基于复合核模的物流需求预测研究

Prediction Analysis for Logistics Demand Based on Multiple Kernels

物流需求预测在制定发展规模、资源整合、政策法规拟定中起着至关重要的作用。为提高预测性能，提出一种基于复合核模的预测模型。利用全局核函数与局部核函数分别训练参训样本，根据训练结果动态提取复合核模底层函数，将底层核函数进行非线性组合；利用该复合核模对训练样本进行在线解析，并对测试样本进行预测检验。利用我国运输物流流量物流统计数据进行案例分析，实验结果表明，基于复合核模的预测模型能从数据源头增强样本的解析性能与非线性学习能力，并能提高预测模型的预测精度，增强泛化性能。

Abstract. Logistics demand prediction plays a vital role in resource integration, logistics development,and policies and regulations making. A SVM prediction model was proposed based on multiple kernels. Training samples have been trained by global kernels and local kernels respectively, and then multiple kernels have been dynamically extracted according to the training results. Composite kernels have been used to parse training samples and to validate test samples with Chinese transportation data. The result shows that multiple kernels model could enhance sample analyticity and nonlinear learning ability. Also, it could improve the prediction accuracy and strengthen the generalization performance.