ML Renewable Trading - Assignment 1

This repository contains the work for Assignment 1 of the Machine Learning for Energy Systems course, focusing on renewable energy trading in day-ahead and balancing electricity markets. The project combines machine learning, regression models, and optimization techniques to forecast wind power production and optimize trading strategies.

Date: September-October 2024

---

📌 Project Overview

The assignment simulates the role of a wind farm owner aiming to trade energy in the electricity market. It processes historical wind and climate data from Bornholm, Denmark, along with day-ahead and balancing market prices, to develop predictive models and optimize trading strategies.

Two main modeling approaches were implemented:

1. Model 1 – Indirect Regression for Trading

   • Predict wind power production using linear and non-linear regression with L1 (Lasso) and L2 (Ridge) regularization.
   • Evaluate models based on prediction accuracy and expected revenue.
   • Explore data clustering (k-means) to improve prediction.

2. Model 2 – Direct Regression for Trading

   • Use regression or classification models to directly predict the optimal offering strategy for each trading period.
   • Compare performance with Model 1.

---

🛠 Skills & Techniques Demonstrated

• Data preprocessing and feature engineering
• Linear and non-linear regression
• Regularization (L1/Lasso and L2/Ridge)
• Optimization for energy trading (linear programming)
• Unsupervised learning (k-means clustering)
• Model evaluation (RMSE, MAE, R², revenue metrics)
• Python programming

---

📂 Repository Structure

├── data/ # Raw and processed datasets
│ ├── features-targets.csv # Model 1 dataset
│ └── model2_dataset.csv # Model 2 dataset
├── scripts/ # Python scripts for analysis, models, and optimization
│ ├── data_collection/ # Functions for dataset generation
│ │ ├── data_generator.py
│ │ └── model2datageneration.py
│ ├── linear_regression/
│ │ └── linear_regression.py
│ ├── nonlinear_regression/
│ │ ├── nonlinear_regression.py
│ │ ├── nonlinear_regression_metrics.py
│ │ └── weighted_regression.py
│ ├── optimization/
│ │ ├── bid_optimization.py
│ │ └── revenue_calc.py
│ └── regularization/
│ └── regularization.py
├── main.py # Script to run all relevant programs
└── report.pdf # Assignment report

---

📈 Key Results

• Non-linear regression with L2 (Ridge) regularization achieved the highest prediction accuracy for wind power.
• Model 1 (predictive regression approach) generated the highest expected revenue in day-ahead and balancing markets.
• Weighted regression and feature expansion improved accuracy, particularly during periods of high variability.

---

📌 Optional Improvements

• Explore ensemble methods or deep learning models to improve prediction accuracy.
• Extend analysis to multiple wind farms for portfolio-level trading optimization.
• Implement real-time prediction and trading workflow for dynamic market conditions.
• Incorporate additional features such as weather forecasts or intraday market signals.

---

👥 Contributors

This project was developed in collaboration with:

• @nic0lew0ng
• @ZilentKnight (Albert R. H.)
• @s233239 (zoewr)
• @MVKA-hub

---

📝 References

• Machine Learning for Energy Systems course materials.
• Historical wind data from Bornholm wind farms (DK2 market area).
• Day-ahead and balancing market data from Danish DK2 market area.

---