Informs Annual Meeting 2017

WB78

INFORMS Houston – 2017

3 - Sparse Nonlinear Regression Pushing the Envelope using Modern Optimization Bart Paul Gerard Van Parys, Affiliate Researcher, MIT, Room E40-154, 77 Massachusetts Avenue, Cambridge, MA, 02139, United States, vanparys@mit.edu, Dimitris Bertsimas We present a novel method for nonlinear sparse regression which is able to scale to real-life problems consisting of up to 10,000 observations for a few 1,000 inputs. The ability of our method to identify all relevant nonlinear features is shown empirically to experience a phase transition. In stark contrast to existing heuristics, the same transition also presents itself in our ability to reject all obfuscating features as well. In the regime where our method is statistically powerful, its computational complexity is interestingly on par with Lasso based heuristics. The presented work fills a void in terms of a lack of disciplined nonlinear sparse regression methods in high-dimensional settings. 4 - Global Optimization of Mixed Integer Nonlinear Programs via Adaptive Tightening of Convex Relaxations Harsha Nagarajan, Staff Scientist, Los Alamos National Laboratory, 3000 Trinity Drive, Apt 8, Los Alamos, NM, 87544, United States, harsha@lanl.gov We propose a two-stage approach to strengthen piecewise convex relaxations for MINLPs with multi-linear terms. In the first stage, we apply bound contraction methods based on SDP relaxations to iteratively until a fixed point with respect to the bounds are achieved. In the second stage, we partition the variables using an adaptive multivariate partitioning scheme and valid partition elimination strategies. Thus, we construct sparser partitions yet tighter relaxations by adaptive partitioning in combination with cutting plane methods on locally convex regions. We demonstrate the superiority of the algorithm on MINLPLib and Richard Forrester, Professor of Mathematics, Dickinson College, Department of Mathematics, College and Louther Street, Carlisle, PA, 17013, United States, forrestr@dickinson.edu, Peixin Sun Most researchers of the 0-1 quadratic knapsack problem (QKP) assess the performance of their solution methods by performing computational experiments on randomly generated instances. In this talk we examine different factors that can influence the difficulty of randomly generated instances and develop different methods for constructing test problems. In particular, we introduce a technique for generating instances of the QKP with a specified correlation between the objective and knapsack coefficients. We provide a detailed computational study to show how both heuristic and exact solution methods perform on our new problem sets. 381A Plug-in Electric Vehicle Charging Planning and Operations Sponsored: Energy, Natural Res & the Environment Electricity Sponsored Session Chair: Wei Qi, McGill University, McGill University, Montreal, QC, H3A 1G5, Canada, qiwei.0216@gmail.com 1 - Joint Planning of Electric Vehicle Fast-charging Network and Distributed Photovoltaic Generation Wei Qi, McGill University, 1001 Sherbrooke Street West, Montreal, QC, H3A 1G5, Canada, qiwei.0216@gmail.com, Hongcai Zhang, Scott Moura, Zechun Hu, Yonghua Song Integration of plug-in electric vehicles (PEVs) with distributed renewable resources will decrease PEVs’ well-to-wheels greenhouse gas emissions, promote renewable power adoption and defer power system investments. We propose a multidisciplinary approach to jointly planning PEV fast-charging stations and distributed photovoltaic (PV) power plants on coupled transportation and power networks. We develop models of 1) PEV fast-charging stations; 2) highway transportation networks under driving range constraints; 3) PV power plants with reactive power control. Then we formulate a stochastic mixed integer second order cone program and decomposition algorithms to solve this problem. 2 - Electric Vehicle Charging Behavior and Infrastructure Availability are Critical to Accurately Assess the Promise of EVS for the Smart Grid Colin Sheppard, Lawrence Berkeley National Laboratory, Berkeley, CA, 94720, United States, colin.sheppard@lbl.gov, Anand Gopal, Rashid Waraich, Julia Szinai, Nikit Abhyankar, Sangjae Bae, Alexei Pozdnukhov, Andrew Campbell Vehicle electrification is a critical component of most pathways to deeply decarbonize the transportation sector. Through highly detailed simulation modeling with BEAM (the modeling framework for Behavior, Energy Autonomy, and Mobility), we examine the relationship between driver behavior and the use (large) Pooling instances w.r.t commercial global opt. solvers. 5 - Construction of Test Problems for the 0-1 Quadratic Knapsack Problem WB78

