Optimizing Freeway Traffic Sensor Locations by Clustering Global-Positioning-System-Derived Speed Patterns

This paper presents a new clustering-based methodology for sensor placements on freeways for estimating travel times. The proposed methodology is applicable to both freeways with no existing deployment and freeways with existing sensor deployments, where it identifies the critical sensors that need to be regularly maintained. The freeway sections are clustered based on speed data, with neighboring sections with identical speed profiles being grouped into a cluster. A new approach to estimate freeway travel times using the final clusters, which is called the optimal placement method, is also proposed. A family of k-means clustering algorithms and a hierarchical algorithm are then explored using real-world case studies of three freeway segments in Virginia. Speed and travel-time data are obtained using Global Positioning System (GPS)-equipped probe vehicles. The clustering results indicated that the hierarchical and k-means with a priori knowledge algorithms produced the best clusters. The tradeoff plots of travel-time measures (e.g., error) versus the number of freeway sensors were generated for two travel-time estimation methods: 1) optimal placement method and 2) midpoint placement method. The optimal placement method consistently produced better travel-time estimates than the midpoint placement method for all three case studies. The travel times were also estimated using three other methods found in the literature: 1) the zone of influence method; 2) the instantaneous method; and 3) the linear method. The results showed that the optimal placement method outperformed these methods in all three case studies.