Abstract: Short-term traffic flow forecasting is the foundation of urban traffic control and guidance. When prediction period is shorter than 5 min, the precision of normal methods for short-term traffic flow forecasting are often difficult to satisfy the actual requirement. To improve the accuracy of short-term traffic flow prediction, considering the non-linear characteristics of short-term traffic flow, the algorithm based-on fractal theory was applied to shorten prediction time and improve the accuracy of forecasting. On the basis of the Grassberger-Procaccis (G-P) algorithm, Euclidean modular is applied to define the distance between any two points in phase space. Sifting method is adopted to calculate the Euclidean distance of selected points to increase the speed of computing. Thus, it is possible to forecast traffic flow within 2 min. Using the data of 24 hours in the segment of Xizhimen to Fuchengmen in Beijing as instance, the forecasting approach based-on fractal was applied to predict the last 30 points in data sequence with 712 effective points and the prediction precision is more than 92％.

Key words: urban traffic, fractal theory, short-term traffic flow, traffic forecasting, traffic guide and control

摘要： 城市交通诱导与控制需要短时交通流预测作为依据，当预测时间小于5 min时，常用的短时交通流预测方法往往难以满足精度要求. 为了提高短时交通流预测的精度，针对短时交通流的非线性特征，采用基于分形的方法可以缩短预测时间、提高预测精度. 在G-P算法基础上，本文利用欧式模定义相空间任意两点间的欧式距离，并采用筛选法计算备选点的欧式距离，以此提高计算速度，使预测2 min内的交通流成为可能. 以北京西直门至阜成门段一天的断面交通量为实例，应用基于分形的短时交通量预测算法，对712个有效数据点的后30点进行预测，预测精度达到92％以上.

关键词: 城市交通, 分形理论, 短时交通流, 交通预测, 交通诱导与控制