Using Supercomputers to Speed Execution of the CAISO/PLEXOS 33% RPS Study

Abstract

Study Sponsors and Key PlayersThe CAISO/PLEXOS 33% Renewable Portfolio Standard (RPS) study was initiated by the California Public Utilities commission (CPUC) in late 2008. The CPUC tasked the California Independent Systems Operator (CAISO) to manage and perform the study. In turn, CAISO partnered with a team at PLEXOS Solutions (a software vendor) to build the model, and tasked analysts at Southern California Edison (SCE) to run the model. The SCE analysts also worked directly with PLEXOS Solutions to refine and reconfigure the model as appropriate over the course of the study.Lawrence Livermore National Laboratory (LLNL) was contacted by the CPUC after the analysts at SCE observed very slow runtimes in the model, necessitating a shift in the study completion date from April 2010 to February 2011. Specifically, a single month of the model was taking over a day to run, and a yearly calculation thus took several days. The CPUC asked whether LLNL high-performance computing (HPC) resources could be used to speed execution time of the model, and a team at LLNL investigated. An initial feasibility study was performed by LLNL in March-April 2010, followed by a comprehensive effort starting in August 2010 and continuing through mid-2011. Objectives and Phases of the CAISO/PLEXOS 33% RPS StudyThe study's official title is "ISO Study of Operational Requirements and Market Impacts at 33% Renewable Portfolio Standard (RPS)." The stated objectives are twofold: 1) identifying operational requirements and resource options to reliably operate the ISO-controlled grid under a 33% RPS in 2020; and 2) inform market, planning, and policy/regulatory decisions by the ISO, state agencies, market participants, and other stakeholders.The first of these objectives requires the hourly estimates of integration requirements, measured in terms of operational ramp, load following and regulation capacity and ramp rates, as well as additional capacity to resolve operational violations. It also involves consideration of other variables that affect the results, such as the impact of different mixes of renewable technologies, and the impact of forecasting error and variability. The second objective entails supporting the CPUC to identify longterm procurement planning needs, costs, and options, as well as informing other decisions made by the CPUC and state agencies. For the ISO itself this includes informing state-wide transmission planning needs for renewables up to a 33% RPS, and informing design of wholesale markets for energy and ancillary services to facilitate provision of integration capacities. CPUCCAISO PLEXOS SCE LLNL Builds Model Runs Model Initiates StudyManages Study Improves Runtimes

Business case uncertainty of power plants in future energy systems with wind power

“…In reality, these investment decisions would also be based on stochastic distributions of PIs, which can lead to different, suboptimal capacity mixes, also because investors in the power sector are risk-averse (Fan et al, 2010). An integrated stochastic optimization, simulation and business case analysis tool like this requires a supercomputer to solve, however (Meyers et al, 2011). 7.…”

High-performance computing for electric grid planning and operations

“…Mid-project, the CPUC contacted LLNL to investigate using supercomputers to accelerate the exection of the ensemble of simulations required for the study. LLNL partnered with Energy Exemplar, makers of PLEXOS for Power Systems R , to accelerate the solution by running the large computational simulations on LLNL's Hyperion supercomputer and by modifying the problem formulation [9].…”

“…It incorporates resource data, hourly load data for wind, import, hydro, solar, geothermal, and qualifying facility generation. The model covers 12 regions total across the WECC grid (shown in Figure 3), and it includes 42 utilities, 2157 generators, and 104 transmission lines [9]. The basic form of the model is an ensemble of deterministic unit commitment models.…”

“…4. Improvements in the number of simultaneous scenarios able to be executed, September-November 2010 [9]. Linux, and the improvement from Oct. to Nov. was by utilizing the Hyperion supercomputer at LLNL.…”

Time Domain Partitioning of Electricity Production Cost Simulations

“…Traditionally, a single deterministic solution is given based on a PCM simulation horizon. Some studies simulate a suite of scenarios highlighting the system sensitivity to certain input variables (e.g., wind or solar generation profiles) where each simulation is computed simultaneously (Meyers et al 2011). …”