A study on system identification using GA and GMDH network

In designing control systems, it is important to make exact mathematical model of the controlled object. In particular, it is difficult to obtain the mathematical model for nonlinear systems. Therefore, lots of design schemes of nonlinear models have been proposed. As one of them, the group method of data handling (GMDH) network has been proposed as a method to represent such systems. It is a kind of multilayered networks with a structure which is determined through training, and has a feature that the nonlinear dynamics are expressed as a mathematical model. However, since the mathematical model generated by the GMDH network includes several needless terms, it becomes considerably large. In this paper, a new nonlinear modeling scheme using a GMDH and two kinds of GA is proposed. According to the newly proposed scheme, necessary terms in the GMDH network are chosen by using the binary-coded GA, and system parameters are adjusted by using the real-coded GA. Finally, the newly proposed modeling scheme is numerically evaluated on a simulation example.