Bachelor and Master Theses

Title: Dynamic modelling of ship behavior using recurrent neural networks
Subject:
Level: Advanced
Description: Dynamic process modeling plays an important role in the design and synthesis of ship control systems. With strong learning and parallel processing abilities, artificial neural networks have been proved as a powerful means to capture the input-output relationships of complex systems. However, feedforward neural networks for dynamic modeling require as inputs (to the network) a number of past values for each physical input and output of a system. Moreover, this number of past values has to be determined by trial-and-error to reach a desired modelling accuracy.
This thesis will explore the application of a new type of networks: recurrent neural networks, for ship modelling. Compared with traditional forward network, a recurrent network can memorize previous information in its hidden units such that no past values are need any more as inputs to the network. The structure of the network makes it very suitable to depict and learn the dynamic characteristics of multi-input and multi-output systems.
The thesis project will start with a literature study about the state-of-the art, followed by comparative investigation of various learning algorithms for recurrent networks. Finally, a new software tool for ship modeling will be developed based on customization of an existing learning algorithm or aggregation of multiple learning methods to best fit the underlying application domain. The project will be jointly supervised by MDH and Q-TAGG.
A student needs to have solid knowledge in math and machine learning, as well as good programming skills for conducting this project. Q-TAGG will follow a procedure of qualification testing to select a qualified candidate.

Please contact George Fodor (george.fodor@qtagg.com) if you are interested.
Company: Q-TAGG, kontaktperson: Gerge Fodor
Proposed: 2018-06-04
Prerequisites: Solid knowledge in math and machine learning, good programming skills
IDT supervisors:
Ning Xiong
ning.xiong@mdh.se, +46-21-151716
     
Examinator:


Rapport och bilagor

Size

Senaste uppdatering


  • Mälardalen University |
  • Box 883 |
  • 721 23 Västerås/Eskilstuna |
  • 021-101300, 016-153600 |
  • webmaster |
  • Latest update: 2018.05.24