The invention relates to the technical field of
traffic engineering and intelligent traffic systems, in particular to a traditional car-following
model parameter calibration method and
system based on a variational auto-
encoder. Comprising the following steps: S1, acquiring vehicle following trajectory data including basic
time sequence information such as speed, acceleration and vehicle spacing; s2, carrying out preprocessing and sample construction on
original data, segmenting a track, and forming a training and
verification data set; s3, constructing a calibration model based on a variational auto-
encoder, wherein the model is composed of an
encoder based on a cross attention mechanism, a parameter generator and a differentiable car-following model module; and S4, training the model by adopting a joint optimization method of reconstruction loss, KL
divergence loss and prediction loss. And S5, inputting to-be-calibrated data into the trained model, outputting time-varying car-following
model parameters by a parameter generator, and constructing a calibration process through a differentiable car-following model to realize high-precision fitting of an actual track. The method has the advantages of high calibration precision, excellent efficiency and physical
interpretability of parameter results.