Abstract
This article presents a combined neural network/genetic algorithm technique for bioprocess modeling and optimization. The approach is illustrated using a published lipase-catalyzed esterification process as an example. A neural network model is developed on the basis of a small data set obtained from a central composite rotatable design. The experimental design provides information of the influence of five process variables and their interactions on a response variable (ester yield). The performance of the resulting neural network model is compared to that of a quadratic polynomial model. Modeling results show that the neural network can outperform the quadratic polynomial in correlating the data set used in the training of the neural network and regression of the polynomial. Both models however give similar predictions of unseen data. A genetic algorithm is successfully used to identify the optimal settings of the five process variables that result in a maximum ester yield based on the neural network. The combined neural network/genetic algorithm technique can accomplish objectives similar to those of response surface methodology.
