Wolters, A.P. (1999) Design and performance aspects of a neuro-fuzzy system. Master's Thesis / Essay, Computing Science.
|
Text
Infor_Ma_1999_APWolters.CV.pdf - Published Version Download (2MB) | Preview |
Abstract
In the design process of a fuzzy system it can be difficult to find optimal membership functions. After the initializing process the membership functions must be tuned. A neuro-fuzzy system can help in that process. A neuro-fuzzy system is a combination of a fuzzy system with learning capabilities of a neural network. In this proposed neuro-fuzzy system a fuzzy system is mapped on a neural network architecture. The neuro-fuzzy system will be initialised with membership functions and fuzzy rules extracted from the explicit knowledge of an expert. In the learning phase implicit knowledge, recorded from a process, will be used to find optimal membership functions through the adaptation of the position of the membership functions. At this moment there are different neuro-fuzzy architectures like ANFIS, FALCON, NEFCON, NEFCLASS and GARIC. All architectures have their own learning algorithms/paradigms, membership function types and inference methods. Two types of experiments are presented with the FALCON architecture with linear membership functions. The first types of experiments are function approximation experiments with crisp inputs and a crisp output calculated with the Center Of Maximum (COM) defuzzification method. The second types of experiments are classification experiments with crisp inputs and fuzzy outputs. Topic of research in these experiments is the performance of the neuro-fuzzy system, which will be compared with the performance of a multilayer perceptron, and design aspects of a neuro-fuzzy system. Investigation points are; single and multiple rule blocks, the number of fuzzy rules, learning rates and the number of epochs.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Degree programme: | Computing Science |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 15 Feb 2018 07:29 |
| Last Modified: | 15 Feb 2018 07:29 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/8821 |
Actions (login required)
 |
View Item |