A framework for short-term traffic flow forecasting using the combination of wavelet transformation and artificial neural networks

The main objective of this paper is to develop a framework for short-term traffic flow forecasting models with high accuracy. Due to flow oscillations, the real-time information presented to the drivers through variable message signs, etc., may not be valid by the time the driver reaches the location. On the other hand, not all compartments of the flow signal are of same importance in determining its future state. A model is developed to predict the value of traffic flow in near future (next 5–35?minutes) based on the combination of wavelet transformation and artificial neural networks. This model is called the hybrid WT-ANN. Wavelet transformation is set to denoise the flow signal, i.e., filtering the unimportant fluctuations of the flow signal. Unimportant fluctuations are those that have little or no effect on the future condition of the signal. The neural network is set and trained to use previous data for predicting future flow. To implement the system, traffic data of US-101 were used from Next Generation Simulation (NGSIM). Results show that removing the noises has improved the accuracy of the prediction to a great extent. The model was used to predict the flow in three different locations on the same highway and a different highway in a different country. The model rendered highly reliable predictions. The proposed model predicts the flow of next 5?min on the same location with 2.5% Mean Absolute Percentage Error (MAPE) and of next 35?min with less than 12% MAPE. It predicts the flow on downstream locations for next 5?min with less than 8% MAPE and for the different highway with 2.3% MAPE.