You can find the full code in here

This repository code came from the work of Mario Dagrada, special thanks for the help.

The repository consists of an energy forecasting model using XGboost. The dataset consists of hourly energy consumption rates in kWh for an industrial utility over a period of around 7 months, from July 2019 to January 2020.

The final model has an forecasting horizon (The number of time periods to forecast into the future) of 48 time periods which corresponds to 2 days ahead forecasting.

Details

Main techniques and terms used:

• Trend, Seasonality
• Stationary and Non-Stationary Time Series
• Augmented Dickey-Fuller Test
• Autocorrelation(ACF) and Partial Autocorrelation(PACF)
• Feature Engineering (lag features, standard time series features, endogenous features)
• XGBoost model with Bayesan hyperparameters optimization

The final model can be found inside model/

Usage

The Jupyter-Notebook, dataset and model were saved as Docker Image in GitHub Package. You can install the package locally:

docker pull docker.pkg.github.com/danieldacosta/energy-forecast/energy-forecast_notebook:v1

You can also download it direct from the repository:

git clone https://github.com/DanielDaCosta/energy-forecast.git

Acknowlegments & References

Special thanks to Mario Dagrada Medium post.

References:

• https://www.youtube.com/watch?v=Nm7m92sZZJA
• https://towardsdatascience.com/containerize-your-whole-data-science-environment-or-anything-you-want-with-docker-compose-e962b8ce8ce5