Neural network-based multi-parameter fault identification for hybrid-source transmission lines

Abstract

With the large-scale integration of inverter-based resources (IBRs), the types and operating characteristics of power sources on both sides of transmission lines have changed significantly, rendering traditional fault-type selection methods inadequate. To address this, this paper proposes a neural network-based multi-parameter fault type identification strategy. The method constructs a 16-dimensional feature vector from local three-phase voltage/current magnitudes, phase angles, and zero-sequence components, and designs a lightweight fully-connected neural network with two hidden layers to learn the complex nonlinear mapping between these comprehensive inputs and fault types. Extensive training and testing data are generated using the PSCAD/EMTDC simulation platform, covering multiple scenarios including double-ended synchronous generator (SG), single-ended IBR, and double-ended IBR. The results show that the proposed strategy achieves identification accuracy exceeding 95% across all scenarios, significantly outperforming traditional current-based methods, especially in IBR-dominated cases. Moreover, the method exhibits strong robustness against variations in fault location, transition resistance, and source type, providing a reliable and adaptive protection solution for evolving hybrid power grids.