Warren Powell, Princeton University

Abstract

The analysis of energy systems works on many time scales, spanning fractions of a second to decades.  In this talk I will describe a series of models that span seconds to a year.  The first, SMART-Invest, is policy model that optimizes investments in renewables and storage using a yearly model that captures wind and solar variations in hourly increments.  SMART-Invest can model the full dispatch stack, but does not capture the grid, and does not schedule individual generators.  Next, SMART-ISO is a highly detailed simulator of the stochastic unit commitment process, simulating week-long periods in 5-minute increments using a process that closely matches the nested planning processes used by PJM.  SMART-ISO has a full model of the grid, and schedules fast and slow generators using actual notification times, while capturing the dynamics and uncertainty of renewables.  SMART-Storage uses the unit-commitment decisions from SMART-ISO, but co-optimizes generators and grid level storage over a period of a day to a week, in five-minute increments.  Finally, our newest model co-optimizes a battery being used simultaneously for frequency regulation (responding to signals every 2 seconds) and energy/peak shifting which requires horizons of a day or more.  All of these models represent a challenging form of stochastic optimization, since the models have to balance making robust decisions in the presence of uncertainty, while modeling the complex dynamics of a physical system.  We present a modeling framework that blends four classes of policies that integrate all the fields of stochastic optimization, and show how each time scale draws on the capabilities of different classes of policies.