An Adaptive Forecasting Method for Net Load Ramping Demand Based on Time–Frequency Dual‐Modal Collaboration

With the increasing penetration of renewable energy, the limited ramping capability of power systems within short timeframes has become a critical challenge for maintaining system security and stability. This imposes higher requirements on the accuracy of ramping demand forecasting. Accurate ramping demand prediction relies on reliable net load forecasting. However, traditional methods often struggle to capture complex temporal patterns, including periodic fluctuations, long‐term trends, and abrupt anomalies. To address these challenges, this paper proposes an Adaptive Enhanced model based on Time‐Frequency Fusion Networks (AETFFN) for net load ramping demand forecasting. First, a dynamic frequency selection module is designed to adaptively identify key frequency components by analysing spectral energy and sparsity, which suppresses high‐frequency noise and enhances core periodic features. Second, a gated time‐frequency fusion module is constructed by integrating one‐dimensional convolution and Fast Fourier Transform to extract frequency‐domain features, while dynamic weighting ensures effective fusion of time and frequency information. Additionally, a lightweight convolutional module is introduced, combining multi‐scale convolution with dual attention mechanisms to improve both local detail extraction and global pattern recognition while maintaining computational efficiency. The proposed method is evaluated on multiple datasets, with both ablation and comparative experiments conducted to validate its effectiveness. Results show that AETFFN outperforms other benchmark models across three net load datasets, achieving an average RMSE reduction of 27.91%. This confirms the effectiveness of the proposed approach in net load forecasting and its ability to accurately estimate ramping demand.