Fault Feature Analysis and Diagnosis of Power Systems with Multi-Source Heterogeneous Renewable Energy Integration

Abstract

The high penetration of renewable energy has turned modern power systems into inverter-dominated grids, where the coupling between grid-following and grid-forming converters causes strong nonlinear interactions, distorting fault characteristics and impairing conventional diagnostic methods. To address fault feature distortion and reduced accuracy under grid-following-grid-forming coupling, this paper proposes an FFT-ResNet-based fault diagnosis approach. A 4-distributedgeneration hybrid model is built in Simulink to capture multiconverter dynamics. 17 typical fault types are simulated to obtain data, and fast Fourier transform (FFT) is applied to extract fundamental-frequency voltage/current amplitude and phase as discriminative features. A residual neural network (ResNet) is then designed to enhance feature representation and classification accuracy. Simulation results show the method achieves 99.1% validation accuracy, outperforming conventional CNN and CNNGCN by $\mathbf{1 1. 8 5 \%}$ and $\mathbf{8. 1 2 \%}$ respectively, with strong robustness across fault scenarios, offering an effective solution for intelligent fault diagnosis in renewable-dominated power systems.