Modelling strategies for rail network predictability, human performance and chaotic behaviour appraisal - Network Rail
Closing Date, Monday 6th June 2022
Please quote project application reference: GEO22_NW
Minor disturbances in the performance of train services can have a cascading effect on overall system performance. Being able to predict the impact of delays is essential. However, the quality of predictions is variable.
Network simulation is one way to predict network performance, but current approaches have several issues:
- Simulation efforts struggle to integrate multiple simulation models being used.
- Human performance has a crucial impact on overall network performance and predictability, but is rarely modelled or accurately accounted for in simulation efforts;
- There is a lack of detailed evaluation of the objective assessment of the quality of current predictions against actual network performance.
- The behaviour of the network, or meaningful sections of it, is challenging to model with sufficient levels of confidence as it is highly complex and potentially chaotic in nature;
- Any solution must be scalable, both to different parts of the network, and to look at network integration as a whole.
During the first year, as part of the MRes in Geospatial Data Science, this project will carry out preliminary background research to define the underpinning modelling approaches used to predict rail network performance, including identifying the parameters used, their relevance and the data sources that provide them. This will allow you to assess the predictive accuracy of current practice for forecasting network performance, including data quality and machine learning. You will also perform a preliminary Assessment of the potential of considering rail network prediction as a chaotic problem. You will apply your machine learning and visualisation skills to formulate a hypothesis on how these techniques (e.g. ML and visualisation) could transform predictability accuracy, including gaps in the parameters used. The findings from this MRes will inform the research design of the subsequent PhD programme.
Applicants should have a good (2:1 or better) undergraduate or masters (merit or better) degree in computing science, mathematics, physics, engineering or related disciplines. Programming skills (e.g. in Python or C++) are desirable but not essential as training in programming will be provided via the MRes taught component and support will be available throughout the research programme. Skills and knowledge of complex network modelling and/or railway systems is desirable but not essential. Unfortunately, this project is only available to UK home students.
This project will be co-supervised by Network Rail.
Self-Adaptive Vehicle Tracking using Transfer Learning and Bayesian Inference within Smart Cities - Defence Science and Technology Laboratory (DSTL)
Closing Date, Monday 6th June 2022
Please quote project application reference: GEO22_DSTL
Deep Neural Networks (DNNs) play a pivotal role in data-driven decision-making within smart cities, and vehicle tracking provides attractive opportunities for their application, especially in combination with data from the Internet of Things (IoTs).
This project aims to develop an innovative transfer learning framework to accommodate the heterogeneity of vehicle tracking within smart cities. This aim will be achieved via three objectives: (1) modelling spatial autocorrelation of IoT sensors using spatial tessellation techniques; (2) producing the re-training datasets using the Bayesian inference-based data generative models; (3) design and development of the new transfer learning framework based on (1) and (2).
During the first year, as part of the MRes in Geospatial Data Science, the research will focus on the investigation of state-of-the-art deep neural networks and a literature review in vehicle tracking using deep learning techniques is expected. Following this MRes work, the PhD research, both spatial analytics and Bayesian inference will combine with deep learning to design and develop the new transfer learning framework for real-time vehicle tracking using IoT sensor networks.
A successful applicant will have a minimum of a 2:1 undergraduate degree (or equivalent) in Computing, Engineering, Geography, Geomatics, Environmental Sciences, Mathematics and Statistics, or related disciplines. Unfortunately, this studentship is only available to UK students. Unfortunately, this project is only available to UK home students.
This project will be co-supervised by Defence Science and Technology Laboratory