Abstract: Aiming at the characteristic of single sheet metal, the two-dimensional cutting stock problem of multi-specification sheet metal is studied, which is an NP-complete problem, due to the diversity of plate specifications and the large quantity of blank specifications. According to the characteristics of this problem, the cutting process is designed into two stages: regular stage and non-regular stage. After completing the main cutting task for each type of rectangular roughcast in the regular stage, if there is still remaining roughcast, the non-regular stage will be entered and the BL algorithm will be used to cut the remaining roughcasts. According to the characteristics of the model, a variable neighborhood artificial bee colony algorithm (VNABC) is proposed, and two decoding strategies, STD and SLD, are designed. The operator of the VNABC algorithm is improved. Finally, the response surface analysis method is used to calibrate the parameters of VNABC. In the simulation experiments, the VNABC algorithm is compared with GA, NUS, SA, ABC algorithms. The experimental results demonstrate the superiority of VNABC in solving the two-dimensional cutting stock problem of multi-specification plates.