Research On Infrared Spectrum Analysis Of Methanol Concentration In The Process Of Pichia Pastoris Fermentation

With the wide use of biocatalyst-enzyme, the biocatalyst techniques have a big impact on the medicine and food industry. In the process of Pichia pastoris fermentation, the use of methanol to catalyze the protease expression is one of the most common techniques. It is crucial for users to monitor the concentration of methanol on-line with good accuracy and fast speed. However, there are various flaws existing in the traditional methods. As a result, aiming at the infrared spectrum of methanol solution in the Pichia pastoris medium, this paper does some research in the aspects of spectrum signal denoising, spectrum characteristic analysis, fermentation process classification and methanol concentration prediction. The main job is listed below.Firstly, an experimental platform and a spectrum gradient table are both established to obtain the infrared spectrum data of medium solution with methanol and glycerin inside. Besides, this paper analyzes the frequency characteristic of the infrared spectrum to know the existence of several noises. Then, the wavelet transform is employed to eliminate noises in the spectrum by the means of automatic threshold choosing algorithm with the signal noise rate of 19.786, which has a better performance than the ordinary threshold choosing algorithm.Secondly, this paper introduces the chemical characteristic of the infrared spectrum. And the independent and mixed infrared spectrums of methanol, glycerin, water and medium solution are obtained to analyze the featured peak of them, which helps to get the suitable feature parameter and build a solid foundation for the prediction of methanol concentration.Thirdly, after analyzing the methanol characteristic in different stages of the Pichia pastoris fermentation, it is necessary to classify the stages before concentration prediction. By means of Restricted Boltzmann Machine, the spectrum is analyzed to divide the process stages of fermentation, which could achieve the accuracy of 80%.Last, based on the chemical analysis of the IR spectrum, the improved partial least square regression is introduced to give scores to the principal components, which helps to weaken the irrelevant information in the spectrum data and build the methanol-spectrum prediction model. The prediction result demonstrates the effectiveness of improved partial least square regression.