Deep Learning for Time-Series Analysis

OData support
Supervisor:
Gincsainé Dr. Szádeczky-Kardoss Emese
Department of Control Engineering and Information Technology

We meet with time series problem in all areas of life. Engineering areas are not expection, where the most of measured data are time based. In my thesis I would like to demonstrate, that the neural networks can be effective and good solutions for this problems. I show firstly the general theorethical of neural networks, than I choose the software environment. In the second part of my thesis I introduct three problem, where I show the neural netwoks in practise. Firstly I present a classification problem, what I solve with deep neural network. In this section I show, how can we use the convolution neural networks on audio signals. The second task is the prediction of pollution in Bejing. I write first about dynamic neural networks, then I show my sollution and I talk the setup of the deep neural network (LSTM). In the last task I present an anomaly detection neural network on ECG signals, and I briefly present what other anomaly detections algorithm still exits, then I show my solution capabilities. Finally I sum my experiences and results.

Downloads

Please sign in to download the files of this thesis.