基于PCA-RF预测方便米饭品质模型的建立

    Establishment of a PCA-RF-based prediction model for the quality of instant rice

    • 摘要: 为了改善方便米饭在加工过程中品质不稳定的问题,缩短产品研发周期,采用主成分分析法(PCA)对方便米饭的外观品质(L*a*b*W、孔隙率、比容)、质构特性(硬度、黏性、弹性、咀嚼性、内聚性、回复性)、感官评价(气味、色泽、完整性、黏性、弹性、软硬度、总分)进行降维分类处理,然后利用降维后的数据以工艺参数为输入变量,综合评分为输出变量,建立随机森林(RF)预测模型;采用均方误差(MSE)、平均绝对误差(MAE)以及决定系数(R2)评价回归预测模型的准确性以及模型对数据的拟合程度。结果表明:经过PCA降维后,方便米饭品质可分为优(1~<2)、良(0~<1)、中(-1~<0)、差(-2~<-1)4类;当RF算法中决策树的棵数为800,最大深度为5时,PCA-RF预测模型的预测误差最小、精度最高;PCA法降维得到的预测模型预测准确率优于LDA法;与PCA-BP、PCA-PLS和PCA-ELM预测模型相比,PCA-RF预测模型的MAE、MSE均最小,R2为0.898,高于其他3种模型。所建立的PCA-RF模型预测效果好、误差小、精度高,具备较好的学习能力与泛化能力,可为方便米饭工业化生产的质量控制提供参考。

       

      Abstract: In order to improve the quality instability of instant rice during processing and shorten the development cycle of instant rice, a method to accurately predicted the quality of instant rice based on process parameters was adopted. Principal component analysis (PCA) was used to reduce the dimensionality and classify the appearance quality (L*, a*, b*, W, porosity, specific volume), textural characteristics (hardness, viscosity, elasticity, chewiness, cohesiveness, resilience), and sensory evaluations (smell, color, integrity, viscosity, elasticity, hardness, total sensory score) of instant rice. Subsequently, using the reduced-dimensional data with process parameters as input variables and comprehensive score as output, a random forest prediction model was established. The accuracy of the regression prediction model and its fit to the data were evaluated using mean square error (MSE), mean absolute error (MAE), and the coefficient of determination (R2). The results showed that after PCA dimensionality reduction, the quality of instant rice could be classified into four categories: excellent (1-2), good (0-1), medium (-1 to 0) and poor (-2 to -1). When the number of decision trees in the RF algorithm was 800 and the maximum depth was 5, the prediction error of the prediction model had the smallest prediction error and highest accuracy. The prediction accuracy of the model obtained by PCA dimensionality reduction was better than that of the linear discriminant analysis (LDA) method. When compared with PCA-BP, PCA-PLS and PCA-ELM prediction models, the PCA-RF prediction model had the smallest MAE and MSE, with an R2 of 0.898, which was higher than that of the other three prediction models. The established PCA-RF model exhibited good prediction performance, small prediction errors, high prediction accuracy, and strong learning and generalization abilities, providing a reference for quality control in the industrial production of instant rice.

       

    /

    返回文章
    返回