State Estimation for Unobservable Distribution Systems via Deep Neural Networks

The problem of state estimation for unobservable three-phase unbalanced distribution systems is considered. A Bayesian approach is proposed that combines Bayesian inference with deep neural networks to achieve the minimum mean squared error estimation of network states for real-time applications. The proposed technique learns probability distributions of net injection from smart meter data and generates samples for training a deep neural network. Structural characteristics of the deep neural networks are investigated. Results illustrate the benefits of deep learning and robustness against distribution errors and bad data. Comparing with the pseudo measurement techniques, direct Bayesian state estimation with deep neural networks significantly outperforms existing pseudo measurement techniques, including those using neural-network based pseudo measurement approaches.