Online monitoring and fault identification of submarine cable are fundamental technology for ensuring the normal operation of cross-sea transmission and communication transmission. To avoid signal distortion due to direct denoising, which affects the extraction of target features, in this paper, the variational mode decomposition (VMD) algorithm is applied to extract features directly from noisy vibration signals. Using the Brillouin optical time domain analysis experimental system for monitoring the submarine cable vibration, the vibration signals of submarine cable under the conditions of anchoring, scouring, and friction are obtained. Three types of vibration signals are divided into 200 groups, and the intrinsic mode function components are obtained using the VMD algorithm. Furthermore, the energy, energy entropy, and kurtosis combinations of each component are obtained as eigenvectors. Using 80% and 20% of the feature vectors as the training and test sets, respectively, the data are classified by inputting them into the support vector machine (SVM) based on the bird swarm algorithm (BSA). The experimental results show that compared with other SVMs, the classification accuracy of BSA-SVM is higher, reaching 99.17%, and the running time is shorter.