Distribution Grid Topology and Parameter Estimation Using Deep-Shallow Neural Network with Physical Consistency

To better monitor and control distribution grids, the exact knowledge of system topology parameters is a fundamental requirement. However, information usually incomplete due to limited sensors in grid. Therefore, estimating using partial data critical topic for systems. Due high nonlinearity unobservable quantities noises, Deep Neural Networks (DNNs) are widely utilized accurate estimation. While traditional approaches either treat DNNs as black box or embed little physical into DNNs, they cannot guarantee that DNN model consistent with equations hence lack accuracy interpretability. we propose Deep-Shallow neural Network (DSN) The key create virtual nodes represent without system, denote approximate missing at nodes. Then, Power Flow (PF) can be estimated via shallow network, achieving consistency. Isolated by nodes, whole decomposed set reduced graphs PF equations. Likewise, introduce Reinforcement Learning-based search algorithm connect grids one connected system. Correspondingly, DSN fine-tuned achieve consistency graph. Finally, paper illustrates superiority proposed its Specifically, comprehensive experiments demonstrate performances our over other methods grids.