基于Hankel-DMD的城市交通事故风险时空预测

为解决城市交通事故风险时空分布预测任务中时空关联性捕捉困难的问题，提出基于动态模态分解(DMD)的城市交通事故分析时空预测模型，模型利用总最小二乘法去除交通事故数据中的噪声，应用结合Hankel矩阵的动态模态分解模型(Hankel-DMD)捕捉交通事故风险的时空关联性，对交通事故风险的时空分布进行预测。研究结果表明：DMD框架能够为高维预测任务提供低秩解决方案，从高维数据中捕捉时空关联性；Hankel-DMD模型在预测评价指标平均绝对误差和均方根误差方面的表现明显优于统计学及机器学习等方法；Hankel-DMD模型产生的动态模态和特征值，对事故风险系统的时空动态特征具有一定的可解释性，同时验证Hankel-DMD模型的适用性。

Spatio-temporal prediction of urban traffic accident risk based on Hankel-DMD

To solve the difficulty of capturing spatio-temporal correlation in the task of predicting the spatio-temporal distribution of urban traffic accident risk,a spatio-temporal prediction model of urban traffic accident analysis based on the dynamic mode decomposition (DMD) was proposed.In this model,the noise in the traffic accident data was removed by the total least squares method,and the spatio-temporal correlation of traffic accident risk was captured by the dynamic mode decomposition model with Hankel matrix (Hankel-DMD),thereby the spatio-temporal distribution of traffic accident risk was predicted.The results showed that the DMD framework could not only provide a general low-rank solution for high-dimensional prediction tasks,but also capture the spatio-temporal correlation from high-dimensional data.The Hankel-DMD model performed significantly better than the common methods such as statistics,machine learning and deep learning in terms of the mean absolute error and root mean square error of the prediction evaluation indexes.Moreover,the dynamic modes and eigenvalues generated by the Hankel-DMD model had some interpretability for the spatio-temporal dynamic characteristics of the accident risk system,which demonstrated the applicability of the Hankel-DMD model for the spatio-temporal prediction tasks of urban traffic accident risk.