Electricity Load Forecasting Pipeline (Germany & Hungary)

End-to-end Time-Series Forecasting Project
Direct analogy: Demand Forecasting in FMCG

MAPE 4.1% on 14-day ahead forecast — production-ready accuracy
Holt-Winters + Facebook Prophet on 200,000+ hourly observations

Why This Project Matters for Nestlé

Nestlé's EU-wide forecasting solutions require:

• Maintaining & improving existing time-series models
• Experimenting with new forecasting approaches
• Clear communication of results to non-technical stakeholders

This project demonstrates exactly that — using real European grid data as a perfect proxy for demand forecasting in FMCG.

---

Key Results

Country	Model	MAE (MW)	RMSE (MW)	MAPE
Germany	Holt-Winters	2,131	2,569	4.08%
Germany	Prophet	2,460	3,174	4.72%
Hungary	Prophet	201	255	4.12%
Hungary	Holt-Winters	903	1,082	18.73%

MAPE < 5% on 14-day horizon — excellent for operational planning

---

Pipeline Overview

graph LR
    A[Download Raw Data<br>OPSD 15-min] --> B[Transform & Clean<br>pandas + interpolation]
    B --> C[Validate Quality<br>missing, duplicates, ranges]
    C --> D[Exploratory Analysis<br>seasonality, trends]
    D --> E[Time-Series Forecasting<br>Holt-Winters + Prophet]
    E --> F[Results & Plots<br>MAPE, forecasts, PNGs]

Loading

One command → full pipeline:

python src/main.py

Features

• Automated ETL (extract → transform → validate)
• Two production-grade forecasting models
  • Holt-Winters (Triple Exponential Smoothing)
  • Facebook Prophet (with holidays & changepoints)
• Robust data quality checks
• Visualizations (daily trends, hourly seasonality, forecast vs actual)
• Ready-to-deploy structure — clean, modular, documented

Tech Stack

• Python, pandas, numpy
• statsmodels (Holt-Winters)
• Prophet (Facebook)
• matplotlib, seaborn
• Git + clean project structure

How to Run

git clone https://github.com/yourname/energy-load-forecasting.git
cd energy-load-forecasting
pip install -r requirements.txt
python src/main.py

→ Downloads data → Cleans → Analyzes → Forecasts → Saves plots & metrics

Business Insights

• Strong weekly & daily seasonality
• Winter peaks due to heating demand
• High correlation between Germany & Hungary (~0.82)
• Prophet excels at capturing holidays & long-term trends
• Holt-Winters best for stable weekly patterns

License

MIT License — free to use, modify, and share.