---
language: en
license: mit
library_name: pytorch
tags:
- regression
- synthetic-data
- solar-energy
---

# Solar Linear Regression

Model trained by `_june-17/solar_linear_regression.py`. PyTorch linear regression predicting daily solar energy output (kWh).

## Dataset

Synthetic Colorado solar dataset generated by `_june-17/generate_solar_dataset.py`.

- **Size:** 500 rows
- **Target:** `daily_energy_kwh` (20.44 – 61.03 kWh)

| Feature | Min | Max |
|---|---|---|
| `day_of_year` | 1.00 | 365.00 |
| `sun_hours` | 4.02 | 8.50 |
| `cloud_cover_pct` | 0.40 | 59.90 |
| `panel_temp_c` | 5.30 | 70.00 |
| `system_size_kw` | 3.01 | 11.99 |

## Performance

- Val R²: ~0.997
- Val RMSE: ~0.48 kWh

## Files

- `solar_best.pt` — model weights, normalization stats, feature names
- `solar_linear_regression.py` — training script

## Usage

```python
import torch
from solar_linear_regression import LinearRegressionModel

ckpt = torch.load("solar_best.pt", map_location="cpu")
model = LinearRegressionModel(len(ckpt["feature_cols"]))
model.load_state_dict(ckpt["model_state_dict"])
model.eval()
```

Normalize input with `ckpt["X_mean"]` / `ckpt["X_std"]`, then denormalize output with `ckpt["y_mean"]` / `ckpt["y_std"]`.

## Limitations

Synthetic dataset; not based on real measured solar production.