Transfer learning-based fault location with small datasets in VSC-HVDC

Abstract

Fault location plays a critical role in long-distance voltage source convertor based high-voltage direct current (VSC-HVDC) transmission systems. Although machine learning-based fault location methods were proved to be effective in some simulation scenarios, the small fault data sets from practical transmission lines are always a barrier for its application. Transfer learning can reach a fast convergence with a small data set when the model has learned knowledge with large data sets which have similar distributions. This paper proposed a transfer learning-based fault location method for VSC-HVDC transmission lines and discussed its performance under different conditions. The method uses stacked de-noising auto-encoder to model the relationship between current waveforms of travelling waves and fault locations, and fine-tunes the model with small data sets from the target transmission line through transfer learning. The proposed method is tested with a simulated VSC-HVDC transmission line on real time digital simulation platform. Results show the proposed method can locate faults with small training data sets and stand the influences from fault type, grounding resistance, and noise. The research results can provide technical support for the practical application of artificial intelligence in transmission grid.