On-Line Control of Dynamic Systems Using Feedforward Neural Networks

An artificial feedforward neural network is used for on-line control purposes of a class of single input single output nonlinear dynamic systems. The neural network's task is to provide one step ahead predictions for the unknown input-output behaviour of the dynamic system to the controller in order to drive the system's behaviour according to the specifications given by a reference model. The learning algorithm used to update the weights of the network, allows dynamic adjustment of their values to cope with varying signals coming from the unknown plant.