2016 INFORMS Annual Meeting Program

WE44

INFORMS Nashville – 2016

WE45 209A-MCC Simulation and Optimization Contributed Session Chair: B Vermeulen, Eindhoven University of Technology, Eindhoven, Netherlands, b.vermeulen@tue.nl 1 - Bi-level Stochastic Approximation For Decomposable Stochastic Optimization Formulations Soumyadip Ghosh, IBM TJ Watson Research Center, 1101 Kitchawan Road, Route 134, Yorktown Heights, NY, 10598, United States, ghoshs@us.ibm.com, Ebisa Wollega, Mark S Squillante We propose a bi-level algorithm to solve stochastic optimization formulations with a certain decomposable structure. Consider, for example, the problem of maximizing total revenue by jointly maximizing unit sales, subject to non-linear market conditions, and minimizing costs, which for fixed sales is a two-stage linear program (2SLP). An outer loop runs stochastic approximation (SA) accounting for the non-linear part. An inner loop solves the 2SLP. The gradient of the objective function of the 2SLP is used in the outer SA, and is obtained using parametric programming. We analyze the convergence of this bi-level SA, and provide experimental evidence of its efficacy on an energy-domain problem. 2 - A Decentralized Solution To The Car Pooling Problem Pawel J. Kalczynski, Professor of IS and Decision Sciences, California State University-Fullerton, 800 N State College Blvd., Fullerton, CA, 92834-6848, United States, PKalczynski@fullerton.edu, Malgorzata Miklas-Kalczynska Existing carpool optimization techniques based on the centralized approach serve policy-makers’ goals, but neglect the realities of participants. Moreover, absent strict enforcement, participants often ignore centralized solutions and maximize their own utility. We present a new model (formulated and tested on real-world and simulated problem instances) that mimics a decentralized carpool self- organization process. Our findings reveal savings similar to centralized models, and a potential strategy for improving carpool utilization. 3 - Avoiding Singularities In Parallel Robotic System Design Cameron Turner, Associate Professor, Clemson University, 206 Fluor Daniel EIB, Clemson, SC, 29634-0921, United States, cturne9@clemson.edu, Sean Fry Singular configurations in robotic systems present significant design and control problems. In parallel robotic systems, the complex nature of singularities makes their consideration during the design process even more significant when high precision motions are desired. Using surrogate-based optimization techniques, this paper achieves a solution for the design of a parallel robotic system for engineered material characterization. 4 - MO-COMPASS For Constrained Simulation Optimization Haobin Li, Institute of High Performance Computing, A*STAR, 1 Fusionopolis Way, #16-16 Connexis North, Singapore, 138632, Singapore, lihb@ihpc.a-star.edu.sg, Xiaofeng Yin, Wanbing Zhang, Loo Hay Lee MO-COMPASS is recently developed for multi-objective simulation optimization, which has limitation that only linear constraints on the decisions space can be explicitly handled. Whereas, for non-linear or stochastic constraints as simulation outputs, the default approach that penalizes the constraint violation on the objective values can be inefficient with the “most-promising-area” (MPA) structure. In this study, we propose a novel approach for constraint handling in the multi-objective environment, by considering Pareto optimality in different feasibility layers with generalized MPAs. So we extend MO-COMPASS to solve constrained optimization problems in an efficient manner. 5 - Approximate, Adaptive Maintenance Scheduling Of Maritime Assets Under Different Operating Modes B Vermeulen, Eindhoven University of Technology, Eindhoven, Netherlands, b.vermeulen@tue.nl, Sena Eruguz, Tarkan Tan, Geert-Jan Van Houtum In the maritime sector, vessels are used under different operating modes with different degradation rates, different maintenance setup and downtime costs, and particular maintenance/ replacement options. Given uncertainty on actual degradation rates for components, asset owners follow the OEMs’ hyper- conservative and hence costly maintenance and replacement policies. We provide an approximate dynamic programming model (and proprietary software tool) with online adjustment of degradation rates and adaptive planning of maintenance activities given the schedule of operating modes.

