Article
Management
Sergey S. Ketkov
Summary: This study addresses uncertainty in the objective function parameters of a class of linear mixed-integer programming problems and also considers uncertainty in the underlying data set. By formulating the problem as a three-level distributionally robust optimization problem, it can be reformulated as a single-level MILP problem if the loss functions are restricted to biaffine functions. Furthermore, for certain forms of data uncertainty, the problem can be solved reasonably fast by leveraging the nominal MILP problem.
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH
(2024)
Article
Operations Research & Management Science
Yining Gu, Yicheng Huang, Yanjun Wang
Summary: In this paper, we consider a data-driven distributionally robust two-stage stochastic linear optimization problem. We explore tractable reformulations under different uncertainties and demonstrate that they can be converted into computationally tractable convex optimization problems. Numerical results are presented to illustrate the effectiveness of the proposed models.
JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS
(2023)
Article
Economics
Guowei Zhang, Ning Jia, Ning Zhu, Long He, Yossiri Adulyasak
Summary: This paper studies a highly integrated humanitarian relief network design problem and proposes a distributionally robust optimization model. Through experimentation, it is demonstrated that jointly optimizing network strengthening and inventory pre-positioning decisions can significantly reduce shortage penalties.
TRANSPORTATION RESEARCH PART B-METHODOLOGICAL
(2023)
Article
Engineering, Multidisciplinary
Zhongming Wu, Kexin Sun
Summary: A new distributionally robust mean-variance model is proposed in this study to solve the multi-period portfolio selection problem, utilizing the Wasserstein metric to capture the uncertainty of returns. The model is transformed into a tractable convex problem using duality theory, and the radius of the Wasserstein ball is estimated using a nonparametric bootstrap method. Analyzing the in-sample data indicates that the portfolio's return and risk are relatively insensitive to parameter values. A series of out-of-sample experiments demonstrate that the proposed model outperforms other models in terms of final wealth, standard deviation, and Sharpe ratio.
APPLIED MATHEMATICAL MODELLING
(2023)
Article
Computer Science, Interdisciplinary Applications
Dimitri J. Papageorgiou
Summary: The paper investigates a data-driven distributionally robust optimization approach for decision making in industrial settings. It compares the performance of different models and demonstrates the potential benefits of the DRO approach for elasticity-aware price-taking decision makers.
COMPUTERS & CHEMICAL ENGINEERING
(2023)
Article
Management
Yunqiang Yin, Zunhao Luo, Dujuan Wang, T. C. E. Cheng
Summary: Recent research on distributionally robust (DR) machine scheduling has explored different approaches to deal with uncertain processing times. One approach is to use statistical metrics to measure the distance between probability distributions. In this study, we focus on Wasserstein distance-based DR parallel-machine scheduling, where we minimize the worst-case expected total completion time-related cost over all distributions within a Wasserstein ambiguity set.
OMEGA-INTERNATIONAL JOURNAL OF MANAGEMENT SCIENCE
(2023)
Article
Mathematics
Luyun Wang, Bo Zhou
Summary: In this paper, a modified conjugate gradient method is proposed to solve the distributionally robust Logistic regression model over the Wasserstein ball, achieving improved numerical efficiency. The method consists of two phases: a conjugate gradient descent step and an instantaneous optimization problem with regularization term minimization and proximity to the interim point obtained in the first phase. The modified conjugate gradient method is proven to attain the optimal solution with a convergence rate of 1/T under nonsummable steplength. Numerical experiments affirm the effectiveness of the proposed method over existing solvers and first-order algorithmic frameworks.
Article
Engineering, Chemical
Wangli He, Jinmin Zhao, Liang Zhao, Zhi Li, Minglei Yang, Tianbo Liu
Summary: This work proposes a novel data-driven Wasserstein distributionally robust optimization framework to tackle the uncertainty in refinery planning operations. A data-driven ambiguity set is constructed based on the Wasserstein metric to represent the distributional uncertainty. Robust kernel density estimation technique is adopted to establish the support set for reducing the impact of potential outliers. A data-driven two-stage distributionally robust optimization model and its computationally tractable robust counterpart are developed for refinery planning. A real-world case study on a petroleum refinery demonstrates the effectiveness and applicability of the proposed framework.
CHEMICAL ENGINEERING SCIENCE
(2023)
Article
Multidisciplinary Sciences
Daniel Bartl, Samuel Drapeau, Jan Obloj, Johannes Wiesel
Summary: The study examines the sensitivity of stochastic optimization problems to model uncertainty, using a non-parametric approach with Wasserstein balls. It provides explicit formulas for first-order corrections to both the value function and the optimizer, with applications in various fields and specific examples like LASSO regression coefficients and option pricing. This research also explores sensitivity analysis in finance and proposes measures for neural network robustness against adversarial examples.
