STUDY ON THE ACCELERATION OF TRAFFIC FLOW BASED ON THE EMPIRICAL DATA

By extracting data from six clips of traffc flow videos taken from Yan'an Viaduct in Shanghai, totally 4132 pieces data of velocity, headway, acceleration and velocity difference of car-following were obtained. Statistical analyses show that the value domain of acceleration is symmetric with respect to zero. In the synchronized or congesting flow, the acceleration obeys the normal distribution. While in the free flow, the distribution of acceleration has strong randomness and more amount of data with large absolute value. At different traffc flow densities, the impacts of headway, velocity and velocity difference are of different importance. Moreover, even in the same situation, these impacts on the acceleration and deceleration di er. The qualitative and quantitative levels of these impacts were summarized. The GM model and Bando model were optimized by using the empirical data. In the GM model, the parameters βand γhave little influence on the optimization result, therefore we proposed a simplified GM model without them. In order to overcome the asymmetry of the value domain of acceleration in Bando model, we proposed an improved model introducing a new parameter to reflect the desired headway. Both of the average fitting errors of these two new models are lower than 6%.