README.md

---
license: bsd-3-clause
task_categories:
- other
language:
- en
- zh
tags:
- mixed-integer-programming
- power-systems
- optimization
- unit-commitment
- mps
- benchmark
size_categories:
- 1K<n<10K
---

# UnitCommitment Trajectory Dataset

This repository stores generated Unit Commitment trajectory benchmark data.

The project is split into two repositories:

- Code repository: `EridanusQ/UnitCommitment_Trajectory`
  - generation scripts
  - modified UnitCommitment.jl source code
  - Matpower instance download and preprocessing workflow
  - documentation for reproducing the MPS generation process
- Dataset repository: `EridanusQ/UnitCommitment_Trajectory_Dataset`
  - this Hugging Face Dataset repository
  - generated `.mps` benchmark files
  - one data case per Git branch
  - each branch stores the case as a compressed `.tar.gz` archive

## Repository Layout

The `main` branch is the landing page for the dataset repository. Data are stored on separate branches.

| Branch | Archive | MPS files | Archive layout |
| --- | --- | ---: | --- |
| `case14` | `case14.tar.gz` | 1460 | direct variant layout |
| `case30` | `case30.tar.gz` | 1460 | direct variant layout |
| `case57` | `case57.tar.gz` | 1448 | direct variant layout |
| `case118` | `case118.tar.gz` | 1460 | direct variant layout |
| `case3375wp` | `case3375wp.tar.gz` | 96 | horizon layout |
| `case6515rte` | `case6515rte.tar.gz` | 96 | horizon layout |
| `case9241pegase` | `case9241pegase.tar.gz` | 96 | horizon layout |

More case branches can be added using the same convention:

```text
branch name:   <case_name>
archive name:  <case_name>.tar.gz
archive root:  <case_name>/
```

## Archive Contents

After extracting a case archive, the archive root is always the case directory:

```text
case_name/
├── hourly_noline/
├── hourly_withline/
├── subhourly_noline/
└── subhourly_withline/
```

There are two layouts inside the four variant directories.

### Direct Variant Layout

Smaller full-year cases use this layout:

```text
case_name/
├── hourly_noline/*.mps
├── hourly_withline/*.mps
├── subhourly_noline/*.mps
└── subhourly_withline/*.mps
```

Current direct-layout branches:

```text
case14
case30
case57
case118
```

### Horizon Layout

Larger sampled cases use an additional horizon layer:

```text
case_name/
├── hourly_noline/{3_days,7_days,14_days}/*.mps
├── hourly_withline/{3_days,7_days,14_days}/*.mps
├── subhourly_noline/{3_days,7_days,14_days}/*.mps
└── subhourly_withline/{3_days,7_days,14_days}/*.mps
```

Current horizon-layout branches:

```text
case3375wp
case6515rte
case9241pegase
```

The four variants represent:

| Directory | Description |
| --- | --- |
| `hourly_noline` | Hourly UC instances without transmission line constraints |
| `hourly_withline` | Hourly SCUC instances with transmission line constraints |
| `subhourly_noline` | Sub-hourly UC instances without transmission line constraints |
| `subhourly_withline` | Sub-hourly SCUC instances with transmission line constraints |

File names follow:

```text
<case>_<date>_<resolution>_<network>.mps
```

For example:

```text
case3375wp_2017-05-01_s_withline.mps
case6515rte_2017-08-01_h_noline.mps
case9241pegase_2017-02-25_h_noline.mps
```

where `h` means hourly and `s` means sub-hourly.

## Download

### Download With Hugging Face CLI

```powershell
huggingface-cli download EridanusQ/UnitCommitment_Trajectory_Dataset `
  case3375wp.tar.gz `
  --repo-type dataset `
  --revision case3375wp `
  --local-dir data/case3375wp
```

Replace `case3375wp` with the target branch/case name.

### Download With Git LFS

```powershell
git lfs install
git clone -b case3375wp https://huggingface.co/datasets/EridanusQ/UnitCommitment_Trajectory_Dataset
```

Then extract the archive:

```powershell
tar -xzf case3375wp.tar.gz
```

## Dataset Generation

The generated data come from Matpower Unit Commitment instances and are exported as standard `.mps` files for mixed-integer programming solver benchmarking.

The generation workflow is maintained in the separate code repository, not in this dataset repository. Please use the code repository for:

- environment setup
- downloading source Matpower instances
- modifying or inspecting UnitCommitment.jl model code
- regenerating MPS files
- reproducing the trajectory constraints and preprocessing logic

This dataset repository is intended to be lightweight to browse: each branch contains only the compressed generated output for one case.

## Storage Policy

Do not commit expanded `.mps` directories to this repository. For each case:

1. Generate the case directory locally.
2. Compress it as `<case_name>.tar.gz`.
3. Commit the archive to the matching branch.
4. Track the archive with Git LFS.

This keeps each branch easy to download and avoids thousands of large text files in the Git tree.

## Citation

The original UnitCommitment.jl project is available at:

[10.5281/zenodo.4269874](https://doi.org/10.5281/zenodo.4269874)