Optimization of the natural frequencies of exponential functionally graded plates is a critical issue in the engineering field, playing a vital role in enhancing the dynamic performance of plate structures. In response to this challenge, this paper presents an innovative optimization approach that synergistically integrates the smoothed finite element method (SFEM) with surrogate models, aiming to address the problem with both high efficiency and precision. Based on the first order shear deformation theory, a SFEM is established for free vibration analyses of functionally graded plates. During the computation of the system stiffness matrix, gradient smoothing operations are applied to bending strains within smoothed domains, effectively improving computational accuracy. To overcome the shear locking phenomenon, different interpolation forms are adopted to treat bending strains and shear strains separately. For natural frequency optimization, a series of sample points are selected, and their corresponding natural frequencies are calculated using the SFEM. Subsequently, a surrogate model is established to map the relationship between the gradient index and the natural frequencies. The golden section method is employed to determine the optimal gradient index that achieves preset natural frequency targets. Numerical examples demonstrate that the surrogate model based on piecewise cubic Hermite interpolation exhibits high computational accuracy. Moreover, the surrogate model-based optimization significantly reduces the number of SFEM frequency calculations required, substantially enhancing optimization efficiency. This approach provides an efficient and practical method for optimizing natural frequencies of exponential functionally graded plates.