xpertsystems/enr002-sample

---
license: cc-by-nc-4.0
task_categories:
- tabular-classification
- tabular-regression
- time-series-forecasting
tags:
- synthetic-data
- renewable-energy
- solar-pv
- wind-energy
- onshore-wind
- offshore-wind
- hybrid-plant
- battery-storage
- bess
- weather
- irradiance
- ghi-dni-dhi
- power-forecasting
- probabilistic-forecasting
- p10-p90
- ramp-detection
- curtailment
- grid-integration
- pcc
- iec-61400
- iec-61724
- nerc
- nrel
- iea-wind-task-36
- power-curve
- capacity-factor
- inverter-efficiency
- soiling
- wake-loss
- yaw-error
- lvrt
- duck-curve
- climate
pretty_name: ENR002 — Synthetic Renewable Energy Generation Dataset (Sample)
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files: enr002_renewable_data.parquet
---

# ENR002 — Synthetic Renewable Energy Generation Dataset (Sample Preview)

**XpertSystems.ai | Synthetic Data Factory | Energy & Climate Vertical**

A single-table, **wide-schema renewable energy telemetry dataset** spanning
solar PV, onshore wind, offshore wind, and hybrid (solar+wind+BESS) plants
with **5-minute SCADA resolution**. Each row joins weather, irradiance,
power production, probabilistic forecast, grid integration at PCC, and
battery dispatch on a single timeline. Calibrated benchmark-first against
IEC 61400-12 (wind power curves), IEC 61724 (PV monitoring), NERC
TOP-001-5 / BAL-003-2 (grid voltage/frequency), and NREL 2023 ATB
(capacity factors).

This is the **sample preview** — 10 sites × 3 days at 5-min cadence (~8,640
rows × 81 columns). The full product covers 1,000 sites × 8,760 hours
(~88M rows) with N-1 grid stress, storm scenarios, and high-renewable
duck-curve dispatch.

---

## Dataset summary

| Property | Value |
|---|---|
| Rows | ~8,640 |
| Columns | 81 |
| Cadence | 5-minute SCADA |
| Time span | 3 days (2024-01-01 → 2024-01-04) |
| Sites | 10 |
| Technology mix | Solar PV / Onshore Wind / Offshore Wind / Hybrid |
| Hybrid sites | Solar + Wind + BESS dispatched on smoothing logic |
| File formats | Parquet (preferred) + CSV |

The 81 columns are grouped into **eight blocks** that join on `site_id` ×
`timestamp_utc`: site metadata, weather, solar irradiance, solar power,
wind speed, wind power, probabilistic forecast, grid (PCC), and BESS.

---

## Calibration sources

All ten validation metrics target named industry sources, not generator
self-metrics:

- **IEC 61400-12** — wind turbine power curve standard (cubic ramp,
rated zone, cut-out)
- **IEC 61724-1** — photovoltaic system performance monitoring
- **Betz limit (1919)** — fundamental wind power extraction bound (16/27 ≈ 0.593)
- **NERC TOP-001-5** — interconnection voltage limits at PCC
- **NERC BAL-003-2** — frequency response and nominal frequency
- **NREL 2023 ATB** — utility-scale solar / onshore wind / offshore wind capacity factors
- **NREL TR-65-72701 / IEA Wind Task 36** — probabilistic forecast verification
(P10/P90 coverage, MAE-as-percent-of-nameplate)
- **Sandia inverter model** — efficiency curve for utility PV inverters
- **ISO 2533:1975** — International Standard Atmosphere (air density)

---

## Validation scorecard (seed = 42)

10/10 PASS · **Grade A+ (100%)** across all six canonical seeds (42, 7, 123, 2024, 99, 1).