PROCEEDINGS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES
(2021)
Article
Management
Rui Gao, Xi Chen, Anton J. Kleywegtc
Summary: Wasserstein distributionally robust optimization (DRO) is an approach to optimization under uncertainty that hedges against a set of probability distributions. It facilitates robust machine learning and has a wide range of applications. This paper presents a general theory for the variation regularization effect of Wasserstein DRO, covering various types of losses and spaces, and emphasizes the importance of the bias-variation tradeoff. Additionally, specific applications and new generalization guarantees are provided.
OPERATIONS RESEARCH
(2022)
Article
Computer Science, Information Systems
Ningning Du, Yankui Liu, Ying Liu
Summary: This article introduces a new method for portfolio optimization in the face of distribution uncertainty, utilizing a robust optimization approach. By developing a fuzzy mean-CVaR portfolio optimization model, theoretical conclusions are drawn and practical examples from the Chinese and US stock markets are considered. Numerical experiments demonstrate the effectiveness of this data-driven distributionally robust portfolio optimization method.
Article
Mathematics
Ruoxuan Li, Wenhua Lv, Tiantian Mao
Summary: In this paper, the authors investigate a distributionally robust optimization (DRO) problem with affine decision rules. They propose a new family of Wasserstein metrics called shortfall-Wasserstein metrics to construct an ambiguity set and summarize the transportation cost random variables using normalized utility-based shortfall risk measures. The authors demonstrate that the multi-dimensional shortfall-Wasserstein ball can be projected onto a one-dimensional one through affine projection. Furthermore, they provide a dual formulation for computational tractability and verify the strong duality that allows for a direct and concise reformulation of the problem.
Article
Engineering, Electrical & Electronic
Hossein Saberi, Cuo Zhang, Zhao Yang Dong
Summary: This paper proposes a new hierarchical coordination method for Home Energy Management (HEM) which optimizes the operation efficiency and robustness by coordinating central and local controllers and using data-driven methods. Numerical simulations validate the high solution robustness and cost-saving advantages of the proposed method.
IEEE TRANSACTIONS ON SMART GRID
(2021)
Article
Thermodynamics
Zhuoya Siqin, DongXiao Niu, Xuejie Wang, Hao Zhen, MingYu Li, Jingbo Wang
Summary: This paper proposes a P2G-CCHP microgrid system integration framework and solves the economic dispatch problem through a two-stage distributionally robust optimization model. The model considers environmental cost and improves the system stability and economy by introducing P2G devices.
Article
Computer Science, Artificial Intelligence
Zhongming Wu, Ye Song, Ying Ji, Shaojian Qu, Zaiwu Gong
Summary: This study investigates a solution for uncertain multiple criteria sorting problems by using stochastic programming and distributionally robust optimization. A preference learning process based on a distributionally robust support vector machine model is proposed. Numerical experiments demonstrate the efficiency and robustness of the proposed method.
APPLIED SOFT COMPUTING
(2023)
Article
Operations Research & Management Science
Ran Ji, Miguel A. Lejeune
Summary: This study focuses on distributionally robust chance-constrained programming optimization problems with data-driven Wasserstein ambiguity sets, proposing algorithmic solutions for various uncertainties. The proposed linear programming and mixed-integer second-order cone programming formulations are evaluated for scalability and tightness on a distributionally robust chance-constrained knapsack problem, showcasing their effectiveness in different uncertainty scenarios.
JOURNAL OF GLOBAL OPTIMIZATION
(2021)
Article
Engineering, Manufacturing
Dmitry Anokhin, Payman Dehghanian, Miguel A. Lejeune, Jinshun Su
Summary: This study investigates the possibility of providing resilience delivery through mobile services during natural disasters, introducing mobile power sources as future restoration technology. Automated decision-making solutions are presented to coordinate the utilization of mobile power sources and repair crew schedules.
PRODUCTION AND OPERATIONS MANAGEMENT
(2021)
Article
Computer Science, Interdisciplinary Applications
Omid Hashemi-Amiri, Fahimeh Ghorbani, Ran Ji
Summary: Due to the global outbreak of COVID-19, perishable product supply chains have been impacted, increasing the risks of food insecurity in affected countries. Supply and demand uncertainty significantly affect supply chain networks, highlighting the importance of food provision and distribution. This study proposes a bi-objective optimization model for a three-echelon perishable food supply chain network, aiming to mitigate supply and demand uncertainties and optimize network costs and suppliers' reliability. The model uses distributionally robust modeling and chance-constrained approaches, and is reformulated as a mixed-integer linear program. A real-world case study in the poultry industry is conducted to validate the model's performance and applicability.
