Prediction of traffic state variability with an integrated model-based and data-driven Bayesian framework

Abstract

Deriving statistical description of uncertainties associated with prediction of traffic states is essential to development of reliability-based intelligent transportation systems. This paper presents a Bayesian learning approach framework for predicting evolution of both traffic states and the associated variability. The proposed framework ensures the interpretability and stability of the predictions with an underlying state space model, and captures sophisticated dynamics of traffic variability via a data-driven recurrent neural network component. By maintaining the filtering structure in the specialized neural network component, the proposed integrated model overcomes the key limitations of deep learning systems by improving the data efficiency and providing interpretability. The framework is trained with a multivariate Gaussian negative log-likelihood loss function for quantifying both model and stochastic uncertainties. It is implemented and tested with actual traffic data collected from a Hong Kong Strategic Route. The case study shows that the proposed prediction framework can simultaneously retain the interpretability of the results while capture the complex dynamics of the evolution of traffic variability with the recurrent neural network component. This study contributes to the development of reliability-based intelligent transportation systems through the use of advanced statistical modeling and deep learning methods.