Challenge 1

The Challenge 1 SCOPF problem utilized sets of unique datasets. Each dataset consisted of a collection of power system network models of different sizes with associated operating scenarios (snapshots in time defining instantaneous power demand, renewable generation, generator and line availability, etc.). Some datasets were open source and included models generated by the ARPA-E GRID DATA program. Some datasets that are not from the ARPA-E GRID DATA program were used; these datasets were proprietary actual industry cases that will not be released. The Original Dataset (OD) was released at the start of Challenge 1 in order to allow entrants to start developing solution methods. Entrants were able to download the dataset in order to test algorithms within their own development environment. Entrants could also submit software to be scored against the OD dataset using the official competition platform at any time. Aggregate scores (as well as individual scores for objective function value for each individual power system network/scenario within OD) were generated after each algorithm submission and returned to the submitter but not posted publicly. This was followed by a series of Trial Events (1, 2, and 3) where submitted software was run against different datasets not publicly available before the Trial Events. The results were posted on public leaderboards and the datasets were released after each Event. The Trial Event datasets consisted entirely of the synthetic datasets generated by three GRID DATA teams specifically for Challenge 1.

Entrants could choose to remain anonymous on the leaderboards. Two entrants did so but did not participate in any of the Trial or Final Events.

The sandbox was used for quick turn-around algorithm testing of input and output using small datasets and the Original Datasets. Entrants were required to demonstrate the ability to generate solution files using the sandbox datasets before being allowed to participate in any of the Trial or Final Events.

The Final Event used 17 synthetic network models with 20 scenarios each and 3 Industry network models with 4 scenarios each.

The 12 teams that placed in the top 10 of each Final Event Division had the following characteristics:

  • 4 submitted C++ code
  • 4 submitted Python code (2 using Python 2.7.13 and 2 using Python 3.7.2)
  • 3 submitted binary executables (2 using C/C++ and one using MATLAB)
  • 1 submitted Julia 1.2.0 code (this was the ARPA-E benchmark)
  • 7 used intelmpi
  • 3 used openmpi
  • for code 1
    • 6 used 6 nodes (up to 144 cores)
    • 6 used 1 node (up to 24 cores)
  • for code 2
    • 10 used 6 nodes
    • 2 used 1 node
  • 6 used Ipopt 3.12.13 (2 in combination with Gurobi 8.1.1; 1 in combination with CPLEX 12)
  • 1 used CPLEX 12 without additional solver libraries
  • 1 used the solvers in AMPL
  • 4 did not request any additional solver libraries