Mining Dynamic Network-Wide Traffic States
Traffic sensors collect data rich in patterns and helpful information. Data analysis techniques are required to discover these patterns and study different relationships that can be used to develop or improve various traffic management and control strategies. A common objective is to optimize the system performance while trying to prevent the occurrence of undesired traffic states. Previous studies have used transition probabilities across traffic states to model different management and control strategies. However, these models considered one link of a traffic network at the time. These types of solutions might not be applicable to manage a traffic network where several links need to be analyzed simultaneously. In addition, only two of the time intervals of a day were considered when calculating transition probabilities. To address these limitations, this study proposes a mathematical programming formulation and solution algorithm for the generation of network-wide dynamic traffic states. The proposed framework enables the calculation of transition probabilities across traffic states considering multiple time intervals.