【作者】 钟泉；

【导师】 潘涛；

【作者基本信息】 暨南大学 ， 应用数学， 2009， 硕士

【摘要】 采用FTIR/ATR光谱和化学计量学方法,建立、优化甘蔗初压汁锤度定量分析的数学模型,并研究模型的稳定性。基于全体样品的最优单点模型预测效果给出定标和预测样品集的一种划分方法,在此基础上,分别采用单点回归、直接PLS、Savitzky-Golay平滑PLS、移动窗口PLS和等间隔移动窗口MLR方法建立定量分析的数学模型,以预测均方根误差(RMSEP)和预测相关系数(R_p)为优化目标,进行多种相关模型参数的联合优化设计,经过大量计算机数值实验,在大范围内进行模型优选,并讨论模型的稳定性。结果表明:所采用的定标和预测样品集的划分方法具有良好的建模代表性、客观性和稳定性;Savitzky-Golay平滑可以较大提高模型的预测精度;移动窗口PLS方法可以遴选高信噪比的光谱特征波段;等间隔移动窗口MLR模型可以遴选简便有效的特征波数组合模型并具有最好的预测精度,大大降低模型的复杂性。最优模型预测效果的RMSEP、RRMSEP、R_p的值分别达到0.4690°Bx、2.38%、0.8456,为FTIR/ATR光谱技术应用于甘蔗初压汁定量分析提供优秀稳定的数学模型。更多还原

【Abstract】 The mathematical models of quantitative analysis for Brix in sugarcane initial-pressed juice were established and optimized by FTIR/ATR spectroscopy and chemometrics method, and the stability of the models were also discussed. A sort of classified method for the calibration and prediction samples was gotten, which based on predictive effect of the best single-point model with all samples. On this condition, the calibration models were established by single-point regression, direct PLS, Savitzky-Golay smoothing PLS, moving window PLS and moving window MLR with fixed interval, respectively. The model parameters were designed according to the root-mean-square error (RMSEP) and the correlation coefficient (R_P) which were the optimum objective. Through a large number of computer experiments, we selected out the optimum models on a large scale and discussed the stability of models.The results showed that the classified method for the calibration and prediction samples was representative, objective and stable. Savitzky-Golay smoothing method can improve model’s predictive accuracy greatly. Moving window PLS method can be used to select the high Signal to Noise Ratio spectra bands. MLR method with fixed interval moving window can be used to select combination wavenumbers effectively and have the best predictive accuracy, so that the complexity of the model was reduced greatly. The RMSEP, RRMSEP and R_P for the best predictive effect of optimum models reached 0.4690°Bx, 2.38%, 0.8456, respectively. The results provided excellent and stable models for quantitative analysis Brix in sugarcane initial-pressed juice by FTIR/ATR spectroscopy technology.更多还原