engleski

A Neuro-predictive Controller for Industrial Processes

This paper presents an application of a neuro-predictive controller for the control of non-linear, time invariant industrial processes. For modelling the process dynamics a MLP neural network is used. The neural network training is performed off-line and the model parameters obtained are constant during the control. The predictive controller based on such a model is realised as a non-linear optimiser, using a Levenberg-Marquardt unconstrained optimisation procedure. A new approach is proposed, which enables the usage of unconstrained optimisation procedures, even in case when hard constraints on the control signal are present. To demonstrate the applicability and satisfactory performance of the proposed controller, an application on a laboratory pilot pump plant is given.

predictive control; nonlinear system identification; MLP neural networks; NARX; nonlinear optimisation; Levenberg-Marquardt

nije evidentirano