A novel calibration approach based on recurrent neural network for vehicle Weigh-In-Motion system

The data processing in Weigh-In-Motion(WIM) is much more complicated, especially in data calibration of the system, because of complicated external environment. In this paper, we proposed a novel calibration approach for vehicle weigh-in motion system, based on a recurrent network, Elman network. Aiming at this complexity, a recurrent network-Elman network was applied to implement data fusion of the main factors influencing the measuring precision in WIM signals for removing environmental interference and correcting non-linearity. Elman network being discussed in this paper contains context layer with a feedback branch from hidden layer to context layer. The simulated results proved that using Elman network for WIM data correction has faster convergence speed and superior dynamic character than using Back Propagation(BP) and Radial Basis Function(RBF) networks. And the accuracy satisfies the requirement of the ASTM WIM system classification III (standard E1318-94).