Data-Driven Distributed Grid Topology Identification Using Backtracking Jacobian Matrix Approach

The topology identification of the distribution system is the foundation of state estimation and system analyses. However, the lack of sufficient measurement devices in practice makes full-scale identification of distributed grids hardly achievable. On the other hand, the high-scale distributed renewable energy grid-connected aggravates the complex associations between nodal injection power and voltage. Taking full advantage of it, this article proposes a novel data-driven approach to identify the topology by backtracking the Jacobian matrix from the voltage and power data from grid terminal units. Specifically, a data-driven power-to-voltage mapping model is proposed. It uses the adaptive online deep learning algorithm based on hedge backpropagation to model the mapping relationship of terminal active power, reactive power, and voltage. Then, backtracking the mapping model, a Jacobian matrix consisting of the partial derivative of voltage to the active and reactive power is obtained, which reflects the connectivity between nodes. Finally, through the Jacobian matrix, a maximum connectivity screening algorithm is developed to generate the topology of medium distribution systems. The effectiveness of the proposed method is verified on IEEE 33-node, IEEE 123-node, and actual distribution systems.