| # | Metric | Observed | Target | Tol | Type | Source |
|---|---|---:|---:|---:|---|---|
| 1 | `solar_daytime_capacity_factor` | 0.283 | 0.25 | ±0.10 | two-sided | NREL 2023 ATB — utility PV |
| 2 | `solar_daylight_nonzero_rate` | 1.000 | 0.95 | ±0.05 | FLOOR | Structural PV chain |
| 3 | `wind_capacity_factor_when_generating` | 0.410 | 0.30 | ±0.10 | FLOOR | NREL 2023 ATB — wind |
| 4 | `wind_betz_compliance_rate` | 1.000 | 0.99 | ±0.01 | FLOOR | Betz limit (1919) |
| 5 | `pcc_voltage_mean_pu` | 0.996 | 1.000 | ±0.015 | two-sided | NERC TOP-001-5 |
| 6 | `pcc_frequency_mean_hz` | 60.000 | 60.000 | ±0.010 | two-sided | NERC BAL-003-2 |
| 7 | `air_density_mean_kg_m3` | 1.231 | 1.20 | ±0.07 | two-sided | ISO 2533:1975 |
| 8 | `inverter_efficiency_mean_pct` | 95.50 | 95.5 | ±2.5 | two-sided | IEC 61724 / Sandia |
| 9 | `forecast_p10_p90_interval_coverage_pct` | 77.35 | 80.0 | ±8.0 | two-sided | NREL / IEA Wind Task 36 |
| 10 | `forecast_mae_pct_of_capacity` | 6.53 | 7.5 | ±3.5 | two-sided | NREL TR-65-72701 / IEA T36 |

---

## Schema highlights (81 columns)

**Site metadata (6):** `site_id`, `timestamp_utc`, `technology` (solar_pv /
onshore_wind / offshore_wind / hybrid), `capacity_mw`, `latitude`, `has_bess`.

**Weather (12):** `temperature_C`, `humidity_pct`, `pressure_hPa`,
`air_density_kg_m3`, `cloud_cover_pct`, `cloud_type` (Clear / Thin_Cirrus /
Cumulus / Stratus / Cumulonimbus), `cloud_state` (0–4 Markov index),
`precipitation_mm_hr`, `fog_flag`, `storm_flag`, `extreme_heat_flag`,
`icing_risk_flag`.

**Solar irradiance (7):** `ghi_w_per_m2`, `dni_w_per_m2`, `dhi_w_per_m2`,
`poa_irradiance_w_m2`, `solar_elevation_deg`, `clearness_index`,
`tracker_angle_deg`. (Zeroed on wind-only sites.)

**Solar power (10):** `cell_temp_C`, `dc_power_kW`, `ac_power_kW`,
`capacity_factor_solar`, `inverter_efficiency_pct`, `clipping_flag`,
`soiling_loss_pct`, `bifacial_gain_pct`, `degradation_pct`,
`ramp_rate_kW_per_min`.

**Wind speed (5):** `wind_speed_hub_m_per_s`, `wind_direction_deg`,
`turbulence_intensity`, `wind_shear_exponent`, `hub_height_m`.

**Wind power (10):** `farm_power_kW`, `turbine_power_kW`,
`capacity_factor_wind`, `cp_power_coefficient`, `wake_loss_pct`,
`yaw_error_deg`, `pitch_angle_deg`, `rotor_swept_area_m2`, `rpm_rotor`,
`downtime_flag`, `fault_code` (None / Overtemp / Grid_Fault / Yaw_Error /
Pitch_Fault).

**Forecast (12):** `solar_forecast_kW`, `actual_power_kW`,
`forecast_error_kW`, `mae_kW`, `rmse_kW`, `skill_score`, `p10_forecast_kW`,
`p50_forecast_kW`, `p90_forecast_kW`, `interval_coverage_pct`,
`ramp_event_flag`, `ramp_magnitude_kW`, `forecast_horizon_hr`.

**Grid integration / PCC (10):** `pcc_voltage_pu`, `pcc_frequency_hz`,
`active_power_export_kW`, `reactive_power_kvar`, `power_factor_pcc`,
`grid_curtailment_kW`, `frequency_response_flag`, `islanding_detection_flag`,
`fault_ride_through_flag`, `interconnect_status` (CONNECTED / CURTAILED /
TRIPPED / ISLANDED).

