Open Theses

Bachelor Thesis

Supervisor: Sina Hajikazemi
Earliest start: immediately
Type: Bachelor Thesis


Energy planning models are essential for analyzing energy and climate policies at national and global scales. However, these models face various uncertainties, categorized into uncertainties in input parameters, such as future fuel prices, and uncertainties in the structure of the model, such as the complexities and constraints inherent in different technologies. While methods such as global sensitivity analysis, stochastic programming, and Monte Carlo simulation address parameter uncertainties, they often overlook uncertainties in the model structure. In addition, policymakers are faced with considerations outside the scope of conventional modeling, such as political feasibility, regulatory challenges, and the timing of actions. As a result, policymakers may choose feasible but suboptimal solutions due to the challenges of quantifying intangibles in energy optimization models. Modeling to Generate Alternatives (MGA), a technique borrowed from the operations research literature, is a valuable approach to address structural uncertainties inherent in energy planning models as well as uncertainties in input parameters. MGA efficiently explores the feasible region around the optimal solution and generates alternative solutions with maximum diversity. By providing a spectrum of viable options beyond the conventional optimal solution, MGA provides invaluable insights for policy makers. These alternative solutions shed light on trade-offs and considerations often overlooked in conventional energy planning models, enabling policymakers to make more nuanced and informed decisions amid uncertainty and real-world constraints. This thesis focuses on implementing this approach in a German energy transition model (see github.com/EINS-TUDa/CESM) and exploring the results and insights it can provide to decision makers.
Project Tasks:
  1. Understand the Modeling to Generate Alternatives (MGA) methodology.
  2. Apply the MGA methodology to the German energy transition model using the Compact Energy System Modelling Tool (CESM).
  3. Investigate the outcomes of the MGA implementation and identify the insights it offers for policymakers.
  4. Evaluate the strengths, weaknesses, and obstacles associated with the MGA methodology.
  5. Prerequisites: Proficiency in Python programming.

Learning Objectives: Through completion of this thesis, you will:
  1. Gain basic knowledge of mathematical programming techniques necessary for basic optimization tasks relevant to energy planning models.
  2. Develop a fundamental understanding of energy planning models, including their components, basic methodologies, and applications in energy policy analysis.
  3. Learn basic skills in reporting and justifying the outcomes of energy planning models, including simple interpretation of findings and basic assessment of model validity.
These learning objectives are tailored to provide essential skills and knowledge suitable for bachelor-level students to engage meaningfully in energy policy analysis and decision-making processes.
References:
[1] DeCarolis, Joseph F. "Using modeling to generate alternatives (MGA) to expand our thinking on energy futures." Energy Economics 33.2 (2011): 145-152.
[2] Brill Jr, E. Downey, Shoou-Yuh Chang, and Lewis D. Hopkins. "Modeling to generate alternatives: The HSJ approach and an illustration using a problem in land use planning." Management Science 28.3 (1982): 221-235.

Supervisor: Kirill Kuroptev
Earliest start: immediately
Type: Bachelor Thesis


This thesis deals with the critical area of frequency reserve mechanisms, which are essential for maintaining the stability of electricity grids. The objective is to present and partially simulate the activation of frequency restoration reserve providers in the European control reserve market. The activation function optimizes reserve providers' use to restore the electric grid's frequency in case of deviations.

The study includes the investigation of the frequency restoration reserve, where the technical requirements, the activation processes, and the participation of the market participants are elaborated.

Furthermore, the paper describes the formulation and constraints of the activation optimization function derived from the preceding technical and economic considerations. The resulting mathematical optimization problem is then implemented for a simplified example using programming languages such as Python, Julia, or Matlab.

To illustrate the identified activation optimization function, the thesis includes a numerical case study with different scenarios, such as limited cross-border capacity and increased volatility of control power demand. The results of this case study are analyzed to draw meaningful conclusions.

Supervisor: Sina Hajikazemi
Earliest start: immediately
Type: Bachelor Thesis


Electric transmission grids are critical energy infrastructures in every country. Intelligent attackers may attempt to damage specific components of the grid to cause maximum load shedding, and grid operators respond by solving the power flow problem to minimize load shedding using the remaining intact components. This raises the question: How vulnerable is the grid to adversarial attacks?

