Neural Network Based Feedforward Adapter for Batch Process Control

In this paper, we propose a control scheme with multiple neural network based feedforward adapters. Each individual neural network is trained with a particular setpoint value and a time sequence. We point out that a single neural network is not feasible to be used in the on-line batch operation. We have also shown that a linear feedforward adapter is practically impossible to implement. We utilize the feedforward neural network as a universal non-linear mapping to find a non-linear feedforward adapter. We construct a different learning error, which requires some modification of the standard back-propagation learning algorithm to include the dynamics of the plant.