Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MD11

2 - Nonlinear Programming Reformulations of Chance Constraints Alejandra Peña-Ordieres, Northwestern University, Evanston, IL, United States, Andreas Waechter, Jim R. Luedtke We present a new method for solving nonlinear continuous optimization problems with chance constraints. We introduce reformulations of the probabilistic constraints based on quantile functions and examine the theoretical and statistical guarantees of the approximation. To handle joint chance constraints, we propose an exact penalty function and use it to design an Sl1QP- type trust-region method. We demonstrate the efficiency of the method in numerical experiments. 3 - A Unifying Scheme of Primal-dual Algorithms for Distributed Optimization Fatemeh Mansoori, Northwestern University, Evanston, IL, United States, Ermin Wei We study the problem of minimizing a sum of local convex objective functions over a network of processors/agents. Many of the existing distributed algorithms with constant stepsize can only converge to a neighborhood of optimal solution. To circumvent this shortcoming, we propose to develop a class of distributed primal-dual algorithms based on augmented Lagrangian. To improve convergence speed, we design algorithms with multiple primal updates per iteration. We can show that such algorithms converge to the optimal solution under appropriate constant stepsize choices. The proposed class of algorithms can be extended to the general form of linearly-constrained convex optimization problems. 4 - Applying Model-based Derivative Free Methods in Reinforcement Learning Liyuan Cao, Lehigh University, Bethlehem, PA, 18015, United States Model-base derivative free methods are algorithms designed to optimize black- box functions. We use them in reinforcement learning, where the problems are modeled as black-box functions whose input is the set of parameters that define the policy, and the output is the reward. We test these methods in the OpenAI Gym environment and show that they are capable of learning high quality policies. Issues in the objective functions such as noise and non-smoothness are addressed. We compare our approach to recently proposed methods based on randomized finite differences and show connections between them. Chair: Selvaprabu Nadarajah, University of Illinois at Chicago, 601 South Morgan Street,, UH 2406, Chicago, IL, 60607, United States Co-Chair: Nicola Secomandi, Carnegie Mellon University, Pittsburgh, PA, 15213, United States 1 - Pathwise Optimization for Merchant Energy Production Modeled as a Switching Option Bo Yang, Carnegie Mellon University, Pittsburgh, PA, 15213, United States, Selvaprabu Nadarajah, Nicola Secomandi Merchant energy production modeled as a switching option gives rise to an intractable Markov decision process (MDP). It is common to combine least- squares Monte Carlo (LSM) and information relaxation and duality techniques to compute a feasible operating policy and bounds on the optimal policy value for this MDP. We extend the less explored pathwise optimization (PO) approach for this purpose, developing preconditioning and decomposition methods to make the resulting linear program solvable. On realistic instances, PO leads to substantially better optimality gaps but requires longer run times than LSM, slightly improving on the near-optimal LSM based operating policies. 2 - Analysis of Competitive Energy Markets under Capacitated Transportation Ian Yihang Zhu, University of Toronto, 144A, 60 Harbord Street, Toronto, ON, M5S3L1, Canada, John R. Birge, Timothy Chan, Michael Pavlin This paper considers the interplay between the transportation network, market structure and equilibrium prices of energy markets. We derive analytical methods drawing on inverse optimization and clustering methods that allows us to obtain insights into market structure and integration using only easily available regional pricing data. We show through several case studies of the North American gasoline market how our methodology yields results that are consistent with the underlying transportation network and market structure. n MD10 North Bldg 125A Joint Session MSOM/Practice Curated: Commodity and Energy Operations Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session

3 - Operations and Investment of Energy Storage in a Tree Network Owen Wu, Indiana University, Kelley School of Business, 1309 E. 10th Street, Bloomington, IN, 47405, United States, Roman Kapuscinski Matching electricity supply with demand is challenging due to inherent variabilities in demand and renewable power generation. Energy storage, though expensive, is operationally beneficial to electricity systems. In this talk, we discuss the optimization of energy storage investment and operations. In particular, we consider the problem of siting (where to invest) and sizing (how much capacity to invest) storage facilities in a tree network that has one power generation node and many demand nodes. The objective is to minimize the total storage investment and long-run power generation costs. We provide new insights on the key factors that affect storage investment decisions and the total costs. 4 - Managing Shutdown Decisions in Merchant Commodity and Energy Production: A Social Commerce Perspective Selvaprabu Nadarajah, College of Business, University of Illinois at Chicago, 601 South Morgan Street,, UH 2406, Chicago, IL, 60607, United States, Alessio Trivella, Stein-Erik Fleten, Denis Mazieres, David Pisinger Merchant commodity/energy production assets operate in markets with volatile prices and exchange rates. Plant closures often adversely affect the parent company and the local community. We study if mitigating these effects is financially viable by designing two types of policies that delay or avoid a plant closure. The first policy is grounded in anticipated regret theory and determined using approximate dynamic programming (ADP), while the second extends production margin-based heuristics used in practice and is computed using ADP and binary classification. Our operating policies significantly delay/avoid shutdowns for small asset value losses on real aluminum production instances. n MD11 North Bldg 125B Joint Session MSOM/Practice Curated: Finance and Risk Management Applications in OM Sponsored: Manufacturing & Service Oper Mgmt/iFORM Sponsored Session Chair: Yuan-Mao Kao, Duke University, Durham, NC, 27708, United States Co-Chair: N. Bora Keskin, Duke University, Durham, NC, 27708-0120, United States 1 - Government Financing for Clean Technology Development Seung Hwan Jung, Texas A&M University-Kingsville, 3118 La Rochelle Way, Corpus Christi, TX, 78414, United States, Lingxiu Dong The goal of this paper is to study government financing for a firm’s clean technology development under a financial constraint. In this paper, we investigate the impact of government financing on environment and the firm’s bankruptcy risk when market uncertainty exists. 2 - Operational Risk Management: Optimal Incentive Contract Yuqian Xu, University of Illinois at Urbana-Champaign, Wohlers Hall 487, 1206 S. 6th St, Champaign, IL, 61820, United States, Lingjiong Zhu, Michael L. Pinedo In this paper, we study how a financial firm can offer incentive bonus contracts to its employees so as to incentivize them to exert efforts in reducing potential operational risk losses. Each employee then needs to balance the trade-off between the effort based bonus and the cost of the efforts to him or her (in a non- monetary form). We characterize the equilibrium strategy between the firm and its $n$ employees, and then discuss the conditions under which incentive bonuses would be issued. 3 - Comparison of Integrated Risk Management Frameworks for Newsvendors Panos Kouvelis We study a newsvendor problem with profit risk control using VaR constraints. When a firm’s demand correlates with the price of a tradable financial asset, both financial tools (derivatives) and operational tools (inventory) can be used for profit risk management. Such integrated risk management (IRM) approaches have been studied using various optimization frameworks to reflect the risk aversion of decision-makers. To the best of our knowledge, we are the first to study IRM in a newsvendor setting using profit maximization under VaR constraints (mean-VaR). We compare different IRM frameworks and find that only under mean-VaR, inventory and financial hedging decisions are separable.

217