Based on the fiber network built by fiber Bragg grating (FBG), the PSO algorithm combined with the least squares support vector machine (LSSVM) was applied to the damage identification problem of 304 steel plate. Information feature of FBG center wavelength variation was used as input quantity and the damage location of steel plate structure was used as output quantity. LSSVM-based damage identification prediction model was constructed. The model was compared with the back-propagation (BP) neural network prediction model constructed under the same conditions. Damage location of steel plate structure was identified by kernel parameter σ and regularization parameter γ of LSSVM damage identification model, which was optimized by PSO algorithm. In the experimental area of 300 mm×300 mm×1 mm steel plate, 34 groups of samples were tested for damage location identification. The results show that the injury position of 33 groups is accurately identified in the 34 samples, and the accuracy rate is 97.06%. The PSO-based LSSVM damage prediction model has a self-diagnostic function.