Graph Signal Denoising in Power Systems and its Effects on GSP-Based State Estimation

Noise in measurement data is not a new challenge, but the rise of graph-aware data processing tools, such as Graph Signal Processing (GSP), offers new opportunities to improve noise handling by capturing the relationships and interactions among data points. Due to the physical structure of the power systems, the measurements collected from these systems can be represented in the form of graph signals and various GSP-based techniques are emerging to process such data. This work first examines the sensitivity of a GSP-based state estimation technique to measurement noise. It then presents various denoising techniques, ranging from classical signal processing methods to novel GSP-based graph filters, and studies their impact on the quality of graph signals. Finally, the effects of these denoising filters on state estimation in power systems are evaluated. The results show that denoising processes that consider both the underlying graph of interactions and the temporal relationships in data outperform other denoising techniques. While the denoising techniques presented here are examined in the context of power systems, they are general enough to be applied to any graph signal.