Due to its periodic structure, the regular planar array has problems, such as wide beamwidth, high global side lobe in beam synthesis. So, an array optimization method based on the improved genetic algorithm was proposed. A planar grid microphone array was designed to satisfy the requirement of element spacing. An objective function with the main lobe width as the constraint and the global side lobe level as the fitness was constructed. The strategy of free intersection among individuals and random element number forced mutation was adopted to increase the searching range of the population on the basis of conventional genetic algorithms. A number of optimized arrays were obtained through simulation. Compared with several regular planar arrays, the random arrays optimized by the improved genetic algorithm have better performance under different signal-to-noise ratio inputs; compared with several conventional optimization algorithms, the improved genetic algorithm has stronger search ability, the number of random arrays is more, and the performance is better, which proves the feasibility of the proposed method.