COMPUTERS & INDUSTRIAL ENGINEERING
(2023)
Article
Genetics & Heredity
Sainan Zhang, Mengyue Li, Yilong Tan, Juxuan Zhang, Yixin Liu, Wenbin Jiang, Xin Li, Haitao Qi, Lefan Tang, Ran Ji, Wenyuan Zhao, Yunyan Gu, Lishuang Qi
Summary: This study identified genetic mutations associated with the prognosis of early-stage lung adenocarcinoma patients and developed a mutational prognostic signature that can improve individualized treatment planning and predict the response to immunotherapy.
JOURNAL OF MOLECULAR MEDICINE-JMM
(2022)
Article
Operations Research & Management Science
Zhenlong Jiang, Ran Ji, Zhijie Sasha Dong
Summary: We propose a two-stage distributionally robust joint chance-constrained (DRJCC) model for designing a resilient humanitarian relief network in the post-disaster environment. The model considers uncertainties in demand and unit allocation cost of relief items and determines the locations of supply facilities and transportation plans. We investigate the problem under two types of ambiguity sets and propose reformulations and algorithms to solve them. Through numerical experiments and a case study in the Gulf Coast area, we demonstrate the superiority of the proposed model compared to other models in terms of cost and network reliability.
Article
Green & Sustainable Science & Technology
Zhenlong Jiang, Yudi Chen, Ting-Yeh Yang, Wenying Ji, Zhijie (Sasha) Dong, Ran Ji
Summary: Effective household and individual disaster preparedness is crucial for minimizing physical harm and property damage. However, it is important for relief agencies to understand the level of public preparedness to assess risk and vulnerability. To enhance the comprehensiveness of disaster preparedness assessments, we develop a framework that integrates machine learning and simulation. By utilizing machine learning algorithms and Monte Carlo simulation, we accurately predict the level of disaster preparedness and improve understanding for relief agencies.
Article
Green & Sustainable Science & Technology
Omid Hashemi-Amiri, Ran Ji, Kuo Tian
Summary: In recent decades, the increase in waste generation rate and environmental impacts have made waste management crucial in urban areas. This study proposes a novel integrated model to improve the municipal solid waste system by considering facility location, shift scheduling, and vehicle routing decisions. The model aims to optimize sustainable development goals such as total profit, air pollution emissions, citizen satisfaction, and social risk factors. The findings demonstrate that the proposed framework allows decision makers to address three critical sustainability aspects and maintain the resilience of the waste system.
Article
Computer Science, Interdisciplinary Applications
Nilay Noyan, Gabor Rudolf, Miguel Lejeune
Summary: This study introduces a new class of optimization problems that simultaneously address distributional and decision-dependent uncertainty, providing a unified modeling framework and discussing ways to specify key model components. Computational challenges in solving complex problems are discussed, with a focus on identifying settings and problem classes where these challenges can be mitigated. Model reformulation results, including robustified risk measures, are provided, along with insights from a computational study on a novel risk-averse machine scheduling problem with controllable processing times.
INFORMS JOURNAL ON COMPUTING
(2022)
Article
Computer Science, Interdisciplinary Applications
Janiele E. S. C. Custodio, Miguel A. Lejeune
Summary: The study presents a spatiotemporal dataset of all out-of-hospital cardiac arrest dispatches for Virginia Beach, along with a toolkit for generating random instances based on user input. This detailed data can be utilized in healthcare emergency situations and facility location models, aiding researchers in data-driven models and optimization algorithms.
INFORMS JOURNAL ON COMPUTING
(2022)
Article
Computer Science, Interdisciplinary Applications
Ran Ji, Miguel A. Lejeune
Summary: Fractional distributionally robust optimization problems with uncertain probabilities aim to maximize ambiguous fractional functions and have a semi-infinite programming formulation. A new closed-form is derived to compute a bound on the Wasserstein ambiguity ball size, along with a data-driven reformulation and solution framework involving modular bisection algorithms. The computational study demonstrates the framework's applicability and scalability in quickly solving large, industry-relevant problems.
INFORMS JOURNAL ON COMPUTING
(2021)
Article
Computer Science, Interdisciplinary Applications
Roya Karimi, Jianqiang Cheng, Miguel A. Lejeune
Summary: We proposed a novel partial sample average approximation (PSAA) framework to solve chance-constrained linear matrix inequality (CCLMI) problems, developed computationally tractable approximations, analyzed their properties, derived conditions ensuring convexity, and designed a sequential convex approximation method. Our comprehensive numerical study demonstrated the superiority of the PSAA reformulation and algorithmic framework over standard benchmarks.
INFORMS JOURNAL ON COMPUTING
(2021)
