Monitoring Service for Updating Power System Digital Twins

A digital twin (DT) is a dynamic replica of its physical counterpart, maintained through continuous data flows that ensure ongoing comparability between the two. Measure-ment and communication failures may introduce errors in the data flows from the physical twin (PT), potentially leading to significant deviations in the DT and undermining the reliability of DT-based applications. The absence of a framework or service to evaluate the PT data errors delays an early detection of the DT divergence. When the PT data contains anomalies or unacceptable errors, updates to the DT are prevented, avoiding impacts on its applications. This paper proposes a practical monitoring service supported by an unsupervised learning (UL) algorithm for the PT data in the DT updates. As the core part of the monitoring service, the development of an auto-encoder is presented using the captured datasets from a virtual physical twin (VPT), and offline evaluation outcomes are provided. The results demonstrate that the auto-encoder can effectively identify different deviation levels in sampled VPT datasets in order to maintain satisfactory DT accuracy. The monitoring service has potential to guarantee the reliability and effectiveness of the DT and its applications. The proposed service creates an opportunity to expand the DT-based solutions in broader and more critical applications of future power systems.