Metasurface absorbers (MA) typically exhibit a single type of absorption function due to their regular structures. In this study, we propose an irregular MA structure with octagonal meta-atoms. The presence of eight vertices in each meta-atom allows for tunable coordinates and offers a multitude of degrees of freedom in terms of geometry. As a result, the proposed MA exhibits diverse functionalities, including perfect absorption, multi-peaks absorption, and high absorption with a filtering window. To predict the geometric parameters of the MA structure based on a given target absorption spectrum, as well as the inverse design of the structure using the absorption spectrum as input, we employ a deep neural network combined with the particle swarm optimization algorithm. Remarkably, the mean-square error for spectrum prediction and inverse design of the MA structure is found to be as low as 0.0008 and 0.0031, respectively. This study opens up new possibilities for designing irregular electromagnetic structures and holds great potential for applications in multifunctional metasurfaces and metamaterials.