Academic site of Carl Ellis

/content/projects/lit_review/teeter1998

Application of functional link neural network to HVAC thermal dynamic system identification - J. Teeter, M.Y. Chow

@inproceedings{dada2008,
abstract = {Several brand owners are calculating the carbon footprint of sample products and intend to make the information available to their consumers as a label on the product. A physical label on the items or on the retail shelf will not be flexible enough to show the carbon footprint because of the dynamic nature of carbon emissions and the potential difference in footprint between instances of the same product. In this demonstration, we show an alternative to a static physical label, namely an NFC-enabled mobile phone that displays the carbon footprint of tagged products. In addition to demonstrating the dynamic nature of carbon footprints, our prototype shows how consumers can be empowered with knowledge about the products they buy.},
address = {Sydney, Australia},
annote = {demo abstract. main idea is that an NFC reader phone displays the current carbon footprint for a given product. rationale is that carbon footprints are in flux, and can vary even for the same product, if originating from different places.},
author = {Dada, Ali and von Reischach, Felix and Staake, Thorsten},
booktitle = {Adjunct Proceedings of Pervasive Computing 2008: Sixth International Conference on Pervasive Computing},
editor = {Mayrhofer, R and Quigley, A and Kay, J and Kortuem, G and Ardon, S and Rukzio, E and {Vande Moere}, A and Adowd, G and Suginuma, K},
keywords = { feedback, non-domestic, poo, zfile-techmediated,zfile-techmediated},
pages = {119--121},
title = {{Displaying Dynamic Carbon Footprints of Products on Mobile Phones}},
volume = {Advances i},
year = {2008}
}
        

Key Points

  • Wide sample input required to get a reasonable NN

Using an artificial neural network to rimulate an RC-network (or plant) and provides a learning algorithm.

Forward system identification = process driven modelling.

Inverse system identification = data driven modelling.

For a NN to be effective, sample input must cover a wide range of inputs. It can only model what it has been trained to do.