Background

Day-ahead market prices play a crucial role in managing flexible assets in energy systems. For calculating an optimal baseline that reflects the cost-efficient operation of flexible assets, such as battery storage, precise predictions of these prices are essential. Since day-ahead prices are not available at the time of baseline calculation, they must be forecasted based on historical data and other influencing factors.

Various methods can be employed for forecasting. Simple statistical approaches, such as calculating averages or trends from comparable days, provide a quick starting point but are often limited in their predictive accuracy.

More advanced methods from the field of machine learning, such as time series analysis or neural networks, can deliver more accurate results but require greater computational effort and data preprocessing.

Objective

The objective of this thesis is to design, implement, and evaluate at least two approaches for forecasting day-ahead market prices. These forecasts will later be used in calculating optimal baselines for flexible assets.

1. Familiarization with the topic (basics of day-ahead markets and their importance for energy systems, introduction to methods of time series analysis and machine learning, research on existing approaches to price forecasting, especially for energy markets)
2. Conception (development of at least two forecasting approaches, definition of the relevant influencing factors (e.g. historical price data or weather forecasts) and selection of suitable training and test data sets)
3. Implementation (implementation of the forecasting approaches in a programming language (e.g. Python))
4. Evaluation (comparison of the forecasting accuracy of the developed approaches using suitable metrics, analysis of the advantages and disadvantages of the individual methods)

• Interest in energy markets, data analysis and machine learning
• Basic knowledge of programming (preferably Python)

Optional:

• First experience with data processing and machine learning (e.g. time series analysis, regression)