Abstract:

Speed time series collected as vehicles approaching an intersection can be used to predict several speed values as they subsequently entering it. Then, traveling tracks and spacing distances of the conflict vehicles are calculated by the predicted speed values, and the collision risk of them can be estimated. Because the speed distribution of a vehicle approaching to an intersection closes to the characteristics of random sequences, auto-regressive moving average (ARMA) theory is introduced to model the vehicle speed prediction. The modeling process includes time series data correlation test, p-q orders determination, formula coefficient estimation and model adaptability test. Test result shows that the ARMA model built by the previous 40 data of the observed speed time series could predict 20 values which are closed to the 20 observed ones. The other evidences of that are the normalized mean absolute errors of the conflict vehicles, which respectively equaled to 0.006 56 and 0.003 4. Further, the model built by all the 60 data of the observed time series is necessarily more applicable to predict vehicle speed, just as all the result values of the residual auto-correlation function test are less than 0.258 2.

Key words: intelligent transportation, collision risk estimation, auto-regressive moving average (ARMA), intersection, vehicle speed prediction

摘要：

车辆进入交叉口前的速度时间序列可用于预测车辆进入交叉口后若干步数速度值，利用车速预测值推算冲突方向车辆在交叉口内的行驶位移及其车间距离，可评估车辆发生碰撞的风险.针对交叉口附近车速分布符合随机序列特征，采用自回归滑动平均 (ARMA)理论进行车速时序预测建模，步骤包括时序数据相关性检查、模型p-q 定阶、解析式系数估计、适用性检验.试验结果表明：利用实测车速中的前40 个时序数据建立ARMA 模型，预测出的20 个车速值与实测值贴近，冲突方向两车车速归一化平均绝对误差分别为0.006 56 和0.003 4；利用全部60 个实测数据建立预测模型，检测预测值残差自相关函数发现其绝对值均小于0.258 2，表明所建车速预测方法适用.

关键词: 智能交通, 碰撞风险评估, 自回归滑动平均建模, 交叉路口, 车速预测