Wind farm layout optimization (WFLO) plays a critical role in maximizing wind energy output by strategically minimizing wake losses. Assessing various wind farm layouts using Computational Fluid Dynamics (CFD) models is both computationally intensive and time-consuming. To mitigate this challenge, surrogate models are utilized as substitutes for high-fidelity CFD models. Moreover, the MLPA-GASA algorithm is introduced, which integrates a Multilayer Perceptron (MLP) surrogate model and leverages Simulated Annealing (SA) to enhance local search capabilities. Numerical experiments were conducted across three distinct wind scenarios to validate the efficacy of the proposed algorithm. The findings illustrate that the algorithm markedly decreases computational time while achieving superior energy production performance.