Current assembly process planning of complex products depends largely on the experience of technical personnel, which results in low design efficiency, low intelligence, and difficult to extract the assembly knowledge. To address these problems, this paper proposes an intelligent decision-making system for complex products assembly process planning through a comprehensively improved BP neural network with considering the assembly unit variability. Firstly, the characteristics, variations and similarities of assembly process for complex products are analyzed. A hierarchical model of product assembly process is established, which decomposes assembly process decision-making for overall structure into several units. A training set of each assembly unit is built considering the effect of assembly unit variability, based on which, an assembly process structure tree can be constructed quickly by connecting the nodes in the assembly process hierarchy model and the assembly process decision results. Secondly, to improve the efficiency and accuracy of decision-making, an ant lion optimizer (ALO)-BP neural network is proposed for complex products assembly process planning, which can optimize the selection of initial weights, thresholds, learning rate and feedback process of BP neural network automatically. Finally, with the case study of assembly process of the marine diesel engines, the proposed approach is compared with other algorithm models. The comparison results demonstrated that the proposed intelligent decision-making system is more precise than the other models.