This project focuses on implementing and understanding an iterative optimization algorithm proposed by Javier Salmeron et al. [1] for the Electricity Network Interdiction problem. The algorithm formulates the problem as a bilevel programming problem, where the attacker aims to maximize load shedding, and the grid operator aims to minimize load shedding through optimal power flow in the attacked network.
Project Tasks:
1. Implementation: Implement the optimization algorithm in Python, ensuring clean and well-structured code.
2. Documentation: Provide clear and concise documentation for the implemented code, explaining key functions and algorithms.
3. Testing: Develop and execute test cases to validate the functionality of the implemented code.
4. Version Control: Utilize Git for effective code management and version control.
5. Presentation: Prepare a concise presentation that explains the project's objectives, methodology, and findings.

Prerequisites:
• Proficiency in Python programming.
• Basic understanding of mathematical optimization concepts.

Reference:
[1] Salmeron, Javier, Kevin Wood, and Ross Baldick. "Worst-case interdiction analysis of large-scale electric power grids." IEEE Transactions on Power Systems 24.1 (2009): 96-104.

Supervisor: Andreas Bott
Earliest start: immediately
Type: Bachelor Thesis


In order to reduce CO2 emissions, district heating grids will have to utilise decentralised heat sources such as large-scale heat pumps and industrial waste heat. Operating these so called 4th generation heating grids requires the development of new algorithms. If a grid has multiple feed ins, at one point the water from different supplies has to mix. Identifying this point is of special interest, as it is most likely to violate grid constrains, e.g. for the water to be to cold or the pressure to be to low. This thesis analyses a special measurement arrangement in order to identify this mixing point.

We assume to a grid segment with multiple uncertain demands. The position of the mixing point depends on the realisation of the demand. We assume, that the pressures and the mass flows at both inlets to the grid section are measured. This setup is especially interesting, as it allows to formulate the power estimation problem as a quadratic optimisation. The thesis should formulate and evaluate the results of this optimisation problem to evaluate the measurement setup.

Supervisor: Kirill Kuroptev
Earliest start: immediately
Type: Bachelor Thesis


Disruptive Technologien wie das "Internet of Everything" und das "Metaverse" führen zu einer erhöhten Auslastung der Telekommunikationsknotenpunkte, die sich im Energieverbrauch niederschlägt. Ein relevanter Anteil der eingesetzten Energie wird dabei für die Kühlung der Server verwendet. Da Server häufig über eine interne Kühlung, z.B. durch Lüfter, verfügen, diese jedoch nicht ausreichend und im Detail schwer zu bestimmen ist, werden verschiedene externe Kühltechnologien eingesetzt, um eine Überhitzung der Server zu vermeiden. Es stellt sich daher die Frage, ob der Gesamtenergiebedarf von Central Offices durch eine intelligente Wahl der externen Kühltechnologien reduziert werden kann, da die interne Kühlung der Server weniger Energie benötigt und somit Energieeinsparungen erzielt werden können.

Ziel der Arbeit ist es, eine empirische Studie für die Ermittlung des Zusammenhangs zwischen verschiedenen Kühlstrategien sowie dem Raumtemperaturniveaus und dem Gesamtenergieverbrauch von Telekommunikationsknotenpunkten zu konzipieren und zu erproben. Dazu sollen zunächst die für den Gesamtenergieverbrauch relevanten Parameter mit Fokus auf die klimatechnischen Komponenten recherchiert werden. Basierend auf den Recherchen soll ein holistisches Simulationsmodell der Telekommunikationsknotenpunkten erstellt werden, anhand dessen es möglich sein soll den Effekt verschiedener Kühlungsstrategien auf den Gesamtenergieverbrauchs abzubilden. Des Weiteren soll eine empirische Studie zur Ermittlung der Effektstärken der relevanten Parameter der Komponenten konzipiert werden. Die Methodik der empirischen Studie soll anhand des implementierten Simulationsmodells erprobt werden.

Die Konzeption der Studie soll unter Betrachtung eines Teststandorts eines großen deutschen Telekommunikationsunternehmens durchgeführt werden.

