Fast State Estimations for Large Distribution Systems using Deep Neural Networks as Surrogate

Traditional iterative-based numerical methods for distribution systems state estimation suffer from slow convergence on large systems. Confronted with increasing renewable integration and its fast power variations, distribution systems call for faster state estimations. In this context, a combination of a machine learning surrogate and a numerical method is developed for distribution system state estimations in this paper. First, a deep neural network model is trained as a fast yet coarse surrogate to the iterative state estimator. Then, the surrogate is fed into a forward/backward sweep estimator as initial values for refinement. By starting from a better “guess” using the surrogate model, the iterative method converges much faster. The proposed method is validated on IEEE 123-bus and 8500-node three-phase unbalanced test systems.