道路交通事故的非线性偏最小二乘回归建模

Road traffic accident regression modeling based on nonlinear partial least squares method

就有效预防交通事故、提高道路交通效率而言,借助高精度的道路交通事故预测模型,准确分析事故原因是重要的基础性工作。首先基于偏相关分析方法,对影响事故起数、死亡人数和受伤人数这3个事故指标的11个因素进行相关性分析,确定最相关的影响因素及其线性相关性;然后利用偏最小二乘回归方法,对事故指标与影响因素之间的线性关系进行建模;进而基于非线性偏最小二乘回归方法,建立两者之间的非线性关系模型。通过对回归模型的精度分析,用偏最小二乘回归方法仅能对事故指标与影响因素之间线性关系准确建模,测定系数最大为0.98,相对误差最大为21.77%。用非线性偏最小二乘回归方法,对事故指标与影响因素之间的线性和非线性关系均能准确建模,测定系数最大为1.相对误差最大为4.23%。

For preventing road traffic accidents and improving the road traffic efficiency, accurate cause analysis of road traffic accident is important basic work according to a high-accuracy traffic accident model. Firstly, the most relevant factors and the linear correlations between influence factors and the traffic accident indexes are determined, after correlation analysis of 11 influence factors with 3 accident indexes, which are the number of accidents, death toll and the number of injured people, based on the partial correlation analysis method. Then, the linear relationship between each of the accident indexes and the influence factors is established based on the partial least squares regression method. And further, the nonlinear relationship between each of the accident indexes and the influence factors is established based on the nonlinear partial least squares regression method. By regressive accuracy verifying, the partial least squares regression method can be used only to model accurately the linear relationship between the influence factors and the accident index, the maximum determination coefficient is 0.98 and the maximum relative error is 21.77%. The nonlinear partial least squares regression method can be used to model the linear and the nonlinear relationships between them, the maximum determination coefficient is 1 and the maximum relative error is 4.23%.