Supervised Learning Approach for State Estimation of Unmeasured Points of Distribution Network

Abstract

This thesis presents a new approach to state estimation (SE) of distribution networks, which becomes more complex when there is lack of monitoring. Several studies have been carried out on SE to compensate for the lack of monitoring; however, the observability of the distribution system is poor compared to the transmission system. In the proposed approach, the representative load profile and the electricity charges of consumers are required to obtain the load profile of each consumer. In addition, the uncertainty was considered owing to the poor accuracy of these obtained load profiles, and the results were analyzed according to the uncertainty. The obtained load profiles were used to calculate the voltage magnitudes and angles by power flow calculations, and the calculated voltage magnitudes and angles were used to train the used supervised learning algorithms including the feed-forward neural network (FFNN), linear regression (LR), and support vector machine (SVM). IEEE 13-, 34-, and 37-node test feeders were used to verify the proposed approach. The proposed approach is not applicable to a terminal bus; however, the voltage magnitudes and angles of consecutive unmeasured buses more than two can be estimated. In addition, the maximum magnitude of the absolute error of the estimated voltage magnitude was 0.01 p.u. or less, and the maximum magnitude of the absolute error of the estimated voltage angle was lower than 0.19°.