Analyza chaotickych biosignalu pomoci neuronovych siti.

There are two main subjects covered in this work. The first one is analysis chaotic time series by means of neural networks. Neural networks are applied for chaotic time series classification, defect detection, chaotic system modeling - generation of chaos, chaotic attractor reconstruction, detection of qualitative dynamic changes and their transitions. Neural networks are suitable in cases where standard methods do not converge or when they are not stable and also in cases where not enough experimental data is available. Designed methods were tested on simulated chaotic systems and the results were compared with standard methods of deterministic chaos theories. The work also mentions the basic limitation of neural network synthesis - the absence of criteria quality of the model. The second subject of the work is the application of neural networks on qualitative bio-signal analysis. The focus was put on analysis chaotic part of the cardio-signal what is the main characteristic of cardiovascular system in terms of non-linear dynamics. In the application part of the work is reviewed the use of global non-linear prediction, represented by neuron model, for classification and diagnostics of chaotic courses. Appropriate choice of model generalization may influence on the amount of generalized classes of similar chaotic processes and thus create the base of recognition rule. Application of neural networks for chaotic process multiatractor dynamics decomposition is mentioned in final part of the work.Available from STL, Prague, CZ / NTK - National Technical LibrarySIGLECZCzech Republi