Nonlinear system identification with on-line learning self organization neural networks

There exits two approaches to identify the unknown system. One is the parameter identification and the other is a functional method. The self organization neural network (SONN) derived from GMDH works as a functional method. The original SONN needs a lot of previous input and output data to build up the model. We proposed a modification on SONN such that SONN can do online learning to identify the unknown system without gathering a large amount of data at first. In this paper, we introduce the original concepts of SONN, give the details of our modified SONN, and explain the reason for developing the online learning algorithm.<>