README.md

---
license: cc-by-4.0
task_categories:
  - reinforcement-learning
  - tabular-classification
  - time-series-forecasting
language:
  - en
pretty_name: SolarChain-Eval
tags:
  - agentic-ai
  - cyber-physical-systems
  - decentralized-energy
  - reinforcement-learning
  - trustworthiness
  - false-data-injection
  - solar-energy
size_categories:
  - 1K<n<100K
---

# SolarChain-Eval Dataset

SolarChain-Eval is a physics-constrained benchmark dataset for evaluating trustworthy economic agents in decentralized peer-to-peer solar energy markets.

The dataset supports the benchmark repository:

```text
https://github.com/GreenComp-ERC/SolarChain-Eval
```

## Intended Use

The dataset is intended for:

- evaluating autonomous economic governors in decentralized energy markets,
- studying trustworthiness metrics for agentic AI in cyber-physical systems,
- reproducing SolarChain-Eval experiments,
- testing reward-misspecification behavior under physics-informed constraints.

The dataset should not be used for real grid dispatch, financial trading, or operational market decisions.

## Dataset Contents

```text
data/
  datasets_2026_04_month/
    urban_energy_nodes.csv
    spatiotemporal_generation.csv
    market_liquidity.csv
    p2p_trades.csv
  datasets/
    urban_energy_nodes.csv
    spatiotemporal_generation.csv
    market_liquidity.csv
    p2p_trades.csv
  cache/
    open_meteo_weather_2026-04-01_2026-05-01.json
dataset_summary.json
checksums.sha256
DATA_LICENSE.md
```

`data/datasets_2026_04_month/` is the main paper dataset. `data/datasets/` is a small smoke-test dataset.

## Main Dataset

- Time window: `[2026-04-01, 2026-05-01)`
- Cities: Beijing, Shanghai, Chengdu, Shenzhen, Hangzhou
- PV nodes: 50
- Hourly timestamps: 720
- Generation rows: 36,000
- Market rows: 720
- FDIA rows: 1,800
- Generation seed: 20260511

## File Descriptions

### `urban_energy_nodes.csv`

Synthetic urban PV node metadata:

- `node_id`
- `city`
- `latitude`
- `longitude`
- `panel_area_m2`
- `efficiency`
- `temp_coefficient`
- `install_date`

### `spatiotemporal_generation.csv`

City-node-hour solar generation records:

- `timestamp`
- `hour`
- `node_id`
- `city`
- `latitude`
- `longitude`
- `irradiance_Wm2`
- `air_temp_C`
- `P_max_W`
- `P_reported_W`
- `fdia_detected`
- `verification_status`

### `market_liquidity.csv`

Hourly aggregate market liquidity and slippage signals:

- `timestamp`
- `hour`
- `total_verified_MW`
- `solarchain_liquidity_MW`
- `baseline_liquidity_MW`
- `slippage_solarchain_pct`
- `slippage_baseline_pct`

### `p2p_trades.csv`

Synthetic peer-to-peer trade demand records:

- `trade_id`
- `timestamp`
- `hour`
- `factory_id`
- `city`
- `energy_purchased_MW`
- `tokens_burned`
- `exergy_dissipated_MJ`

### Open-Meteo Cache

`data/cache/open_meteo_weather_2026-04-01_2026-05-01.json` stores weather API responses used to generate the main monthly dataset. It is included to improve reproducibility and reduce dependence on live API availability.

## Provenance

The dataset was generated by the SolarChain-Eval repository script:

```text
scripts/generate_monthly_datasets.py
```

Weather inputs are derived from the Open-Meteo Historical Weather API. PV nodes, false-data injection labels, reported generation, market liquidity, and P2P trades are synthetic benchmark records generated deterministically from the stated seed.

## Responsible AI Metadata

- **Synthetic data:** Yes.
- **Personal or sensitive information:** None. The dataset does not contain personal user records or real private transactions.
- **Limitations:** The benchmark covers five Chinese cities and one month. It is intended for controlled evaluation, not deployment.
- **Biases:** Results depend on city selection, the April 2026 weather window, synthetic PV node placement, synthetic trade demand, and the FDIA injection procedure.
- **Use cases:** Trustworthy agent evaluation, safety benchmarking, fairness analysis, reproducible RL experiments.
- **Non-use cases:** Real energy trading, grid dispatch, financial decisions, or operational market control.
- **Social impact:** The dataset is designed to support safer evaluation of agentic AI in cyber-physical markets. Misuse could occur if simulated outputs are mistaken for real operational data.

## Checksums

Use `checksums.sha256` to verify that downloaded files match the release package.

Example:

```bash
sha256sum -c checksums.sha256
```

On Windows PowerShell, compare against:

```powershell
Get-FileHash -Algorithm SHA256 <path>
```

## License

Dataset files are released under CC-BY-4.0. Benchmark code in the GitHub repository is released under MIT.