Abstract
The car-following model is essential in the traffic simulation
model. The performance of traffic simulation model is depend on
whether the car-following model is correct or not. There is a
few study on freeway traffic simulation model in Taiwan. Before
this study no other research of driver''s car-following behavior
on freeway is made. The car-following model of Taiwan freeway is
made using driver simulator system constructed by virtual
reality technique to collect data.
The variables effect car-following behavior including : 1.
distance between lead car and follow car, 2.speed differential
between lead car and follow car, 3. follow car''s speed, 4.
traffic condition. Therefore, the input variables of proposed
model of back propagation neural networks model are distance
between lead car and follow car, speed differential between lead
car and follow car, and follow car''s speed. The output variable
is follow car''s acceleration. Total six model is constructed
according to different situations.
The emperor outputs show that the nonlinear correlation between
car-following behavior''s input variables and output variable can
present by one hidden layer''s back propagation neural networks
model. The proposed mode is superior to liu''s ML matrix function
of Taiwan in mean root standard error. Besides, Traffic accident
will happen when uses output of liu''s model because some
situation the output is positive big value where correct
acceleration is near zero or negative. The proposed model is
safe than the liu''s model because no accident will occur
according the output of proposed model.