A Wavelet Network Model for Short-Term Traffic Volume Forecasting

Abstract

Wavelet networks (WNs) are recently developed neural network models. WN models combine the strengths of discrete wavelet transform and neural network processing to achieve strong nonlinear approximation ability, and thus have been successfully applied to forecasting and function approximations. In this study, two WN models based on different mother wavelets are used for the first time for short-term traffic volume forecasting. The Levenberg-Marquardt algorithm is used to train the WN models because it has better efficiency than the other algorithms based on gradient descent. Using the traffic volume data collected on Interstate 80 in California, the WN models are compared with the widely used back-propagation neural network (BPNN) and radial basis function neural network (RBFNN) models. The performance evaluation is based on mean absolute percentage error (MAPE) and variance of absolute percentage error (VAPE). The test and comparison results show that the WN models consistently produce lower average MAPE and VAPE values than the BPNN and RBFNN models, suggesting that the WN models are a better predictor of accuracy, stability, and adaptability.