Informs Annual Meeting 2017

WC17

INFORMS Houston – 2017

2 - A Multi-agent Based Simulation Model for Pricing Vehicle Routes in the Physical Internet Long Zheng, University of Louisville, 1407 Nightingale Road, Apt 4, Louisville, KY, 40213, United States, long.zheng@louisville.edu In vehicle routing under the protocol of Physical Internet (PI), it is a critical decision making problem for carriers to set the price of shipment loads to maximize their revenues. In this study, we propose a multi-agent system structure for the PI-vehicle routing problem (PI-VRP). A multi-agent based simulation model is developed to demonstrate the pricing approach. For inputs of the model, we use real data in a PI-enabled logistics networks. To an end, results of the simulation model are compared with a mathematical pricing approach based on dynamic programming optimization so as to valid our modelling framework. 3 - Optimal Fire Station Relocation to Minimize the Response Time: Case of the City of Kingsville Rutang Vijaykumar Shah, Texas A&M.University Kingsville, 1100 W. Corral Ave, Apt #14, Kingsville, TX, 78363, United States, rutang.shah@yahoo.com, Joon-Yeoul Oh, Hua Li It is very critical to response on fire and emergency calls quickly. In cases of fire and emergency call, the average response time in the city of Kingsville is 278 seconds, which is 32 seconds more than the established NFPA standard, 240 seconds. The aim of this research is to reduce the response time by analyzing the data and ARENA simulation models. This research shows that a new location of the fire station can meet the NFPA standard with covering wider area than existing location. The new location is selected using NLP technique. Simulation results shows that the new proposed location provides a less average response time in more areas.Key words: Emergency response time, Fire station relocation, NLP. 4 - Simulation of Out-of-control Process Observations for Evaluating Effectiveness of Phase I Studies Murat Caner Testik, Hacettepe University, Muhendislik Fakultesi, A Phase I study is an essential component of many statistical process control implementations. At this stage, a reference set of observations is searched for analysis. Accordingly, a Shewhart control chart with trial control limits is often used iteratively for identification of out-of-control observations. This dataset is then used for estimation of unknown parameters. Although this is a common practice recommended in textbooks, here we investigate its effectiveness by simulating different types of out-of-control processes. It is shown that one should be careful in using this approach since many out-of-control process observations may go unnoticed, resulting in poor parameter estimates. 340A Stochastic Networks and Queueing in Applied Probability II Sponsored: Applied Probability Sponsored Session Chair: Amarjit Budhiraja, University of North Carolina, Chapel Hill, NC, 27599, United States, budhiraj@email.unc.edu 1 - Asymptotically Optimal Control of N-systems with H_2^* Service Times under Many-server Heavy Traffic Arka P. Ghosh, Iowa State University, Department of Statistics, 3216 Snedecor Hall, Ames, IA, 50011, United States, apghosh@iastate.edu, Keguo Huang We are seeking an optimal control policy for a “N-system”, under the Halfin- Whitt regime. A N-system has two customer classes and two server pools, where servers from one pool can serve both customer classes, while servers from the other pool can only serve one customer class. Also, impatient customers will renege from his queue if the waiting time exceeds his patient time. The service time distribution is only pool-dependent, not class-dependent, and it follows a special case of hyper-exponential distributions, H*2 distribution. A static priority policy is shown to be asymptotically optimal in minimizing the total cost caused by queue length growth and customer abandonment. 2 - Serve the Shortest Queue and Walsh Brownian Motion Asaf Cohen, University of Michigan, Ann Arbor, MI, 48109, United States, shloshim@gmail.com Motivated by internet applications, we study a multi-class M/M/1 queueing system under heavy-traffic with heterogeneous customers where priority is given to the shortest queue. Upon arrival, customers are kept in N buffers with infinite capacity in accordance to their classes. At any time instant, the first customer in the nonzero shortest queue is being served. We show that the workload process converges to a Walsh Brownian motion living in the union of the N non-negative coordinate axes in RN. Roughly speaking, this process behaves like a Brownian motion when it is away from 0, and when it reaches 0, it chooses its ray according to a fixed probability measure. Endustri Muhendisligi Bolumu, Ankara, 06800, Turkey, mtestik@hacettepe.edu.tr, Christian H. Weiss, Yesim Koca, Ozlem Muge Testik WC17

