Application Of Wavelet Support Vector Machine In Modeling For Fermention Process

Fermentation process is a strongly nonlinear, time-variant and correlative process. Reasonable model must be set up to carry out further optimizing and control strategy for the fermentation process. However, reasonable model can not be built for the reasons that nowadays only certain physical and chemical parameters can be measured online while more complicated key biological parameters can not be measured online effectively and expeditiously due to development level restrictions of bio-sensing technology and price. Therefore, the introduction of soft-sensor technique for the fermentation process has great significance.Based on the detailed analysis of the current methods of biomass soft-sensor, the method of biomass soft-sensor based on Support Vector Machine (SVM) algorithm is studied in this thesis, and the model based on standard SVM algorithm is constructed for the glutamic fermentation process. Multi-scale Wavelet Support Vector Machine (WSVM) algorithm is proposed to solve the problem of SVM algorithm with low prediction accuracy. Compared with SVM algorithm, glutamic concentration model, value of OD model and residual glucose concentration are set up for glutamic fermentation process, the precision of the models set up by WSVM are improved.Then, to solve the difficulty of modeling speed of SVM and WSVM algorithms, glutamic fermentation process model is set up by Least-squares Support Vector Machine (LS-SVM). Compared with WSVM algorithm, LS-SVM algorithm has great advantage at speed, but the precision of model is not good. Multi-scale Least-squares Wavelet Support Vector Machine (LS-WSVM) algorithm is proposed for glutamic fermentation process. Compared with SVM, WSVM and LS-SVM algorithms, LS-WSVM algorithm not only keeps good modeling performance, but also has quick modeling speed. LS-WSVM is more suitable for modeling prediction online.