This study addresses the robust scheduling problem in distributed flowshop involving coupled uncertainties in processing times and sequence-dependent setup times. A reinforcement learning-driven iterated local search algorithm (QILS) is proposed. Firstly, an expectation-risk robust scheduling model is established with the objective of minimizing makespan to effectively balance the stability and optimality of the solution. Secondly, the NEHUPT heuristic is designed for uncertain environments, where scheduling priorities are generated based on scenario analysis, and a fine-tuning strategy is applied to enhance the initial solution quality. Additionally, a collaborative optimization framework integrating $Q $-learning and ILS is developed, where reinforcement learning and dynamic decay methods guide the adaptive selection of perturbation strategies to balance exploration and exploitation capabilities. Finally, a robustness contribution-based local search method is introduced to further enhance solution quality. Comprehensive simulation experiments and comparative analysis with multiple state-of-the-art algorithms demonstrate the superior performance of the proposed method in solving distributed robust flow shop scheduling problems.