A neural network-based technique for structural identification of SISO systems

This paper presents a simple technique for the structural identification of single-input, single-output (SISO) dynamic systems, based on the use of a neural network. The network is trained to recognize some significant features of the process dynamics starting from a simplified representation of its unit step response, which in turn is obtained by a convenient I/O experiment. In addition, the network classifies the process with respect to a convenient set of possible model structures, which represent the most common situations arising when a process model needs to be identified for control purposes.<>