基于LIBSVM和时间序列的区域货运量预测研究

摘要

To the problem of excessive affecting factors and small sample size in regional freight volume forecasting, the LIBSVM support vector regression model and state space time series model with mutual information technique are proposed. In this approach, the MI is adopted to reduce the dimensionality of the high dimensional features, and then the new lower dimensional subspace is treated as the sample input to establish the LIBSVM support vector regression model and the state space time series model. The experimental results of Chongqing freight volume forecasting and comparative analysis show that the method can improve the prediction accuracy while accomplishing a valid forecast, and the relative error is about 0.06.%针对区域货运量预测中影响因素多、样本数量小的问题，提出了互信息MI与LIBSVM支持向量回归以及状态空间时间序列相结合的预测方法，采用MI进行高维度特征降维后，以新的低维空间作为样本输入，分别建立LIBSVM支持向量回归和状态空间时间序列预测模型。通过重庆市货运量预测实验结果及对比分析表明，该方法在进行有效预测的同时能够改善预测精度，相对误差约为0.06。