Skip to main content

Neural Networks for Forecasting and Diagnosis Applications

Objective

At the heart of the project is the development of a unified description language (i.e. a unified framework). A number of applications will be thoroughly analysed to provide input for the development of the framework. With the help of the unified language, topological and architectural heuristics will be formulated.
The project aims to apply neural networks to forecasting and diagnosis problems.
At the heart of the project is the development of a unified description language (ie a unified framework). Its goal is to classify and describe different well studied paradigms, as well as novel ones, within the domain considered. Such a language is crucial in order to get a consistent framework for applications and a deeper understanding of neural networks. A number of existing applications will be thoroughly analyzed to provide input for the development of the framework. With the help of the unified language, topological and architectural heuristics will be formulated.

3 existing and 1 novel application are being considered and are progressing. The existing applications are phoneme recognition, integration of heart scan data, and the forecasting of water demand and consumption.
1 novel application is examined in particular: the problem of disturbance analysis in electrical networks. This is also intended to help in refining the prototypes of the other applications.
3 major architectural aspects of neural networks are currently investigated with promising results:
architectures that can handle temporal sequences of input patterns;
the preprogramming of knowledge into networks;
architectures that can deal with a changing environment.
The unified description language for describing neural network paradigms has been developed and successfully used within the project for the development of customised applications. The methodology has also been used for benchmarking the major neural network paradigms (i.e. classification with regard to specific applications).

Promising and innovative results have been obtained in the following areas of research: (1) architectures that can handle temporal sequences of input patterns; (2) the pre-programming of knowledge into networks; (3) architectures that can deal with changing environments.

Prototype systems have been developed for the following applications: (a) forecasting of water demand and consumption; (b) heart-scan data interpretation, (c) electrical network fault detection, (d) rainfall prediction; (e) sales prediction; and (f) traffic prediction.

The NEUFODI-results are available: (i) for end-users, to be integrated in user-specific applications: customised versions of the existing prototype systems have to be developed; (ii) for suppliers of software systems, to be integrated in analysis, management, or planning software products, etc.

Coordinator

THE BABBAGE INSTITUTE FOR KNOWLEDGE AND INFORMATION TECHNOLOGY
Address
Plateaustraat 22
9000 Gent
Belgium

Participants (6)

AUSTRIAN RES.INST.OF ARTIFICIAL INTEL.
Austria
Address
Schottengasse, 3
1010 Wien
COPENHAGEN TELEPHONE COMPANY
Denmark
Address
Noerregade, 21
1199 Copenhagen
ELORDUY SANCHO Y CIA
Spain
Address
Av Ejercito, 11-4
48014 Bilbao
LABORATORIO DE ENSAYOS E INVESTIGACIONES INDUSTRIALES
Spain
Address
Cuesta De Olabeaga, 16, 1234
48013 Bilbao
RHEA
France
Address
11 Vieux Pont
92000 Nanterre
SOCIETE LYONNAISE DES EAUX
France
Address
42 Rue De Lisbonne
75008 Paris