基于主成分分析的豆粕物理特性评价

Evaluation on Physical Properties of Soybean Meal Based on Principal Component Analysis

为对豆粕的物理特性进行分析及量化评价,本文以42个豆粕样品为试验材料,测定分析其水分、容重、热物理特性、摩擦特性及粉碎特性指标参数,并运用主成分分析法简化物理特性指标,构架评价体系,得到每种样品的综合得分。结果表明,样品各物理特性指标之间存在不同程度的相关性,其中比热与水分和热传导系数呈显著正相关,与热扩散系数、休止角和摩擦系数呈极显著负相关;热传导系数与热扩散系数呈显著正相关,与休止角和摩擦系数呈极显著负相关;热扩散系数与容重呈显著负相关;摩擦系数与水分呈显著正相关,与容重和休止角呈极显著正相关;平均粒径与水分呈显著正相关,与休止角呈显著负相关。通过主成分分析提取了前4个主成分,累积方差贡献率达到88.898%(>85%),能够代表原来8个物理特性指标中的绝大部分的信息,并得到评价公式Z=0.363Z_1+0.234Z_2+0.150Z_3+0.142Z_4,计算出豆粕物理特性指标综合得分。

In order to analyze and quantify the physical properties of soybean meal, 42 samples were taken as materials to evaluate water content, volume weight, thermophysical property, frictional characteristics and rushing characteristics. Principal component analysis （PCA）was used to simply physical characteristic indicators, establish a comprehensive evaluation system and get the comprehensive score of each sample. The results showed that there were correlations between different indexes. The specific heat had a strong positive correlation with the moisture content and the thermal conductivity, and it was significantly negative correlation with the thermal diffusivity, the angle of repose and the friction coefficient. The thermal conductivity had a strong positive correlation with the thermal diffusivity, and it was significantly negative correlation with the angle of repose and the friction coefficient. The thermal diffusivity had a strong negative correlation with the density. The friction coefficient had a strong positive correlation with the moisture content, and it was significant positive correlation with the density and the angle of repose. The geometric mean particle size had a strong positive correlation with the moisture content, and it was strong negative correlation with the angle of repose. The first four principal components ,which contained 88. 898% accumulative variance contribution rate（ 〉 85% ）that standed for most information in 8 physical characteristic indicators, and the comprehensive evaluation model Z = 0. 363Z1 ＋ 0. 234Z2 ＋ 0. 150Z3 ＋ 0. 142Z4 was established. Score of each sample can be obtained by this model.