Data-driven analysis and forecasting of highway traffic dynamics

A M Avila,I Mezić

doi:10.1038/s41467-020-15582-5

Abstract

The unpredictable elements involved in a vehicular traffic system, like human interaction and weather, lead to a very complicated, high-dimensional, nonlinear dynamical system. Therefore, it is difficult to develop a mathematical or artificial intelligence model that describes the time evolution of traffic systems. All the while, the ever-increasing demands on transportation systems has left traffic agencies in dire need of a robust method for analyzing and forecasting traffic. Here we demonstrate how the Koopman mode decomposition can offer a model-free, data-driven approach for analyzing and forecasting traffic dynamics. By obtaining a decomposition of data sets collected by the Federal Highway Administration and the California Department of Transportation, we are able to reconstruct observed data, distinguish any growing or decaying patterns, and obtain a hierarchy of previously identified and never before identified spatiotemporal patterns. Furthermore, it is demonstrated how this methodology can be utilized to forecast highway network conditions.

Highlights

The unpredictable elements involved in a vehicular traffic system, like human interaction and weather, lead to a very complicated, high-dimensional, nonlinear dynamical system
A large portion of these methods belong to the class of algorithms known as dynamic mode decomposition (DMD)[45,48,49,50,51,52,53,54]
In what follows we demonstrate how the Koopman mode decomposition (KMD) can decompose the Generation Simulation (NGSIM) traffic data into dynamically important subpatterns, hidden within the data, and identify their temporal characteristics

Summary

Introduction

The unpredictable elements involved in a vehicular traffic system, like human interaction and weather, lead to a very complicated, high-dimensional, nonlinear dynamical system. Even if the state-of-the-art traffic models were accurate, typically they would require an unrealistic amount of data collection and parameter tuning to function across differing highways[5,21,25]. 25 validate our view, in that some of the shortcomings of previous research approaches are not primarily their lack of accuracy but more so their heavy dependence on parameters and large amounts of training data This renders many state-of-the-art techniques being developed today unfeasible for a large scale global implementation across differing highways[21]. Many state-of-the-art techniques often times require extensive parameter tuning and the proper pre-processing of raw data to perform adequately[21] This has lead to the common practice of removing previously computed seasonal averages, aggregating and smoothing raw data[2,20,35,36]. Many state-of-the-art benchmarks have been obtained at the highway corridor (single lane) level, over a limited number of sensor locations and are incapable of generalizing to handle the multi-lane network scenario without extensive re-training

Methods

Results

Conclusion