<p>The cloud manufacturing (CMfg) environment exhibits complex characteristics involving multi-resource collaboration and concurrent disturbances, resulting in the quality of service (QoS) for cloud services exhibiting fluctuations during the service composition process. Existing service composition optimization models fail to simultaneously consider the uncertainty of QoS attributes of cloud services and the adaptability of service compositions to dynamic CMfg environments. Moreover, when solving high-dimensional interval multi-objective optimization models, existing algorithms still face challenges such as insufficient accuracy in interval number ranking methods and poor convergence performance. To resolve these limitations, an interval multi-objective service composition optimization model is established, with time, cost, quality, and flexibility as objectives. To enhance algorithmic performance in solving high-dimensional interval multi-objective models, an improved interval-based fast non-dominated sorting genetic algorithm (<i>i</i>-NSGA-II) is proposed. In <i>i</i>-NSGA-II, an improved interval-based α-dominance strategy is designed to improve the accuracy of interval objective value rankings and accelerate convergence. This strategy incorporates a possibility degree calculation model for interval number ranking under nonuniform distributions. Further, an adaptive parameter adjustment method is introduced to enhance the search capability of the algorithm. Finally, simulation experiments are conducted using problem instances of different scales and a CMfg case of customized automobile fuel tank production. The results show that <i>i</i>-NSGA-II significantly outperforms the chosen comparison algorithms regarding the quality, convergence, and distribution of the optimal solution set. The case study results demonstrate that the proposed model can generate robust service composition schemes. Both the model and algorithm exhibit strong industrial applicability.</p>