Application And Research Of PSO In Fed-batch Fermentation Process Control

The fermentation is a kind of time-variable biochemical process,which can be label as complex,uncertain,unlinar.Fed-batch fermentation is semi-continuous.Nutrients fed intermittently during the process, which have wide range of applications, almost throughout the fermentation industry. Early in the fermentation industry the instruction of feeding entirely empirically determined.In large-scale fermentation industry,The way of feeding have develop from singe-stage to multi-stage,one nutrient to several nutrients. Nutrients-feeding configuration program will greatly affect protein production Thus,It is of great significance to informed a set of intelligent feeding program to Fed-batch fermentation process.Particle swarm algorithm is a new approachto optimization, And it is very suitable for solving the optimization problem in complex environment.But traditional PSO have problems like easy fall into local optimal as well as rapid convergence. We put forward a modify PSO by loacal esacpe mechanism with Scale-free topology(LESF-PSO).The Specific improvements invoIves,Firstly, By Analysising neighborhood interaction. We propose dynamic topology based on sacle-free network,which enhanced the neighborhood exploration ability;secondly, We point out local-escape mechanism to improve the escape capacity of local optimal;At last,We record the displacement of best particle and update velocity mask boundary,which fasten search speed,the problem that Particle swarm gathered at the border is avoided. The proposed aglorithm display a relatively acceptable performance.We apply the proposed alogrithm to practical problems, By Analysising control variables,growth characteristics,feeding methods of proteins such as interleukin,A fermentation expert system is designed, the LESF-PSO is applied to optimize sugar supplement at each time.The proposed algorithm exhibit good prefromance in optimize the multi stage realtime controlling process that the traditional PSO may not work,which displays a better solution as well as shorter time consuming. The above results indicated the propose algorithm have considerable optimize ability especially to multi-dimension complex problem that has a certain reference value.