GOC Challenge 3 Benchmark Algorithm

The ARPA-E Benchmark Algorithm for Challenge 3 of the Grid Optimization Competition. This repository contains Julia code and a command line-executable Julia script for solving the GOC-3 AC Unit Commitment problem.

The GOC-3 AC Unit Commitment problem formulation may be found on the GO Competition website. A solver for the competition problem accepts input data as a JSON file and produces its output as another JSON file. Solutions may be evaluated using the C3DataUtilities Python package. The JSON files accepted and produced by the solver are described here. Problem data files used for testing are included in the test/data directory of this repository. All problem files used in the competition are available here.

Installing this package

This package has the following dependencies:

• JuMP.jl
• HiGHS.jl
• Gurobi.jl (only necessary for using the command line-executable script)
• Ipopt.jl
• MathOptSymbolicAD.jl
• ArgParse.jl
• JSON.jl
• Printf.jl

This package is not registered and must be installed from this repository. For example:

julia> ]
(@v1.X) pkg> add https://github.com/lanl-ansi/GOC3Benchmark.jl.git

To make sure that the package is installed correctly, please run the tests with

julia> ]
(@v1.X) pkg> test GOC3Benchmark

Using the solver

Challenge 3 of the GO Competition expects Julia submissions in the form of a single MyJulia1.jl file, which contains a function MyJulia1. This file is provided in the top-level of this repository. The APIs expected by the competition evaluation platform are documented here. The solver may be run with:

$ julia --compiled-modules=no -e 'include("MyJulia1.jl"); MyJulia1(ProblemFile, TimeLimitInSeconds, Division, NetworkModel, AllowSwitching)' &>MyJulia1.log

where arguments to MyJulia1 are as follows:

Argument	Julia type	Example
ProblemFile	String	"test/data/C3E4N00073D1_scenario_303.json"
TimeLimitInSeconds	Int	600
Division	Int	1
NetworkModel	String	"C3E4N00073"
AllowSwitching	Int	1

For example:

$ julia --compiled-modules=no -e 'include("MyJulia1.jl"); MyJulia1("test/data/C3E4N00073D1_scenario_303.json", 600, 1, "C3E4N00073", 1)' &>MyJulia1.log

When running via this API, solutions are written to solution.json in the working directory, as in the competition. Note that, in this solver, the "basename" of the ProblemFile string must contain the network name as a 6-character substring N#####. In this solver, the number of buses is extracted from this string and used to set some parameters.

For a more realistic representation of the benchmark code used by the competition, set the JULIA_NUM_THREADS environment variable before running the solver. For the competition, JULIA_NUM_THREADS=50 was used to allow all 48 ACOPF subproblems to run in parallel for Division 2.

Alternatively to using MyJulia1.jl, the solver may be run via the scripts/ac-uc-solver.jl script. The only required argument to this script is -c CASE, which must specify the path to the input data file. For example:

$ julia scripts/ac-uc-solver.jl -c test/data/C3E4N00073D1_scenario_303.json

For a list of all options, run

$ julia scripts/ac-uc-solver.jl --help

This script writes the solution file to the input file's directory, with the same name as the input file other than .json replaced with _solution.json. The script will remove solution files if the --remove-solution argument is set.

If the --evaluate-solution argument is set, this script will use the C3DataUtilities Python package to evaluate the solution and display results. This relies on PyCall.jl, and on the ability to find a Python installation with access to the C3DataUtilities package.

The default solver for the unit commitment subproblem (when running via the command line) is Gurobi. If Gurobi is not available, and you would like to run with an open-source MIP solver, use the --mip-solver=highs option. Note that the tests do not use Gurobi.

Structure of this repository

Of the Julia "library" code contained in this package, the run_ac_uc_solver function in the src/ac_uc_solver.jl file is the primary driver. This driver operates primarily by calling "subroutines," which accept input data and return output data for the three subproblems: copper-plate unit commitment, ACOPF, and reserve allocation. The "subroutines" may be found in the scheduler.jl, opf.jl, and reserves.jl files. Each of these subproblems build and solve one or more JuMP models. The code to construct the models themselves may be found in the scheduling_model.jl, opf_model.jl, and reserves.jl files.

License

This software is provided under a BSD license as part of the Grid Optimization Competition Solvers project, C19076. See LICENSE.md.