Real-Time Power System State Estimation and Forecasting via Deep Unrolled Neural Networks

Contemporary power grids are being challenged by rapid voltage fluctuations\nthat are caused by large-scale deployment of renewable generation, electric\nvehicles, and demand response programs. In this context, monitoring the grid's\noperating conditions in real time becomes increasingly critical. With the\nemergent large scale and nonconvexity however, the existing power system state\nestimation (PSSE) schemes become computationally expensive or yield suboptimal\nperformance. To bypass these hurdles, this paper advocates deep neural networks\n(DNNs) for real-time power system monitoring. By unrolling an iterative\nphysics-based prox-linear solver, a novel model-specific DNN is developed for\nreal-time PSSE with affordable training and minimal tuning effort. To further\nenable system awareness even ahead of the time horizon, as well as to endow the\nDNN-based estimator with resilience, deep recurrent neural networks (RNNs) are\nalso pursued for power system state forecasting. Deep RNNs leverage the\nlong-term nonlinear dependencies present in the historical voltage time series\nto enable forecasting, and they are easy to implement. Numerical tests showcase\nimproved performance of the proposed DNN-based estimation and forecasting\napproaches compared with existing alternatives. In real load data experiments\non the IEEE 118-bus benchmark system, the novel model-specific DNN-based PSSE\nscheme outperforms nearly by an order-of-magnitude the competing alternatives,\nincluding the widely adopted Gauss-Newton PSSE solver.\n