Neural Networks implemented as feedforward controllers

OData support
Supervisor:
Dr. Srp Ágoston Mihály
Department of Control Engineering and Information Technology

In my thesis work, I attempted to shed some light on the topic of artificial neural networks, and show their practical uses. In the last few decades neural networks have advanced in a rapid rate, both their network theory and the systems that are usable to implement them. I also found it important to highlight their possible flaws, and not just their very nummerous benefits. For the demonstration I solved a simple problem. This particular problem was the simulation of a thermostat, which I chose so that I may be able to compare the effectivenes of the neural network to the classical solutions. The model has a lower complexity and as such the size of the neural network and the training time is modest, also the problem is more transparent.

To acquire the necesseary data, I used MATLAB. My neural network was constructed using a simple function followed by theoretical modell.(ezt nézd meg) I studied the effect of the numbers of input neurons and hidden neurons on the learning skills and punctuation of the system. Following a similar mindset I also studied the effects of changing learning rate.

Downloads

Please sign in to download the files of this thesis.