In special aluminum alloy production, smelting for aluminum ingots is the first process that affects production efficiency and product quality in subsequent processes directly. There exists two problems that charging plans cannot be made efficiently and furnace-grouping results are not optimal in the smelting process due to product variety and difference of batch size. To solve them, a furnace-grouping optimization model is established. The furnace-grouping problem is formulated with two objectives of minimizing the number of charging plans and the percentage of scrap metal with some constraints such as capacity of melting furnace and ingot-grouping rules in this model. According to the feature of this model, real number coding rule is employed that takes the percentage of order allocation as decision variable. A specialized multi-objective approach combining multi-swarm cooperative artificial bee colony is proposed to solve this optimization model. Decomposition strategy and multi-swarm strategy with information learning is employed to improve optimizing ability of the algorithm. The simulation experiment is designed on the basis of the truthful data of special aluminum alloy production. The numerical results demonstrate that this optimization model meets the requirements of manufacturing enterprises and the proposed algorithm is a powerful search and optimization technique for the furnace-grouping problem of special aluminum ingots.