Abstract:
This research presents the implementation of an optimal control with neural network predictor to control a batch crystallizer. Due to the limited understanding of nonlinear and complicated dynamics of crystallization processes, the optimal control which is a model-based control strategy may not perform well as expected. A further difficulty in batch process control is that product quality variables usually cannot be measured on-line and can only be obtained through laboratory analysis at the end of batch run. To overcome such difficulties, an artificial neural network model is developed based on input and output process data and integrated with the optimal control strategy for controlling a batch crystallizer. The formulated optimal control problem is solved by a sequential method in which the control profile is parameterized by using a piecewise constant function. The neural network is applied to predict the moment variables that represent a crystal product quality and the solution concentration within batch crystallizer. The batch crystallization of potassium sulfate production is chosen as a case study to demonstrate the proposed control strategy. The simulation results have shown that the recursive neural network-based optimal control gives a better control performance compared to a conventional linear cooling control technique.