A Framework to Capture Dynamic Traffic Trends from Historical Sensor Data
The typical approach of increasing infrastructure to alleviate traffic issues such as congestion is becoming unviable due to limited space, high cost, and associated externalities. Control and management strategies using Intelligent Transportation Systems (ITS) seek to maximize the use of existent infrastructure. Many ITS strategies, such as the deployment of information, require or benefit from knowledge about traffic patterns and trends. This study proposes a mathematical programming formulation and solution algorithm that considers multiple time intervals for the estimation of network-wide traffic states and calculation of the corresponding transition probabilities. The proposed solution enables the determination of sections of the network with high traffic variability. This enables the location of congested zones and the determination of reliable traffic flow characteristics. Results from analyzing network level data suggest a trend for congested periods and predominant traffic states in the time intervals considered. It is observed that limited route choices during these periods affected the number of traffic states. From the results set a forecasting system that considers the traffic conditions of multiple time intervals simultaneously was developed and validated with the 10-fold cross validation method. This system presents one of the applications of the joint analysis of clustering of traffic characteristics and associated transition probabilities.