Application of Multi-Layer Artificial Neural NetworkBased Adaptive Learning Algorithm for UnknownMotor/Load Dynamics

This paper tackles the problem of the speed controlof a dc motor in a very general sense. Use is made of the power offeedforward artificial neural networks to capture and emulatedetailed nonlinear mappings, in order to implement a fullnonlinear control law. The random training for the neuralnetworks is accomplished online, which enables better absorptionof system uncertainties into the neural controller. An adaptivelearning algorithm, which attempts to keep the learning rate aslarge as possible while maintaining the stability of the learningprocess is proposed. This simplifies the learning algorithm interms of computation time, which is of special importance in realtime implementation. The effectiveness of the control topologieswith the proposed adaptive learning algorithm is demonstrated.It is found that the proposed adaptive learning mechanismaccelerates training speed Promising results have also beenobserved when the neural controller is trained in an environmentcontaminated with noise

کلیدواژه ها:

Learning rate adaptation ، neural networks ، nonlinearities ، online identification and control