Spatial-temporal Data Mining for Traffic Speed Clustering and Prediction

Bing Zhang

doi:https://doi.org/10.21985/N2X11M

Work

Spatial-temporal Data Mining for Traffic Speed Clustering and Prediction

Public Deposited

Download PDF

Download All Files (.zip)

Spatial-temporal data mining, with data driven model and machine learning techniques, significantly benefit the traditional transportation research. This dissertation focus on three problems related to uncertain location data, lane-level traffic speed clustering and anomalous traffic speed prediction. We take a first step towards combining the uncertain location data i.e., fusing the uncertainty of moving objects location obtained from both GPS devices and roadside sensors. We develop a formal model for capturing the whereabouts in time in this setting and propose the Fused Bead (FB) model, extending the bead model based solely on GPS locations. We also present algorithms for answering traditional spatio-temporal range queries, as well as a special variant pertaining to objects locations with respect to lanes on road segments. We address the problem of efficient spatio-temporal clustering of speed data in road segments with multiple lanes. We postulate that the navigation/route plans typically reported by different providers as a single-value need not be accurate in multi-lane networks. Our methodology generates lane-aware distribution of speed from GPS data and agglomerates the basic space and time units into larger clusters. In addition, we address the problem of incorporating uncertain location data in the generation of speed profiles for vehicles on roads with multiple lanes. Moving objects location data can be obtained from different/multiple sources e.g., GPS on-board the moving objects, roadside sensors, cameras. However, each source has inherent limitations that affect the precision from pure measurement-errors, to sparsity of their distribution. Incorporating such imprecisions is paramount in any query/analytics oriented system that deals with location data. The difficulties multiply when one needs to reason about localization with lane-awareness and attempts to use the location-in-time data to enable effective navigation systems. We improve the accuracy of short-term traffic speed prediction with an novel prediction framework that is adaptive to anomalous events. A new demand feature is proposed with an anomalous events detection algorithm to collect features when anomalies occur. An artificial neural network based prediction model is introduced to incorporate demand features and traffic speed features. Our experiment demonstrate that the proposed prediction framework offer a more accuracy short-term traffic speed prediction.

Last modified