Modeling Of Burkholderia Cepacia Lipase Fermentation Process Based On Neural Network

As an important part of biotechnology, microbial fermentation engineering is used in most areas of the national economy such as chemical industry, agriculture, food industry, pharmaceutical industry and energy sectors. Fermentation process has the characteristics of highly nonlinear, time-varying and basic biomass hardly measured on line. Therefore, how to control and optimize fermentation process effectively on line has become a research hot spot in fermentation engineering.There are some very important parameters which control the entire fermentation process. These parameters such as cell concentration, substrate concentration, and product concentration are hard to be directly measured on line. Therefore, in this study, we have employed soft-sensing technology, which can make use of real-time measured parameters (e.g. time, pH, dissolved oxygen) to estimate the parameters which can not be directly measured on line, to simulate the entire fermentation process.On consideration of its advantages of simple structure, various adjustable parameters, multiform training algorithm, and versatile manipulation, BP neural network well adapts to non-linear biofermentation process modeling.Taking the Burkholderia cepacia lipase fermentation process as a research object, we used BP neural network to establish a soft sensor model of the lipase fermentation process, and applied genetic algorithms to optimize the initial weights and thresholds of neural networks, so as to speed up the convergence rate and to achieve the global optimum results. The soft-sensing model realized key biomass to be indirectly measured on line for the lipase fermentation process. This laid a foundation for real-time controlling, management and optimization of the lipase fermentation process.Through Matlab6.5, we used its built-in Neural Network Toolbox and installed gaot Genetic Algorithm Toolbox to construct the soft-sensor model of the B. cepacia lipase fermentation process. Then, we tested the model by the experimental data, they both fited well, suggesting that the model possesses good reliability and generalization ability.