Probabilistic State Estimation Approach for AC/MTDC Distribution System Using Deep Belief Network With Non-Gaussian Uncertainties

Abstract

The increasing complexity of distribution grids due to widespread deployment of renewable resources and/or power electronic devices, e.g., voltage source converters, has necessitated the needs of distribution system state estimation (DSSE) for efficient control relying on an accurate picture of the system states. This paper therefore explores the application of using the deep belief network (DBN) for pseudo measurements modeling in the context of DSSE. Two DBNs are trained respectively, for active and reactive power injection outputs, with load profiles and limited number of real measurements. Given the non-Gaussian behavior of the estimated pseudo measurements, we model the error by Gaussian mixture distribution and accordingly design a state estimator based on the Gaussian component combination method (GCCM). This method is able to handle the non-Gaussian measurement uncertainty while retaining the framework of the classic weighed least square (WLS) algorithm. The effectiveness of the proposed DSSE is demonstrated on a modified US PG&E69 distribution network in terms of the accuracy for both the estimated quantities and the associated uncertainty distributions.