Real-time power system state estimation via deep unrolled neural networks

Contemporary smart power grids are being challenged by rapid voltage fluctuations, due to large-scale deployment of electric vehicles, demand response programs, and renewable generation. To secure grid operations, it becomes increasingly critical to obtain accurate estimates of power system AC states, namely the complex voltages at all buses, in real time. With the emergent nonconvexity however, past optimization based power system state estimation (PSSE) schemes are either sensitive to initialization, or computationally expensive. To bypass these hurdles, this paper advocates deep neural networks (DNNs) for PSSE. Different from model-agnostic NNs, that are difficult to tune and train, the novel model-specific DNN is obtained by unrolling state-of-the-art physics-based prox-linear PSSE solvers. The proposed prox-linear net requires a minimal tuning effort, and is easy to implement. Simulated tests show improved performance of the proposed prox-linear net relative to 'plain-vanilla' NN-, and the 'workhorse' Gauss-Newton-based PSSE solvers.