https://github.com/nahom32/energy-consumption-forecasting
This repository contains sequence modelling for energy consumption forecasting.
https://github.com/nahom32/energy-consumption-forecasting
deep-learning deep-neural-networks rnn tensorflow
Last synced: 3 months ago
JSON representation
This repository contains sequence modelling for energy consumption forecasting.
- Host: GitHub
- URL: https://github.com/nahom32/energy-consumption-forecasting
- Owner: Nahom32
- License: mit
- Created: 2025-06-24T06:18:19.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-06-24T07:47:42.000Z (4 months ago)
- Last Synced: 2025-06-24T08:33:46.625Z (4 months ago)
- Topics: deep-learning, deep-neural-networks, rnn, tensorflow
- Language: Jupyter Notebook
- Homepage:
- Size: 72.3 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# 🔮 Future Forecasting with Autoregressive LSTM/GRU
This Streamlit web app allows you to upload a preprocessed time series CSV dataset with 11 specific features and generate future forecasts using two deep learning models: LSTM and GRU. You can select which feature to predict and compare the forecasts visually and in tabular form.
Link: https://energyconsumptionforecast.streamlit.app/
---
## Features
- Upload a CSV file containing the required features.
- Select the target feature to predict from the 11 available.
- Configure the input sequence length (window size) and number of future steps to predict.
- Autoregressive forecasting using pretrained LSTM and GRU PyTorch models.
- Visualization of historical data alongside forecasts from both models.
- Side-by-side comparison of forecasted values.
---
## Required Features in CSV
Your uploaded CSV must contain these columns:
- `Global_active_power`
- `Global_reactive_power`
- `Voltage`
- `Global_intensity`
- `Sub_metering_1`
- `Sub_metering_2`
- `Sub_metering_3`
- `hour`
- `dayofweek`
- `month`
- `is_weekend`
---
## Getting Started
### Prerequisites
- Python 3.8 or above
- PyTorch
- Streamlit
- pandas
- numpy
- scikit-learn
- matplotlib
Install dependencies via:
```bash
pip install torch streamlit pandas numpy scikit-learn matplotlib
````
### Models
Place your pretrained model weights (`lstm_model.pt` and `gru_model.pt`) in the `../models/` directory relative to the app script.
---
## Running the App
Run the Streamlit app with:
```bash
streamlit run your_app_script.py
```
---
## Usage
1. Upload a CSV file with the required features.
2. Select the data slice to use (start and end row indices).
3. Choose the feature you want to predict.
4. Set the input sequence length and number of future steps to predict.
5. View the forecast plot comparing LSTM and GRU predictions.
6. Review forecasted values in the comparison table.
---
## Notes
* The app uses MinMaxScaler internally to scale features before prediction.
* The forecasting is autoregressive: each predicted step is fed as input to predict the next.
* Make sure your CSV contains no missing values for the required features.
---
## License
This project is open source under the MIT License.