预测交通流量时间序列的组合动态建模方法

为了预测交通流量,提出一种预测交通流量的组合动态建模方法。考虑交通流量的特征,将流量时间序列分解成周期项、趋势项和混沌扰动项。采用季节性指数平滑法预测周期项和趋势项之和。该计算过程取周期为一天和一周,并用带遗忘因子的递推最小二乘法确定权重,采用邻域法预测混沌项。对实际交通流量序列的预测结果表明,交通流量与前一天和前一周的状态均存在相关性,且季节性指数平滑预测后的残差是混沌的。一周的不同统计间隔的交通流量序列预测的平均相对误差在9%以下。

Combined dynamic modeling to forecast traffic volume time series

A combined dynamic modeling was proposed to forecast the traffic volume time series. Taking the characteristic of the road traffic volume into account, the traffic volume time series was decomposed into the cyclic item, the tendentious item, and the chaotic disturbing item. The sum of cyclic and tendentious items was forcast by the seasonal index smoothing method. The cycle of the computation was set at one day and one week, and the weights were determined by the recursive least square method with the forgetting factor. The chaotic item was forecast by the adjacent domain method. The results of forecasting the real traffic volume time series show that the traffic volume is interrelated with the states of the preceding day and preceding week. The remnant after removal by the seasonal index smoothing forecast is chaotic. The averag relative error of the one week traffic volume forecast with different statistic gaps is less than 9%.