A novel deep learning architecture for distribution system topology identification with missing PMU measurements

Abstract

With the increased penetration of distributed energy resources (DERs) in the distribution networks, it is critical to accurately identify the topology in real time. However, the loss of physical measurements due to meter failures or denial of service (DoS) attacks on communication channels creates the major challenges in topology identification (TI), and it also impairs the convergence or solution of the distribution system state estimation (DSSE). Addressing the issues of missing measurements, a novel informer model is proposed for the TI in this paper. The performance of the proposed approach is evaluated by randomly missing the phasor measurement unit (PMU) voltage measurements. The compensation of missing measurements using the proposed approach is compared with the robust attention based approaches at different contamination levels. Further, the TI is evaluated after filling the lost data with the compensated measurements given by the proposed model. The corresponding topology classification accuracy is compared with other attention-based approaches. Based on the experimental results, the proposed approach outperformed the comparison attention based models in the compensation of missing measurements, and the identification of accurate topology. The root mean square error (RMSE) of the proposed approach has an improvement of 23%, and 25% over the second-best model for the respective modified IEEE 13 node and 37 node test systems in the missing measurement compensation.