Commutation Failure Identification Of UHVDC Transmission System Based On Empirical Mode Decomposition

UHVDC transmission technology has solved the problem of disproportionate distribution of energy and load in China with its unique advantages of high voltage level, far transmission distance and large transmission capacity. Commutation failure is one of the most frequent faults in UHVDC transmission system which can cause the mutations of electrical quantities such as AC voltage and DC current. Therefore, it is a prerequisite to ensure the safe and stable operation of UHVDC transmission system. In this paper, using the linear analog component of the inverter side DC current signal as the analysis object, the main research contents are as follows:The EMD method combined with neural network for recognition of commutation failure is studied. By calculating the energy and the correlation coefficient of IMF component of the DC current signal linear mode component by the EMD decomposition, then composing two feature vectors, using Elman neural network classifier to fault recognition and comparison.The simulation results show that using Elman neural network combined with energy method can accurately determine the fault type.Proposed EMD improved method of Ensemble Empirical Mode Decomposition (EEMD), and for the direct current signal linear mode component decomposition, calculate approximate entropy and auto regression (AR) model coefficients were grouped feature vectors and as the Elman neural network inputs identify faults. After simulation, approximate entropy and neural network combination method can distinguish commutation failure, line failure, normal state.Research on the decomposition of empirical mode decomposition - Extreme-Point Symmetric Mode Decomposition ESMD), for the current signal linear mode component ESMD decomposition, extracted from the energy component of the IMF, the correlation coefficient,coefficient of AR model, and approximate entropy composed of four kinds of features vector, Particle swarm optimization method of least squares support vector machine system to normal state, line fault, commutation failure fault identification, further use of support vector machine cause commutation failure four side AC inverter system failure identification. The simulation results show that the diagnosis in the case of small sample, ESMD, approximate entropy method and least squares support vector machines combination of methods to identify high failure rate and good stability.