Supervisor: Carolin Ayasse
Earliest start: immediately
Type: Bachelor Thesis


The Heat Planning Act, which came into force on 1 January 2024, requires all municipalities to draw up a municipal heat plan. A wide range of data is required. This includes, for example, the potential for generating energy from renewable sources, data on the existing energy infrastructure or energy consumption. As real data is often not available, models are used to estimate the missing values by generating synthetic data. In this work, a data set with synthetic data on the heat consumption of buildings is to be compared with real electricity and gas consumption data obtained from the local energy supplier. The synthetic data and the real consumption data are then compared. Is the data consistent? Which deviations exist, and how can they be explained? Data science methods should be used to better understand the data and derive information about the heat generators in buildings. The comparison should be made for all buildings in a municipality near Darmstadt.
Tasks:
  • Preprocessing of real consumption data from the local energy distributor
  • Use data science methods to compare synthetic data on heat demand and real consumption data
  • Analyze the results. Is the data consistent? How can deviations be explained?

Master Thesis

Supervisor: Sina Hajikazemi
Earliest start: immediately
Type: Master Thesis


Energy Planning Models (EPMs) are essential tools for simulating and guiding decisions about multimodal energy systems in specific regions or countries. These models provide strategies to meet future demands and environmental targets, which are often used by large energy companies to negotiate with the government and other investors. However, there is a concern that these profit-driven companies could manipulate the input data of EPMs to align with their interests.
This thesis aims to address this problem by formulating it as a bilevel mathematical programming model. The upper-level problem represents the adversary who attempts to alter the input data, while the lower-level problem corresponds to the EPM. While some algorithms can effectively solve the linear upper and lower-level form of this problem, more advanced and detailed EPMs require the use of binary variables, resulting in a much more challenging Bilevel Mixed Integer Linear Programming.
The student will review available methods for solving bilevel programming problems with binary variables and tailor these methods to the aforementioned adversarial attack, taking into account the size, special form, and other specifications of the EPM. The designed algorithms will be implemented, and the results will be interpreted. This study will contribute to developing robust energy planning models that are resistant to adversarial attacks, ensuring that decision-making processes are fair and transparent.
• Familiarity with energy planning models
• Proficiency in python programming
• Experience with mathematical programming languages

Supervisor: Andreas Bott
Earliest start: immediately
Type: Master Thesis


In order to reduce CO2 emissions, district heating grids will have to utilise decentralised heat sources such as large-scale heat pumps and industrial waste heat. Operating these so called 4th generation heating grids requires the development of new algorithms. Classical approaches are computational costly, due to the complex non-linear behaviour of district heating grids. AI based algorithms on the other hand are promising to deliver reliable and fast results.

This Thesis should investigate the optimal dispatch problem, i.e. the question which plant should produce heat at which temperature, such that all grid constrains are satisfied. This problem resembles a mixed integer, nonlinear optimisation problem, which is difficult to solve. The goal of this thesis is therefore, to learn the solution of the optimal dispatch problem directly based on the current demand situation.

The true cost function and boundary conditions are (practically) not differentiable. This leads to different approaches, which can be analysed within the scope of the Thesis:

-       Use Reinforcement algorithms to train the networks on the true loss function. As evaluating the cost function is still costly, this would require pre-identifying critical points to reduce computational costs.

-       Approximate the true cost function and boundary conditions with a differentiable model (e.g. another neural network) and use this ass a loss function during the training of the dispatcher

-       Learn the optimal dispatch and the description of the district heating grid simulations by employing a physics-aware Loss function

 

Supervisor: Mario Beykirch
Earliest start: immediately
Type: Master Thesis


An established approach to obtain operation schedules for energy systems under uncertainty is to first predict the uncertain residual demands and energy prices with a probabilistic forecast model that minimizes a statistical score such as the logarithmic score or the continuous ranked probability score. Then, in a second, separate step, an optimization problem is solved to determine the optimal schedule given the demand and price forecasts. However, this method is imperfect, particularly for typically finite, small sample sizes, where errors cannot be eliminated, and the statistical score does not necessarily accurately reflect the real task-loss after the optimization step. Consequently, a forecast selected (or even trained) using this task-loss may outperform another forecast with a comparable or worse statistical score in the actual optimization task.

Approaches to use the optimization result’s performance directly as a loss function in the model training exist, e.g., stochastic end-to-end learning, but are computationally expensive because the optimization must be solved in every training iteration. Task-specific score functions, conversely, only approximate the forecast’s performance in optimization tasks. They are simpler to compute since they allow us to assess the forecast quality without solving the actual optimization task. A task-specific score function can be estimated with a machine learning model and then employed as the loss function in the forecast training.

