As the core equipment of high-voltage direct current (HVDC) transmission system, the health status of the converter valve directly affects the stability of power grid operation. In this paper, we propose a fusion of modal analysis and multi-source data-driven method for extracting the time-varying laws of electrical features and health monitoring of the converter valve. The multidimensional vibration feature map is constructed, and the multi-scale features of the vibration signal are extracted by combining the time-domain trajectory image, Markov variation field and wavelet packet transform. The short-circuit fault characteristics and protection principle of the converter valve are analyzed, and a valve state discrimination model based on current timing characteristics is established. The effectiveness of the method is verified through simulation and engineering experiments, and the proposed method can realize the state discrimination of converter valve fault characteristics. Considering all 405 power modules of the whole bridge arm, the error of capacitance capacity estimation based on field data is within 1%, and the accuracy of fault detection for short circuit on AC side of converter valve, short circuit on converter valve and short circuit on DC side of converter valve reaches 99.88%, 99.86% and 99.91%, respectively.
