Independent Component Analysis and Manifold Learning with Applications to Fault Diagnosis of VSC-HVDC Systems

A fault diagnosis scheme for voltage source converter-high voltage direct current transmission(VSC-HVDC) system is proposed based on the independent component analysis(ICA) and the locally linear embedding(LLE) in the present work.The measured signals in the VSC-HVDC can not be used directly to detect system fault due to the heavy inference noise.The FastICA is hence employed to eliminate the disturbed noise and recover the fault sources from the measured DC line voltage and current observation signals.Then the LLE algorithm is applied to extract distinct characteristics hiding in the recovered fault signals.To enhance the fault pattern recognition,the support vector machine(SVM) is adopted to learn the relationship between the fault features and the system operation conditions.The ability of the proposed ICA-LLE method to detect VSC-HVDC system fault is evaluated with the simulated data.The analysis results demonstrate the feasibility and effectiveness.The distinguished features of the fault signals,such as AC line-to-line fault and AC line-to-ground fault,and the compound faults,can be extracted efficiently and then isolated in the 3-D feature space correctly.The classification rate of the SVM with the proposed scheme is increased by 20% compared with LLE.