Modelling and Optimization of Fed-Batch Processes using Recurrent Neural Networks

Fed-batch processes have been used to produce high value products in the chemical, biological, food, pharmaceutical and semiconductor industries. General features of fed-batch biological processes include strong nonlinearity, no steady state operation, instinctive time variation, batch-to-batch variation and uncertainty caused by drifting of raw materials. These features complicate modelling and control.

Recurrent neural networks(RNNs) have a dynamic memory and can process temporal context information, RNNs are highly promising tools used for solving complex temporal, nonlinearity, time variation and uncertainty tasks.

 In this project, RNNs are used to model the fed-batch processes and optimized by covariance matrix adaption evolutionary strategy.