4 - A Sequential Learning and Decision-making Methodology and its Application Amirhossein Meisami, University of Michigan, Ann Arbor, MI, United States, meisami@umich.edu, Henry Lam, Mark P. Van Oyen We study a sequential learning and decision-making methodology that integrates robust/stochastic optimization with regression-based prediction. We also study the uncertainty quantification of the resulting decision and optimality level. We discuss applications in digital marketing, patient discharge planning and readmission control. 332E Performance Measurement Contributed Session Chair: Krishna Kalyanam, InfoSciTex Corporation, Dayton, OH, United States, krishnak@ucla.edu 1 - A New Approach for the Empty Container Repositioning with Uncertainties Shaorui Zhou, Assistant Professor, Sun Yat-Sen University, 135 Xingang Xi Road, Guangzhou, China, zshaorui@126.com In this paper, we propose a new hybrid learning algorithm for two-stage stochastic programs for empty container repositioning with uncertainties. This algorithm is a hybrid of a piecewise linear approximation method and a stochastic subgradient method. The piecewise linear functions are updated using stochastic sub-gradient information and sample information on the objective function itself. To achieve global optimum, a projection step implementing the stochastic sub- gradient method is performed to jump from local optimums. This algorithm is proven convergent for general two-stage stochastic programs. Numerical experiments demonstrate that the algorithm has a fast rate of convergence. 2 - Clustering Algorithms in Real-time Risk-averse Portfolio Optimization Sitki Gulten, Stockton University, School of Business, 101 Vera King Farris Drive, Galloway, NJ, 08205, United States, sitki.gulten@stockton.edu This study examines the application of risk-averse optimization techniques to high-frequency trading (HFT) in real-time portfolio management. First, I develop efficient clustering methods for scenario tree construction with large scale of returns data. Then, I construct a two-stage stochastic programming problem with conditional measures of risk. Finally, I present an extensive simulation study on both interday and high-frequency intraday real-world data of the methodology. 3 - Sequential Weapon Target Assignment under Error Prone Battle Damage Assessment Krishna Kalyanam, Sr. Research Scientist, InfoSciTex Corporation, Dayton, OH, 45385, United States, krishnak@ucla.edu, David Casbeer A single shooter with homogenous weapons engages, sequentially, a collection of targets that may vary in value. Upon firing, the target is either killed or not; there is no partial or cumulative damage. The shooter adopts a shoot-look-shoot strategy, where an error-prone BDA is available at each look. The BDA is a classifier that returns the live or dead status of the target and it is prone to both Type I and II errors. This is a generalization to Manor & Kress (NRL, 1997) wherein only a kill is confirmed with a certain probability. We provide a lower bound on the maximal difference between the probabilities of correctly identifying a killed and live target, such that a greedy shooting strategy is optimal. 332F Simulation and Optimization Contributed Session Chair: Murat Testik, Hacettepe University, Ankara, Turkey, mtestik@hacettepe.edu.tr 1 - Efficient Simulation for Expectations Over the Region Outside a Convex Polytope Dohyun Ahn, Korea Advanced Institute of Science and Technology, #2111, E2-2, KAIST, 291 Daehak-ro, Yuseong-gu, Daejeon, 34141, Korea, Republic of, dohyun.ahn@kaist.ac.kr, Kyoung-Kuk Kim We consider the problem of estimating the expectation over the region outside a convex polytope. Such expectation arises in many applications including option pricing, stochastic activity networks, and financial networks. Assuming that random variables follow a multivariate elliptical distribution, we develop a conditional Monte Carlo method and prove its asymptotic efficiencies. We then demonstrate how this method can be applied to the above three examples. WC16 WC15

509