LANS Informal Seminar: Various Speakers

Seminar Title: LANS SASSy – Summer Argonne Students’ Symposium 2016
Speaker: Various Speakers, LANS Summer Students

Date/Time: 2016-08-19 13:00
Location: Bldg 240 rm 1404-1405

---

Description:
Summer students in LANS give talks on their research.

Session I
Chair: Wendy Di

Luo, Fengqiao 1:00PM
Title: Coordinated Vehicle Platooning with Multiple Speed Options
Abstract: Vehicle platooning is a fuel-saving technology under test. Its fundamental idea is to coordinate several automatic transmitted vehicles such that they form a train-like fleet while driving with the same speed and short inter-vehicle distance on a highway. In this platooning mode, trailing vehicles in a fleet can save fuel due to experiencing less aerodynamic drag compared to driving individually. The current vehicle platooning model is formulated as mixed-integer-programming that intends to minimize the total fuel consumption via searching for potential platooning opportunities achieved by properly routing and scheduling involved vehicles. We extend the current vehicle platooning model by introducing multiple speed options that each vehicle can take on edges of the highway network. We build numerical experiments for instances based on 25 vehicles distributed in a 10×10 grid network and a 4553-node greater-Chicago highway network. For instances with large number of vehicles, we propose a clustering-based heuristic method that decomposes the problem into sub-problems and each deals with a subset of all vehicles.

Lu, Shan 1:15PM
Title: Modeling weekly electricity demand
Abstract: We propose a static framework based on principal component analysis to model electricity demand over a long time horizon. The data is partition into weeks, and following the PCA analysis, we calibrate a Gaussian mixture to reconstruct the data.The forecaster is then used to model the exogenous demand process of a long term power system planning optimization problem, avoiding to compute the full cost function by sampling random weeks along a broken time horizon. In so doing, each generated week includes appropriately chosen overlapping sections, from which we characterize the relative memory, or forgetfulness, of the underlying decision process.

Jalving, Jordan 1:30PM
Title: Graph Abstractions for Scalable Modeling
Abstract: Optimization problems are often treated as static objects solved once to obtain results. In real applications however, the model is a basic component used to perform higher level functions. These functions range from solving a model many times to facilitate sensitivity analysis, or evaluating the performance of a network with different types of constraints and objectives over a range of scenarios. This work introduces a modular graph framework that abstracts the representation of algebraic models and the computational tasks associated with them into a hierarchal structure of nodes and edges. We describe the modular design and demonstrate its application to stochastic programming problems and infrastructure analysis. We will also discuss how we may convey structure to parallel optimization solvers and link to high-performance simulation packages such as PETSc.

Maldonado, Daniel Adrian 1:45PM
Title: Software library for scalable multi-physics multi-scale network simulation: application to water distribution systems
Abstract: We will present DMNetwork, a framework included in the scientific computing library PETSc that provides a set of abstractions to represent general unstructured networks. We use DMNetwork for the development of a water network simulator. Implementation details are discussed as well as scalability simulations on ANL’s MIRA.

BREAK

Session II
Chair: Julie Bessac

Patel, Vivak 2:30PM
Title: Estimation of Some Latent Parameters in Power Networks
Abstract: In power networks, the estimation of states and latent parameters is an important task to ensure reliable and efficient operation. As higher frequency observations of the network become available, previously inestimable latent quantities can potentially be estimated. As an example, we systematically develop the maximum likelihood estimators for the states of the network and the inertia of a wind farm over a fixed period of time. If time permits, we will also discuss efficient calculation of sensitivities using variational methods. Ultimately, the goal is to provide quickly-computable, consistent estimates of latent parameters which can inform decisions about network operations.

Kahvecioglu, Gokce 2:45PM
Title: On Implementation of Parallel Dual Decomposition for Stochastic Mixed-Integer Nonlinear Programming
Abstract: We consider computing solutions for stochastic mixed-integer nonlinear programs (SMINLP) via employing dual decomposition method. The dual decomposition method considers the Lagrangian relaxation of nonanticipativity constraints, which, in turn, enables to decompose the problem into mixed-integer nonlinear subproblems. There already exists an open-source and parallel package DSP by Argonne that implements decomposition methods for stochastic mixed-integer programming problems. We aim to contribute to DSP by expanding the range of problems it can solve, via adding the capability of solving SMINLPs. To this end, we aim to link DSP with Minotaur, an open-source toolkit by Argonne for solving mixed-integer nonlinear problems. On the modeling front, we use StructJuMP, a Julia package, as our choice of algebraic modeling language. We also aim to add on a few options to this package in order to handle the structure of SMINLPs.

Li, Juan 3:00PM
Title: Forecasting restaurants with critical violations in Chicago
Abstract: A predictive model, based on the real-time data from the City of Chicago, is designed to identify the presence of a critical violation in a particular food establishment. The goal of this model is to optimize food inspection processes to identify the riskiest restaurants earlier, thereby reducing the length of exposure of risky restaurants to patrons. Critical violations were identified approximately 7.44 days earlier over a 60 day period compared to current operations in the out-of-sample test. And we try to dig out more useful variable or effective model to find violations as early as possible.

Miksis, Zachary 3:15PM
Title: MHD with Nek5000
Abstract: We have worked to continue the development of the capabilities of Nek5000 for magnetohydrodynamics (MHD), and to apply this to a motivating problem of a Couette flow. This presentation gives an overview of the discretized formulation of the MHD equations and how Nek solves these, presents a current working example problem that exists within the repository of Nek along with an analysis of the solution, and an overview of further work for the motivating problem.