Thesis description

Background

Accurate forecasting of electricity markets has become increasingly important as renewable energy sources continue to transform power systems. At the same time, energy forecasting has developed into a highly active research field, with thousands of new studies on electricity prices, load, and renewable generation published each year.

While a large body of academic literature proposes forecasting models for electricity prices, load, and renewable generation, the practical value of these approaches often remains difficult to assess. Most studies rely on static historical datasets and retrospective evaluations that do not reflect real-world operational constraints. As a result, reported performance is often difficult to compare across studies and may not translate into actual market value.

The Energy-Arena (developed at KIT IIP, Link to Paper) addresses this challenge by providing an open, real-time benchmarking platform for operational energy forecasting. Participants submit forecasts before predefined market deadlines, and forecasts are evaluated once actual observations become available. The platform aims to establish transparent and continuously updated benchmarks for forecasting performance under realistic conditions.

Objective

The objective of this thesis is to develop an open and operational forecasting system that can be integrated into the Energy-Arena benchmark platform and contribute to the transparent evaluation of forecasting performance over time.

Rather than focusing solely on forecasting accuracy, the thesis aims to develop a reproducible and continuously operating forecasting service that remains available for long-term benchmarking and comparison. The developed system should automatically generate forecasts under realistic operational constraints, be publicly documented, and support open and reproducible research.

Possible Application Areas

The forecasting system may address one or multiple forecasting tasks, including:

• -> Day-ahead electricity price forecasting
• -> Intraday electricity price forecasting
• -> Imbalance price forecasting
• -> Electricity load forecasting
• -> Solar generation forecasting
• -> Wind generation forecasting
• -> Additional forecasting tasks within the Energy-Arena framework

You can choose from a broad range of forecasting methodologies, ranging from classical statistical approaches and machine learning models to deep learning architectures and foundation models for time series forecasting. Depending on the research focus, forecasts may be developed as deterministic point forecasts or probabilistic forecasts, including quantile, interval, ensemble, and multi-horizon predictions.

The developed system will be designed as a reproducible and continuously operating forecasting service that can be benchmarked under real-world conditions within the Energy-Arena framework. Expected outcomes include an operational forecasting pipeline, automated forecast generation and deployment, integration into the Energy-Arena benchmarking platform, open-source implementation and documentation, and a scientific assessment of forecasting performance under realistic market conditions.

Scientific Contribution

The thesis contributes to the transition from retrospective forecasting studies towards open and continuously operating forecasting systems. By creating a reproducible forecasting service within the Energy-Arena ecosystem, the work supports a central objective of modern energy forecasting research: making cumulative scientific progress visible and measurable.

Application

Via email with CV and grades (BA&MA) to max.kleinebrahm∂kit.edu.