Distribution System State Estimation with Convolutional Generative Adversarial Imputation Networks for Missing Measurement Data.

In distribution systems, the state estimation task is very complicated due to the problems arising from the unbalanced operation and network configuration changes. Additionally, the convergence process of state estimation gets affected by the loss of physical measurements due to meter failure, communication failure, and large communication time delays. To address the issue of loss of measurement data, a new deep learning approach, namely a convolutional generative adversarial imputed network (CGAIN) with modified loss functions, is proposed. A branch current-distribution system state estimation (BC-DSSE) with different percentages of missing measurement data has been implemented and studied. Furthermore, the analysis has been carried out for different types of uncertainties involving several kinds of measurement devices and pseudo-measurements for different classes of consumer loads. Also, the effect of imputed data on state estimation accuracy is tested at different operating conditions of load. The imputation performance of the CGAIN model is tested by comparing it with statistical methods, and ML/DL approaches. The benefits of the proposed model are validated by Monte Carlo simulations on the modified IEEE 13-node and IEEE 37-node unbalanced distribution test systems. Simulation results show that the proposed model can effectively impute the missed data under any operating condition. Further, the robustness of the algorithm is checked with the largest normalized residual (LNR) test for minimizing the effect of bad data. [ABSTRACT FROM AUTHOR]