SeqESR-GAN-Based Sparse Data Augmentation for Distribution Networks

作者：Tao Xu, Jiadong Zhang, Meng He, Lutong Liu, Kaiqi Wang, Qiao Ji, Zixuan Zhao, Zhu Hong, Wendi Wang · 发表于：IEEE Transactions on Industrial Informatics · 年份：2024 · DOI：10.1109/tii.2024.3431009 · 被引用次数：8 · 研究领域：Smart Grid and Power Systems

The increasing integration of distributed energy resources brings considerable uncertainties to power distribution systems. Limited redundancy in measurement not only constrains dynamic state estimation but also inevitably overlooks the electricity generation/utilization characteristics of typical prosumers. To enhance the system observability, a three-stage sparse data augmentation framework, namely sequence-to-sequence enhanced super-resolution generative adversarial network (GAN) is established. A novel data image encoding method is proposed to reflect the periodic electricity utilization patterns of renewable energy sources, loads, and energy storage systems, and enhances measurement data resolution by learning energy flow behaviors. Different learning strategies are employed in the initial two stages, enabling the GAN-based model to capture the spatiotemporal characteristics of the energy tensor and restore high-frequency elements. The third stage involves a gated recurrent unit-based Seq2Seq model to eliminate time lags and invalid details in the super-resolved data. Case studies and analyses are carried out to validate the effectiveness of the proposed approach.