Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC42

2 - Multi-Tasks Bayesian Optimization Approach to Demand Response Management in Smart Grid Jinkyoo Park, PhD, Korea Advanced Institute of Science and Technology, Daejeon, 34141, Korea, Republic of No abstract available. 3 - Forecasting Variability in Local Area Photovoltaic Power Generation by Sparse Identification Nonlinear Dynamics of Cumulus Clouds Jeff Manning, University of Texas, RASTRAC, 13809 Research Boulevard, Austin, TX, 78750, United States We propose and investigate the tractability of prediction of the temporospatial evolution of fair-weather cumulus clouds affecting the power production of a local photovoltaic generation site, by discovering a governing partial differential equation from data alone, in a convenient sequential photographic measurement basis. Clouds in such conditions have lifetimes ranging from five to thirty minutes, so a predictor must account for cloud evolution in addition to mere horizontal advection. The design is challenged by observability limitations, so we investigate alternative, but similar approaches. n TC44 North Bldg 227C Improving System Flexibility and Robustness Through Advanced Computation Sponsored: Energy, Natural Res & the Environment/Electricity Sponsored Session Chair: Shmuel S. Oren, University of California-Berkeley, Berkeley, CA, 94720-1777, United States Co-Chair: Georgios Patsakis, University of California Berkeley, Berkeley, CA, 94702, United States 1 - Recovery of Locally Optimal Solutions From Convex Relaxations of the AC Optimal Power Flow Andreas Venzke, Technical University of Denmark, Kgs. Lyngby, Denmark, Daniel K. Molzahn, Spyros Chatzivasileiadis Convex relaxations of the AC optimal power flow (OPF) have recently attained significant interest as in several test cases they provably yield the global optimum to the original non-convex problem. If the relaxation is inexact, i.e. it is infeasible to the original non-convex problem, it is observed that the optimality gap is often small, and the obtained solution is close to feasibility. We propose a method based on sequential quadratic programming to recover a locally optimal solution from the solution of the inexact convex relaxation and show the benefits compared to penalization techniques. We evaluate the applicability of our method for security and chance constrained AC-OPF formulations. 2 - Solving Large-scale Unit Commitment with Asynchronous Parallel Decomposition Anthony Papavasiliou, CORE, UCL, Voie du Roman Pays 34, L1.03.01, Office b.114, Louvain la Neuve, 1348, Belgium We present an asynchronous algorithm for solving a large-scale unit commitment problem. Dual iterations are performed using a block-coordinate subgradient method that works with delayed information, while primal solutions are recovered from the solutions of scenario subproblems using heuristics. We implement the algorithm in a high performance computing cluster and we conduct numerical experiments for instances of the Western Electricity Coordinating Council system and of the Central Western European system. The algorithm solves all instances within operationally acceptable tolerances and time limits. 3 - Mobilizing Grid Flexibility for Renewables Integration through Topology Control and Dynamic Line Ratings Shmuel S. Oren, University of California-Berkeley, Etcheverry Hall, Room 4119, Berkeley, CA, 94720-1777, United States, Jiaying Shi We propose an optimization model that mobilizes the inherent flexibility in the transmission grid through topology control and dynamic line ratings. We determines, in a stochastic unit commitment framework, when and which lines should be switched off or adopt temporarily higher ratings as part of the recourse actions. Such recourse actions in the second stage help mitigate the adverse impact of renewable resources uncertainty and enable less conservative first stage commitments. The potential gains are demonstrated for the IEEE 118 bus test case and a more realistic test system representing the central European grid. 4 - Mobilizing Renewable Resources for Power System Restoration Georgios Patsakis, University of California Berkeley, 1460 Cedar St, Berkeley, CA, 94702, United States, Deepak Rajan, Ignacio Aravena, Shmuel S. Oren Renewable energy currently serves a significant proportion of the electricity demand in the US. However, power system restoration is still considered predominantly a task for conventional generation, due to the intermittent nature

Antonie J. Jetter, Portland State University, ETM Department, 1900 SW 4th Ave, Portland, OR, 97207, United States, Ahmed Alibage Participatory modeling often occurs when the system under study is poorly understood and quantified. A case in point is safety culture, which describes the values, routines, and work processes that allow an organization to prevent disasters by avoiding and quickly bouncing back from mistakes. It is particularly relevant in oil and gas industry, where initially small errors can have devastating impacts. Relevant quantitative data on safety culture is virtually non-existent and practitioners have to rely on research findings in other context and their own observations: our work uses thematic analysis, t-coefficient, and Fuzzy Cognitive Mapping to create a model based on these knowledge sources. 4 - Twelve Questions for the Participatory Modeling Community Steven Gray, Michigan State University, Amherst, MI, United States, Antonie J. Jetter, Karen Jenni Participatory modeling engages the implicit and explicit knowledge of stakeholders to create formalized and shared representations of reality, and has evolved into a field of study as well as a practice. Participatory modeling researchers and practitioners who focus specifically on environmental resources met at the National Socio-Environmental Synthesis Center (SESYNC) in Annapolis, Maryland, over the course of two years to discuss the state of the field and future directions for participatory modeling. We will present a description of 12 overarching groups of questions that could guide future inquiry. n TC42 North Bldg 227A Joint Session Analytics/Practice Curated: Impactful Applications of Analytics in Industry No Abstract Available. 2 - Operations Research and Mathematical Sciences becomes Pervasive on a Networked, Collaborative Platform Ben Amaba, IBM Ben Amaba, University of Miami, Miami, FL, United States Mathematical modeling has provided a competitive advantage for institutions to aid in understanding, learning, decision making, forecasting and operations. As the number of sensors, smart phones, Internet of Things (IoT), robots, blockchain and other data sources grow, the discipline and methods become foundational on a platform to manage complexity. The power of processors, better networks, less expensive memory and disk space, and better user interfaces are making data sciences accessible to more individuals on an open, secured platform where reuse and collaboration allows users to combine artificial intelligence and other technologies to advance operations research application. n TC43 North Bldg 227B Machine Learning to Facilitate Renewable Integration in Power Grids Emerging Topic: Energy and Climate Emerging Topic Session Chair: Vaidyanathan Krishnamurthy, University of Pittsburgh, PA, United States Co- Chair: Jeff Manning, MSEE, University of Texas, Austin, TX, 78750, United States 1 - Multi Energy Storage Systems Control Using Multi Agent Reinforcement Learning Heechang Ryu, Korea Advanced Institute of Science and Technology, 3120 Ho, E2-2,, KAIST,291, Daehak-Ro, Daejeon, 34141, Korea, Republic of, Jinkyoo Park Energy storage systems (ESSs) can resolve the temporary imbalance in microgrid with a renewable energy source. Finding the optimal charging and discharging schedule of an ESS is essential. In particular, it is very challenging to optimally control each ESS when multiple households have their own ESSs and renewable energy sources. In this study, we show that these ESSs are controlled by using multi-agent reinforcement learning as a data-driven method. In addition, it is validated through a numerical simulation study with real data. Sponsored: Analytics Sponsored Session Chair: Subrat Sahu, Caterpillar Inc, Peoria, IL, United States 1 - Operationalizing a Customer Retention Framework in Logistics Industry Swapnil Srivastava

322