Thesis Description

Background

Accurate forecasting of electricity prices, renewable generation, and demand has become increasingly important as renewable energy sources transform modern power systems. While energy forecasting has developed into a highly active research field, with thousands of new studies published every year, considerably less attention has been paid to how forecasts are ultimately converted into operational decisions.

In practice, market participants do not earn value from forecasts themselves but from the decisions they make based on those forecasts. Energy traders, utilities, aggregators, storage operators, and system operators continuously use forecasts to determine bidding strategies, dispatch decisions, portfolio allocations, and risk management actions. As renewable penetration increases, uncertainty becomes a central feature of power system operation, making decision-making under uncertainty increasingly important.

Despite this, most academic studies evaluate forecasting models in isolation using statistical accuracy metrics such as MAE, RMSE, or CRPS. Such metrics provide only an indirect measure of practical usefulness. A more accurate forecast does not necessarily lead to better operational outcomes, and small improvements in forecasting accuracy may generate little value compared to improvements in decision-making.

The Energy-Arena (developed at KIT IIP, Link to Paper) provides an open and operational benchmarking platform for energy forecasting. Building on this infrastructure, the next step is to benchmark not only forecasts but also the decisions derived from them.

Objective

The objective of this thesis is to extend the Energy-Arena framework from forecasting to operational decision-making by developing a benchmark for energy trading strategies under realistic market conditions. Instead of comparing forecasting models directly, the benchmark evaluates the quality of decisions made based on available information at predefined submission deadlines. A possible use case is the operation of a virtual PV-battery portfolio that participates in electricity markets. Participants submit trading and battery dispatch decisions, which are subsequently evaluated using realized market outcomes. The platform maintains continuously updated leaderboards and enables transparent comparison of competing strategies over time. The thesis will investigate how such decision-making challenges can be designed, implemented, and evaluated within an open and reproducible benchmarking environment.

Scientific Contribution

While most existing energy forecasting benchmarks focus on prediction accuracy, the practical value of forecasts ultimately depends on the decisions derived from them. This thesis contributes to a shift from benchmarking forecasts to benchmarking operational value. Using the example of a PV-battery portfolio, the developed framework may evaluate strategies for day-ahead trading, intraday trading, battery charging and discharging, or renewable production hedging. Approaches can range from rule-based systems and stochastic optimization to reinforcement learning and AI-based decision support. Performance may be assessed using economic, technical, and sustainability-related metrics such as profit, risk-adjusted returns, constraint violations, and carbon emissions. The resulting benchmark would provide a transparent and continuously updated environment for comparing operational energy management strategies and could serve as a foundation for future research on autonomous energy trading agents and decision-making under uncertainty.

Application

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