A Composed Machine Learning Algorithm for Fault Classification in Hvdc Lines

Protection system is an integral part of an electrical transmission systems. Previously, various protection methods has been proposed for AC systems and has been hammered for years to get a better performance. In the last few decades, HVDC transmission has been introduced as a solution for long distance transmissions and offshore transmission, thanks to power electronics technology developments. A fast and flexible control, large capacity of transmission, economic justification for distances over 500km (depends on power electronics technology) and less occupied Right of Way (RoW) comparing to HVAC transmission for a certain power capacity[1], are some advantages of utilizing HVDC transmission. A statistical analysis on a HVDC transmission system in china shows that 36.8% of 114 valve group outages were caused by line protection zone faults[2]. Thus a reliable protection method can prevent wrong fault detection and decreases whole power outage as a result. Differential protection is one of traditional solutions which has been employed to AC transmission. in [3] a SIEMENS typical HVDC line protection configuration has been studied which has introduced differential protection as a backup protection. An improved differential protection of Current Source Converter (CSC)-HVDC transmission lines has been proposed in [4] which a compound of a blocking unit and a newly defined differential current criteria has been used for fault detection. [5] Has defined a signal distance between rectifier side and inverter side currents which external and internal fault can be distinguished.

The effect of capacitive current and the problem communication channel requirements has been mitigated in [6], [7]. Transient power and other combinations of two sides voltage and current measurements has been studied in [8]–[10]. The presence of smoothing reactor and DC side filters at both ends of a CSC-HVDC link, allows some selective non-unit protection methods[11] to be implemented such as Using the impedance characteristic of smoothing reactors and DC side filters in a faulty conditions which has been carried out in [2], [12], [13]. Travelling Wave Protection (TWP), which is used as primary protection [14], can benefit from some methods such as Principle Component Analysis (PCA), Wavelet Transformation (WT) and etc [15]–[17], but the attenuation and distortion of traveling wave which is caused by fault resistance and fault location, should be considered. Although some protections has been suggested for VSC-HVDC systems, they can be applied to CSC-HVDC transmission lines too. (methods based on current and voltage derives can be mentioned too, which in case, references in introduction part reaches above 23). It should be noted that almost all mentioned researches have used the concept of threshold as a criteria which is usually obtained non analytical(based on worst case). Usually there is a trade-off between security and dependability in protection systems which causes inaccuracy in threshold based methods. at march 21st ,2005, primary protection of Tian-Gaung HVDC system could not detect a high impedance fault because amplitude of did not reach its threshold value [18] and the backup protection refused to act because of current fluctuations. To overcome the mentioned problem, we present a composed machine learning algorithm based on one side measurements, which omits the use of threshold in protection. The suggested solution contains two different machine leaning methods, a binary Support Vector Machine (SVM) as a starting unit and a K-Nearest Neighborhood (KNN) classifier for fault classification. In section Ⅱ the basis of employed classifiers is introduced. Section Ⅲ includes the characteristics of faults on a bipolar CSC HVDC test model, and selected features of the system as the classifier inputs are also studied in this section. In section Ⅳ, the proposed algorithm is applied to the test model, simulation is carried out and results and verification tests are presented. Finally section Ⅴ is dedicated to the conclusion.