Informs Annual Meeting 2017

WB28

INFORMS Houston – 2017

WB26

6 - Efficient Computing Budget Allocation for Optimal Subset Selection with Correlated Sampling Si Zhang, Shanghai University, Shang-Da Road No. 599, Shanghai, China, zhangsi817@sina.com, Loo Hay Lee, Ek Peng Chew, Chun-Hung Chen This study mainly considers the problem selecting the top-m alternatives under the environment with correlation. Correlation demonstrates the relationship between alternatives and can provide more information when comparing two alternatives. Hence, it is sometimes intentionally introduced into simulation to reduce the variance of the difference between two alternatives. We model the optimal subset selection problem with correlated sampling based on the optimal computing budget allocation framework and derive the respective optimal allocation rules. 350D Supply Chain, Decision Analysis Contributed Session Chair: Shuxiao Sun, Peking University, Beijing, China, sunshuxiao@pku.edu.cn 1 - Game Theoretic Modeling of a Biofuel Supply Chain with Existence of Incentives to Promote Cellulosic Based Biofuel Production Amirsaman Hamzeh Bajgiran, PhD Candidate, University of Wisconsin-Milwaukee, 1320 E Capitol Drive, Apt 205, Milwaukee, WI, 53211, United States, hamzehb2@uwm.edu, Jaejin Jang, Xiang Fang We consider a biofuel supply chain (BSC) problem in which a farmer supplies two downstream refineries with non-identical crops (corn and energy crop) to study the feasibility of reaching the goal of RFS. The problem has been modeled as a multi-leaders one follower game to capture the farmer’s decisions on land use as well as refineries’ proposed prices to the farmer. We have considered subsidizing the farmer and the refinery which uses energy crop to study whether it is capable of enhancing advanced biofuel production. This work can provide insights for governments and help them to make effective policy-making decisions to achieve their goals. 2 - Optimal Advertising and Pricing in a Dynamic Omnichannel Supply Chain Yang Bai, Ajou University, Suwon, Korea, Republic of, b198716y@hotmail.com, Byeong-Yun Chang Channel coordination and advertising promotion strategy play an important role in modern lives. In this research, we examine an effective and efficient coo- petition model and consider advertising decisions for a retailer directing omni-channel. We derived the N-A model in dynamic analysis of coo-petition in an omni-channel. Using the optimal omni-channel model, we obtain the optimal dynamic pricing and advertising decisions. 3 - Leadership and Information Flow in a Supply Chain with Coopetition Maosen Zhou, Shenzhen University, No.3688, Nanhai Road, Shenzhen, China, moussin@gmail.com, Qingyu Zhang We study the issue of coopetition in a two-echelon supply chain where competing manufacturers purchase through a common GPO to achieve economies of scale. Our game-theoretical modeling generalizes information structures about uncertain demand with horizontal and vertical information flows, and quantifies horizontal coopetition under different vertical leadership. By analyzing the effects of coopetition, leadership, information flow, and their interplay, managerial insights into exploiting the relationship and information structures for coopetitive supply chains will be provided. 4 - Entry Deterrence and Price Competition under Asymmetric Information Jooyol Maeng, Assistant Professor, Pacific Lutheran University, School of Business, 12180 Park Avenue S, Tacoma, WA, 98447-0003, United States, maengjy@plu.edu, Sungyong Choi An incumbent has an incentive to deter entry of a potential entrant by lowering pre-entry price. We study limit pricing in a price-based duopoly market under asymmetric demand information. We present a separating perfect Bayesian equilibrium, which indicates the incumbent with private information can successfully deter entry. WB28

350B Simulation and Optimization Contributed Session Chair: Si Zhang, Shanghai University, Shanghai, China, zhangsi817@sina.com 1 - Resilience Based Network Design under Uncertainty Xiaoge Zhang, Vanderbilt University, 6501 Harding Pike, Apt. P-25, A nonlinear function is introduced to characterize the component restoration behavior in the presence of a disruptive event. The introduction of such a function facilitates the modeling of absorptive ability, restoration capability and recovery speed related with each system component. A system resilience constraint is formulated to impose that the system resilience at a given time meet a desired target. We use the maximum flow through the network as an indicator of system performance. Our objective is to design an optimal network that incurs the least cost, but meets a system resilience constraint. Two numerical examples are used to illustrate the effectiveness of the proposed method. 2 - A New Reinforcement Learning Algorithm for Average Reward with Numerical Results on Total Productive Maintenance Angelo Encapera, Missouir University of Science and Technology, 15512 Tamarac Ct, Wichita, KS, 67230, United States, amet3b@mst.edu, Abhijit Gosavi We present a new Reinforcement Learning (RL) algorithm for average reward Markov decision processes (MDPs) and semi-Markov decision processes (SMDPs). The existing RL algorithm for SMDPs, namely R-SMART, requires exploration decay. Our new algorithm, named iSMART, is fully exploratory and hence eliminates the need for turning the exploration decay. iSMART is designed using principles of deep learning for RL. We present numerical results on small cases, as well as with a larger case study on Total Preventive Maintenance. The new algorithm is more robust than its precursor R-SMART and generates near-optimal solutions on all test cases studied. 3 - Actor-critics Revisited: A New Bounded Algorithm and Numerical Results with Revenue Management Ryan J. Lawhead, Missouri University of Science and Technology, 113 West 16th St, Rolla, MO, 65401, United States, rjlm97@mst.edu, Abhijit Gosavi Actor-Critics are Reinforcement Learning algorithms that can solve large-scale Markov Decision Processes (MDPs) and semi-MDPs. The classical actor-critic has a critical drawback in that the actor’s values get unbounded. The artificial bounding required as a result limits the exploration of the state-action space, which can cause sub-optimal behavior even on small-scale problems. We propose a new version that keeps the actor’s values bounded, allowing the algorithm to search the state-action space more thoroughly and generate optimal solutions. We also present results on a large-scale problems from airline revenue management where our algorithm outperforms a standard industrial heuristic. 4 - Heuristic Algorithm for Initial Sample Size in Ranking and Selection Problem Ruijing Wu, Phd Student, Shanghai Jiao Tong University, 2334 Willowbrook Drive, Apt 258, Shanghai, 200030, China, wu_ruijing@yeah.net Ranking and selection problem is to choose the best system from a set of systems. When variance are unknown, it is necessary to take some initial samples to estimate it. In most procedures, initial sample size (ISS) is predetermined and fixed. In this paper, we present a heuristic algorithm to calculate ISS, which is a generalization of the idea in Hartmann (1991). To check the validity and effectiveness of our algorithm, we implement it on KN procedure and compare it with KN procedure and Hartmann’s method. The numerical experiments show that our procedure can (1) guarantee the probability of correct selection; (2) reduce the total sample size by 20%-80%. 5 - Path Planning of Multiple Unmanned Surface Vehicle under Communication Constraint Md Saiful Islam, Texas Tech University, 2500 Broadway, Lubbock, TX, 79409, United States, md-saiful.islam@ttu.edu, Timothy I. Matis Unmanned Surface Vehicles (USV) in current application has limited payload and endurance. Multiple USV’s are often used to overcome these limitations but, path planning of multiple USV’s mostly uses either leader - follower approach or formation fleet which leads to a myopic optimization problem. Only the leader vehicle have the ability to make an independent decision. To avoid this problem, USV’s in this study will be considered as mobile agents where each USV can make a decision independently and will maintain communication with other members of the team. Nashville, TN, 37205, United States, zxgcqupt@gmail.com, Sankaran Mahadevan, Shankar Sankararaman, Kai Goebel

483