Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SA41

n SA41 North Bldg 226C Joint Session DAS/Practice Curated: Risk and Decision Analytics Sponsored: Decision Analysis Sponsored Session Chair: Sayanti Mukhopadhyay, PhD, Purdue University, West Lafayette, IN, 47907, United States 1 - Optimizing Inspection Routes in Pipe Networks Thomas Ying-Jeh Chen, University of Michigan, Ann Arbor, MI, 48105, United States, Connor Riley, Pascal Van Hentenryck, Seth Guikema The inspection of aging water distribution pipes is a vital process for utilities to aid better decision making for risk-based management. To facilitate cost-efficient deployment of inspection robotics, a process that finds high risk pipe while accounting for the tool limitations is needed. We formulate the problem as an integer program, and explore a variety of methods to identify optimal routes: branch and bound, constraint generation, breadth-first search, depth-first search, and depth-first search with pruning. While only three factors are used to characterize tool limitations, the formulation can be extended to include technology-specific complexities in real world applications. 2 - Optimizing Design for Hybrid Renewable Energy Systems Under Long-range Uncertainties Ramin Giahi, Iowa State University, Ames, IA, 50010, United States Understanding the potential for new applications and different environments under which a system will operate is important in engineering design. This presentation focuses on design with long-range uncertainty. We identify and model significant uncertainties that will impact the use and lifespan of a system. This research explores designing a hybrid renewable energy system design while taking into account long-range uncertainties of 20 years. 3 - Modeling the Impact of Natural Hazards on the Serviceability of Infrastructure Systems: A Bayesian Approach Jin-Zhu Yu, Vanderbilt University, Nashville, TN, United States, Mackenzie G. Whitman, Hiba Baroud The ability to make accurate predictions of failure or recovery measures for infrastructure systems is often hindered by the lack of data and uncertainty of hazards. To address this challenge, this study presents a framework that incorporates a Bayesian updating mechanism of network component fragility into the evaluation of the overall serviceability of an infrastructure network under multiple hazard scenarios. This framework allows for better understanding of the subsequent performance of the individual components and the entire network as new data becomes available. A case study of a water distribution network is presented to illustrate the framework. 4 - Multi-stage Prediction for Zero-inflated Hurricane Induced Power Outages Sara Shashaani, University of Michigan, Ann Arbor, MI, 48105, United States, Seth Guikema, Chengwei Zhai, Jordan V. Pino, Steven M. Quiring Predictive models on hurricane power outages can be built via statistical learning methods that use past hurricanes data to capture the effects of several climatological and environmental variables on the power systems. Classical data minim methods and accuracy metrics are misleading for datasets with the majority of their response variables being zero. To deal with the zero-inflation in the power outage datasets, we develop and validate a 3-stage framework for three historic hurricanes and predict the outages of three recent hurricanes in the central Gulf region. The results show improvement over the traditional approaches. 5 - Impact of Climate Model Uncertainties in Projecting Long Term Regional Energy Demand Sayanti Mukherjee, Purdue University, West Lafayette, IN, United States, Panteha Alipour, Roshanak Nateghi The uncertainties in projecting long-term energy demands are not only associated with stochasticity in future socio-economic conditions, population changes or technology infusion, but also with climate projections as provided by the General Circulation Models (GCMs). Uncertainties attributed to climate projections mostly arise from the structure and processes within the GCMs, leading to projection divergence across the models. The purpose of this study is to provide an assessment of how the divergence in climate projections influence the long-term regional energy demand projections. The proposed framework will help the stakeholders in risk-informed long-term utility planning.

n SA42 North Bldg 227A Joint Session ISim/APS: Rare Events and Stochastic Optimization Sponsored: Simulation Sponsored Session Chair: Raghu Pasupathy, Purdue University, West Lafayette, IN, 47907, United States 1 - Simulation of Bipartite or Directed Graphs with Prescribed Degree Sequences Using Maximum Entropy Probabilities Enrique Lelo de Larrea, Columbia University, 403 Uris Hall, Columbia Business School, New York, NY, 10027, United States, Paul Glasserman We propose an algorithm for simulating bipartite or directed graphs with given degree sequences, motivated by the study of financial networks with partial information. Our algorithm sequentially computes certain “maximum entropy matrices, and uses the entries of these matrices to assign probabilities to edges between nodes. We prove the correctness of the algorithm, showing that it always returns a valid graph and that it generates all valid graphs with positive probability. We illustrate the algorithm in an example of an inter-bank network. 2 - Asymptotically Optimal Chance Constrained Optimization Approximations in Rare Event Regimes Fan Zhang, Stanford University, 450 Serra Mall, Stanford, CA, 94305, United States, Jose Blanchet, Bert Zwart Chance constrained optimizations are known to be NP hard, but it can be approximated by scenario approach when the tolerance-violation is small. The number of sampled constraints in conventional scenario approach is inversely proportional to the tolerance-violation, which is computationally expensive when the chance constraints become increasingly rare. In this talk, we applied importance sampling to construct a rare event simulation estimator, and the number of sampled constraints logarithmically grows with respect to the tolerance-violation 3 - Feasibility Check with Recycled Observations Yuwei Zhou, Georgia Institute of Technology, Atlanta, GA, United States, Sigrun Andradottir, Seong-Hee Kim, Chuljin Park We consider the problem of repeatedly finding a set of feasible or near-feasible systems among a finite number of simulated systems in the presence of stochastic constraints when the values of thresholds in the constraints change. Instead of restarting feasibility check from scratch for a new set of constraint threshold values, we recycle every observation collected from the previous feasibility check. We show that feasibility check with recycled observations achieves higher statistical guarantee and higher efficiency in terms of the number of observations than feasibility check with restart. 4 - A Decomposition Approach for Generating Correlated Random Vectors Oscar O. Guaje, Universidad de los Andes, Cra. 1 #18a-12, Bogotá, 111021, Colombia, Andres L. Medaglia, Jorge A. Sefair Current methods for generating correlated random numbers often rely on specific properties of the probability distributions of the input variables. We present a method based on mixed-integer programming that induces correlations to a set of random vectors without making any strong assumptions on the distribution of the random variables nor on the correlation structure. To improve the computational performance of our method we propose a column generation procedure. Our results compare favorably against the state-of-practice in terms of speed and solution quality. Finally, we illustrate its use in simulation where multiple input variables share a correlation structure within a flow process.

16