A neural methodology for batch process optimizing control

The purpose of this paper is to provide an artificial-neural network based optimizing control approach for batch chemical processes. Nowadays batch process are growing important in the chemical industry and more precise tracking or programmed control techniques are required. However, control and optimization techniques for batch processes seem to be still unsatisfactory because the batch process is highly nonlinear and time varying and there exists a strong coupling between the dynamic behavior of the direct control layer and optimizing control layer which makes us have to deal with a difficult nonlinear problem. Another difficulty for control and optimization of the batch process is that the dynamic behavior of optimizing control has a very long time delay because the performance criterion is generally a final yield of product. Using the neural network technique which is developed in this article we express a batch process as a stair-shape structure of neural networks which can be trained by the practical operation record data. This modelling method does not need much theoretical knowledge about the real process. And also a nonlinear system identification which may be the most difficulty procedure for control and optimization in batch processes are overcome. In this investigation the optimization problem of a fed-batch fermentation reactor is considered in a novel view point. The simulation on the microcomputer shows that the method can give out an excellent result