Register now After registration you will be able to apply for this opportunity online.
Data-Driven Demand-Side Flexibility Quantification
The integration of distributed renewable energy sources into electric power grids is essential for transitioning to low-carbon energy systems. However, the intermittent nature of distributed renewable energy poses challenges to grid stability. Demand-side flexibility has emerged as a key solution, allowing consumers to adjust their electricity usage to help balance supply and demand. Buildings, as major energy consumers, offer substantial demand-side flexibility potential by shifting or reducing their energy use without compromising occupant comfort. To harness this potential, predictive energy management systems have been developed to optimize energy usage and quantify flexibility, typically represented as flexibility envelopes. These envelopes are used by distribution system operators (DSOs) for effective grid coordination.
However, existing methods for flexibility quantification are largely optimization-based, requiring significant computational resources—especially problematic for real-time or rolling updates, which involve repeatedly solving complex models under varying conditions. This limits their scalability and responsiveness in practice.
This research aims to develop a machine learning-based approach to predict flexibility envelopes using historical data. The goal is to provide real-time flexibility estimates with significantly reduced computational cost, making this method more practical for integration into smart energy systems.
Keywords: advanced machine learning models
Demand side management
Flexibility envelops
The primary objective of this project is to develop a machine learning-based approach to replace the repeated optimization procedures currently used for real-time flexibility quantification in buildings. Specifically, the research focuses on predicting flexibility envelopes, which quantify a building’s energy flexibility and are represented as a three-dimensional surface spanned by lead time, feasible power levels, and the corresponding maximum sustained duration (see Figure for an example).
In our previous work, each point on this 3D surface was predicted individually—i.e., the maximum sustained duration was estimated for specific combinations of lead time and power level. While effective, this point-wise prediction approach is limited in efficiency and scalability.
The aim of this project is to predict the entire 3D grid (or image-like structure) at once using advanced machine learning models, such as Transformers or Convolutional Neural Networks (CNNs). These models are well-suited for structured data and offer the potential for both high accuracy and fast inference.
The primary objective of this project is to develop a machine learning-based approach to replace the repeated optimization procedures currently used for real-time flexibility quantification in buildings. Specifically, the research focuses on predicting flexibility envelopes, which quantify a building’s energy flexibility and are represented as a three-dimensional surface spanned by lead time, feasible power levels, and the corresponding maximum sustained duration (see Figure for an example). In our previous work, each point on this 3D surface was predicted individually—i.e., the maximum sustained duration was estimated for specific combinations of lead time and power level. While effective, this point-wise prediction approach is limited in efficiency and scalability. The aim of this project is to predict the entire 3D grid (or image-like structure) at once using advanced machine learning models, such as Transformers or Convolutional Neural Networks (CNNs). These models are well-suited for structured data and offer the potential for both high accuracy and fast inference.
**Literature Review:** Conduct a comprehensive review and comparison of machine learning techniques suitable for image or 3D grid prediction tasks.
**Model Development:** Design and implement a machine learning algorithm capable of predicting the full flexibility envelope surface from historical data.
**Evaluation:** Assess the performance of the proposed method using real-world data from the NEST building in Dübendorf, Switzerland, through simulation-based experiments.
**Literature Review:** Conduct a comprehensive review and comparison of machine learning techniques suitable for image or 3D grid prediction tasks.
**Model Development:** Design and implement a machine learning algorithm capable of predicting the full flexibility envelope surface from historical data.
**Evaluation:** Assess the performance of the proposed method using real-world data from the NEST building in Dübendorf, Switzerland, through simulation-based experiments.
Master_thesis_proposal_MAGNIFY_empa.pdf
