Deep Generative Model-Aided Communication-Efficient and Cyber-Resilient Power System Dynamic State Estimation

Abstract

Fast and robust dynamic state estimation (DSE) is critical for capturing the internal dynamics of cyber-physical power systems. Traditional model-based approaches, such as Kalman filtering and particle filtering, require accurate parameters, control inputs, phasor measurement unit (PMU) data, and reliable centralized communication. These dependencies make them vulnerable to accuracy bottlenecks under structural or process errors, unknown control vectors, cyberattacks, and communication contingencies. In contrast, deep learning-aided DSE eliminates parameter dependence but often suffers from limited generalization under unforeseen operating conditions. To overcome these challenges, this paper proposes a deep generative model-aided framework with AI-generated content (AIGC) for error-bounded DSE. The framework integrates an encoder-decoder architecture to estimate unknown control inputs, a robust encoder to suppress PMU anomalies, and a latent diffusion model to mitigate communication issues in centralized DSE. In addition, a lightweight adaptor is introduced to rapidly adjust latent vector distributions under unforeseen contingencies and moderate topology variations. Extensive experiments on the IEEE 39-bus and NPCC 140-bus systems demonstrate that the proposed method effectively addresses modeling imperfections, measurement uncertainties, communication contingencies, and distribution shifts, while also reducing storage and communication requirements. Overall, the proposed error-bounded approach provides a powerful complement to model-based DSE methods dominated by Kalman filtering (KF).