Research on power dynamic data sample generation technology based on brain-like computation and its efficient computation methods

The paper aims to address the issue of insufficient fault samples in UHV converter equipment, which hinders their intelligent operation and inspection. For the operation and inspection of UHV converter equipment, this paper suggests a multimodal brain-like learning sample spatio-temporal correlation generation method. This method grabs typical fault samples from the defect-fault development time sequence process and creates samples using the nearest-neighbor generating segment technique by fusing the time sequence evolution law and the similarity of the adjacent samples. Based on the physical model of the converter and converter valve, we analyze the fault development laws of partial discharge, high temperature overheating, and micro-motion wear. The multimodal fault sample generation model with an embedded fault mechanism is established by integrating the time-sequence fault evolution mechanism and the spatial correlation between multimodal state quantities. The simulation demonstrates that brain-like learning generates samples embedded in the fault evolution laws of converter partial discharge and converter valve IGBT micromotion wear in 539 columns, encompassing 376 converter and 163 converter valve cases. The consistency between the generated samples and the actual samples exceeds 90%, thereby facilitating the training of brain-like models for health assessment of extra-high-voltage converter equipment, fault diagnosis, and trend prediction.