4 présentations

• 
  08h30 - 08h52

  Alternating Methods for Large-Scale AC Optimal Power Flow with Unit Commitment Decisions

  • Matthew Brun, prés., MIT Operations Research Center
  • Thomas Lee, MIT
  • Dirk Lauinger, MIT
  • Xin Chen, Texas A&M
  • Xu Andy Sun, Massachusetts Institute of Technology

  Unit commitment with alternating current optimal power flow and (UC-ACOPF) is a central problem for power grid operations. These models are large scale, involving multiple time periods and networks with thousands of buses, and need to be solved efficiently. We propose a decomposition scheme and penalty alternating direction method that separates the problem into a set of mixed integer linear programs for each device and a set of nonlinear programs for each time period. To improve performance of the algorithm, we introduce a variety of heuristics, including restrictions of temporal linking constraints and a second-order cone relaxation. We construct a dual bound on the optimal objective value, allowing for quantification of the quality of feasible solutions. Evaluation on large scale test cases from Challenge 3 of the Grid Optimization Competition yield an average optimality gap of 1.33%, demonstrating that this approach yields near-optimal solutions within stringent time limits.

• 
  08h52 - 09h14

  Global optimality of optimal power flow problems

  • Fengyu Zhou, Two Sigma
  • Steven Low, prés., Caltech

  Optimal power flow (OPF) problems are nonconvex and NP-hard. Though hard in theory, they seem to be "easy" in practice in the sense that local algorithms, such as Newton-Raphson or interior methods, tend to yield global optima and convex relaxations tend to be exact. We prove a Lyapunov-like condition for a general nonconvex optimization problem to both have no spurious local optima and have exact convex relaxations. The condition is sufficient, and almost necessary. We illustrate this global optimality condition through application to an OPF problem in the branch flow model.

• 
  09h14 - 09h36

  Impact of power flow representation on stochastic nodal capacity expansion planning

  • Tomas Valencia Zuluaga, prés., Lawrence Livermore National Laboratory
  • Amelia Musselman, Lawrence Livermore National Laboratory
  • Jean-Paul Watson, Lawrence Livermore National Laboratory
  • Shmuel Oren, University of California Berkeley

  Accurately capturing the climate impacts faced by power systems due to increasingly weather-dependent electricity demand and generation necessitates models that handle their inherent uncertainty and have high spatial resolution. A stochastic nodal capacity expansion planning (CEP) model can address both but has scalability computational challenges. The use of power transfer distribution factors (PTDF) in the representation of power flow constraints can help addressing them but requires special consideration since the topology of the system is changing as part of the optimization process. In this work, we extend an existing PTDF formulation for transmission expansion planning to our stochastic nodal generation, transmission and storage CEP model. We show that the method proposed consistently outperforms the more commonly used b-theta formulation of linear DC power flow in terms of the resulting optimal cost and computational solution time using realistic test systems based on California and South Carolina.

• 
  09h36 - 09h58

  Wholesale Electricity Pricing under Extreme Events

  • Yury Dvorkin, prés., Johns Hopkins University

  Extreme weather events pose significant risks to renewable-dominant power systems, often causing beyond-design failures and prolonged supply interruptions. Current electricity market designs inadequately account for these events, to the extent that wholesale pricing was infamously suspended during extreme events in Australia, Texas, and Europe. This talk will describe a novel market design framework that integrates large deviation theory into wholesale market design (using so-called “rare” chance constraints), which enables the completion of wholesale markets relative to extreme events and provides incentives for market participants to contribute to system resiliency. However, the vanilla “integration” yields a non-convex optimization problem and returns a conservative solution. We address these challenges by using weighted chance constraints and customized cutting planes. We also present a multi-period extension of this framework to evaluate the contributions of energy-limited resources (such as battery storage) to system resiliency. Numerical experiments demonstrate the improvements to system reliability and efficiency.