基于ARIMA与信息粒化SVR组合模型的交通事故时序预测  

Urban traffic accident time series prediction model based on combination of ARIMA and information granulation SVR

作　　者：孙轶轩[1] 邵春福[1] 计寻 朱亮[2] 

机构地区：[1]北京交通大学城市交通复杂系统理论与技术教育部重点实验室,北京100044 [2]中国铁道科学研究院运输及经济研究所,北京100081

出　　处：《清华大学学报（自然科学版）》2014年第3期348-353,359,共7页Journal of Tsinghua University(Science and Technology)

基　　金：国家"九七三"重点基础研究项目(2012CB725403);国家自然科学基金国际合作重大项目(71210001)

摘　　要：该文基于自回归滑动平均(ARIMA)模型和支持向量回归机(SVR)模型,构建时间序列组合预测模型,对道路交通事故相关指标进行趋势预测。通过ARIMA预测模型进行线性拟合;基于模糊信息粒化方法,将ARIMA预测模型残差季度变化趋势映射为包含最小值Low、中值R、最大值Up三个参数的模糊信息粒;并以其为输入构建SVR模型,对季度残差变化趋势进行预测;最后根据SVR残差预测值修正ARIMA模型预测值。实证研究结果表明:时间序列组合预测模型精度优于单一ARIMA模型,由模糊信息粒子确定的预测区间较好描述了实证数据的季度变化趋势。A hybrid prediction model was established to implement the time series forecasting of traffic accident statistical index based on the ARIMA model and the SVR model. The ARIMA model was used to complete the linear fitting of original time series, with the residual error of the ARIMA model then transformed into fuzzy information granulation particles made up of Low, R and Up. An SVR model was developed to describe the seasonal trend of the residual error with Low, R and Up as input. The predicted value of the ARIMA model was fixed based on the SVR regression result of the seasonal residual error, with the predicted value of the hybrid model being calculated. Empirical research results show that the accuracy of the hybrid model is higher than that of the single ARIMA model and that the seasonal trends of empirical time series are precisely represented by fuzzy information granulation particles.

关 键 词：事故预测 时间序列 自回归移动平均法(ARIMA) 模糊信息粒化 支持向量回归机(SVR) 

分 类 号：U491.31[交通运输工程—交通运输规划与管理]