A fast data-driven topology identification method for dynamic state estimation applications

This paper proposes a fast topology identification method to avoid estimation errors caused by network topology changes. The algorithm applies a deep neural network to determine the switching state of the branches that are relevant for the execution of a dynamic state estimator. The proposed technique only requires data from the phasor measurement units (PMUs) that are used by the dynamic state estimator. The proposed methodology is demonstrated working in conjunction with a frequency divider-based synchronous machine rotor speed estimator. A centralized and a decentralized approach are proposed using a modified version of the New England test system and the Institute of Electrical and Electronics Engineers (IEEE) 118-bus test system, respectively. The numerical results in both test systems show that the method demonstrate the reliability and the low computational burden of the proposed algorithm. The method achieves a satisfactory speed, the decentralized approach simplifies the training process and the algorithm proves to be robust in the face of wrong input data.