On Using Gated Recurrent Units for Nonlinear System Identification
Alexander Rehmer, Andreas Kroll · 2019
During recent years Deep Learning (DL) methods facilitated impressive progress on various fields of research: Deep Convolutional Neural Networks (CNN) enabled object classification with to this day unmatched precision, while state of the art results in speech recognition and natural language processing (NLP) were achieved via gated units such as the LSTM. Although recurrent neural network architectures are long established in the field of system identification as a realization of an internal dynamics approach [1] [2], little research has yet been dedicated towards gated units. The purpose of this paper is to evaluate the architectures of recurrent gated units from the viewpoint of system identification and test their performance on a nonlinear system identification task.