Applied Research Of Mixture Kernel Support Vector Machine In Modeling For Fermentation Process

Microorganism fermentation engineering is the foundation of Bioengineering and biotechnology. With the quick development of the fermentation industry, the request of the automatization extent on the producing process is becoming hard. As a result, there is an urgent need for the process parameters on-line detection and optimal control. However, because of the technology and price factors, most sensors used in fermentation process are to measure the physical and chemical parameters while some more crucial biological parameters can't be detected online effectively in recent days. Therefore, applying the soft-sensor technique to the fermentation process has great practical value.In view of the glutamate fermentation process, this paper makes a detailed research about soft sensor modeling for the biological parameters in fermentation process.In order to solve the difficulties of modeling for the fermentation process which has the features of non-liner, time varying and uncertainty, a modeling method based on the Support Vector Machine (SVM) is proposed based on detailed analysis of the current soft sensor methods. In SVM's theory, choosing a suitable kernel is very essential. Through the study of kernel function and the Mercer Theorem, a mixture kernels function which was linearly combined by a local characteristic kernel function and a global characteristic kernel function was proposed. The influence of the two kinds of kernel functions can be tuned by parameter, thus the model can get high precision and wide applicability. The material step of the mixture kernels SVM modeling method was shown. The soft sensor models for glutamic concentration, OD value and residual glucose concentration are set up, and the practical results show that the SVM model with mixture kernels has higher learning and generalization abilities.On the basis of glutamate forecast model of mixture kernels SVM, taking the prediction accuracy as a goal, a new method is put forward to optimize the parameters of mixture kernel SVM by using chaotic particle swarm optimization (CPSO) which has better global search ability. The algorithm includes judging and handling the local convergence which is of strong ability to avoid being trapped in local minima. The material step of the method was shown. The simulated experiment proved the effectiveness of the algorithm above. The method for the research of modeling in the glutamic acid fermentation process obtains higher accuracy.