Informs Annual Meeting 2017

MB66

INFORMS Houston – 2017

MB67

4 - An Iterated Beam Search & Column Generation Hybrid Solution to the Green Vehicle Routing Problem Mesut Yavuz, The University of Alabama, Alston Hall Box 870226, Tuscaloosa, AL, 35487, United States, myavuz@cba.ua.edu, Isil Koyuncu This talk addresses the routing problem of a homogeneous fleet of green vehicles, where each vehicle has a finite driving range before it must visit a refueling station, and all vehicles must return to the depot by end-of-day. We develop a column generation algorithm based on optimally routed customer sets. An iterated beam search algorithm is developed to generate the customer sets. The talk discusses static and dynamic strategies to integrate the two algorithms. Computational evaluation of the proposed algorithm is also presented.

371B Computer Experiments and Uncertainty Quantification Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Ying Hung, Rutgers State University of New Jersey, Piscataway, NJ, 08854-8019, United States, yhung@stat.rutgers.edu 1 - A Generalized Gaussian Process Model for Computer Experiments with Binary Time Series Chih-li Sung, Georgia Institute of Technology, Atlanta, GA, United States, csung33@gatech.edu, Jeff Wu Conventional analysis for computer experiments is based on Gaussian process (GP) models. Non-Gaussian observations such as binary responses are common in some computer experiments, but the extensions of GP models to these cases have received scant attention in the literature. Motivated by the analysis of a class of cell adhesion experiments, we introduce a generalized Gaussian process model for binary responses. The methodology is applied to study two different cell adhesion mechanisms, which were conducted by computer simulations. The fitted models reveal important biological differences between the two mechanisms in repeated bindings, which cannot be directly observed experimentally. 2 - Sequential Learning of Deformation Models in Additive Manufacturing through Calibration of Simulation Models. Tirthankar Dasgupta, Rutgers, Piscataway, NJ, United States, tirthankar.dasgupta@rutgers.edu Successful prevention of geometric shape deformation of products manufactured by additive manufacturing depends on building predictive deformation models. However, resource constraints impose restrictions on the number of test shapes of a particular type, making the use of meta models inevitable. To build meta models of deformation with good predictive power, calibration of existing models with data from physical experiments is necessary. We propose a sequential procedure for designing physical experiments and calibrating an ensemble of existing simulation models with data obtained from such experiments. 3 - Active Subspaces: Emerging Ideas for Dimension Reduction in Computational Science and Engineering Models Paul Constantine, Colorado School of Mines, Golden, CO, 80401, United States, paul.constantine@colorado.edu Engineers use simulations to study relationships between a model’s inputs and outputs. However, thorough parameter studies (e.g., UQ) are challenging when the simulation is expensive and the model has several inputs. The engineer may try to reduce the dimension of the model’s input parameter space. Active subspaces identify important directions in the input parameter space. I will (i) describe computational methods for discovering active subspaces, (ii) propose strategies for exploiting the reduced dimension, and (iii) review results from engineering applications. Visit activesubspaces.org 371C Functional Data, Image and Shape Analysis: Methodology and Applications Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Hao Yan, Atlanta, GA, 30319, United States, yanhao@gatech.edu Co-Chair: Kamran Paynabar, Georgia Institute of Technology, Atlanta, GA, 30332-0205, United States, kamip@umich.edu 1 - An Image Data Based Manifold Learning Approach for Online Diagnosis of Additive Manufacturing Processes Chenang Liu, Virginia Tech, Blacksburg, VA, 24060, United States, lchenang@vt.edu, Zhenyu Kong Online quality monitoring and control is one of the major challenges in additive manufacturing (AM). A potential effective approach for online quality control of AM is based on image analysis. This work proposes an online manifold learning based classification method for analyzing real-time image data collected by digital microscopes embedded in the AM machine. The resulting system can diagnose the surface quality status, i.e., to identify the type and severity of the printing defects, in a real-time manner. The case study shows that the proposed method is very effective and it can also be integrated into an online closed-loop control system. 2 - Sparse Subspace Learning of Dynamic Multivariate Functions Hao Yan, Georgia Institute of Technology, 3525 Highgrove Way MB68

MB66

371A School Bus Routing and Scheduling Sponsored: Transportation Science & Logistics Sponsored Session

Chair: Dimitris Bertsimas, Massachusetts Institute of Technology, 77 Massachusetts Avenue, E40-147, Cambridge, MA, 02139, United States, dbertsim@mit.edu 1 - Solving Multi-school Bus Routing and Scheduling Problem Zhongxiang Wang, Graduate Student, University of Maryland, 406 Ridge Rd, Apt 7, Greenbelt, MD, 20770, United States, zxwang25@umd.edu, Ali Shafahi, Ali Haghani School bus routing and scheduling is usually treated as two separated problems. Such separation will lead to a worse solution than solving them together with respect to the number of buses and travel time. The key point connecting routing and scheduling problem - trip compatibility - is thus deeply studied. A Mixed Integer Programming model is proposed along with a School Decomposition Algorithm. The model and algorithm are tested on eight sets of randomly- generated mid-size problems in comparison to the existing models. The results show that the proposed model and algorithm can find a better solution using up to 30% fewer buses than the best traditional models in a reasonable amount of time. 2 - Dynamic Electric Buses Scheduling under Stochastic Traffic Conditions Xindi Tang, Tsinghua, Beijing, China, txd16@mails.tsinghua.edu.cn, Fang He, Xi Lin Our research proposes a dynamic electric bus scheduling model under pre- determined timetable which effectively handles the subsistent stochasticity of traffic conditions and range limitation of electric vehicles. To expeditiously solve the problem which suffers from exponential expansion of variables, we embed it into a branch and price framework. We conduct numerical experiments based on real data in Beijing, simulate potential scenarios with delay propagation considered, and do sensitivity analyses, proving effectiveness of our models and performances of our algorithms. 4 - To School and Back Again an Optimization Approach to Large Scale School Bus Routing Dimitris Bertsimas, Massachusetts Institute of Technology, 77 Massachusetts Avenue, E40-147, Cambridge, MA, 02139, United States, dbertsim@mit.edu, Arthur J. Delarue, Sebastien Martin We present a novel optimization-based methodology for the School Bus Routing Problem (SBRP) that maximizes bus re-use between schools, in order to minimize overall costs to the city. Our method is tractable at the scale of a large city, simultaneously optimizing routes for 200 schools and 25,000 students, in a network with tens of thousands of edges. In the real-world setting of the Boston Public Schools (BPS) district, we show that our method reduces the overall number of buses needed each day by 18%, the total daily driving distance by 40%, and the total daily driving time by 37%.

188