Seminar Papers and Theses
Guidelines
Master and Bachelor
- Registration: This is done with a form that the student receives from the Examination Office after account clarification. The student submits the form to the supervisor for discussion of the topic. Topic is agreed, supervisor signs. The student informs the chair's office and registers his or her work at the examination office within one week.
-
Processing Time: For bachelor's theses, this period is 16/11 weeks, and for master's theses, it's 20 weeks.
- Extent: The written work should have a length of approx. 40 pages for Bachelor's theses. For master's theses, it is correspondingly approx. 50 pages.
- Arrangement: For the content and formal design of the thesis, you can follow the recommendations of the chair . recommendations of the chair or the guidelines of the Faculty of Economics, Department of Economics. Please note that Master's theses can be written in English as well as in German, whereby English is generally preferred.
- Interviews and Data: If an (online) survey is conducted as part of the thesis, a link to this survey must be published on the chair's website. Please send this link to Ms. Undeutsch. The data from the survey will also be posted on the webseite after completion of the thesis (password protected). Furthermore, it is possible to use data from former surveys for theses. Please contact a member of the chair.
- Interim presentation: In the middle of the processing time, the current status of the work and the still open work steps are to be presented in a 20-minute interim presentation. The date has to be announced to Ms. Undeutsch in time and should be within the framework of the dates for the presentation of seminar papers on a Wednesday starting from 9:20 am.
- Submission: The written work is to be submitted in double bound form (preferably printed on both sides) with all digital sources on CD-ROM or another suitable data carrier. Digital sources include the Word or LaTeX document of the written work, all self-created Excel worksheets, all data files (as appropriate as .dta, .r, .sav, .csv, .txt, .etc.), scripts (do-files, R-script, etc.), figures, etc., as well as pdf printouts of all internet sources. In addition, the written work must be uploaded to the publication server Qucosa hochzuladen and the link sent to Ms. Undeutsch. If there is a blocking notice for the work, it does not need to be published on Qucosa.
- Defense: In the case of Master's theses, the results of the work are to be presented in a final presentation approximately four weeks after submission of the written work: time budget for presentation and discussion together maximum 40 minutes. The date has to be announced to Ms. Undeutsch in time and should be within the dates for the presentation of seminar papers and final theses on a Wednesday starting from 9:20 am. The presentation must be uploaded to Opal no later than the evening before the appointment and will then be available on our chair notebook. Bachelor theses will not be defended.
Seminar (Research)
In the research seminar of the master's program, students can deal in depth with a topic from transport econometrics and statistics or transport modeling and simulation and practice writing, presenting and defending a research paper. In-depth knowledge of quantitative methods is required to take the seminar. To do so, first choose a topic from the topic catalog and contact the supervisor of the topic. You are welcome to suggest your own topics. Please ask a member of the department of your choice. Consultation and submission dates will be arranged individually with the respective supervisor.
Please note: There is no introductory event for the research seminar! The course of the seminar will be arranged individually with the respective supervisor. This allows you and us a lot of flexibility.
- Registration: If the registration takes place in the semester provided for in the study plan, then online, otherwise report to the secretary's office.
- Processing time: current semester
- Length: approx. 20 pages
- Design: For the content and formal design of the paper, you can follow the recommendations of the chair or the guidelines of the Faculty of Economics, Department of Economics.
- Surveys and Data: If an (online) survey is conducted as part of the research paper, a link to this survey must be published on the chair's website. For this purpose, please send this link to Ms. Undeutsch. The data from the survey will also be posted on the website after completion of the work (password protected). Furthermore, it is possible to use data from former surveys for research work. For this purpose, please contact a member of the chair.
- Submission: one copy of the written paper in bound form (preferably printed on both sides) with all digital sources on a suitable data carrier (e.g. USB stick).
- Defense: Approximately two weeks after submission of the written work, the results of the work are to be presented in a final presentation on a Wednesday starting at 9:20 am. Time budget for presentation and discussion together maximum 40 minutes. The date has to be announced to Ms. Undeutsch in time and should be within the dates for the presentation of seminar papers and final papers on a Wednesday starting at 9:20 am. The presentation has to be uploaded to Opal the evening before the appointment at the latest and will then be available on our chair notebook.
