André P. Schoorl, Nicolaos P. Kourounakis, Caedmon D. A. Somers, Nikitas J. Dimopoulos,
Department of Electrical and Computer Engineering, University of Victoria |
P.O. Box 3055, Victoria, B.C., Canada V8W 3P6 |
{aschoorl, nkouroun, csomers, nikitas}@ece.uvic.ca |
Abstract
In this work, large-scale fault detection in cable television amplifier networks is considered. The status monitoring information measured in these networks is affected by noise, temperature, and other external elements making closed-form solutions impractical. Existing techniques utilizing recurrent neural networks have been developed to analyze and reliably detect faults in such a system. However, due to changes in amplifier behaviour such neural networks often need to be retrained, introducing significant computational expense when used on a large data set. To solve this problem a technique is needed to quickly scan through data for such a system and only pass questionable amplifiers on to more time consuming and rigorous routines. This work presents a means by which statistics and feed-forward neural networks can be used to categorize the forward pilot signal to make implementation of a large-scale model-based fault detection system feasible.