A frequency spectrum-based method for detecting and classifying faults in HVDC systems

•Intelligent approach for protecting the whole HVDC system.•Only local signals are used for detecting and classifying faults.•Additional hardware or communication links are not needed.•Able for protecting VSC and CSC-HVDC systems.•Accurate performance for different fault characteristics. This paper proposes a method based on Artificial Neural Networks - ANNs for detecting and classifying faults in HVDC systems. An analysis of the frequency spectrum of different fault types in a HVDC system allowed finding different patterns of frequency spectrum for different fault types. As a result, the proposed method uses only the spectrum of the voltage signals to identify and classify faults in any point of the system, that is, upstream of the rectifier substation, in the DC line, or downstream of the inverter substation. The method is composed of an online step, to detect the fault, and an offline step, to classify it. The set of ANNs employed in the proposed scheme uses the same topology for all ANNs, which simplifies its practical implementation. The proposed algorithm was validated using VSC and CSC based technologies, being able to precisely detect and classify faults in HVDC systems.