Abstract: Appearance of metasurfaces/metagratings makes anomalous control of wavefronts more and more convenient and flexible. However, most of the existing metasurfaces/metagratings are designed based on experience. As a result, their wavefront manipulation performance is often not optimal, and their working frequency bandwidth is narrow, which seriously limits their applications. Meanwhile, researchers found that as the incident angle is greater than a critical value, the generalized Snell’s law fails to estimate behavior of metasurfaces. In order to solve the above problems, based on the high-order diffraction theory, we propose a genetic optimization algorithm based method to design broadband wave-splitting metagratings. Based on the above technique, we specifically design three wave-splitting metagratings for flexural waves in thin plates, in which the supercells are composed of two subunits with a phase shift of \textπ . First, extensive numerical simulations are carried out to characterize the performance of our proposed metagratings and the optimized subunits. Then, a 3D printing technology is employed to fabricate metagratings and subunits to conduct experimental verification. Finally, our designed metagratings are compared with similar metagratings designed by two other methods. The results show that our metagratings work well as the designed functionality in the prescribed broad frequency range. However, the metagratings designed by two other methods only work well within a narrow frequency range. Although only flexural waves are considered in this work, our proposed technique is also applicable to other elastic waves. The results in this work provide a possible and effective way to design broadband metagratings for other waves.