ATR-FTMIR在线检测关键技术研究

As a process analysis technique, Attenuated Total Reflection-Fourier Transform Mid-infrared (ATR-FTMIR) spectroscopy owns the unique advantages of quickness, accuracy, no destruction and no pollution, and hence can realize the online detection on primary biochemical indicators of fermentation process. However, the characteristics of Attenuated Total Reflection (ATR) sampling technique and fermentation process have brought online ATR-FTMIR detection some technical problems. This project takes glutamic acid fermentation as the research object, and do some research on the key technical problems. The research contents include: (1) To eliminate the nonlinear effects in spectral measurement caused by the inherent defects in ATR sampling technology, a new ATR correction algorithm with independent intellectual property rights in our country is developed from its optical characteristics; (2) based on the optimal combination between among different nonlinear modeling strategies or methods, the mixed nonlinear calibration models with higher robustness and prediction precision is studied to solve different nonlinear phenomena in fermentation process; (3) considering the time-varying characteristic of fermentation process and the samples composing of massive unlabeled ones and a small amount of labeled ones, online modeling methods based on incremental semi-supervised learning are studied.

衰减全反射傅里叶变换中红外（ATR-FTMIR）光谱技术作为一种过程分析手段，以其快速、准确、无损、无污染的独特优势，可以实现对发酵过程主要生化指标的在线检测。然而衰减全反射（ATR）采样技术以及发酵过程自身的特点给ATR-FTMIR在线检测带来了一些技术难题。本项目以谷氨酸发酵为研究对象，对这些关键性技术难题展开研究。研究内容包括：(1) 针对衰减全反射（ATR）采样技术本身固有的缺陷会给光谱测量带来的非线性效应，从其光学特性出发，研究具有我国自主知识产权的ATR校正算法；(2) 针对发酵过程中不同的非线性现象，研究通过不同非线性建模策略或方法之间的优化组合，构建鲁棒性和预测精度更高的混合非线性校正模型；(3) 针对发酵过程的时变性以及大量未标定和少量标定样本的构成特点，研究基于增量式半监督学习的在线建模方法。

衰减全反射傅里叶变换中红外（ATR-FTMIR）光谱技术作为一种过程分析手段，经常用于工业过程在线检测。本项目针对ATR-FTMIR在线检测过程中面临的一些技术难题展开研究，具有重要的科学意义和应用价值。主要包括下列四个方面：（1）系统研究了ATR校正算法涉及到的关键参数的理论推导和优化计算，进而构建了ATR校正算法的基本框架；（2）研究了如何通过将多种非线性策略相互融合的方式来构建预测精读更高的非线性多元校正模型，提高ATR-FTMIR的定量分析精度。（3）分别进行了基于增量式监督学习和增量式无监督学习的在线建模方法的研究，实现了对有标签数据和无标签数据的在线模型更新。（4）分别针对有标样和无样本的条件下，研究了能够有效消除谱图间非线性差异的模型传递新算法，成功提高了多元校正模型在不同测试条件下的通用性。