Data Anomaly Detection in Modern Transmission Grids Based on GRU-Attention-VAE

With the increasing proportion of renewable energy in power grids and the growing frequency of climate changes, traditional power system state estimation methods tend to misidentify sudden load changes as anomalies, thereby affecting the quality and speed of decision-making in subsequent energy management tasks. Conducting research on anomaly detection techniques for the power dispatch cloud to adapt to the multi-source nature, long time-scale, and high timeliness requirements of dispatch cloud data holds significant application value. This paper proposes a deep learning-based anomaly detection algorithm for electrical data. The model integrates the weakly supervised learning of Variational Autoencoders (VAE), the temporal feature extraction of Gated Recurrent Units (GRU), and the cost-effective extraction of core rules/characteristics in complex systems enabled by the Attention mechanism. The effectiveness and accuracy of the algorithm are verified through case studies, and comparisons with baseline algorithms demonstrate its cost-effectiveness.