Enhancing the Temporal Resolution of Energy Time Series Using Machine Learning

Energy-related time series, such as household smart meter data or renewable energy output, are recorded with varying time resolutions. For instance, smart meter data are often available in 15-minute intervals to balance the need for understanding the dynamics of energy demand with manageable data sizes. However, demand peaks at higher time resolutions (e.g., less than 15 minutes) can differ significantly from those in 15-minute intervals. Since the peak demand determines the required secured connection capacity, it is crucial to conduct energy system analysis at higher resolutions.

To maintain manageable data sizes while still enabling energy system analysis at high temporal resolutions where necessary, the goal of this thesis is to develop a machine learning-based framework capable of enhancing the temporal resolution of energy-related time series. This framework will involve training machine learning algorithms (such as neural networks, neural operators, etc.) on high-resolution time series to learn the temporal stochasticity of energy-related time series. The trained algorithms will then be used to enhance time series with lower resolutions.

The first part of the thesis will involve a comprehensive literature review, comparing and contrasting suitable algorithms for the described problem (Neural Operators, GANs, Neural Networks). Next, the experimental setup will be described, and a framework will be developed that allows for training algorithms with high-resolution data and subsequently upsampling datasets with lower resolutions. The final section will critically evaluate the developments and provide an outlook. The code developed during this thesis will be made freely available as open-source.

All necessary data for this thesis is readily available.

Prerequisites:

• Experience with programming languages such as Python
• Initial experience with machine learning and big data
• High motivation to contribute to open-source energy modeling

If you are interested in the topic, I look forward to hearing from you via email.
Please apply with your CV and Transcript of Records.