基于KPCA-LSSVM的两阶段纺纱质量的预测模型

A Two-stage Predictive Model for Yarn Quality Based on KPCA-LSSVM

纺纱是复杂的多环节生产过程,纺纱产品的质量等级评价大多需要依赖领域专家的个人经验,有必要从定量分析的视角建立纱线质量预测模型。选取原棉的含水、含杂、主体长度、主上长度、均匀度、短绒率、成熟度、公制支数、强力、疵点总数10个影响因素作为纱线质量等级预测的评价指标,第一阶段对影响因素进行相关分析,用KPCA方法提取影响因素的主成分;第二阶段用LSSVM算法进行训练及预测,建立了基于KPCA-LSSVM的两阶段纺纱质量预测模型。通过对实测18组纺纱数据作为训练样本数据集对模型进行训练,13组数据作为该预测模型的测试数据,进行纺纱质量预测,并且与其他预测模型进行对比。结果表明:利用核主成分分析法对因素进行筛选,消除因素相关性,使得该组合算法具有较高的预测精度。

Yarn production is a complex industrail process, and its quality grade evaluation is mostly depended on the domain expert′s experience. So it is necessary to make the model of yarn quality prediction model from the perspective of quantitative analysis. Ten influencing factors（including body length, Lord length and uniformity etc.） of cotton are put forward. In the first phase, the above influencing factors has been analyzed by using correlation analysis and screened by using KPCA theory. In the second stage, LSSVM prediction model of yarn quality is built. Eighteen groups of data are taken as training data. Additional thirteen groups are used as test data to verify it and compare with the other prediction method. The results show that KPCA may reduce the correlation among the factors, and efficiency and precision of LSSVM prediction model are raised. The combination algorithm has higher prediction accuracy.