Templates
The following templates are intended only as a recommendation. You are welcome to modify them or ignore them altogether. The decisive factor is that the form of the paper is based on current scientific standards.
Proposed Topics
General Information
- Suggestions for topics of your own are welcome.
- We are open to joint support with external partners.
- The presented topics can (with a few exceptions) be worked on in diploma, master, bachelor as well as seminar theses. The adaptation of the focus of the work to the required scope is to be discussed with the respective supervisor in the specific case.
- The topics can generally also be worked on in groups.
Statistics and/or Machine Learning
| Topic (contact person) | Short description |
|---|---|
| Autonomous robot car with Deep Reinforcement Learning (Dianzhao Li) | Are you interested in working with robotic cars and Deep Reinforcement Learning approaches? With multiple sensors such as the camera and Lidar sensor as input, the autonomous car should be able to perform different driving behaviors. The robot cars and current working progress can be found here. |
|
Applicability of statistical methods for evaluating the quality of interaction between overhead contact line and pantograph (Luise Wottke / Prof. Ostap Okhrin) |
The aim is the identification of significant influencing variables on the driving quality by means of statistical tests (e.g. anova or correlation) and the analysis of methods for forecasting possibilities (regression) of the driving quality. The core of the task is the identification of statistical test possibilities, as well as their implementation and application to available data sets. Guidelines: Knowledge in the field of overhead contact line systems is not mandatory. The processing takes place in coordination with the Chair of Electrical Railways. |
| Temperature modeling (Prof. Dr. Ostap Okhrin) | Model temperature in time and space. Methods: Regression, Shape Invariant Model, Spatial Models. |
| Transport-based footprint (Prof. Dr. Ostap Okhrin) |
Investigate how the ecological footprint develops in different countries. Methods: Claster analysis, factor analysis and/or time series. |
| Analysis of the spread COVID-19 (Prof. Dr. Ostap Okhrin) | There are methods that model the spread of a disease. One should investigate whether these are suitable for COVID-19, or investigate which external factors have an influence (e.g. weather conditions, lockdown). Methods: (non) linear regression, time series. |
| Adjustment tests as estimation method (Prof. Dr. Ostap Okhrin) | Investigate whether maximizing the p-value of an adjustment test can be used as an estimation method. Compare with ML and LQ estimation methods. This requires simulation study as well as empirical application. |
| Autonomous Driving with Deep Reinforcement Learning (Dianzhao Li) | Deep Reinforcement Learning approaches are more and more popular these days. With our autonomous driving car, you need to use camera images as input of a Deep Reinforcement Learning approach and control the car drive in different environments. |
|
Maximizing Ensemble Diversity in Supervised Learning (Martin Waltz) |
Ensemble-methods for supervised learning tasks train several approximators for a common target function. A potential limitation of the procedure is the collapse of the ensemble into a single highly similar estimate despite being trained on different datasets. Objective is to identify and apply approaches which maximize the ensemble diversity. |
| t-SNE: Dimension Reduction, Classification and Visualization (Martin Waltz) | t-Distributed Stochastic Neighbor Embedding is a powerful technique for dimension reduction and classification tasks. Task is to thoroughly analyze the core methodology, find existing modifications and perform an overall comparison on different benchmark problems. |
| Explainability in Deep Reinforcement Learning (Martin Waltz) | Investigate and apply current methodological approaches to explainability in Deep Learning, especially with focus on Deep Reinforcement Learning. Can we generate insights into the learned behaviour of agents by analyzing the trained neural network? |
| Machine Learning in the field of autonomous shipping (Fabian Hart) | Hyperparameter optimization of machine learning algorithms for decision making of overtaking maneuvers in autonomous vessel traffic. The goal is to find and test a suitable hyperparameter configuration. |
