Energy A.I. Hackathon 2025 – Predictive Energy Modeling

View on GitHub
Check out the presentation here!

Overview

As part of the Energy A.I. Hackathon 2025, our team developed a machine learning pipeline to predict energy consumption (Grid, Diesel, and CNG) for hydraulic fracturing operations. The project integrated robust uncertainty modeling, advanced feature engineering, and domain-specific insights, earning 1st Place out of 15 teams.

Key Features

• High Prediction Accuracy: Achieved R² scores of 0.9918 (Grid), 0.9766 (Diesel), and 0.9692 (CNG) using XGBoost and Random Forest models.
• Uncertainty Modeling: Quantified prediction uncertainty with 100 realizations per well.
• Feature Engineering: Leveraged CatBoost encoding, feature ranking, and temporal modeling to enhance predictions.
• Custom Libraries: Integrated ozbayrak_goodness for model evaluation and UTuning for hyperparameter optimization.

Tools Used

• Programming Languages: Python
• Libraries/Frameworks: scikit-learn, XGBoost, Pandas, NumPy, CatBoost
• Custom Libraries: ozbayrak_goodness, UTuning

Achievements

• First Place Winners: Secured top position among 25 teams in accuracy (MAPE), code quality, and presentation.