Addressing intermittency in medium-term photovoltaic and wind power forecasting using a hybrid xLSTM-TCCNN model with numerical weather predictions

Accurate medium-term forecasting of wind and solar power generation is essential for optimizing renewable energy utilization, stabilizing power grids, and supporting electricity market operations. However, achieving high-accuracy predictions remains challenging due to the intermittent and nonlinear nature of renewable energy sources. This paper proposes a novel hybrid forecasting model, NWP-Time2Vec-xLSTM-TCCNN, which integrates numerical weather prediction (NWP) data with advanced periodic feature analysis to address these challenges. The model incorporates an enhanced xLSTM framework, applied for the first time in hybrid power forecasting, to capture complex temporal correlations, with a detailed analysis of sLSTM and mLSTM layer pairings on performance. Additionally, a Time-series Cascade Convolutional Neural Network (TCCNN) is introduced to mitigate feature loss in deep CNNs and enhance the ability to model nonlinear relationships among multiple variables. Experimental validation on wind–solar datasets from power plants of different scales in Natal and Belgium shows that the proposed model significantly outperforms state-of-the-art methods, including Time2Vec-WDCNN-BiLSTM, LSTMformer, and IEDN-RNET, reducing mean absolute error by 12.28 %–20.00 % for photovoltaic power and 10.33 %–11.53 % for wind power. These findings highlight the model's superior accuracy, robustness, and scalability, providing a powerful tool for advancing renewable energy forecasting and supporting efficient management of sustainable energy systems and electricity markets.