CCF Early Warning Method of LCC-HVDC Based on XGBoost Algorithm

Zhizhong Li,Bin Wang,Haigang Wang,Bin Yu,Min Xie,Shi Chen,Xinzhou Dong

DOI: https://doi.org/10.1049/cp.2019.0459

2019-01-01

Abstract:The lack of analytical mechanism of continuous commutation failure (CCF) process hinders the development of CCF early warning algorithm and further obstructs the coordination strategy of relay protection between AC and DC size in hybrid grid. However, summarizing the protection threshold of CCF in the form of electrical quantity analytically under different fault types is a nontrivial pattern recognition problem. To overcome the difficulty of directly analysing the electromagnetic process of CCF, a statistical learning modelling approach is adapted to propose a CCF warning system, in which XGBoost algorithm was used as the backbone to regress on the relationship between AC/DC electrical quantities and the possibility of the occurrence of CCF. The design of the feature vector in XGBoost-based classifier considers not only the analytical expression of key electrical quantity during single commutation failure but also the important state variable of control system. Additionally, a special data windowing method was proposed to enrich of information encoded by the feature vector while ensuring the speed requirement posed by the protection system. Validation on multi-fault scenes simulation shows that the proposed early warning system can trigger warning signal correctly with sufficiently small measured data required, proving the practicability in engineering practice.