Abstract:Traffic flow uncertainty quantification is important for making reliable decisions in transportation operations. Compared with well-studied level prediction or point prediction models, the study of uncertainty quantification that can capture the second-order fluctuations of traffic observations is still in its infancy. Current traffic flow uncertainty quantification approaches can be classified in general into distribution- or nondistribution-based. For the former, generalized autoregressive conditional heteroscedasticity (GARCH) model and stochastic volatility (SV) have been widely applied to quantify traffic flow uncertainty in terms of prediction interval, usually under a parametric Gaussian distribution assumption. However, a parametric model relies on a prespecified model structure and cannot meet the requirement raised by the time-varying traffic condition patterns. Therefore, this paper proposed a real-time traffic condition uncertainty quantification approach based on a nonparametric probability density function (PDF) estimation. For this approach, the real-time nonparametric kernel density estimation method is applied to capture the time-varying probability density of traffic flow data based on which prediction intervals are constructed in real time using the quantiles computed from the estimated time-varying nonparametric PDF. Real-world traffic flow data are applied to validate the proposed approach. The results show that the proposed approach outperforms the comparative models of an online GARCH filter and three lower and upper bound estimation (LUBE) models based on multilayer perceptron (MLP), spiking neural network (SNN), and long short-term memory networks (LSTM). The findings indicate that the quantification of traffic condition uncertainty is complementary to the conventional traffic condition level modeling, and combined, traffic level modeling and traffic uncertainty quantification can support the development of proactive and reliable transportation applications.

Real-time traffic flow uncertainty quantification: a seasonal kernel density estimation approach

Real-Time Traffic Flow Uncertainty Quantification Based on Non-Parametric Probability Density Function Estimation

Real-time Freeway Traffic State Estimation with Fixed and Mobile Sensing Data

Multimodel Ensemble for Freeway Traffic State Estimations

Real-time Traffic Condition Uncertainty Quantification Using Adaptive Grey Prediction Interval Model

Freeway Travel-Time Estimation Based on Temporal–Spatial Queueing Model

Towards a Unified Understanding of Uncertainty Quantification in Traffic Flow Forecasting

Traffic Condition Uncertainty Quantification under Nonnormal Distributions

Measurement Analysis of Traffic Flow Uncertainty on Chinese Highway Network

Real-Time Prediction of Seasonal Heteroscedasticity in Vehicular Traffic Flow Series.

One-dimensional Cellular Automaton Simulated Model for Reproducing Uncertainty of Traffic Flow

Real-Time Short-Term Traffic Speed Level Forecasting and Uncertainty Quantification Using Layered Kalman Filters

The Combined Short-Term Forecasting Approach to Traffic Flow Based on Non-Parametric Regression

Integrated Heteroscedasticity Test for Vehicular Traffic Condition Series

Adaptive Modeling of Uncertainties for Traffic Forecasting

How predictable are macroscopic traffic states: a perspective of uncertainty quantification

Study of a method for real-time dynamic forecast of traffic volume

Fuzzy State Transition and Kalman Filter Applied in Short-Term Traffic Flow Forecasting.

Modeling Seasonal Heteroscedasticity in Vehicular Traffic Condition Series Using a Seasonal Adjustment Approach

Traffic State Estimation and Uncertainty Quantification at Signalized Intersections with Low Penetration Rate Vehicle Trajectory Data

Reliable Short-Term Traffic Flow Forecasting for Urban Roads: Multivariate Generalized Autoregressive Conditional Heteroscedasticity Approach