**BESS (6):** `bess_soc_pct`, `bess_charge_kW`, `bess_discharge_kW`,
`bess_round_trip_eff_pct`, `hybrid_dispatch_mode` (Charging /
Storage_Discharge / Solar_Only), `smoothing_activation_flag`.

---

## Suggested use cases

- **Solar irradiance → power chain** — train a regressor mapping POA
irradiance + cell temp + soiling → AC kW, with cloud type as a categorical
feature
- **Wind power curve learning** — fit nonparametric power curves per site
from `wind_speed_hub_m_per_s` → `farm_power_kW`, conditioned on
`turbulence_intensity` and `air_density_kg_m3`
- **Probabilistic forecasting evaluation** — benchmark new forecast models
against the included P10/P50/P90 baseline using `actual_power_kW`,
`forecast_error_kW`, and `interval_coverage_pct`
- **Ramp event detection** — classifier for `ramp_event_flag` given
weather predictors (cloud_cover_pct, wind_speed, storm_flag)
- **Hybrid (solar + wind + BESS) dispatch ML** — learn `hybrid_dispatch_mode`
from net power + ramp + SoC features
- **PV soiling and degradation modeling** — fit decay curves from
`soiling_loss_pct` and rainfall-triggered cleaning resets
- **PCC voltage / frequency response** — train LVRT and FFR classifiers
from `fault_ride_through_flag`, `frequency_response_flag`, and PCC
voltage/frequency timeseries
- **Curtailment prediction** — model `grid_curtailment_kW > 0` given local
generation and PCC voltage stress
- **Carbon intensity / merit-order integration** — combine with ENR001 grid
dispatch data for net-renewables analysis

---

## Loading examples

```python
from datasets import load_dataset

ds = load_dataset("xpertsystems/enr002-sample", split="train")
print(ds.column_names[:10], "...")
print(ds.shape)
```

```python
import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
repo_id="xpertsystems/enr002-sample",
filename="enr002_renewable_data.parquet",
repo_type="dataset",
)
df = pd.read_parquet(path)

# Filter to solar PV sites during daylight only
solar_day = df[
(df["technology"] == "solar_pv") &
(df["solar_elevation_deg"] > 5)
]
print(f"Daytime solar rows: {len(solar_day):,}")
print(f"Mean CF: {solar_day['capacity_factor_solar'].mean():.3f}")

# Wind power curve for one site
site_id = df[df["technology"] == "onshore_wind"]["site_id"].iloc[0]
site = df[df["site_id"] == site_id]
print(site[["wind_speed_hub_m_per_s", "farm_power_kW", "cp_power_coefficient"]].describe())
```

```python
# Probabilistic forecast evaluation
import numpy as np
from huggingface_hub import hf_hub_download
import pandas as pd

df = pd.read_parquet(hf_hub_download(
"xpertsystems/enr002-sample",
"enr002_renewable_data.parquet",
repo_type="dataset",
))

# Compute empirical P10-P90 coverage by technology
for tech, sub in df.groupby("technology"):
coverage = ((sub["actual_power_kW"] >= sub["p10_forecast_kW"]) &
(sub["actual_power_kW"] <= sub["p90_forecast_kW"])).mean()
print(f"{tech:<15} P10-P90 coverage: {coverage*100:.1f}%")
```

---

## Limitations and honest disclosures

This sample is calibrated for **structural fidelity, not bit-exact reproduction
of any specific fleet's SCADA archive.** Specifically:

