A hybrid optimization algorithm based on shuffled frog leaping algorithm (SFLA) and least squares support vector machine (LSSVM) algorithm is proposed and applied to the feature extraction of multi-peak Brillouin scattering spectra. The penalty factor C and kernel width σ of kernel function in LSSVM algorithm are optimized by SFLA, which reduces the Brillouin frequency shift error caused by the local optimization. Multi-peak Brillouin scattering spectra in the same signal-to-noise with different line width and the same line width with different signal-to-noise ratio are presented by simulation analysis and experimental verification. The fitting fitness of the experimental data is 0.0067, the fitting degree is 99.99%, and the Brillouin frequency shift error is 0.18 MHz. The results show that the SFLA-LSSVM algorithm can precisely fit the multi-peak Brillouin scattering spectrum. The proposed algorithm has the advantages of high fitting precision, small mean square error and fast running speed. The SFLA-LSSVM algorithm is an effective fitting method in the feature extraction of multi-peak Brillouin scattering spectrum.