Data Imputation Using Self Attention Based Model for Enhancing Distribution Grid Monitoring and Protection Systems

The availability of high-fidelity time-series data is essential for distribution grid operations such as state estimation, prediction, protection, and scheduling of distributed energy resources. However, disruptions in the metering systems, such as latency, equipment malfunctions, and communication congestion, could lead to missing data points. In this article, we propose a fully data-driven approach for the distribution grid aimed at imputing the missing data from distributed-phasor measurement units (D-PMUs). An advanced method is proposed named self-attention-based imputation for time series (SAITSs) which is adapted to handle the complex and diverse data found in digital substations. The proposed method has the ability to accurately impute data even during fault events, thereby mitigating the potential misoperation of protection devices. This enables accurate data imputation and reconstruction of missing data points, thereby preserving the continuity and integrity of the dataset. We have validated the performance of the proposed method by comparing it with various traditional methods. By implementing the proposed imputation technique, the accuracy, and reliability are significantly improved, which enhances the decision-making processes for digital substations.