Day-Ahead Power Forecasting of Renewable Energy Sources Using Neural Networks and Machine Learning

The paper considers the problem of day-ahead forecasting the power of renewable energy sources such as solar panels and wind turbines. To solve this problem, the short-term predictor principle was used. Well-known neural network architectures and machine learning algorithms were used as forecasting models. Numerical experiments have shown that to improve the forecast stability and quality, additional features should be used along with power values. The paper shows that mode values, obtained using the Empirical Wavelet Transform and introduced as additional features, significantly improve the forecast quality. Neural networks such as Multilayer Perceptron and machine learning algorithms Random Forest and Yandex’s CatBoost showed the best forecasting accuracy. The results obtained can be used for optimal energy consumption scheduling of enterprises powered by renewable energy.