Machine Learning for Unit Commitment - Assignment 2

This project applies machine learning to predict power plant commitment decisions in day-ahead generation scheduling.
By training classification models on optimized/historical unit commitment results, it explores how data-driven approaches can complement traditional optimization for faster scheduling.

Date: October-November 2024

---

📌 Project Overview

The study investigates how linear and non-linear classifiers (SVMs) can approximate the binary on/off status of generation units from a system operator perspective.
Data include load, wind, and weather features, used to train and evaluate models predicting optimal commitment states.
The workflow covers:

• Data preprocessing and feature generation
• Model training and comparison across kernel types
• Evaluation and interpretation of prediction accuracy

---

🛠 Skills & Techniques Demonstrated

• Unit commitment modeling and operational planning
• Linear and non-linear SVM classification
• Feature engineering with meteorological and load data
• Model evaluation using precision, recall, and F1-score
• Comparative analysis of kernel performance and regularization

---

📈 Key Results

• Classification models can accurately replicate unit commitment results using only historical data.
• Non-linear kernels (e.g., RBF) generally outperform linear ones for capturing operational boundaries.
• Weather and demand features significantly improve model performance.
• Classification can accelerate scheduling when integrated into optimization workflows.

---

🚀 Optional Improvements

• Address class imbalance to better predict rare commitment states.
• Explore hybrid ML–optimization schemes for real-time applications.
• Extend feature space to include market prices and reserve margins.

---

🧠 Key Files

• A2.py: Implements the mathematical formulation and optimization model for unit commitment, including startup, ramping, and min up/down time constraints.
• assignment2_datapreprocessing.ipynb: Data cleaning, feature engineering, and exploratory analysis.
• assignment2_step3.ipynb: Training and evaluating linear vs. non-linear classifiers, with visual comparisons.

---

👥 Contributors

@strenchev
@nic0lew0ng
@raullabarthes

---

📝 References

• Machine Learning for Energy Systems course materials.
• Assignment instructions are provided in Assignment 2.pdf.

---

🔖 Tags

#machinelearning #optimization #svm #classification #unitcommitment #energysystems