Extending Data-Driven Merit Order Models for Multi-Country Electricity Price Forecasting with Flow and Storage Integration
The recent paper by Ghelasi and Ziel (2025) proposes a hybrid approach that learns fundamental electricity market parameters from historical data, bridging the gap between classical fundamental models and pure data-driven methods like machine learning. However, its current scope is limited to a single-country market and does not explicitly model cross-border flows or storage operation.
About the thesis:
Objectives
This project aims to extend the existing framework in two key directions:
1. Multi-Country Modelling: Include multiple EPEX SPOT SDAC countries; shift from regression-based import/export to implicit flow modelling.
2. Storage Integration: Add pumped hydro and other storage as market players; use deterministic/stochastic optimization with a price-forward curve.
Research Questions
3. Can the hybrid model outperform Axpo’s operational model on price and flow forecasts?
4. How accurate can it capture inter-country flows and storage operation without explicitly modelling FBMC like EUPHEMIA?
5. What is the value-add of integrated storage optimization for forecast accuracy and realism?
Methodology
6. Extend to a multi-node framework with transmission constraints and flow variables.
7. Integrate storage dispatch via optimization using price-forward curves.
8. Benchmark against Axpo’s operational model used in asset-backed trading with out-of-sample backtests and error metrics for prices and flows.
Expected Impact
9. Demonstrate feasibility and benefits of the models for interconnected European markets.
10. Provide actionable insights for Axpo’s traders and analysts on price and flow forecasting.
Your profile:
11. Master’s student with experience in Energy Markets or a related field (e.g. Engineering, Mathematics, Energy Science, Data Science), enrolled at a Swiss university or university of applied sciences (FH)
12. Proficient in Python
13. Solid understanding of optimization methods
Starting Date: As soon as possible