Skip to content

Latest commit

 

History

History
95 lines (79 loc) · 4.42 KB

README.md

File metadata and controls

95 lines (79 loc) · 4.42 KB
Raw

Milepost 7

Milepost 7 is an application of the ECP ExaSGD project to solve multiperiod security-constrained optimal power flow (MPSCOPF) at a large-scale on GPU architectures.

Dependencies

Milepost 7 depends on the following packages:

Installation

These packages and sub-dependencies will be pulled automatically when installing Milepost 7. To instantiate all the submodules, run the following command:

] up

Manifest files for Polaris (ALCF), Summit (OLCF), and Frontier (OLCF) are provided that have been used to run on these systems.

Quickstart

To run Milepost 7, run the following command:

$ julia --project main.jl
required argument case was not provided
usage: main.jl [--profile] [--rhopq RHOPQ] [--rhova RHOVA]
               [--configuration CONFIGURATION]
               [--proxal_iter PROXAL_ITER]
               [--exatron_inner_iter EXATRON_INNER_ITER]
               [--exatron_outer_iter EXATRON_OUTER_ITER] case load T K
  • case is the case file to use
  • load is the load files to use
  • T is the number of periods taken from the load.Pd and load.Qd files
  • K is the number of contingencies taken from the case.Cg file
  • --configuration is the configuration of the block solver in ProxAL given as an integer
    • 1: use JuMP and Ipopt,
    • 2: use ExaAdmm/ExaTron on CPU,
    • 3: use ExaAdmm/ExaTron on GPU with CUDA,
    • 4: use ExaAdmm/ExaTron on GPU with KernelAbstractions.jl, targeting either NVIDIA, AMD, or Intel GPUs, depending on the detected hardware.
  • --profile enables profiling of the code using PProf.jl
  • --rhopq is the ADMM penalty parameter for pq
  • --rhova is the ADMM penalty parameter for va
  • --proxal_iter is the maximum number of ProxAL iterations
  • --exatron_inner_iter is the maximum number of ExaTron inner iterations
  • --exatron_outer_iter is the maximum number of ExaTron outer iterations

Example

To run Milepost 7 on the ACTIVSg10k case with 4 MPI ranks, 2 periods, 1 contingency, amounting to 4 blocks, using the GPUs and KernelAbstractions.jl, with 500 ProxAL iterations, 400 ExaTron inner iterations, and 1 ExaTron outer iteration, with profiling enabled, run the following command:

./mpiexecjl --project -n 4 julia --project main.jl cases/case_ACTIVSg10k.m cases/case_ACTIVSg10k 2 1 --configuration 4 --proxal_iter 500 --rhopq 3e3 --rhova 3e4 --exatron_inner_iter 400 --exatron_outer_iter 1 --profile

We assume that each MPI rank is associated to one GPU. This will assign one block problem to each GPU. Note that multiple block problems can be assigned to a single GPU. In this case, the GPU will solve the block problems sequentially.

Data format

Contingencies

Milepost 7 supports N-x contingencies, though no convergence is guaranteed in this case. The format for the contingencies is (lines, columns) = (contingencies, lines).

Load data

Load data is stored in two files for the P and Q load, respectively. The format is (lines, columns) = (buses, periods).

The load data for the ACTIVSg2000 and ACTIVSg10k case have been generated by NREL and provided in CSV format. To convert these load files into the format used by Milepost 7, run the following command in the cases directory environment:

julia --project pdqdfromcsv.jl --help
usage: pdqdfromcsv.jl [--plot] [-h] case Pd Qd

positional arguments:
  case        Case file
  Pd          Pd load file
  Qd          Qd load file

optional arguments:
  --plot      Create a plot of the load
  -h, --help  show this help message and exit

Here is the plot of the load for the ACTIVSg10k case as defined in ACTIVSg10k.Pd and ACTIVSg10.Qd.

load

Funding

This research was supported by the Exascale Computing Project (17-SC-20-SC), a joint project of the U.S. Department of Energy’s Office of Science and National Nuclear Security Administration, responsible for delivering a capable exascale ecosystem, including software, applications, and hardware technology, to support the nation’s exascale computing imperative.