Design and Application of Intelligence Algorithms in Continuous Fermentation of Glycerol

The bioconversion of 1,3-propanediol from glycerol by Klebsiella pneumoniae can be described by a nonlinear dynamic system. Some work has been done on the identification and optimization of the system, in which the dilution rate of glycerol is considered as a constant. However, the demand of glycerol may vary at different fermentation stages, it is reasonable to view the glycerol metabolic system with dilution rate varying with time. In this paper, we model the glycerol metabolic process as a fourteen-dimensional nonlinear dynamical system, where the dilution rate is considered varying with time. Then an optimal discrete-valued control problem for maximizing the average concentration of 1,3-propanediol in the fermentation process is established. To solve the optimization problem, auxiliary control and an exact penalty function are used to convert this problem into a large-scale parameter optimization problem. For better balancing local and global search ability, a competitive particle swarm algorithm with time-varying control factors is proposed which is proved to be faster and more stable than the traditional competitive particle swarm algorithm. Numerical experiments are conducted to show the rationality, effectiveness and applicability of the method proposed.