Welcome to the project repository for our Responsible Data Analytics final project! This work presents a comprehensive, four-phase analytics pipeline designed to assess urban sustainability and renewable energy potential across three Baltic capital cities: Vilnius, Riga, and Tallinn.
Urban areas face mounting pressure to integrate renewable energy sources into their power grids while ensuring resilience against climate variability. Our project addresses this challenge by:
- Analyzing five years of meteorological data from NASA POWER
- Forecasting daily solar irradiance with a robust stacking ensemble
- Translating forecasts into actionable grid-management strategies
By combining advanced machine learning with ethical data practices, we pave the way for climate-resilient, energy-efficient cities.
The complete project report is under inspection for publication, but you can access part of the report: 📘 View part of the Report (PDF)
- Multi-City Analysis: Comparative study of Vilnius, Riga, and Tallinn.
- Comprehensive Pipeline: Four phases: Descriptive, Diagnostic, Predictive, Prescriptive.
- Stacking Ensemble: Random Forest, Gradient Boosting, SVR, and MLP with RidgeCV meta-learner.
- High Accuracy: Achieved RMSE 0.58–0.66 kW·h/m²/day and R²≈0.93.
- Actionable Insights: Maintenance scheduling, battery dispatch, demand-side load shifting, market bidding.
- NASA POWER: Daily meteorological variables (solar radiation, temperature, wind speed, humidity) for 2018–2022.
- City Metadata: Geographic coordinates and urban parameters for each capital.
- Visualized seasonal trends and correlations among key variables.
- Identified interdependencies and potential outliers.
- Engineered time-aware features (lagged variables, rolling statistics).
- Trained and validated a heterogeneous stacking ensemble.
- Evaluated performance via RMSE and R² metrics.
-
Translated irradiance forecasts into:
- Dynamic Battery Dispatch: Optimized charge/discharge cycles.
- Maintenance Scheduling: Scheduled during low-production periods.
- Demand-Response Load Shifting: Recommended peak load adjustments.
- Market Bidding: Informed pricing strategies based on forecast confidence.
- Ensured data privacy and ethical use.
- Documented bias checks and model interpretability assessments.
| City | RMSE (kW·h/m²/day) | R² Score |
|---|---|---|
| Vilnius | 0.60 | 0.92 |
| Riga | 0.58 | 0.94 |
| Tallinn | 0.66 | 0.93 |
Our ensemble consistently delivered high predictive accuracy, enabling reliable operational planning.
We welcome contributions! Please open issues or submit pull requests for bug fixes, enhancements, or new features.
- Milad @milad
- Jennifer j.hu-16@student.tudelft.nl