Abstract: Commutation failure is one of the most common faults in high voltage direct current（HVDC） transmission systems. In order to mitigate the inverter commutation failure in DC transmission systems, the link of commutation failure prediction control（CFPREV） is widely used in practical projects at present. However, this link is susceptible to the influence of the initial fault angle.At some initial fault angles, the variation trend of zero-sequence voltage index is contrary to the variation trend of fault severity,which will seriously restrict the starting speed of CFPREV and hinder the suppression of commutation failure. Based on CFPREV,an improved CFPREV method based on waveform similarity detection is proposed in this paper. Pearson correlation coefficient is used to represent the changes of zero-sequence voltage waveform in real time, so as to quickly detect faults and avoid the shortcomings of the original CFPREV which starts slowly at some initial fault angles. The simulation results based on the CIGRE standard test model show that the proposed control strategy can eliminate the influence of the initial fault angle on the original CFPREV, improve the operation sensitivity of CFPREV link, and inhibit the initial commutation failure of HVDC transmission system to a certain extent.