中药过程质量控制的近红外光谱高性能模型融合

Near-infrared (NIR) spectroscopy combined chemometrics provide an effective means for qualitative and quantitative analysis of complex samples. In the analysis of NIR datasets, a reliable model is a matter of primary importance for prediction of composition contents or characterization of unknown samples. Traditional models are not always convictive when the calibration set is relatively small, noise or outliers are present. This project proposes a novel modeling method based on model fusion. Taking pharmaceutical raw materials, intermediates and final products of solid preparation (Compound Danshen Dripping Pill) and liquid preparation (Shuanghuanglian Injection) as research carriers, the project intend to perform a series of innovative works on model fusion including the production of sub-model data, modeling methods, fusion frameworks and fusion strategies of prediction results by introducing least absolute shrinkage and selection operator (LASSO), Hilbert-Huang transform (HHT), extreme learning machines (ELM), hybrid topologies and induced ordered weighted averaging (IOWA). The goal is to realize high performance NIR model for the process quality control of Traditional Chinese Medicine (TCM), including accuracy, precision, robustness, stability and generalization. It is believed that the implementation of the project will provide the innovative methodology and theory for the application of NIR in the process quality control of TCM.

近红外光谱分析技术结合化学计量学方法是解决复杂样本的定性定量分析难题的有效手段，当中最重要的是建立一个预测未知样本组分含量或性质的可靠模型。常规建模方法在样本数目较少、噪声或奇异样本存在时，模型的预测性能往往达不到满意的效果。本项目针对中药过程质量控制的近红外光谱高性能模型建立提出基于模型融合技术的建模方法，以固体制剂（复方丹参滴丸）和液体制剂（双黄连注射液）原料药、生产过程中间体及成品为载体，引入最小的绝对缩减和变量选择算子、希尔伯特-黄变换、极限学习机、混合式拓扑结构以及基于诱导有序加权平均算子等方法，研究子模型数据的产生、子模型的建模方法、模型融合框架以及预测结果的融合策略，建立预测精度高、泛化能力强、稳定性高、稳健性好的中药近红外模型，为近红外分析技术在中药过程质量控制领域的合理运用提供新方法和新理论。

近红外光谱分析技术结合化学计量学方法是解决复杂样本的定性定量分析难题的有效手段，当中最重要的是建立一个预测未知样本组分含量或性质的可靠模型。常规建模方法在样本数目较少、噪声或奇异样本存在时，模型的预测性能往往达不到满意的效果。本项目针对中药过程质量控制的近红外光谱高性能模型建立提出基于模型融合技术的建模方法，引入最小的绝对缩减和变量选择算子、希尔伯特-黄变换、极限学习机、展开策略等方法，研究子模型数据的产生、子模型的建模方法以及预测结果的融合策略，建立预测精度高、泛化能力强、稳定性高、稳健性好的中药近红外模型，为近红外分析技术在中药过程质量控制领域的合理运用提供新方法和新理论。