Historical traffic flow data reconstrucion applying Wavelet Transform

TL;DR

This paper presents a wavelet-based method to reconstruct detailed historical traffic data from aggregated data, achieving high correlation and low error rates, thus improving data resolution for traffic analysis.

Contribution

It introduces a novel application of Wavelet Transform for disaggregating aggregated traffic data, enabling reconstruction of short-interval data from longer-interval data.

Findings

High correlation between reconstructed and original data (around 0.96)

Low mean absolute error (below 10%) in reconstructed data

Effective reconstruction across various aggregation intervals (10 to 80 minutes)

Abstract

Despite the importance of fundamental parameters (traffic flow, density and speed) to describe the traffic behavior, there still are some difficulties in order to obtain and store this information. Furthermore, given the type of study or the project the resolution analysis interval can vary from less than one hour to annual. To create alternatives in database structures,this article aims to present a method to reconstruct disaggregated historical data from aggregated data using Wavelet Transform. From the proposed method, it is possible to reconstruct data in short intervals from data with longer intervals,since they have the same behavior, for example, data from the same or similar highway. For such, a Detail coefficient is generated through the Discrete Wavelet Transform (DWT) with the disaggregated data. The aggregated data was reconstructed through an Approximation coefficient. After establishing these coefficients, the Inverse Wavelet Transform (IWT) is applied. The results indicated an average correlation between the reconstructed the original data of 0.960; 0.974; 0.968 and 0.960 for data initially aggregated at 10, 20, 40 and 80 minutes interval, respectively. The results also indicated an absolute mean error of 7.13%; 8.74%; 9.83% and 11.23% for data initially aggregated in 10, 20, 40 and 80 minutes, respectively. In other words, results suggest that the reconstituted data have a high correlation and a low percentage of mean absolute error with the original signal. In conclusion, the reconstruction of data from aggregated data and Wavelet Transform presents a good correlation and low average absolute error rate, comparable to traffic estimation studies (CASTRO-NETO et al., 2009; CORIC et al. 2012, LAM et al., 2006; LIM, 2001).