OPFRecourse

Package containing supplementary code for the paper "Statistical Learning For DC Optimal Power Flow" by Ng, Misra, A. Roald, and Backhaus.

Installation Instructions

OPFRecourse.jl is not a registered julia package. To install it, you should first install the following dependencies by running Pkg.add(...) in a Julia REPL:

• Distributions
• JuMP
• MathProgBase
• PowerModels
• ProgressMeter
• Clp

Secondly, you can install the package by running Pkg.clone("https://github.com/lanl-ansi/OPFRecourse.jl.git"), or manually downloading this github repository into your julia package directory (you can run Pkg.dir() in a Julia REPL to locate it).

Finally, you should clone a copy of the PGLIB OPF repository into your OPFRecourse/test/data folder. The current tests are based on release v17.08.

To test if it is successfully installed, run Pkg.test("OPFRecourse") in a Julia REPL.

Quick Example

We begin by importing the relevant packages, and getting the location of a matpower file, e.g.

using OPFRecourse, JuMP, Clp

data_file = string(Pkg.dir(),"/OPFRecourse/test/data/nesta_case30_ieee.m")

NetworkReference

This package provides a NetworkReference object that uses PowerModels to parse the matpower file, before pulling out the relevant information for the DC-OPF problem. To construct it, run

ref = OPFRecourse.NetworkReference(data_file, bus_prob = 0.95, line_prob = 0.95, σscaling = 0.05);

where optional parameters (i.e. bus_prob, line_prob and σscaling) can be passed in for the Chance Constraint OPF.

Model Formulations

To construct and solve the DC OPF model, run the following code:

m = OPFRecourse.SingleScenarioOPF(ref, Clp.ClpSolver());
@time JuMP.solve(m.model)

OPFScenarios

The ref::NetworkReference also contains the distributional information for generating scenarios of forecast deviations. In additional, we identify the optimal bases for for each scenario by solving single-scenario OPFs, and store them all within an OPFScenarios object:

@time scenarios = OPFRecourse.OPFScenarios(ref, m, nsamples = 1000);

The set of all unique column bases can be accessed using scenarios.cbases, and the set of all unique column bases can be accessed using scenarios.rbases respectively. These correspond to the power generation values and line flow constraints respectively, and contain information about system configurations (provided by scenarios.whichbasis) in which a subset of them are set to be active. Read the paper for more details.

Recourse Policies

SingleScenarioOPF implements a get_opf_solution(model, scenario) method that computes the power generation plan for the given scenario. So the following code

get_opf_solution(m, scenarios.scenarios[1,:])

would get the power generation plans corresponding to the first scenario.

BasisRecourse and EnsembleRecourse

This package also provides a BasisRecourse object that implements the recourse policy corresponding to a basis of the OPF problem. You can construct it, e.g. by

br1 = BasisRecourse(ref, m, scenarios.cbases[2], scenarios.rbases[1])
br2 = BasisRecourse(ref, m, scenarios.cbases[1], scenarios.rbases[2])

and be able to call get_opf_solution(br, scenarios.scenarios[i,:]) just like the other models.

You can also combine them into a EnsembleRecourse object that implements an ensemble of policies. For example,

ensemble = OPFRecourse.EnsembleRecourse(ref, br0, [br1, br2])

would construct an ensemble of the policies by br0, br1 and br2, falling back to the br0 policy (because it provides a probabilistic guarantee) if none of them provided a feasible solution.

Development

Community-driven development and enhancement of OPFRecourse.jl are welcome and encouraged. Please fork this repository and share your contributions to the master with pull requests.

Citing OPFRecourse

If you find OPFRecourse useful in your work, we kindly request that you cite the following technical report:

@misc{1801.07809,
  author = {Yeesian Ng and Sidhant Misra and Line A. Roald and Scott Backhaus},
  title = {Statistical Learning For DC Optimal Power Flow},
  year = {2018},
  eprint = {arXiv:1801.07809},
  url = {http://arxiv.org/abs/1801.07809}
}

We also encourage citing PowerModels.jl and PGLIB OPF if you use the packages in your work.

License

This code has been developed as part of the Advanced Network Science Initiative at Los Alamos National Laboratory, and is provided under a BSD license as part of the Multi-Infrastructure Control and Optimization Toolkit (MICOT) project, LA-CC-13-108.