System Identification Based on Generalized ADALINE Neural Network

摘要

System identification of linear time-varying systems consists of estimation of system parameters that change with time. In this paper, we present an online identification method for such systems based on a generalized ADAptive LINear Element (ADALINE) neural network. It is well known that ADALINE is slow in convergence and, hence, not appropriate for online application and identification of time varying systems. ADALINE is generalized such that the input now consists of a Tapped Delay Line of the system input signal and a Tapped Delay Line of the system output feedback. Two techniques are proposed to speed up the convergence of learning, thus increase the capability of tracking time varying system parameters. One idea is to introduce a momentum term to the weight adjustment during the convergence period and the learning curve is smoothed by turning off the momentum once the error is within a given small number ?C epsilon. The other technique is to train the generalized ADALINE network multiple epochs with data from a sliding window of the system??s input output data. Simulation results show that the proposed method provides a much faster convergence speed and better tracking of time varying parameters. The low computational complexity makes this method suitable for online system identification and real time adaptive control applications