University of Southern California

Ugur Demiryurek

Cyrus Shahabi

Farnoush Banaei-Kashani

Bei Pan

Yan Liu

Linhong Zhu

Dingxiong Deng

Afsin Akdogan

Rose Yu

A method for traffic prediction of a road network includes receiving past traffic information corresponding to multiple locations on the road network. The method further includes determining, by a processor and based on the past traffic information, temporal characteristics of the past traffic information corresponding to changes of characteristics over time and spatial characteristics of the past traffic information corresponding to interactions between locations on the road network. The method further includes predicting predicted traffic information corresponding to a later time based on the determined temporal and spatial characteristics of the past traffic information. The method further includes receiving detected additional traffic information corresponding to the later time. The method further includes updating the temporal characteristics of the traffic information and the spatial characteristics of the traffic information based on the predicted traffic information and the detected additional traffic information.

Latent space model for road networks to predict time-varying traffic

Systems and techniques for enhancing accuracy of traffic prediction include a system of one or more computers operable to receive a request relating to traffic prediction, compare a first prediction error for a first (moving average) traffic prediction model with a second prediction error for a second (historical average) traffic prediction model, calculated using a historical data set selected from previously recorded traffic data in accordance with a day and time associated with the request, select use of the first model or the second model based on the comparison of prediction errors, and provide an output for use in traffic prediction, wherein the output comes from applying the first traffic prediction model when the first prediction error is less than the second prediction error, and the output comes from applying the second traffic prediction model when the first prediction error is not less than the second prediction error.

Traffic prediction using real-world transportation data

The present disclosure relates to a short-lived throwaway index structure for generating an index from scratch in a short period of time rather than updating an index with every location change of moving objects. Rapid index construction results from the generation of Voronoi diagrams in parallel using multiple cloud servers simultaneously.

Throwaway spatial index structure for dynamic point data

Real-time high-fidelity spatiotemporal data on transportation networks can be used to learn about traffic behavior at different times and locations, potentially resulting in major savings in time and fuel. Real-world data collected from transportation networks can be used to incorporate the data's intrinsic behavior into a time-series mining technique to enhance its accuracy for traffic prediction. For example, the spatiotemporal behaviors of rush hours and events can be used to perform a more accurate prediction of both short-term and long-term average speed on road-segments, even in the presence of infrequent events (e.g., accidents). Taking historical rush-hour behavior into account can improve the accuracy of traditional predictors by up to 67% and 78% in short-term and long-term predictions, respectively. Moreover, the impact of an accident can be incorporated to improve the prediction accuracy by up to 91%.

The class of k Nearest Neighbor (k NN) queries in spatial networks has been studied in the literature. Existing approaches for k NN search in spatial networks assume that the weight of each edge in the spatial network is constant. However, real-world edge-weights are time-dependent and vary significantly in short durations, hence invalidating the existing solutions. The problem of k NN search in time-dependent spatial networks, where the weight of each edge is a function of time, is addressed herein. Two indexing schemes (Tight Network Index and Loose Network Index) are proposed to minimize the number of candidate nearest neighbor objects and reduce the invocation of the expensive fastest-path computation in time-dependent spatial networks. We demonstrate the efficiency of our proposed solution via experimental evaluations with real-world data-sets, including a variety of large spatial networks with real traffic-data.

Efficient K-nearest neighbor search in time-dependent spatial networks

With real-world spatial networks the edge travel-times are time-dependent, where the arrival-time to an edge determines the actual travel-time on the edge. To speed up the path computation, exact and approximate techniques for computation of the fastest path in time-dependent spatial networks are presented. An exact fastest path computation technique based on a time-dependent A* search can significantly improve the computation time and storage complexity of existing approaches. Moreover, for applications with which approximate fastest path is acceptable, the approximate fastest path computation technique can improve the computation time by an order of magnitude while maintaining high accuracy (e.g., with only 7% increase in travel-time of the computed path on average). With experiments using real data-sets (including a variety of large spatial networks with real traffic data) the efficacy of the disclosed techniques for online fastest path computation is demonstrated.

Growing community of inventors

Redondo Beach, CA, United States of America

Ugur Demiryurek