状态空间时间序列的区域物流需求预测研究

区域物流需求是制定区域物流发展政策、基础设施建设和物流系统规划的重要依据，由区域各项相关经济指标共同决定。针对区域物流需求预测中样本数量小的问题，提出了互信息高维度特征降维方法，在保证相关综合信息完整性基础上降低原始数据维度，在此基础上建立了状态空间时间序列预测模型，同时采用局部线性小波神经网络和LIBSVM支持向量回归模型进行对比实验。算例分析及实验结果表明，采用互信息降维后的预测模型相对误差平均减少54.8%，互信息与状态空间时间序列模型相结合的预测方法对于区域物流需求预测问题预测精度较高，相对误差约为0.08。

Research of regional logistics demand forecasting using state space time series

Regional logistics demand is an important evidence of the regional development policy formulating, infrastructure construction and logistics system programming. It is decided jointly by all the related regional economic indicators. On account of the small sample size when forecasting the regional logistics demand, the feature dimension reduction method of Mutual Information（MI） is proposed to reduce the original data dimensions without destroying the integrity of the relevant synthesis information. And on this basis, the state space time series forecasting model is established, together with the local linear wavelet neural network and LIBSVM support vector regression models as comparisons. The results of example analysis and experiment show that, a 54.8% mean decrease of the relative errors in the forecasting models can be obtained by using the Mutual Information to reduce the data dimensions, and the approach of combining the MI and state space time series model has a higher forecasting accuracy in regional logistics demand forecasting problem. The relative error is about 0.08.