Research On Multi-objective Cooperative Optimization Controlmethodand System For Fermentation Process

Fermentation is the basis of bioengineering, modern biotechnology andtheir industrialization. With the rapid development of modern biotechnologyand the continually expanding of production scale of fermentation industry, itis urgent to improve the productivity and production quality of fermentationprocess by optimization control methods. The early single objectiveoptimization control for fermentation process can not deal with a number ofproduction indexes such as production output, substrate cost, fermentationtime in fermentation process at the same time. The Multi-ObjectiveOptimization (MOO) control for fermentation process is an effect way toimprove the production quality and benefits of fermentation process. Theexisting MOO control method for fermentation process based on linearweighted summation has the problem that the artificial weights of objectivesmay have negative impact on the results of optimization and only beappropriate for the case that the objectives are not competing with each other.Multi-Objective Evolutionary Algorithm (MOEA) applies large size population to search optimal solutions and needs a large number ofevolutionary generations, which lead to require substantial evaluations ofobjective functions and can not deal with MOO control problem which haslarge-scale data in fermentation process. Beside, in MOO control forfermentation process, the MOO which generates approximate Pareto front andthe Decision-Making (DM) which selects the preferred solution are two mainsub-tasks of solving MOO problem. However, the existing integrationmethods of MOO and DM have the problem of heavy computation burden,long running time and omitting optimal solutions. Therefore, the research ofmore efficient method for integrating MOO and DM in order to realize MOOcontrol for fermentation process has important theoretical significance andpractical applications value for further improving the production levels offermentation process.On the basis of analyzing the review and the current state of research inoptimization control for fermentation process and methods for solving MOOproblem, a multi-objective cooperative optimization control method andsystem for fermentation process is studied in this paper.A multi-objective cooperative optimization control method forfermentation process combining interactive MOO and DM withmulti-controller cooperative control is proposed. The discrete, continuousapproximation of Pareto front and Multi-Attribute Decision-Making (MADM)are invoked interactively, then a preferred solution which has high accuracy can be generated based on complete approximate continuous Pareto front withsmall computation and short running time; the control precision, flexibilityand adaptability of optimization control for fermentation process is improvedby multi-controller cooperative control based on switch strategy. On that basis,the architecture and each function module’s realization of multi-objectivecooperative optimization control system are given.A method for continuous approximation of Pareto front based onGeometric Support Vector Regression (GSVR) is proposed. At first, a methodfor discrete approximation of Pareto front based on Swarm EnergyConservation Multi-Objective Particle Swarm Optimization (SEC-MOPSO) isgiven. The swarm energy conservation mechanism is introduced for boostingthe exploration capability of swarm and improving the distribution anddiversity of small size discrete Pareto front. Then the continuousapproximation of Pareto front is realized by establishing the GSVR model ofsmall size discrete Pareto front. Considering the distribution characteristic ofPareto optimal points, a method for generating the augmented training samplesets by shifting the original training samples along multiple coordinated axis isgiven to solve the problem of unable to generate separable augmented trainingsample sets by shifting original training samples only along single coordinatedaxis. The experiments are carried out to classical test functions, the resultsshow that the small size approximate discrete Pareto front generated bySEC-MOPSO has good distribution and diversity. The method for continuous approximation of Pareto front based on GSVR can generate complete andaccurate approximate continuous Pareto front.For the optimization control problem, considering the time variety anduncertainty of fermentation process, a multi-controller cooperative controlmethod for fermentation process based on a predictive switching strategy isproposed. A predictive control method for fermentation process based onNon-Linear Quadratic Gaussian (NLQG) is proposed, the NLQG controller iscomposed of a Extended Kalman Filter (EKF) and a Non-Linear QuadraticRegulator (NLQR) which are connected in series; the switched law of anumber of controllers is predicted according to a control performance indexand the multi-controller cooperative control can be realized. The experimentsare carried out to a classical non-linear system, the results show that theproposed fermentation process predictive control method has good trackingeffect for the change of set value and strong robustness in different noisyenvironment. The multi-controller cooperative control method forfermentation process can enhance the control precision effectively.An industrial fed-batch yeast fermentation process and a penicillinfed-batch fermentation process are introduced to carry out the experimentalresearch on the multi-objective cooperative optimization control method andsystem for fermentation process. The results show that, the proposedmulti-objective cooperative optimization control method for fermentationprocess can solve the problem of large computation amount, long running time, omitting optimal solutions. The method can realize the optimal tradeoff amonga number of production indexes in fermentation process, obtain good controleffect and has wide application prospects in optimization control forfermentation process.