Aiming at the lack of accuracy of traveling wave head calibration and the performance of intelligent location model affected by parameters, a short-circuit fault location method based on the beluga whale algorithm (BWO) is proposed to optimize variable mode decomposition (VMD) and kernel extreme learning machine (KELM) for MMC-HVDC transmission lines. Firstly, the BWO is used to optimize the parameters of VMD， combined with wavelet soft threshold denoising method for noise reduction and decomposition of the collected fault signals. Then the arrival moment of the initial traveling wave is calibrated by combining the Hilbert transform (HT). Next, the arrival moments of traveling waves are used as eigenvalues to construct the feature dataset. The KELM localization model is optimized using BWO. Finally, the dataset is substituted into the optimized localization model to achieve fault localization. The results show that the localization model of the method fits 99.4% with high localization accuracy and good robustness. The proposed method is highly tolerant to noise and transition resistance, and the localization error is within 500m.