Taking into account the multiple types of floating cars and the different vehicles’ performance among distinct vehicle types

in order to obtain more accurate link speed

this paper distinguishes vehicle types and introduces artificial neural network technology to model the velocity of the floating vehicles and link speeds on highway. We apply the floating car data from Guangzhou Airport highway for model verification

and compared with the Bayesian Network based method. The results show that the average absolute relative error (MAPE) of the link speed before correction is about 20%

the average absolute error (ABS) is about 8 km/h

the MAPE of the link speed after correction is less than 10%

and the ABS is less than 5 km/h

which shows that the proposed method has good effect.