README.md

This repository contains a fine-tuned **Patch Time Series Transformer (PatchTST)** model designed for predicting cooling load metrics from occupancy and weather time-series data. The model simultaneously predicts:

- `predicted_cooling_load_kWh`
- `optimized_cooling_load_kWh`
- `actual_cooling_load_kWh`

## Model Description

The model leverages historical occupancy, zone, and rich weather features to perform multivariate, multi-output time-series regression. The architecture is based on Microsoft's PatchTST transformer adapted for multivariate forecasting.

## Intended Use

This model is intended for:

- HVAC energy consumption forecasting
- Smart building climate control optimization
- Research on occupancy and environmental impact on cooling load

## Limitations & Ethical Considerations

- Model performance depends heavily on quality and representativeness of input data.
- Deployment in real environments should consider additional safety and calibration factors.
- Biases in historical occupancy or weather data may influence model predictions.

## Training Details

- Trained on timestamped structured data with occupancy, airflow, and weather inputs
- Sequence length: 48 hours historical context
- Prediction horizon: 1 hour ahead
- Library used: TSai (Time Series AI)
- Loss function: Mean Squared Error (MSE)
- Metrics: MAE and RMSE reported on validation data

## Evaluation Results

| Metric | Value (Validation Set) |
|--------------------------------|-----------------------|
| Mean Absolute Error (predicted_cooling_load_kWh) | 24.29 |
| Root Mean Squared Error (predicted_cooling_load_kWh)| 31.65 |
| Mean Absolute Error (optimized_cooling_load_kWh) | 26.13 |
| Root Mean Squared Error (optimized_cooling_load_kWh)| 34.26 |
| Mean Absolute Error (actual_cooling_load_kWh) | 26.44 |
| Root Mean Squared Error (actual_cooling_load_kWh) | 34.60 |

## How to Use

1. Load the model using TSai or PyTorch.
2. Prepare input data consistent with training normalization and encoding.
3. Run inference to obtain multivariate cooling load forecasts.

## Citation

Please cite this model if used in academic or commercial projects.

---

README.md

@@ -0,0 +1,21 @@
+---
+license: apache-2.0
+language:
+- en
+metrics:
+- mae
+- rmse
+pipeline_tag: time-series-forecasting
+library_name: transformers
+
+tags:
+- time-series
+- forecasting
+- passenger_occupancy
+- airport
+- weather
+- patchtst
+- transformer
+- cooling-load-prediction
+- multi-output
+---