Distribution System State Estimation Via Data-Driven and Physics-Aware Deep Neural Networks

摘要

Massive integration of renewables and electric vehicles comes with unknown dynamics - what exemplifies the need for fast, accurate, and robust distribution system state estimation (DSSE). Due to limited real-time measurements however, optimization-oriented DSSE faces major challenges related to convergence, as well as multiple global/local minima. To address these challenges, this paper puts forth a novel deep neural network (DNN)-based computational framework for DSSE that consists of two modules: a deep recurrent neural network (RNN) based pseudo-measurement postulating module, and a prox-linear net-based real-time state estimation module. Both RNN and prox-linear nets learn complex nonlinear functions, and can afford efficient training by leveraging existing deep learning platforms. Numerical tests with semi-real load data demonstrate the merits of the DNN-based DSSE approach.