基于组合预测优化模型的交通事故预测研究

Research on optimal combination model for forecasting traffic accident

为提高交通事故预测精度,基于熵值法构建UGM(1,1)-Holt组合预测模型,将滑动转移概率思想引入马尔科夫链模型,实现组合预测优化。应用该模型拟合分析2003—2011年湖北省交通事故死亡人数的历史数据,并以2012—2014年数据验证其有效性。通过实例对比UGM(1,1)模型、Holt指数平滑模型、组合预测模型和组合预测优化模型的预测精度。结果表明:相比前3种模型,提出的组合预测优化方法拟合值平均相对误差(MRE)为0.45%,3年预测值MRE为1.25%,能有效获取单一模型优势,预测精度更高。

To predict death toll caused by traffic accident accurately,a UGM（1,1）-Holt combination forecasting model was built by using the entropy method,and the Markov model was improved by introducing the slide transition probability thinking into it to realize optimization of combination forecasting.Finally,Hubei traffic accident death toll in 2012-2014 was predicted,and a comparison of 4 models＇ predicted values with the measured value was made.The results show that mean relative error（MRE） of the optimal combination forecasting model＇s fitted value is further reduced to 0.451 0%,MRE of predicted values of 2012-2014 reduced to 1.25%,and that the model can gain advantages of the single models,and can ensure greater prediction accuracy.