乙醇固态发酵过程参数及状态的近红外光谱检测方法研究

发布时间：2018-03-26 11:31

本文选题：乙醇　切入点：固态发酵　出处：《江苏大学》2017年硕士论文

【摘要】：为实现乙醇固态发酵过程的实时检测,本论文开展了基于近红外光谱技术的乙醇固态发酵过程参数的定量检测和状态的定性判别研究。具体研究工作如下:(1)对玉米粉生料固态发酵乙醇的工艺进行了试验研究。开展了玉米粉生料固态发酵产乙醇的试验,采用理化试验分析方法得到样品中乙醇和还原糖的含量,并采集样本的近红外光谱,为后续建立固态发酵过程参数及状态检测模型提供试验数据。(2)探讨了乙醇固态发酵过程关键参数(乙醇和还原糖)的近红外光谱定量检测方法。使用联合区间偏最小二乘法从标准正态变量变换预处理后的光谱中选择关于参数乙醇和还原糖含量的最优联合子区间;再使用迭代保留信息变量法从最优联合子区间中分别筛选出关于参数乙醇和还原糖的特征波数变量,并与传统方法遗传算法和竞争自适应重加权采样法进行对比;最后,建立关于乙醇和还原糖含量的偏最小二乘预测模型。试验结果显示:迭代保留信息变量方法选择的关于参数乙醇和还原糖含量的特征变量数分别为45个和43个;由这些特征波数变量建立的偏最小二乘模型关于参数乙醇含量的测试集均方根误差和预测相关系数分别为0.2485和0.9937,关于参数还原糖含量的测试集均方根误差和预测相关系数分别为0.1418和0.9949;迭代保留信息变量方法选择的特征波数变量个数是最少的,而且由这些特征波数变量建立的偏最小二乘模型具有最好的预测结果。研究结果表明,利用近红外光谱技术结合适当的化学计量学方法可以有效对乙醇固态发酵过程关键参数进行快速检测。(3)探讨了乙醇固态发酵过程状态的近红外光谱定性检测方法。使用联合区间偏最小二乘法从标准正态变量变换预处理后的光谱中选择关于发酵状态的最优联合子区间;再分别使用迭代保留信息变量法、竞争自适应重加权采样法和遗传算法从最优联合子区间中筛选出关于发酵过程状态的特征波数变量;最后,建立发酵过程状态的主成分分析模型和极限学习机模型。试验结果显示:迭代保留信息变量方法筛选的光谱特征波数变量建立的主成分分析模型前2个主成分累计贡献率为96.0236%,均高于其它主成分分析模型;迭代保留信息变量方法筛选的光谱特征波数变量建立的极限学习机状态识别模型校正集和测试集正确率分别为99.8182%和97.2728%,均高于其它极限学习机预测模型。研究结果表明,利用近红外光谱技术结合适当的化学计量学方法可以有效对乙醇固态发酵过程状态进行快速识别。本研究为乙醇固态发酵过程的近红外光谱在线检测提供新的思路,为乙醇固态发酵过程在线检测的便携式近红外光谱装备研发奠定理论和技术基础。
[Abstract]:In order to realize the real-time detection of ethanol solid-state fermentation process, In this paper, the quantitative detection of the parameters of ethanol solid-state fermentation process based on near-infrared spectroscopy and the qualitative discrimination of the state were carried out. The specific research work is as follows: 1) the technology of solid state fermentation of ethanol from raw corn meal was tested. The experiment of producing ethanol by solid state fermentation of raw corn meal was carried out. The contents of ethanol and reducing sugar in the samples were obtained by physicochemical analysis, and the near infrared spectra of the samples were collected. The key parameters (ethanol and reducing sugar) of solid-state fermentation process were determined by Near-Infrared Spectroscopy (NIR). The combined interval bias was used to determine the key parameters (ethanol and reducing sugar). The least square method selects the optimal joint subinterval of the parameter ethanol and reducing sugar content from the spectrum of the standard normal variable transformation pretreatment. Then the characteristic wavenumber variables about the parameter ethanol and reducing sugar are screened out from the optimal joint subinterval by iterative preserving information variable method, and compared with the traditional genetic algorithm and the competitive adaptive re-weighted sampling method. Finally, A partial least square prediction model for ethanol and reducing sugar content was established. The experimental results showed that the number of characteristic variables on the content of ethanol and reducing sugar selected by iterative retention information variable method was 45 and 43 respectively. The RMS error and predictive correlation coefficient of the test set for the parameter ethanol content are 0.2485 and 0.9937, respectively, and the RMS error and prediction of the parameter reducing sugar content are 0.2485 and 0.9937 for the partial least squares model based on these characteristic wavenumber variables, respectively. The correlation coefficients are 0.1418 and 0.9949, respectively, and the number of characteristic wavenumber variables selected by iterative preserving information variable method is the least. Moreover, the partial least squares model based on these characteristic wavenumber variables has the best prediction results. Near-infrared spectroscopy (NIR) combined with appropriate chemometrics can be used to detect the key parameters of solid-state ethanol fermentation. Using the joint interval partial least square method to select the optimal joint subinterval about the fermentation state from the spectrum pretreated by the standard normal variable transformation; Then the iterative preserving information variable method, the competitive adaptive re-weighted sampling method and the genetic algorithm are used to screen out the characteristic wavenumber variables about the state of fermentation process from the optimal joint subinterval. The principal component analysis model and ultimate learning machine model of fermentation process were established. The experimental results showed that the first two principal component analysis models were established by the spectral characteristic wavenumber variables screened by iterative retention information variable method. The total contribution rate was 96.0236, which was higher than that of other principal component analysis models. The correct rates of calibration set and test set of state recognition model of LLM are 99.8182% and 97.2728%, respectively, which are higher than those of other prediction models. Near-infrared spectroscopy (NIR) combined with appropriate chemometrics can be used to identify the state of solid-state ethanol fermentation. This study provides a new idea for on-line detection of solid-state ethanol fermentation by near-infrared spectroscopy (NIR). It lays a theoretical and technical foundation for the development of portable near infrared spectrum equipment for the on-line detection of ethanol solid state fermentation process.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TQ223.122;O657.33

【参考文献】