Bioprocess Modeling using a robust genetic programming

Using genetic programming, we can model the avermectin fermentation process based on some training data. However, these training data obtained from practical industrial processes generally are corrupted by large noise or outliers whose effect on the performance of the classical genetic programming is adverse. Furthermore, the performance of the most current robust learning algorithms depends on the suitable choice of cutoff parameters in their estimators and the values of cut-off parameters are difficult to define. To overcome these problems, a robust genetic programming based on a tanh-estimator with several populations is proposed. This approach can perform multi-directional search on the whole problem space for different cut-off parameters, so it can get mixed information from several different directions and has more chance to rind an acceptable solution. In addition, the proposed approach is less sensitive to the values of the cut-off parameters.