- **Raw `wind_speed_hub_m_per_s` has fat upper tails.** The generator samples
the wind shear exponent per-timestep (`rng.uniform(0.10, 0.40, n_steps)`)
rather than as a per-site constant, which inflates hub-height wind speed
variance and yields nonphysical peaks (occasional 50+ m/s). The IEC 61400-12
power curve clips downstream at rated_wind (12 m/s onshore, 13 m/s offshore)
and cut_out (25 m/s), so `farm_power_kW`, `capacity_factor_wind`, and `cp`
remain physically valid — but **do not use raw wind speed for distribution
studies**. Use `farm_power_kW` instead.
- **Capacity factors are instantaneous (per-5-min-interval), not annualized.**
The bare `capacity_factor_solar` mean across all rows includes nighttime
zeros AND zero filler from wind-only sites. Use the technology-conditional
daytime CF metric (solar tech × solar_elevation_deg > 5°) — landing 0.25–0.30
in line with NREL ATB.
- **`ramp_rate_kW_per_min` is overwritten by wind ramp for hybrid and wind-only
sites** (last-dict-wins in generator assembly). For solar-only sites the
column holds solar ramp; for wind/hybrid sites it holds wind ramp. Treat as
*plant-level net ramp* rather than tech-specific.
- **`interconnect_status` (CONNECTED / CURTAILED / TRIPPED / ISLANDED) is
sampled independently of `active_export_kW` and `grid_curtailment_kW`.** Do
not use status × export joint distributions for ML training; use
`active_export_kW > 0` and `grid_curtailment_kW > 0` directly as filters.
- **Curtailment events fire rarely at sample scale** (curtail_factor ≈ 0.05 ×
cf_norm > 0.95 gate, only ~0–10 events at 72h). The full product activates
curtailment scenarios via `high_renewable` and `n1_grid_stress` configs.
- **All sites use the same "solar noon = 12:00 UTC"** regardless of longitude
— the generator models latitude but not longitude/timezone offsets. This
is fine for fleet-aggregate ML, but don't expect timestamp ↔ local-clock
alignment for any specific geography.
- **`forecast_skill_score` is a per-site scalar broadcast to all timesteps,**
and varies widely (0.0 to 0.7 across seeds) because the synthetic forecast
model adds Gaussian noise on top of true power — when persistence is
strong (calm wind days, smooth solar), persistence wins. We validate
`forecast_mae_pct_of_capacity` (consistently 5–8% across seeds) instead.
- **5-min cadence forecast horizons are anchored at 1-hour ahead** (column
`forecast_horizon_hr` = 1.0 throughout the sample). The full product
generates 15-min, 1-hr, 4-hr, and 24-hr horizons.

The full ENR002 product addresses these by per-site shear exponent constants,
DC-OPF curtailment dispatch, longitude-aware solar noon, and four forecast
horizons — contact us for the licensed commercial release.

---

## Companion datasets in the Energy & Climate vertical

- **ENR-001** — Synthetic Power Grid Operations Dataset (bus telemetry, line
flows, generation dispatch, frequency, contingency, weather/renewable)
- **ENR-002** — Synthetic Renewable Energy Generation Dataset (you are here)

Use **ENR-001 + ENR-002** together for full grid-plus-renewables ML
workflows: dispatch decisions from ENR-001 conditioned on plant-level
renewable telemetry from ENR-002.

For the broader catalog, see:

- [Oil & Gas](https://huggingface.co/xpertsystems) — OIL-001 through OIL-004
- [Materials & Energy](https://huggingface.co/xpertsystems) — MAT-001
- [Insurance & Risk](https://huggingface.co/xpertsystems) — 10 SKUs
- [Cybersecurity](https://huggingface.co/xpertsystems) — 11 SKUs

---

## Citation

```bibtex
@dataset{xpertsystems_enr002_sample_2026,
author = {XpertSystems.ai},
title = {ENR002 Synthetic Renewable Energy Generation Dataset (Sample Preview)},
year = 2026,
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/xpertsystems/enr002-sample}
}
```

---

## Contact

- **Web:** https://xpertsystems.ai
- **Email:** [e-mail]
- **Full product catalog:** Cybersecurity, Insurance & Risk, Materials & Energy,
Oil & Gas, Energy & Climate, and more

**Sample License:** CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0)
**Full product License:** Commercial — please contact for pricing.