Data-Mining Techniques Based Relaying Support for Symmetric-Monopolar-Multi-Terminal VSC-HVDC System

Abstract

Considering the advantage of the ability of data-mining techniques (DMTs) to detect and classify patterns, this paper explores their applicability for the protection of voltage source converter-based high voltage direct current (VSC-HVDC) transmission systems. In spite of the location of fault occurring points such as external/internal, rectifier-substation/inverter-substation, and positive/negative pole of the DC line, the stated approach is capable of accurate fault detection, classification, and location. Initially, the local voltage and current measurements at one end of the HVDC system are used in this work to extract the feature vector. Once the feature vector is retrieved, the DMTs are trained and tested to identify the fault types (internal DC faults, external AC faults, and external DC faults) and fault location in the particular feeder. In the data-mining framework, several state-of-the-art machine learning (ML) models along with one advanced deep learning (DL) model are used for training and testing. The proposed VSC-HVDC relaying system is comprehensively tested on a symmetric-monopolar-multi-terminal VSC-HVDC system and presents heartening results in diverse operating conditions. The results show that the studied deep belief network (DBN) based DL model performs better compared with other ML models in both fault classification and location. The accuracy of fault classification of the DBN is found to be 98.9% in the noiseless condition and 91.8% in the 20 dB noisy condition. Similarly, the DBN-based DMT is found to be effective in fault locations in the HVDC system with a smaller percentage of errors as MSE: 2.116, RMSE: 1.4531, and MAPE: 2.7047. This approach can be used as an effective low-cost relaying support tool for the VSC-HVDC system, as it does not necessitate a communication channel.