Bayesian Dynamic Linear Model with Switching for Real-Time Short-Term Freeway Travel Time Prediction with License Plate Recognition Data

Abstract

This paper presents a Bayesian inference-based dynamic linear model (DLM) with switching based on three-phase traffic flow theory to predict online short-term travel time with plate recognition data. The proposed method combines the DLM model with a Hidden Markov Model (HMM) to capture the probability of flow breakdown and delays associated with congestion. By viewing travel time fluctuations as a time-varying stochastic process due to unforeseen events (e.g., incidents, accidents, or bad weather), the proposed dynamic linear model with Markov switching (SDLM) employs the HMM to determine the optimal traffic state sequence corresponding to a given travel time and flow rate observation sequence. The experimental results based on automatic license plate recognition data of a Jingtong Expressway stretch in Beijing City suggest that the proposed method can provide accurate and reliable travel time prediction under various traffic conditions.