Two-stage risk-neutral stochastic optimization model has been widely studied recently. The goal of our research is to construct a data-driven two-stage distributionally robust optimization model with risk aversion and to extend it to multi-stage case. We use a coherent risk measure, Conditional Value-at-Risk, to describe risk under probability uncertainty. Due to the computational complexity of the nonlinear objective function of the proposed model, two improved decomposition methods based on cutting planes algorithm are proposed to solve the two-stage and multi-stage distributional robust optimization problems, respectively. To verify the validity of the two models, we give two applications on multi-product assembly problem and portfolio selection problem. Compared with the risk-neutral stochastic optimization models, the proposed models are more robust.

Shaojian Qu received the Ph.D. degree from Xi’an Jiaotong University, China, and was a Research Fellow at National University of Singapore for 4.5 years. He is now a Professor of the Business School at University of Shanghai for Science and Technology. His current research interests include: Decision theory and method, game theory, supply chain management, and computer vision.