4 Presentations

• 
  08:30 AM - 08:52 AM

  Large-Scale Experimentation and Analysis of HPC-Based Scenario Decomposition via Progressive Hedging with mpi-sppy

  • Jean-Paul Watson, presenter, Lawrence Livermore National Laboratory
  • David L. Woodruff, UC Davis
  • Knueven Bernard, National Renewable Energy Laboratory

  Parallel implementations of scenario-based decomposition strategies are now scalable to thousands and millions of scenarios, thanks to advances in modeling systems (e.g.., Pyomo) and supporting meta-solvers (e.g., mpi-sppy). We describe challenges and their solution considering a large-scale power grid capacity expansion model, where scenarios represent individual days of weather. We discuss in detail various algorithmic tuning and approaches to rapidly finding high-quality solutions to these stochastic programs, utilizing the mpi-sppy meta-solver library.

• 
  08:52 AM - 09:14 AM

  2 New MPI-SPPY features: AML Agnosticism and Stochastic Consensus ADMM

  • David L. Woodruff, presenter, UC Davis

  In this talk we will describe two new features in the MPI-SPPY software library: AML Agnosticism and Stochastic Consensus ADMM

• 
  09:14 AM - 09:36 AM

  Performant Optimization Strategies for Stochastic Expansion Planning Models

  • Zachary Kilwein, presenter, Sandia National Labs
  • Rachael Alfant, Sandia National Labs
  • Matthew Viens, Sandia National Labs
  • Kyle Skolfield, Sandia National Labs
  • William Hart, Sandia National Labs

  We explore algorithmic strategies for rapidly generating high quality solutions for large-scale generation and transmission expansion planning models. We consider variations of progressive hedging that decompose the problem into bundles containing high and low fidelity model representations. For example, possible fidelities include physical, temporal and network resolutions, as well as alternative representations of uncertain scenarios. Multi-fidelity progressive hedging promises to accelerate search for near optimal solutions by leveraging inexpensive, low-fidelity models. We demonstrate the methods with realistic electrical grid planning problems, incorporating uncertainty in future forecasts. These multi-fidelity demonstrations leverage the idaes-gtep software package, which allows for facile creation and customization of expansion planning models.

• 
  09:36 AM - 09:58 AM

  A Julia Package for Modeling Linear Decision Rules

  • Bernardo Freitas Paulo da Costa, presenter, FGV
  • Joaquim Dias Garcia, PSR

  We present a JuMP extension, LinearDecisionRules.jl, that allows users to declare uncertain parameters in an optimization problem, in addition to variables. The package builds automatically the uncertainty set from constraints involving only the uncertain parameters, and the corresponding optimization problems for primal and dual linear decision rules.