Informs Annual Meeting 2017

SD59

INFORMS Houston – 2017

SD58

4 - Understanding the Impact of Complexity on Facility Location Mozart Batista de Castro Menezes, Kedge Business School - Bordeaux, Office: 1449, Talence, 33405, France, mozart.menezes@me.com, Diego Ruiz-Hernandez Facility location problems are well known problems in the combinatorial optimization field of study, where the objective is to minimize the cost incurred to serve customer from a set of facilities. Our study brings to the field of facility location the concept of operations complexity, opening up a new research line within the field. When defining facility location one should aim not only to reduce operational costs but also to keep complexity in what we call complexity comfort range in which tactical and operational decisions are at their bests. The preliminary (empirical) results suggest that ignoring complexity matters may hurt that same bottom line that the locational problem is trying to improve. 362D Enhanced Estimation and Sampling in Multi-model Setting Sponsored: Simulation Sponsored Session Chair: Giulia Pedrielli, Arizona State University, Tempe, AZ, 85281, United States, giulia.pedrielli.85@gmail.com SD57 It has long been known, when dealing with a finite partition, that stratification can yield significant variance reductions. To apply this idea to debiased multi-level Monte Carlo (MLMC), we extend the framework to infinitely many strata. While new mathematical issues arise in the infinitely stratified setting, we are still able to develop a central limit theorem in this context that permits the construction of asymptotically valid confidence intervals. Furthermore, we suggest and theoretically justify sequential stopping procedures for such estimators that allow for easy recursive updating. This work is joint with Peter Glynn. 2 - Efficient Simulation Sampling Allocation using Multi-fidelity Models Yijie Peng, George Mason University, VA, United States, pengy10@fudan.edu.cn, Jie Xu, Loo Hay Lee, Jian-Qiang Hu, Chun-Hung Chen We present a new framework that integrates information from multi-fidelity models to increase computational efficiency for selecting the best high-fidelity systems. A Gaussian mixture model is proposed to capture important information contained in low-fidelity models that might not accurately approximate the performance of the high-fidelity simulation model. Posterior information obtained by Gaussian mixture model-based clustering analysis incorporates both cluster- wise information and idiosyncratic information for each design. A new budget allocation method is proposed to efficiently allocate high-fidelity simulation replications, utilizing posterior information. 3 - Parallel Simulation for Transportation Networks Feliza Vazquez-Abad, CUNY, New York, NY, United States, felisav@hunter.cuny.edu For simulation of a large public transportation network, we propose an accelerated method using nearly decomposable Markov Chains. This allows parallelization of the simulation, thus achieving great improvement in efficiency. The Kemeny constant is a measure of the strength of links for the stationary Markov chain. Its derivative an be used to assess the sensitivity of individual transitions (in the stationary operation). Using this information, we can decompose a Markov chain into smaller classes approximating the stationary regime. Each class is assigned a processor for simulation, but statistical corrections have to added in order for the parallel simulation to provide good estimates. 1 - Stratification in Debiased Multi-level Monte Carlo Zeyu Zheng, Stanford University, CA, United States, zyzheng@stanford.edu, Peter W. Glynn

362E Location Analysis Sponsored: Transportation Science & Logistics Sponsored Session

Chair: Noor Alam, Northeastern University, md.alam@northeastern.edu 1 - Modeling Interdependent Transportation and Power Systems under Dynamic Supercharging Demand of Electric Vehicles

Liqun Lu, University of Illinois at Urbana-Champaign, 205 N.Mathews Ave., Urbana, IL, 61801, United States, liqunlu2@illinois.edu, Yanfeng Ouyang

With an increasing number of electric vehicles (EVs) and supercharging stations, the daily power load profile is experiencing a noticeable reshape caused by travel demand in the transportation network. This study proposes a modeling framework to find a market equilibrium between EV users and power generators under dynamic travel demand. It is shown that traffic flow on transportation network and electricity loads/generations in power grids have mutual impacts via EV users’ routing decisions. 2 - An Algorithms to Measure Distance Between Irregular Regions Ruilin Ouyang, Northeastern University, Boston, MA, United States, ouyang.ru@husky.neu.edu, Dinghao Ma, md.noor.e.alam This talk will present an algorithm to measure distance between two irregular shapes regions used for location analysis. 3 - Datacenter Location Allocation Models MD Noor. E. Alam, Northeastern University, 334 Snell Engineering Center, 360 Huntington Avenue, Boston, MA, MA 02115, United States, md.alam@neu.edu, Tasnim Ibn Faiz A Datacenter facilitates high volume online transactions. Its location, capacity and configuration decisions are driven by customer demand volume, latency and uptime requirements respectively. An MILP model is presented that provides optimal decisions for meeting customer demand with required service levels. A second MILP model is presented to meet replication demand associated with each customer. A two-stage decision algorithm is proposed for making global optimal decisions by solving two models sequentially. Numerical results show that the models are scalable and the algorithm is efficient.

SD59

362F Joint session RAS/Practice: RAS Problem Solving Competition Sponsored: Railway Applications Sponsored Session

Chair: Yanfeng Ouyang, U. of Illinois at Urbana-Champaign, Urbana, IL, 61801, United States, yfouyang@illinois.edu 1 - RAS Problem Solving Competition Yanfeng Ouyang, U. of Illinois at Urbana-Champaign, 205 N.Mathews Ave, 1209 Newmark Lab, MC-250, Urbana, IL, 61801, United States, yfouyang@illinois.edu This session is reserved for the finalists of the RAS Problem Solving Competition: “Data Analytics for Railroad Empty-to-Load Peak Kips Prediction.” The presenters and their abstracts will be determined the Judging Committee by early October. More information about the Competition is available at http://connect.informs.org/railway-applications/awards/problem-solving- competition.

121