The goal of this thesis shall be to obtain and employ a task-specific score for a simplified case of a market schedule optimization given an uncertain residual demand and a power plant without start-up costs and to quantify the task utility improvement compared to statistical scores.

Supervisor: Sina Hajikazemi
Earliest start: immediately
Type: Master Thesis


This master's thesis explores the potential of harnessing Large Language Models (LLMs) to transform user interactions with energy system models, with a focus on developing a Textual User Interface (TUI). The study investigates the interconnection between LLMs and Application Programming Interfaces (APIs) in order to facilitate a more intuitive and user-friendly means of engaging with complex energy system models. The research aims to bridge the gap between energy system model developers, decision-makers, and the public by providing a user-friendly TUI that enhances transparency and decision-making processes.

Research Objectives:
1. Examine the emerging role of LLMs in the future of user interfaces and their potential to transition from graphical to textual interfaces.
2. Develop a TUI that interfaces with Energy System Models (ESMs) via well-documented APIs.

Methodology:
1. Investigate and select appropriate tools and frameworks for integrating LLMs with APIs, drawing on existing examples and technologies, such as LangChain's Python Library for Use Cases with APIs (https://python.langchain.com/docs/use_cases/apis) and OpenAI's GPT Function Calling (https://platform.openai.com/docs/guides/gpt/function-calling).
2. Define use cases and questions that the LLM-based TUI can address to enhance understanding and decision-making in the context of energy system models.
3. Create well-documented API specifications using standard formats (e.g., OpenAPI) for a reputable energy planning model, such as "Open and Compact Model for the German Energy Transition," (available on GitHub: https://github.com/OCGModel/OCGModel).
4. Establish a functional connection between the API and LLM to develop a working prototype of the LLM-based TUI.

Significance and Contributions:
This research contributes to the evolving field of human-computer interaction by demonstrating how Large Language Models (LLMs) can be integrated with energy system models, providing an innovative approach to user interfaces in the context of energy planning and governance. The development of a TUI for ESMs enhances transparency and public engagement, while also encouraging model developers to improve precision and openness in their assumptions and methods.
By addressing these objectives, this master's thesis aims to facilitate better-informed decision-making in the energy sector, foster transparency in energy model development, and contribute to the broader discourse on the role of LLMs in future user interfaces and decision support systems.

Keywords: Large Language Models (LLMs), Textual User Interface (TUI), Energy System Models (ESMs), Application Programming Interfaces (APIs), User Experience, Transparency, Decision Support, Energy Planning, Human-Computer Interaction

Supervisor: Sina Hajikazemi
Earliest start: immediately
Type: Master Thesis


AC Optimal Power Flow is a fundamental building block for several optimization problems of electrical power systems, such as unit commitment and optimal transmission switching. It is a nonlinear and non-convex optimization problem, and solving it to optimality is an NP-hard problem. In practice, this problem is solved using linearized approximations for computational tractability. However, linear approximations can lead to suboptimal solutions or solutions that are infeasible in the original nonlinear problem. With the introduction of stochastic energy sources into the grid, the need to solve the AC-OPF more frequently has increased.

A recent approach to overcome this difficulty is to combine machine learning techniques and optimization algorithms such as Lagrangian dual methods [1]. The learning model exploits the information in the similar states of the network and the optimization algorithm satisfies the constraints of the problem.

In this thesis, the student will conduct a comprehensive review of the existing literature on these innovative techniques. Subsequently, the most promising approach will be implemented and rigorously evaluated. This will be followed by a comparative analysis comparing the performance of the chosen methodology with alternative strategies. The study aims not only to investigate the potential of using these hybrid methods to solve the AC OPF, but also to shed light on the potential of combining deep learning and nonlinear optimization to address complex challenges in modern power system optimization.

[1] Fioretto, Ferdinando, Terrence WK Mak, and Pascal Van Hentenryck. "Predicting ac optimal power flows: Combining deep learning and lagrangian dual methods." [https://arxiv.org/abs/1909.10461]

Supervisor: Tobias Gebhard
Earliest start: immediately
Type: Master Thesis


Residential electricity consumption represents a significant part of the overall electricity demand. In order to forecast the demand, accurate models that describe the consumption of households are needed. While deterministic models are useful for longer periods (e.g. days, years), consumption patterns seem to get stochastic at shorter timescales. If not just one household but the population of a city is considered, the aggregated demand is of interest. It usually stays within a narrow band because consumers are mostly independent from each other. However, crises or other events (e.g. the TV program [3]) can cause a synchronization of people, which can lead to critically high electricity use. Synchronization can be induced by external factors (like the daily cycle or weather influences) or internal effects emerging from the people (e.g. panic behavior). A general model to describe the synchronization of consumers has been developed in [1].

