Control of Unstable Linear and Nonstationary Systems using LAMSTAR Neural Networks

D. Graupe and M. Smollack (USA)


Neural-network control, LAMSTAR NN, unstable, nonlinear, time-variant, backpropagation, controllability, initialization


The paper discusses the employment of the LAMSTAR neural network (NN) as an intelligent controller that requires no identifier, to control time-varying and nonlinear (NL), possibly unstable systems of unknown and parameters. In contrast to other neural networks, such as Backpropagation (BP), the LAMSTAR NN requires no search for initialization of weights. Search for initialization of weights, even when using fast BP initialization algorithms, prevents controlling systems of unknown parameters that that may be unstable, since by the time an initialization seed is found the system may have already blown up. Similarly, parameter identification, for linear or nonlinear systems of a-priori unknown parameters, requires considerable time to converge before parameters can be used for control purposes, and by that time the system may again blow up. The present LAMSTAR-based NN design, requires no parameter identification per-se, since implicit identification is integrated into the NN-control. Computed results are presented for the LAMSTAR-NN control of SISO unstable time-varying and nonlinear third and fourth order systems whose parameters and orders are unknown to the NN.

Important Links:

Go Back