Abstract:

To reduce the complex structure of least squares support vector machine (LSSVM) in logistics demand modeling and improve the forecasting accuracy of LSSVM for logistics demand further, based on the grey relational analysis (GRA) and the kernel principal component analysis (KPCA), a LSSVM forecasting method is proposed. First, GRA is used to choose the main influential factors of logistics demand. Then, the KPCA is applied to extract the nonlinear principal components, which can eliminate the correlation in the main influential factors. Finally, the extracted nonlinear principal components are selected as the input variables of LSSVM to construct the logistics demand forecasting model. And the parameters of LSSVM are adjusted by the improved particle swarm optimization (IPSO). Using this method, China’s logistics demand is analyzed. The results indicate that the proposed method effectively reduces the number of the input variables in LSSVM and simplifies the structure of the LSSVM. The forecasting accuracy of logistics demand is improved to some degree.

Key words: logistics engineering, forecasting method, LSSVM, logistics demand, forecasting accuracy

摘要：

为降低物流需求建模中最小二乘支持向量机（LSSVM）的结构复杂性、进一步提高LSSVM对物流需求的预测精度，提出一种基于灰色关联分析（GRA）与核主成分分析（KPCA）的LSSVM预测方法.首先利用GRA找出物流需求的主要影响因素；然后利用KPCA提取主要影响因素的非线性主成分，消除因素之间的多重相关性；最后，将提取出的非线性主成分作为LSSVM的输入变量，构建物流需求预测模型，并采用改进粒子群 (IPSO)算法调整LSSVM参数.运用该方法对我国物流需求进行实例分析，结果表明，该方法有效减少了LSSVM输入变量个数，简化了LSSVM结构，并且在一定程度上提高了物流需求预测精度.

关键词: 物流工程, 预测方法, 最小二乘支持向量机, 物流需求, 预测精度