<p>Service composition in cloud manufacturing is a critical optimization problem that must balance multiple conflicting objectives, including service quality, cost, and service association impact. However, existing approaches often overlook the quantitative impact of service associations on composition performance, leading to suboptimal solutions. To address this issue, this study introduces a service association cost function and develops a three-objective optimization model that explicitly accounts for service quality, cost, and service association effects. To efficiently solve this model, we propose an enhanced NSGA-II algorithm with the following key improvements: (1) Good point set-based population initialization, integrating good point sets and random sampling to enhance solution diversity and search efficiency. (2) Reverse learning-based crossover operator, designed to improve exploration capability and prevent premature convergence. (3) Adaptive dynamic elitism strategy, which dynamically adjusts the elite retention ratio and adaptively incorporates local search operators to balance convergence and diversity. Extensive experiments on both benchmark problems and cloud service composition scenarios demonstrate that the proposed algorithm outperforms conventional multi-objective optimization methods in terms of convergence, diversity, and robustness. These findings confirm the effectiveness of our approach and its practical applicability in real-world cloud manufacturing environments.</p>