---

PowerCast: Germany Electricity Price Forecasting Model

Overview

PowerCast is a machine learning-based model designed to forecast electricity prices in Germany, benefiting traders, grid operators, and policymakers. It utilizes an ensemble of XGBoost and Random Forest models with energy price data, generation sources, and grid metrics. The model achieves 62.7% directional accuracy, captures 40% of extreme price spikes, and helps optimize trading strategies, ensure grid stability, and promote renewables.

Key Findings

• Price Trends: Peak at 110–120 €/MWh (morning & evening), drop to 70 €/MWh at night. Weekends average 65–70 €/MWh.
• Wind Impact: High wind generation reduces prices by 32.41 €/MWh, often dropping below 100 €/MWh at 20,000+ MW wind output.
• Correlations: Prices rise with fossil fuel usage (0.67) and unmet demand (0.82), but fall with renewable adoption (-0.78).
• Model Performance: RMSE: 26.00, MAE: 16.40, R²: 0.8350, capturing 40% of extreme events.
• Feature Importance: SHAP analysis highlights renewable_demand_ratio, fossil_ratio, and conventional_vs_renewable as key factors.

Benefits for the Energy Sector

• Traders: Maximize profits by leveraging weekly price cycles (Sunday 65 €/MWh, Tuesday 98 €/MWh).
• Grid Operators: Store energy during low-price periods (32.41 €/MWh savings) for peak demand management.
• Policymakers: Encourage renewables to stabilize prices and reduce volatility.

Repository Structure

• data/: Raw and processed datasets.
• models/: Trained models (XGBoost, Random Forest, Ensemble).
• notebooks/: Step-by-step Jupyter Notebooks for EDA, training, SHAP analysis, and visualizations.
• report/: Final PowerCast_Report.pdf with findings.
• src/: Python scripts for data processing, model training, and interpretation.

Installation

1. Clone the repository:

   git clone https://github.com/[YourUsername]/PowerCast-Germany-Electricity-Price-Forecasting.git  
   cd PowerCast-Germany-Electricity-Price-Forecasting  

2. (Optional) Set up a virtual environment:

   python -m venv venv  
   source venv/bin/activate  # Windows: venv\Scripts\activate  

3. Install dependencies:

   pip install -r requirements.txt  

4. Launch Jupyter Notebook:

   jupyter notebook  

Usage

• Data Exploration: Run notebooks/01_data_preprocessing.ipynb and notebooks/02_eda.ipynb for insights.
• Model Training: Train models via notebooks/03_model_training.ipynb.
• Feature Analysis: Use notebooks/04_model_interpretation.ipynb for SHAP feature importance.
• Visualizations: Generate report charts using notebooks/05_report_visualizations.ipynb.
• Automated Pipeline: Run scripts in src/ directory for end-to-end execution.

Future Improvements

• Remove leaky features to improve model robustness.
• Integrate weather forecasts for better accuracy.
• Develop a real-time API for live electricity price forecasting.

Contributing

Contributions are welcome! Focus areas include:

• Enhancing model performance for extreme price events.
• Weather data integration for improved accuracy.
• Better visualizations for easier interpretability.

License

This project is licensed under the MIT License.

Contact

For questions or collaboration, contact me at [email protected].

---