2 - Sequential Exploration With Geological Dependencies And Uncertainty In Oil Prices Babak Jafarizadeh, Statoil, Sandsliveien 90, SE-SV ND2, Bergen, 5020, Norway, bajaf@statoil.com Economic valuation and analysis of drilling decisions for a cluster of exploration opportunities can be analytically challenging. If oil is found in one location, the probability of finding oil in the nearby prospects may increase. Furthermore, the time required to interpret data and update the geological understanding exposes these investments to hydrocarbon price dynamics. In this work, we develop a framework for valuation of clusters of exploration opportunities where prospects are geologically dependent and uncertainty in oil prices is described as a mean- reverting stochastic process. 3 - Sequential Sample Allocation For Multiple Attribute Selection Decisions Dennis D Leber, NIST, 100 Bureau Drive, MS 8980, Gaithersburg, MD, 20899-8980, United States, dennis.leber@nist.gov, Jeffrey W Herrmann When faced with a limited budget to collect data in support of a multiple attribute selection decision, the decision-maker must decide how many samples to observe from each alternative and attribute. This allocation decision is of particular importance when the observed attribute values contain uncertainty, such as with physical measurements. We present a sequential allocation scheme that relies upon Bayesian updating in an attempt to maximize the probability of correct selection when the attribute values contain Gaussian measurement error. WE44 208B-MCC Perceptions, Behavior, and Decisions Sponsored: Decision Analysis Sponsored Session Chair: Franklyn Koch, Koch Decision Consulting, Eugene, OR, United States, kochfg@gmail.com Co-Chair: Gregory L Hamm, Stratelytics, LLC, Alameda, CA, United States, ghamm@strts.com 1 - Perceived Catastrophic Risks In Sequential Social Networks Shu Huang, Master Candidate, Tsinghua University, Tsinghua Slovic (1987) proposes to use degrees of dread, unknown and personal exposure to describe individual risk perception. These factors depend on whether an individual has undergone disasters or near misses in the past, and are also affected by experiences and perspectives of his or her neighbors in the social network. We model each individual’s risk perception of uncertain consequences and likelihood of personal exposure as a Bayesian learning process to achieve equilibrium of estimation within each neighborhood. We can then construct a utility map over the social network to depict the dynamics of risk perception in response to multiple disasters. 2 - Bipolar Cardinal Ranking For Group Disagreement Evaluations University, Haidian District, Beijing, 100084, China, huang-s15@mails.tsinghua.edu.cn, Chen Wang Public decision making involve decisions that affect many stakeholders who have conflicting opinions. This may cause implementation issues for an authority, and to avoid this and to identify which stakeholder groups that are credited or discredited, it is of interest for the authority to understand how their preferences differ. One way of approaching this is to provide effective means for stakeholders to state preferences and affect towards proposed alternatives. For this we developed a web questionnaire using bipolar cardinal ranking where stakeholders state negative, neutral, or positive affect towards alternatives. The approach is demonstrated in a real-life case in Upplands Väsby, Sweden. 3 - Effects Of Total Cost Of Ownership on Automobile Purchasing Decisions Muhammed Sutcu, Assistant Professor, Abdullah Gul University, Sumer Campus, Erkilet Bulvari, Kayseri, 38060, Turkey, muhammed.sutcu@agu.edu.tr We reveal a complete picture of ownership-related expenses and construct a decision model which helps decision maker to make the optimal choice when purchasing an automobile. The decision model helps the costumers to understand what a car will cost beyond its purchase price when customers consider out-of- pocket expenses like fuel, repair, and insurance. For that purpose, representative joint probability distributions of a decision maker are approximated using maximum cumulative residual entropy (CRE) and CRE based first order dependence tree approaches to elicit decision maker’s preferences to calculate the total cost of ownership of an automobile. Aron Larsson, Associate Professor, Stockholm University, Postbox 7003, Kista, SE-16407, Sweden, aron@dsv.su.se, Tobias Fasth

498