Lightning Strike Identification Algorithm of an All-Parallel Auto-Transformer Traction Power Supply System Based on Morphological Fractal Theory

Lightning strikes are one of the major causes leading to faults in the traction power supply systems. Quick identification of the lightning shielding and backflashover failures is a complex task. In this paper a lightning strike identification model based on morphological fractal theory is constructed. A simulation model of lightning-struck tractive network faults is developed aiming at investigating the influence of the structure of the all-parallel AT tractive network on the waveform characteristics. The differences in the waveform characteristics under lightning shielding and backflashover failures are studied. Further, the fault analysis of lightning strike signals is carried out based on mathematical morphology. The influence of different structural elements and basic operations on the effect of morphological transformation is obtained. The classification effects of single- and multi-scale fractal dimensions of voltage and current waveforms are compared based on fractal theory. Moreover, a neural network framework is employed using the fractal dimensions of samples at different scales as feature vectors and the fault state as the output for establishing a lightning-strike identification model. The result shows that this method can effectively identify lightning strike faults in all-parallel AT traction power supply system with an accuracy of more than 95%.