Markov-Based Time Series Modeling Framework for Traffic-Network State Prediction under Various External Conditions

In this paper, a Markov-based time series model (MTSM) framework is developed to predict traffic network conditions by integrating archived and real-time data under various external conditions, including weather, work zones, incidents, and special events. The model considers a Markov process to explicitly characterize the probabilistic transitions between traffic states with external conditions. Environmental observations (e.g., weather) and external events (e.g., incidents and work zones) are grouped into different scenarios based on archived traffic data, and Markov transition matrices for different scenarios are calculated to predict traffic state evolutions under different scenarios. In the proposed MTSM framework, short-term and long-term time series models are combined to forecast the traffic conditions of normal cases to consider weekly/daily trends, and the outputs are further adjusted with the Markov processes for external conditions. The evaluation results show that the error rates under normal cases are about 6%, and the error rates under external conditions are around 10%. The prediction results of the proposed model are compared with the benchmark (i.e., single short-term series model and weighted time series model without the Markov process). The performances of the proposed framework are superior to the benchmark models for both cases with and without external conditions. The proposed model is able to accurately capture recurring and nonrecurring traffic congestion. It provides a reliable prediction of traffic speeds for traffic management centers to efficiently deploy proactive traffic management strategies.