MIMO system identification with extended MADALINE neural network trained by Levenberg-Marquardt Method

Presented in this paper is an extended version of the Multi-ADAptive LINear Element (MADALINE) neural network, termed EMADALINE, for On-line System identification of Multi-Input Multi-Output (MIMO) linear time-varying (LTV) systems Trained by Levenberg-Marquardt Method. A sliding window on the data set is used in the learning algorithm for the purpose of improving convergence speed during training and thus better tracking system parameters. Based on the input output polynomial model, which can be easily transformed into the row canonical state space model, Tapped delay lines are introduced, so the EMADALINE becomes recurrent in nature and thus is suitable for parameter estimation of such systems. The EMADALINE can then be setup under the assumption that the system structure is known in advance. The estimated parameters are obtained as the weights of trained individual neurons of the EMADALINE. The method is implemented in MATLAB and simulation study was then performed on a few well known examples. Simulation results show that the algorithms offer impressive results on convergence speed improvement. This work is based on author´s previous work on MIMO systems´ identification (Wenle Zhang, 2010).