Short‐Term Load Forecasting Method for Strongly Coupled <scp>AC</scp> / <scp>DC</scp> Distribution Networks Based on <scp>VMD</scp> ‐ <scp>ICOA</scp> ‐ <scp>BiLSTM</scp> ‐ <scp>MTL</scp>

Abstract With the advancement of smart grid technology, load forecasting for AC/DC hybrid distribution networks faces challenges such as strong volatility and high coupling. To address this, this paper proposes a short‐term load forecasting method based on Variational Mode Decomposition (VMD), Improved Coati Optimization Algorithm (ICOA), Bidirectional Long Short‐Term Memory network (BiLSTM), and Multi‐Task Learning (MTL). The method first employs VMD to adaptively decompose the original load series, extracting multi‐scale features and mitigating non‐stationarity. Then, ICOA is introduced to adaptively optimize the key hyperparameters of BiLSTM, enhancing the model's alignment with data characteristics. Furthermore, an MTL framework based on BiLSTM is constructed, which collaboratively models the coupling relationship and independent temporal evolution patterns of AC and DC loads through an uncertainty‐weighted loss function. Experiments based on measured data from a regional distribution network in Guangzhou demonstrate that the proposed model achieves optimal performance in forecasting both AC and DC loads, with prediction accuracy significantly outperforming traditional methods. This validates the method's effectiveness and robustness in complex load scenarios, providing reliable support for the optimal dispatch and planning of AC/DC hybrid grids. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.