| Machine Learning in the field of autonomous shipping (Fabian Hart) | Architecture and design of neural networks for decision making of overtaking maneuvers in autonomous shipping. The goal is to find and test a suitable architecture. |
| Reinforcement Learning in the field of autonomous shipping (Fabian Hart) | Optimization of reinforcement learning algorithms for the generation of ship trajectories. The goal is to generate collision-free trajectories through independent learning. |
| How meaningful is the p-value? (Dr. Martin Treiber) | The p-value indicates the probability of obtaining more extreme values than the current measurement in the presence of the null hypothesis H0. Conversely, one wants to know the probability of H0 in the presence of the measurement. With the help of Bayes' theorem, this is to be analyzed for different distributions and null hypotheses. |
| Covid-19 infection statistics(Dr. Martin Treiber) | The number of reported cases depends not only on the number of infected persons, but also on the frequency of testing, the selection of persons to be tested, and the sensitivity and specificity (or 1st and 2nd type errors) of the test. This will be analyzed descriptively as well as with an epidemic model (corona-simulation.de). |
| Modeling of Covid-19 infections in Excel or R. (Dr. Martin Treiber) | Depending on the level of detail/type of work (seminar to master), you can simulate the spread of infection macroscopically as a SIR model (Susceptible-Infected-Recovered/Removed), as a SEIR model (additionally assuming a non-infective time after infection), as a SEIR model with memory (as in corona-simulation.de) or microscopically with a particle model. |
| Time costs due to congestion: internal or external? (Dr. Martin Treiber) | In most studies, time costs due to delays represent the lion's share of congestion costs, which are sometimes very high. Different approaches as well as the nature of the costs (internal or external) are to be discussed |
| CO2 emissions now and in 2050 (Dr. Martin Treiber) |
Using a model-based forecast, the development of CO2 emissions from the transport sector in Germany or other regions is examined under various scenarios. |
| Life cycle assessment of conventional and electric vehicles (Dr. Martin Treiber) | In addition to operation, pollutants are also produced during the manufacture and disposal/recycling of the various automotive components. This is being investigated using the methods of life-cycle assessment for latest-generation vehicles. |
| Renewable energies: Statistics of energy flows from a demand perspective(Dr. Martin Treiber) | How can economic incentives be used to adjust the temporal demand for energy to the fluctuating supply? One lever, for example, is economic incentives (variable electricity prices). More details on request. |
| Energy consumption of electric vehicles (Dr. Martin Treiber) | Although e-vehicles do not consume fuel, they do consume energy and thus also emit CO2 indirectly - depending on the energy mix. This is analyzed/simulated using regression-based and physics-based consumption models for different driving patterns. |
| Application of econometric methods in vehicle consumption modeling (Dr. Martin Treiber) | Depending on the objective and the level of detail of the consumption modeling, several statistical tools are used, such as factor analysis, regression, lookup tables or simulation (see e.g. chap. 20 of the textbook "Traffic Flow Simulation). It is possible to work on different problems with or without simulation. |
| Renewable energies: supply-side statistics of energy flows (Dr. Martin Treiber) | How does the energy mix influence the temporal fluctuations of electricity supply and how can supply be adjusted to demand? A suggestion may be given e.g. by this diese Simulation . More details on request. |
| Renewable Energies: Energy Storage Statistics(Dr. Martin Treiber) | With the increase in the share of fluctuating renewable energies, the management of the electricity grid is becoming more and more more demanding. Energy storage systems are therefore indispensable. Their interaction will be analyzed statistically. More details on request. |
Traffic Dynamics
| Topic (contact person) | Short Description |
| Weighting of the pre-crash matrix of traffic accident research (Ostap Okhrin) | The Pre-Crash Matrix (PCM) is a specified format that can be used to describe the phase of a traffic accident before the first collision (the so-called pre-crash phase). It can be used to represent and evaluate the parties involved, their dynamics, and the environment. The challenge is to develop a procedure with which the PCM can be weighted to the German accident event. |
