Maximizing the production of light oil in atmospheric distillation column is one of the important ways to improve the benefit of enterprises. In order to adapt to the change of market demand and price, and to produce high-demand and high-value light oil, a multi-objective optimization algorithm for light oil production of atmospheric distillation column based on self-adaptive differential evolution is proposed. The algorithm adopts the decomposition method of the penalty boundary intersection method, and introduces the perferred learning operator in the stage of population mutation to decrease the randomness and blindness brought about by the random selection or the simple selection of the best individual by the traditional operator. The self-adaptive strategy is used to gradually change the crossover mutation operator. The improved algorithm is applied to three kinds of test functions and the actual refinery atmospheric distillation column light oil production optimization. The results show that the proposed algorithm has obvious advantages in testing functions and can effectively improve the atmospheric distillation column light oil production, which verifies the effectiveness of the proposed algorithm.