia, Mohammed, El-Hammouchi, Azeddine, Akchioui, Nabil El, Fathi, Amine El
• Format: Conference Proceeding
• Language: English
• Subjects: Alternative energy; Correlation coefficients; Energy management; Forecasting; Meteorological data; Neural networks; Recurrent neural networks; Renewable energy; Renewable resources; Wind power
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0149216

In order to achieve sustainable renewable energy management and satisfy future needs, the forecast of renewable energy production is obvious. The main objective of this work is to forecast wind energy production using deep learning. In fact, good planning of wind energy requires a forecast of its production. This is necessary to find a reliable balance between production and energy demand. Our prediction was made by recurrent neural networks, specifically networks, which have the ability to learn from current and previous information to find a better solution. The Long Short-Term Memory neural networks are able to store useful information over an arbitrary period of time. In addition, LSTM cells have the ability to learn which data should be read, stored and deleted from memory. These characteristics make LSTM networks very suitable for forecasting wind energy production. Forecasting by an LSTM network requires a database of historical power and meteorological data. These play the role of improving the accuracy of the forecast. In this context a comparison based on correlation coefficients (Pearson, Spearman and Kendall), in order to find the adequate coefficient and the meteorological data that should be taken into account for forecasting wind energy production is done.