• TUCS Publication Series
• How to Publish in TUCS Series
• Publication Search
• Publication Graph
• Publication Input
• Publication Input Guide
• JuFo Browser

Kernel-based Methods for Forecasting of Waste Heat Recovery in Ships

Mikael Manngård, Amir H. Shirdel, Jari M. Böling, Hannu T. Toivonen, Kernel-based Methods for Forecasting of Waste Heat Recovery in Ships. In: Proceedings of Automaatio XXI, The Industrial Revolution of Internet – From Intelligent Devices to Networked Intelligence, SAS Julkaisusarja, 1–6, 2015.

Abstract:

Utilization of waste heat is of great interest to the ship industry, and the need for effective databased
learning methods is steadily increasing. Kernel-based models can easily be fitted to data, but
models tend to become large as the number of data points grows. Model complexity is hence a
problem, and finding sparse solutions is of great importance. We evaluate two identification methods
which promote sparsity in the models. A regularization based approach, using `2-regularization with
iterative reweighting is compared to support vector regression. Both methods are trained and validated
on simulated data from a marine diesel-engine fresh-water cooling system, for modelling the
recovered waste heat energy based on the engine load. No assumptions on the underlying structure
of the system are needed, which make the methods flexible and easy to implement in practice.

BibTeX entry:

@INPROCEEDINGS{aconv25176,
  title = {Kernel-based Methods for Forecasting of Waste Heat Recovery in Ships},
  booktitle = {Proceedings of Automaatio XXI, The Industrial Revolution of Internet – From Intelligent Devices to Networked Intelligence},
  author = {Manngård, Mikael and Shirdel, Amir H. and Böling, Jari M. and Toivonen, Hannu T.},
  series = {SAS Julkaisusarja},
  pages = {1–6},
  year = {2015},
  keywords = {Kernel regression, sparse optimization, iterative reweighting, support vector regression.},
  ISSN = {1455-6502},
}

Belongs to TUCS Research Unit(s): Other

Edit publication