Developing real-time wave-net models for non-linear time-varying experimental processes

This paper investigates procedures for on-line learning and improvement with wave-nets utilizing a stream of data and then applies these methods to an experimental thermal process. Wave-nets are wavelet based neural networks with localized and hierarchical multi-resolution learning. The multi-resolution framework of wave-nets allows non-linear modeling of any complicated systems. The recently developed on-line features to wave-net learning have enhanced their capability in learning and adaptation of any non-linear time varying systems (with low dimensions). A real experimental time varying thermal process is modeled and the on-line learning was implemented to show the applicability of these algorithms. The platform used for the real-time implementation is MATLAB/Real-Time-Toolbox with a DAQ interface. The results show the effectiveness of the methods.