| Time is money - economic impact of a fuel-efficient driving style (Dr. Martin Treiber) | If you drive faster, you save time but spend more on fuel. To which limiting time value does a certain speed correspond? |
|
Analysis of vehicle trajectories ( Dr. Martin Treiber) |
Trajectories, i.e. space-time data of vehicles, are the "gold standard" in the development of traffic flow models. However, they contain a large number of errors. Statistical methods are used to detect and, if possible, correct these errors in the data sets provided. |
| How well do vehicle successor models replicate "real" driving behavior? (Dr. Martin Treiber) | Vehicle following models mathematically represent the acceleration and braking behavior of human drivers and automated vehicles. Different models are to be adapted ("calibrated") to data provided by human drivers and then the predictive power of the models is to be tested ("validated").old after? |
|
To cross or not to cross? That is the question! (Dr. Martin Treiber) |
Empirical data collection of pedestrian crossings at a selected unsignalized crossing. Subsequent analysis of the data using a discrete choice theory model. |
| Collaboration on the interactive simulator traffic-simulation.de (Dr. Martin Treiber) |
Here, extensions of the JavaScript simulator traffic-simulation.de to various intersection or traffic circle scenarios and subsequent simulations are conceivable. For example, one could investigate when a traffic circle makes sense from a traffic perspective. |
| Simulation with or collaboration on the traffic simulator MovSim (Dr. Martin Treiber) |
Depending on the type of work (seminar to master), you can use the open-source traffic simulator MovSim to test a variety of traffic flow models for plausibility in standard situations or implement (in Java) extensions such as traffic circles without programming yourself. |
| To follow the Navi detour recommendations or not? (Dr. Martin Treiber) |
Modern navigation (e.g. Google Maps or TomTom) takes into account the current traffic and suggests, for example, a detour around a traffic jam. When does this result in "routing oscillations", i.e. driving into a freshly developing traffic jam while the original route becomes free? This can be investigated either with an offline version of traffic-simulation.de (no programming required) or with a macroscopic model. |
| External work at TomTom (Dr. Martin Treiber) | Understanding traffic through analysis of large-scale historical floating car data in TomTom's Data Center. This is really about Big Data. For more information contact Dr. Arne Kesting |
Registered Theses
Current
| Name | Type | Title | Supervisor | Interim Presentation | Submission Deadline |
|---|---|---|---|---|---|
| Daniel Höfler | Master | Risk Measures in Transportation: A Long-Term Risk Exposure of Motorized Individual Traffic | Martin Treiber | 07/28/2021 | 08/02/2021 |
| Paul Ziebarth | Diploma | Development of a Reinforcement Learning Agent for Implementing a Ship Following Model | Fabian Hart | 05/26/2021 | 08/02/2021 |
| Fabian Hinze | Master | Anomaly Detection in Time Series with Multiple Seasonal Components | Iryna Okhrin | 05/05/2021 | 07/21/2021 |
Completed Theses
The completed works can be found here.
Presentation Dates
Information
- Venue: Zoom
- Date coordination: Ms. Undeutsch
- Regular attendance of all seminarists is desired.
- The presentation has to be uploaded to Opal the evening before the appointment at the latest and will then be available on our chair notebook.
- The current dates or changes of dates can now also be read as RSS feed. By clicking on the corresponding icon in the bottom left corner of the status bar, the service will be provided and weekly updates will be shown directly.
Dates
| Date | Time | Speaker | Topic |
|---|---|---|---|
|
SS 2021 |
|||
| 28.07.21 | 10:00 | Daniel Höfler | Zw.MA: Risk measures in transport: A long-term risk exposure of motorised private transport |
| 28.07.21 | 9:20 | Fabian Hinze | MA: Anomaly detection in time series with multiple seasonal components |
| 26.05.21 | 9:20 | Paul Ziebarth | Zw.DA: Development of a reinforcement learning agent for the realisation of a ship following model |
| 05.05.21 |
9:20 |
Fabian Hinze | Zw.MA: Anomaly detection in time series with multiple seasonal components |
Lectures of the last semesters Archiv