Application of Gradient-Free Optimization Algorithms in Yield Optimization of Fed-Batch Fermentation Processes

Abstract

The fed-hatch fermentation process of antibiotics is a complex industrial process with high nonlinearity and uncertainty. Due to the difficulty of building the yield model, yield optimization of antibiotic production becomes a challenging task. Aiming to optimize the antibiotic yield of the fed-batch fermentation process efficiently, a kind of gradient-free optimization scheme was introduced, which does not rely on a yield model. Two different gradient-free optimization algorithms, adaptive particle swarm optimization (APSO) as a global algorithm and the Simultaneous Perturbation Stochastic Approximation (SPSA) as a local algorithm, were incorporated and applied to the yield optimization of the penicillin fermentation process. The results had shown that both types of gradient-free optimization algorithms could achieve satisfactory performance. However, as a global optimization algorithm, the APSO may attain better optimal settings than the SPSA, while the SPSA is much beneficial than the APSO in optimization cost. The trade-off should be made according to different application scenarios.