In the thesis, this model shall be tested/validated with measured consumption data by using statistical or machine learning methods. The dataset [2] contains the electric load of 38 households in 10 seconds resolution over more than 2 years. With this, detailed load profiles of the individual as well as the aggregated demand can be constructed, while being able to distinguish between season, weekday, and time of the day. It is to be investigated whether the synchronization model [1] can be reasonably applied to the data. As the dataset also contains weather data, the representation of known/predicted synchronization from external factors could be examined.

 

Requirements:

  • Basic Knowledge in (multivariate) statistics
  • Attendance in lecture “Machine Learning & Energy” helpful
  • Interest in working with data

[1] Tobias Gebhard; Eva Brucherseifer; Florian Steinke:  Monitoring Electricity Demand Synchronization Using Copulas.

[2] Schlemminger et. Al.: Dataset on electrical single-family house and heat pump load profiles in Germany

[3] https://en.wikipedia.org/wiki/TV_pickup

 

Supervisor: Andreas Bott
Earliest start: immediately
Type: Master Thesis


Die Wärmewende bedeutet für den Fernwärmesektor eine grundlegende Umstellung. Statt wie bisher die Wärme aus wenigen zentralen Heizkraftwerken zu verteilen gewinne dezentrale, fluktuierende (Ab-)Wärmequellen zunehmend an Bedeutung. Dieser flexible Betrieb erfordert neue Algorithmen um eine sichere Versorgung zu garantieren.


Eine besondere Herausforderung im Kontext von Wärmenetzen ist die relativ langsame Fließgeschwindigkeit des Wassers. Hierdurch kommt es zu einer erheblichen Verzögerung bis eine Betriebsentscheidung am Kraftwerk den gewünschten Effekt im Netz zeigt. Entscheidungen müssen daher vorausschauend getroffen werden und mögliche Entwicklung unsicherer Größen, beispielsweise die Entwicklung des Wärmebedarfs, vorab mitberücksichtigen.


Aus dieser Problematik lassen sich zwei Voraussetzungen für Netzmodelle ableiten.

1.      Die Modelle müssen die Zeitabhängigkeiten und Laufzeiten korrekt abbilden
2.      Die Rechenzeit muss relativ gering sein um in probabilistischen Algorithmen, z.B. in Monte Carlo Simulationen, eingebunden zu werden.


Klassische Simulationssoftware, wie Modelica, erfüllen zwar den 1. Punkt, sind aber zu rechenaufwendig für Echtzeitanwendungen.


In dieser Arbeit soll eine neuartige datengetriebene Modellierungstechnik reduzierter Ordnung vom Typ der neuronalen ODE angewandt werden, um ein genaues und schnelles Ersatzmodell der Simulationsmodelle zu generieren, das zudem einen geringen Rechenaufwand aufweist. Die Arbeit umfasst dabei einerseits die Konstruktion konventioneller Modelle in Modelica als Benchmark sowie zur Erzeugung von Trainingsdaten, andererseits auch Einbinden und Trainieren der datengetriebenen basierten Modelle in Python. Sie schlägt damit eine Brücke zwischen klassischer Modellierung und modernsten KI-basierten Ansätzen.

 

Other projects

Unfortunately, there is nothing available in the moment.

No open topics? No problem

Even if we have no open topics, but you are interested in writing a thesis at our institut dont hesitate to ask us. Just send us a mail with your latest transcript of records and a current CV.

Pro Seminar / Project Seminar

You would like to complete your Proseminar / Projectseminar with us and think that our published bachelor and master theses sound interesting? 

You have some own ideas and think they would fit in in our institute profile? 

Get in touch. We can surely sit down and talk about it.

Project seminar: Second-life Battery

In collboration with the Darmstadt-based startup ReLi you can participate in the development of a second life battery systems for homes with solar energy. Several tasks can be defined as a project seminar. This includes both programming work and eletrical / electronics hardware development. If you are interested, pease get in touch with Laura ().