Raw measurement supervised learning transformer for anomaly detection of power system digital twin updates

Continuous updates are essential to ensure that a digital twin (DT) remains an accurate representation of its physical counterpart. The performance of DT applications heavily relies on how accurately the DT reflects its physical counterpart. DT updates, however, can be compromised by anomalous PT data stemming from physical twin (PT) measurements, communication malfunctions, and/or external attacks. Detecting such anomalies in PT data is crucial to ensuring the accuracy and reliability of DT, thereby generating only valid outcomes for associated applications. This paper proposes a detection method to identify anomalous PT data before its integration to the DT. The proposed raw measurement supervised learning Transformer (RM-SL-TF) facilitates a straightforward identification of PT data using raw measurements, eliminating the dependency on data preprocessing. The feasibility and effectiveness of the RM-SL-TF are demonstrated by using a power system digital twin (PSDT) that requires frequent updates. The resulting detection accuracy of anomalous PT data is comparable to, or even surpasses, that of other artificial intelligence (AI) algorithms that rely on input feature normalisation. By directly analysing raw measurements without normalising input features, the proposed approach is simpler, more flexible, and expandable, making it suitable for establishing and advancing the development and implementation of DTs for power systems and other industries. • Detection of anomalous PT data is required before DT updates. • Introduction of a new function to enhance DT accuracy. • A Transformer-based solution allows for direct evaluation of raw measurements. • Comparison with algorithms using normalised inputs shows superior accuracy. • Expandable and flexible solution for DT implementation.