<p>During cloud remanufacturing, the overestimated quality of recovered products can lead to a mismatch between platform-assigned services and task requirements, resulting in inefficiencies or task failures. Therefore, this study proposes a stable service composition and optimal selection approach for cloud remanufacturing, which accounts for quality uncertainty to improve execution robustness. A bi-objective model is established by matching a preferred service of equivalent quality and an alternative of superior quality. An improved particle swarm optimization algorithm incorporating reinforcement learning is designed to solve this model. This algorithm introduces an enhanced action selection mechanism and an abandonment strategy to enhance search efficiency and facilitate escape from local optima. Four experiments were conducted to validate the proposed model and algorithm. The results show that, compared to traditional service composition and optimal selection method, the proposed solution is on average 2.16% and 0.57% higher in finish time and remanufacturing cost but improves in robustness by 1.71% and 0.92%, respectively. The effectiveness of the proposed algorithm is validated through comparison with two solvers on small-scale examples. On large-scale examples, the proposed algorithm was compared with nine optimization approaches, including metaheuristic, learnheuristic, reinforcement learning methods, and algorithms incorporating only one of the improved strategies, thereby highlighting the effectiveness of both enhancements. The proposed algorithm demonstrated superior performance under three problems across nine scales. These findings suggest the proposed approach is well-suited for addressing the challenges of stable service composition and optimal selection in cloud remanufacturing.</p>