复杂过程系统不可避免地受到诸如产品需求、原料供给、运行条件等不确定因素的影响。如何有效地处理不确定因素，权衡过程系统设计和运行的优化性能和鲁棒性能不仅是决策优化的核心，也是工业界和学术界面临的重大挑战。本文以提高不确定条件下过程系统设计和运行决策的样本外性能为目标，利用基于沃瑟斯坦距离的模糊集合描述不确定参数，系统性地开展了分布式鲁棒优化方法及应用研究，展示了分布式鲁棒优化方法在不确定参数的小样本历史数据下的优势。针对经典分布式鲁棒优化方法要求决策在模糊集合内均可行而导致的保守性问题，面向短周期生产调度问题，提出了基于沃瑟斯坦模糊集合的分布式鲁棒机会约束规划方法，推导了其基于最差情况条件风险值的安全近似形式。基于赋范向量空间中的强对偶理论，证明了该近似形式与独立/联合分布式鲁棒机会约束规划方法的等价性。通过间歇过程生产调度案例评估了所提出近似形式的有效性，验证了所提出方法样本外性能的优越性。面向中周期石化产品海运库存-路径问题，提出了两阶段分布式鲁棒优化方法，以降低经典分布式鲁棒优化方法的保守性。在1-范数沃瑟斯坦距离度量下，根据强对偶原理和对偶范数原理，推导了两阶段分布式鲁棒优化方法的鲁棒对等式，建立了两阶段分布式鲁棒连续时间弧流模型，并提出了专用Benders分解新算法求解该NP-hard优化问题。与两阶段随机规划方法相比，所提方法给出的调度方案具有更强的抵御等待和航行时间变化的能力。面向长周期过程系统鲁棒优化设计问题，在两阶段分布式鲁棒优化方法基础上，进一步拓展并提出了多阶段分布式鲁棒优化方法。基于不确定参数的阶段独立性，推导了在1-范数沃瑟斯坦模糊集合下多阶段分布式鲁棒优化方法的鲁棒对等式，解决了在产品需求和原料可用量不确定情况下的可再生资源转化过程设计问题，实现了最优拓扑结构鲁棒优化设计，充分展示了在不确定参数的小样本历史数据下多阶段分布式鲁棒优化方法所给出的决策方案的样本外优越性能。最后，以沃瑟斯坦模糊集合描述港口等待时间和航行时间不确定性，基于所提出的两阶段分布式鲁棒优化方法和专用Benders分解算法，成功求解了中国东海岸柴油海运库存-路径优化的工业实例（调度周期一个月，大型混合整数线性规划模型的变量数大于4,490,000，其中二元变量数大于110,000），显著验证了所提Benders分解算法的有效性和分布式鲁棒优化路径决策的样本外性能优势。

The Complex process systems suffer from uncertainties inevitably, for example, product demand, raw materials supply, operation conditions, etc. How to deal with the uncertainties effectively to balance the optimality and robustness of process systems design and operation is not only the core of the decision-making, but also a major challenge faced by both of industry and academia. In order to improve the out-of-sample performance of the decisions concerning process systems design and operation under uncertainty, distributionally robust optimization with ambiguity set constructed with Wasserstein distance is investigated systematically. And the benefits of distributionally robust optimization with limited historical data of uncertain parameters are demonstrated by a series of applications.Aiming at the conservatism due to the restriction that decisions are feasible for any uncertain parameters realization within the ambiguity set in classical distributionally robust optimization, distributionally robust chance constrained programming method and its safe approximation based on worst-case conditional value at risk are proposed for short-term production scheduling problem. According to the strong duality theorem in the normed vector space, the equivalence between the proposed approximation and the individual/joint distributionally robust chance constrained programming method is proved. Short-term batch process production scheduling problems are utilized to evaluate the effectiveness of the proposed approximation and the superiority in terms of the out-of-sample performance.For the medium-term petrochemicals maritime inventory-routing problem, two-stage distributionally robust optimization method is investigated to reduce the conservatism of classical distributionally robust optimization. According to the strong duality and dual norm theorem, the corresponding robust counterpart of two-stage distributionally robust optimization method is derived under the 1-norm Wasserstein ambiguity set. Furthermore, the two-stage distributionally robust continuous time arc-flow model is formulated. A tailored Benders decomposition algorithm is also developed to solve the resulting NP-hard mixed integer linear programming model. Compared with the two-stage stochastic programming method, the scheduling decisions of the proposed optimization method are more capable of hedging against the variations of waiting time and sailing time.For the long-term process system robust design problem, the present work further extends to multi-stage distributionally robust optimization method on the basis of two-stage distributionally robust optimization method. According to the stage-wise independence of the uncertain parameters, the corresponding robust counterpart is derived under 1-norm Wasserstein ambiguity set, and applied to consider the uncertain product demand and raw materials availability in conversion process design problem with renewable sources. The optimal process network topology with excellent out-of-sample performance can be successfully obtained by the proposed multi-stage distributionally robust optimization method even with limited historical data of uncertainties.Finally, a real-world industrial case about the refined diesel maritime inventory-routing optimization problem along the east coast of China is addressed with the proposed two-stage distributionally robust optimization method and the corresponding tailored Benders decomposition algorithm, in which Wasserstein ambiguity set is utilized to characterize uncertain port waiting time and sailing time. The effectiveness of the proposed tailored Benders decomposition algorithm is further demonstrated by solving the resulting large-scale mixed integer linear programming model with one-month time horizon, which involves more than 4,490,000 variables among which more than 110,000 are binary. The distributionally robust optimization model yields routing decisions with improved out-of-sample performance when compared to the two-stage stochastic programming method.