Input Dimension Determination of Linear Feedback Neural Network Applied for System Identification of Linear Systems

Abstract

In the application of a linear neural network (LNN) to linear system identification and parameter estimation, it is important to determine the input dimension of the LNN so that the identification can be performed efficiently. In the LNN for linear system identification, both the input and output data are taken as input of the LNN. The output data are delayed and are fed-back to input of the LNN. The input dimension determination is to determine the right number of past inputs should be applied to its input and the right number of past outputs should be fed-back to its input also. The sampled input and output data are used to train the LNN. The performance errors are collected during training and are used in the evaluation by Akaike’s Information Criterion to determine the input dimension. The advantage of LNN method is its simplicity and effectiveness. Satisfactory results from simulation are provided to show the effectiveness of the proposed algorithm.