Research on Urban Road Traffic Flow Prediction Based on Wavelet Denoising and Multi-layer Perceptron

Abstract

Aims: Develop a novel traffic flow prediction model to improve the accuracy of traffic flow prediction, better assist intelligent traffic management, improve traffic efficiency, reduce traffic congestion, and thus better improve sanitation and quality of life. Study Design:  Develop an urban road traffic flow prediction model with strong predictive power and excellent stability. Place and Duration of Study: Southwest University of Science and Technology, between July 2021 and March 2022. Methodology: Adopting wavelet threshold to denoise, first decompose the original data, then perform noise filtering on the subsequences obtained after decomposing, and finally reconstruct the denoised data. Use denoised data to train a multilayer perceptron and make predictions on future data. At the same time, several representative models are selected to compare with the proposed model to verify the competitiveness of the proposed model. Results: The proposed model has the smallest prediction error in the two training sets with different temporal granularity. In addition, we are using the data after wavelet denoising for training and prediction results in a smaller prediction error than using the data without denoising. Conclusion: The proposed prediction model has strong prediction ability and generalization performance in the field of traffic flow prediction. The wavelet denoising method can effectively improve the prediction accuracy of traffic flow prediction.