Projects / Programmes
Integration of the wavelet multiresolution analysis into soft-computing techniques
Code |
Science |
Field |
Subfield |
2.07.07 |
Engineering sciences and technologies |
Computer science and informatics |
Intelligent systems - software |
Code |
Science |
Field |
P170 |
Natural sciences and mathematics |
Computer science, numerical analysis, systems, control |
wavelets, neural networks, soft computing, process control
Researchers (1)
no. |
Code |
Name and surname |
Research area |
Role |
Period |
No. of publicationsNo. of publications |
1. |
16109 |
PhD Uroš Lotrič |
Computer science and informatics |
Head |
2002 - 2004 |
175 |
Organisations (1)
Abstract
The mathematical formulation of smoothing with wavelet analysis makes it possible to integrate it with neural networks into a novel uniform model. So far, studies have shown that using wavelet multiresolution analysis and multilayered feed-forward neural networks time series prediction can be improved without data pre-processing. The main advantage of the new model is its ability to set the smoothing level dynamically. The automatic noise reduction is of particular importance in on-line modeling of real processes where noise is inherently present in the measurements and the process dynamics does not allow for data pre-processing. With the present study, we aim to expand this approach to more specialized neural networks and other techniques of soft computing. For modeling of stationary processes recurrent neural networks are used since they have the ability of memorizing the process dynamics. These networks will be integrated with noise reduction techniques to gain on their robustness. When non-stationary processes are considered, the problem of continuous parameter adaptation emerges. For both, feed-forward and recurrent neural networks, this adaptation can only be performed by real time parameter setting algorithms. Such algorithms will be expanded for parameters of wavelet based smoothing. The model's ability to adapt to different noise levels can also be used in pattern recognition systems, especially if it is performed in real time. Therefore, the model will also be expanded to cover two-dimensional problems.