Fault Location in VSC-HVDC Using Stacked Denoising Autoencoder

This paper proposed an intelligent algorithm based approach for fault location in a high voltage direct current (HVDC) transmission system. To obtain post-fault signals, the point-to-point HVDC transmission lines, including overhead lines and cables, are modeled on PSCAD/EMTDC. The data set is split into two parts used for training and testing, separately. The proposed method uses stacked denoising autoencoder (SDAE), which takes the raw training data as the input of network and can directly obtain fault locations. SDAE with unsupervised learning is utilized to extract representative features automatically from raw data in pre-training. Then labeled data is applied to network for fine-tuning in a supervised manner. The testing data is used for the evaluation of the proposed method. The simulation results indicate that the SDAE based method performs well in fault location and has robustness against noises, ground resistances, and system parameters.