Super-Resolution Reconstruction Method for Power System Data Considering Weather Impacts based on Self-Attention TimeGAN

High-proportional renewable energy integration and rapid installation of new type loads, such as electric vehicles (EVs), into the power system are challenging the monitoring and operation of the power system. Different from the conventional system operation requirements, high-quality and real-time power grid data become essential for system operators on key power equipment monitoring, power flow dispatching, and operation optimization. However, due to equipment limitations, bandwidth, storage and other factors, power grid data is usually collected at lower sampling frequencies. Therefore, super-resolution reconstruction of these data is necessary to obtain higher frequency data for meeting the system operation requirements. In this paper, a super-resolution reconstruction method based on self-attention-time series generative adversarial network (SA-TimeGAN) is proposed. The proposed method considers the influence of meteorological factors by introducing a self-attention mechanism. The developed generator can more effectively capture long-distance dependencies in power grid data and improve the quality of generated data, which could help the operators obtain higher resolution and more realistic data. The performance of the proposed super-resolution reconstruction method is verified based on the real distribution system data obtained from northern China. The experiment results are validated through indicators such as reconstruction error and timing characteristic evaluation and compared with traditional generative adversarial network (GAN) reconstruction methods. The experiment results indicated that the proposed super-resolution reconstruction method has better performance compared to the traditional methods.