Long-term forecasting of solar and wind energy patterns using deep learning

The significant volatility and randomness of renewable energy sources such as wind and photovoltaic power increase the operational uncertainty of power systems, while short-term forecasting struggles to capture monthly and seasonal characteristics of wind-solar data. To address the challenge of long-term time series prediction in renewable energy integration, this article proposes a wind-solar long-term time series prediction method based on a hybrid forecasting model combining Seasonal-Trend decomposition using Locally Estimated Scatterplot Smoothing (LOESS), Long Short-Term Memory, and Frequency Enhanced Decomposed Transformer. This approach combines the advantages of Frequency Enhanced Decomposed Transformer and Attention-Long Short-Term Memory Networks models, which have demonstrated excellent performance in time series prediction, and achieves complementary benefits through Seasonal and Trend Decomposition Using LOESS data decomposition. Experimental results show that the proposed method significantly improves Auckland wind sequence prediction, with mean squared error and mean absolute error reduced by up to 32.8% and 20.8% respectively, across different prediction horizons compared to existing classical prediction models.