of charging infrastructure in the context of current day and projected plug-in electric vehicles (PEV) adoption. We demonstrate the importance driver behavior when analyzing the spatiotemporal distribution of electricity demand from PEVs, as well as the potential for PEVs to provide flexibility services to the electric grid. 3 - Recharging Electric Vehicle Sharing Fleet Xin Wang, Assistant Professor, University of Wisconsin-Madison, 8916 Red Beryl Drive, Middleton, WI, 53562-4278, United States, xin.wang@wisc.edu, Long He, Guangrui Ma, Wei Qi Electric vehicles have been considered as key to cutting carbon emissions in urban transportation. Several cities have welcomed electric vehicle (EV) sharing services and supported them by building charging infrastructure. Despite the efforts of the cities and operators, it remains challenging to recharge the fleet due to limited charging facilities. Such operational difficulty discourages the use of EV in the sharing fleet and forces operators to switch to gas-powered cars or even cease their operations. In this paper, we address the charging infrastructure planning problem in joint with fleet repositioning and recharging operations in the context of EV sharing. 381B Forestry I - Forest Management Sponsored: Energy, Natural Res & the Environment Forestry Sponsored Session Chair: Peter Rauch, BOKU-Univ of Natural Resources & Life Sciences, BOKU-Univ of Natural Resources & Life Sciences, Wien, 1180, Australia, Peter.rauch@boku.ac.at 1 - Optimal Deployment of Firefighting Aircrafts for Wildfire Initial Attack Planning Joao Zeferino, University of Coimbra, Coimbra, Portugal, zeferino@dec.uc.pt The success of the initial attack is crucial to wildfire fighting efficiency. An optimization model for aerial firefighting fleet deployment planning is presented. The model aims to find a solution for the aircrafts location that maximizes the coverage of different areas according to their wildfire susceptibility and hazard. It takes into account the different forests within a region, the available airport facilities and the aircrafts characteristics. The model potentialities are illustrated through a national-scale case study. 2 - Ascending and Descending Satellite Observation of Forest Fires Aaron Bradley Hoskins, United States Naval Research Laboratory, 7220 Briarcliff Drive, Springfield, VA, 22153, United States, abh318@msstate.edu, Hugh Medal We investigate the monitoring of a forest fire by a constellation of satellites. Each satellite is maneuvered to observe the forest fire twice per day. The location of the forest fire is unknown at the time of launch. The problem is formulated as a stochastic programming problem where the initial orbital configuration is the first-stage decision and the satellite maneuver sequence is the second-stage decision. The problem is solved using a Sample Average Approximation Algorithm. The proposed solution is compared through simulation to the current operational paradigm. 3 - Simulating Multi-modal Wood Supply Chains including Risks Agents Peter Rauch, BOKU-Univ of Natural Resources & Life Sciences, Feistmanteistr 4, Wien, 1180, Austria, Peter.rauch@boku.ac.at, Christoph Kogler Increasing natural disturbances lead to supply chain risks and seasonal irregularities in wood harvest and transport. A discrete event simulation model supporting manager decisions and contributing to a better understanding of the multimodal wood supply chain was developed. Simulating collaborative supply chain control strategies enhances the development of advanced risk management improving supply chain resilience. The supply chain covers wood harvest, precarriage and storage in terminals, transhipment to rail waggons or vessels and final transport woodworking plants. 4 - An Exact Approach for Spatial Prioritization of Conservation Actions Jordi Garcia-Gonzalo, Researcher, Centre Tecnològic Forestal de Catalunya (CTFC)., Solsona, Spain, j.garcia@ctfc.es Limited conservation budgets require prioritizing which management actions to implement and where to maximize the long-term persistence of biodiversity. Recently a heuristic approach has been used to find the combination of actions that remediate threats to species at the minimum cost while considering connectivity among actions through the planning area. We present a mixed integer programming model to prioritize actions to address threats to freshwater fish species in the Mitchell River catchment, northern Australia. We compared our results with the heuristic approaches and found our approach delivered more efficient solutions for different connectivity requirements. WB79

502