Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SC08

2 - Improving Exploration in Population-based Metaheuristics using Fading Consensus: Application to PSO Xin Su, Arizona State University, Tempe, AZ, United States, Theodore P. Pavlic Simple averaging protocols on fault-prone networks can exhibit long periods of quasi-stability punctuated by large group-level jumps. Although communication theorists view these recently observed jumps as pernicious, we propose using them to augment population-based metaheuristics to improve search ergodicity in multiextremal optimization. We demonstrate this approach with faded consensus particle swarm optimization (FC-PSO), which is a multi-agent optimization algorithm that improves the performance of PSO while potentially decreasing the number of required function evaluations as well. 3 - Distributed Non-linear Optimization under Non-separable Constraints: A Low-communication Approach with a Power Systems Example Theodore P. Pavlic, Assistant Professor, Arizona State University, ASU - CIDSE, P.O. Box 878809, Tempe, AZ, 85287-8809, United States Economic dispatch, matching generator operation levels to customer demands over a network, is the optimal diet problem of power systems. Non-linear cost functions and non-separable demand constraints usually require centralized solutions or decentralized approaches with high amounts of communication (e.g., bidding with Lagrange multipliers). In this talk, I introduce a distributed primal- space approach that converges to a bounded set of the optimizer with no direct communication between generators and self adjusts to changes in demands without broadcasting new parameters to every generator. 4 - Efficient LP Algorithms for which the Dense Constraint Matrix has a Sparse Factorization Robert J. Vanderbei, Princeton University, Operations Research & Financial Engineering, 209 Sherrerd Hall, Princeton, NJ, 08544, United States In linear programming, it is sometimes the case that the constraint matrix A is fairly dense but has a known sparse factorization. A few different methods for exploiting this sparsity in the context of interior-point methods will be presented and compared. The same ideas can be extended naturally to interior-point methods for nonlinear programming. 5 - Motion Planning for Autonomous Vehicles using MINLP Hande Benson, Drexel University, Department of Decision Sciences, 3141 Chestnut Street, Philadelphia, PA, 19104, United States We will present a mixed-integer nonlinear programming model, its centralized and decentralized solution, for motion planning in fleets of autonomous vehicles under communication constraints. Uncertainty and Network Resilience Sponsored: Optimization/Network Optimization Sponsored Session Chair: Zhijie Dong, Texas State University, San Marcos, TX, 78640, United States 1 - Interdependent Network Functionality and Recovery for Community Resilience Charles D. Nicholson, University of Oklahoma, 202 West Boyd, Room 124, Norman, OK, 73019, United States, Weili Zhang, Naiyu Wang, Peihui Lin, Xianwu Xue A framework is presented to estimate building functionality loss and utility recovery in a community following a hazard. Analysis includes spatial distribution of physical damages to both buildings and utility infrastructure; utility disruptions deriving from the cascading failures; and the recovery of the utility system. The framework couples stochastic functionality analyses of physical systems and hazard response characteristics to provide a rich array of information for hazard mitigation and resilience planning. An earthquake scenario in Shelby Co.,TN illustrates the framework. 2 - Resilience Quantification in Global Maritime Networks Elise Miller-Hooks, George Mason University, 208 Rosalie Cove Ct, Silver Spring, MD, 20905, United States, Ali Asadabadi Ports are critical components of the global supply chain, supplying key connections between land- and maritime-based transport modes. They operate in cooperative, but competitive environments wherein individual port throughput is linked through an underlying transshipment network. Building on concepts of stochastic equilibrium problems with equilibrium constraints, this presentation models and analyzes protective investment strategies aimed at enhancing resilience to disruption from a host of potential future damage scenarios while protecting each port’s market share. n SC07 North Bldg 123

3 - Restoration Crew Routing Problem under Incomplete Information Kash Barker, University of Oklahoma, 202 W. Boyd St., Room 124, Norman, OK, United States, Kash Barker We consider the problem of restorative capacity enhancement problem in an infrastructure network which is interconnected with a routing network through which restoration crews are dispatched. The output will be a set of synchronized routes formed by planning and scheduling restorative efforts for infrastructure networks. Along with the uncertainty and urgency during post-disruption situations, considering the routing network disrupted may result in the limited information on the requirements for network restoration. To deal with this case, we propose a stochastic restoration crew routing problem to improve the compatibility of the model with the real-time conditions. 4 - Uncertainty-aware Routing of Aerial Sensors for Infrastructure Damage Inspection Andrew Lee, Massachusetts Institute of Technology, Cambridge, MA, United States, Mathieu Dahan, Saurabh Amin, Cynthia Barnhart We present an approach to actively inspect urban networks facing risk of disruptions due to natural events using small Unmanned Aerial Systems (sUAS). Information from fixed sensors and environmental features are used to predict the number and type of failures in different spatial regions. These predictions are used to achieve prioritized plans for routing repair vehicles and sUAS. This entails incorporating uncertainties in the distribution of failure events and travel times into a network inspection problem and a vehicle routing problem, and sequentially solving them. We illustrate our approach using data from Houston’s drainage network inspection in the aftermath of Hurricane Harvey. n SC08 North Bldg 124A Joint Session OPT/Practice Curated: Network Optimization Models in Routing and Communications Sponsored: Optimization/Network Optimization Sponsored Session Chair: Tachun Lin, Bradley University, Peoria, IL, 61625, United States 1 - Consistent Aircraft Fleeting and Routing among Schedule Periods Zhili Zhou, United Airlines, 233 S. Wacker Drive, 5th Floor, Chicago, IL, 60606, United States Commercial airlines invest in international markets, which gain revenues compatible with its domestic counterpart. For international services, airlines change flights and markets in different schedule periods. To reduce the operational burdens, we address the fleet assignment and aircraft routing consistency problem between two schedule periods with the objective to minimize the changes of fleets on served markets and routes. We explore the column-and-row algorithm under a cross-layer network setting for this airline scheduling problem. Preliminary experiment results demonstrate the improvement of scheduling consistency between schedule periods. 2 - On Routing Unmanned Aerial Vehicles for Surveillance and Reconnaissance Activities Cai Gao, University at Buffalo, Buffalo, NY, 14260, United States, Jose Luis Walteros We tackle a variation of the Close-enough Traveling Salesman Problem where the salesman is accounted for visiting a node if he traverses a precalculated distance through a circular area surrounding each node. This variation arises in the context of unmanned aerial vehicle (UAV) routing where a UAV collects information form a set of targets, while minimizing detection risks. We provide a mixed-integer formulation and solve it using Benders Decomposition. We enrich our approach by introducing a set of lifting algorithms to strengthen the optimality cuts generated by the proposed decomposition and a k-opt heuristic in the style of the classic Lin-Kernighan algorithm to generate better lower bounds. 3 - 5G Hierarchical Network Slicing with Uncertain Demands Network slicing, a key enabling technology for 5G development, creates concurrently dedicated and independent virtual networks and virtual network services for tenants on a common physical infrastructure platform. Compared with early works targeting single-domain physical infrastructure and demand- driven virtual network construction, we present in this talk multi-domain network slicing jointly with the construction of network functions’ forwarding graph. We discuss random tenant choices and the respective resource allocation based on traffic/demand uncertainty under a cross-layer network topology. Tachun Lin, Bradley University, 1501 W. Bradley Ave, Bradley Hall 171, Peoria, IL, 61625, United States

63