Continuous Feature Representation of Fault Location Based on Electrical Distance for Generalizable Data-Driven Dynamic Security Assessment

Abstract

Pre-fault dynamic security assessment (DSA) of power systems needs to consider different fault locations. Existing data-driven DSA methods lack generalizable fault location features, failing to accurately assess unlearned fault locations, i.e., fault locations not in the training set. To address this issue, this paper proposes a continuous feature representation method of fault location based on electrical distance, aiming to construct a pre-fault DSA model generalizable to unlearned fault locations. Firstly, fault location is represented continuously as an electrical coordinate with the electrical distance to preselected reference nodes. Then, the representation ability of electrical coordinates to fault locations is optimized by reference node selection and metric learning. Finally, with electrical coordinates and steady-state features as input, a hybrid DSA model composed of convolutional and radial basis function networks is constructed to assess system angular and voltage dynamics with different fault locations. The generalization performance of the proposed DSA method on unlearned fault locations is verified on New England 39-bus system and a provincial system of China.