A Robust Deep Learning Approach for Distribution System State Estimation with Distributed Generation

Abstract

Distribution System State Estimation (DSSE) remains a challenging problem due to the nature of distribution grids. Conventional methods, which are used to solve state estimation on the transmission level, require the grid to be observable. This is not directly applicable to distribution grids. In addition, the high integration of renewable energy introduces uncertainty, which makes the DSSE problem more complex. This work proposes a deep neural network approach that solves the DSSE problem with and without distributed generation, without using highly inaccurate pseudo-measurements. Due to the lack of public frameworks, we create a dataset that emulates real-life scenarios to train and test the neural network. Also, to evaluate the robustness of the algorithms, we test the neural network, without retraining it, on multiple scenarios with noisier data and bad data. The algorithms are tested on three different networks. The proposed approach solves the DSSE problem with limited measurements as inputs, which cannot be solved using conventional state estimation methods. Our approach also achieves highly accurate results, despite the additional noise introduced to the measurements.