NIKOVSKI DANIEL (US)
GENG YANFENG (US)
WO2001069570A2 | 2001-09-20 |
US20080097686A1 | 2008-04-24 | |||
US20080071465A1 | 2008-03-20 | |||
US20110161261A1 | 2011-06-30 | |||
US7894980B2 | 2011-02-22 | |||
US7375649B2 | 2008-05-20 | |||
US20120290204A1 | 2012-11-15 | |||
US20120136561A1 | 2012-05-31 | |||
US20100063715A1 | 2010-03-11 |
TSUYOSHI IDÉ ET AL: "Travel-Time Prediction using Gaussian Process Regression: A Trajectory-Based Approach", SOCIETY FOR INDUSTRIAL AND APPLIED MATHEMATICS. PROCEEDINGS OF THE SIAM INTERNATIONAL CONFERENCE ON DATA MINING, 1 January 2009 (2009-01-01), Philadelphia, pages 1185, XP055108139
[CLAIMS] [Claim 1] A method for predicting future travel times of a target vehicle traveling on a route from a starting point to a destination, comprising the steps of: acquiring real-time probe data by probe vehicles traveling on a set of one or more alternative links from the starting point to the destination; estimating a current travel time for each alternative link using the real-time probe data; and predicting the future travel time of the target vehicle traveling on the route based on the current travel times for the set of one or more alternative links using a set of regression functions, wherein the steps are performed in a processor. [Claim 2] The method of claim 1, wherein each regression function is JV where a LI is a current link traveled by the target vehicle at a current time t, τ L2 is a current travel time estimate of a particular alternative link L2, Δ t is a future time for the prediction, with respect to the current time, w{ is a weight, G is a real valued function with parameters with corresponding parameters & i. [Claim 3] The method of claim 2, wherein G is a Gaussian function. [Claim 4] The method of claim 1, wherein the set of regression functions is learned offline using historical probe data. [Claim 5] The method of claim 1, further comprising: outputting the future travel time to an operator of the target vehicle. [Claim 6] A system for predicting future travel times of a target vehicle traveling on a route from a starting point to a destination, comprising: a probe vehicle configured to acquire real-time probe data for a set of one or more alternative links from the starting point to the destination; and a processor in the target car configured to predict the future travel time of the target vehicle traveling on the route based on the current travel times for the set of one or more alternative links using a set of regression functions. |
METHOD AND SYSTEM FOR PREDICTING FUTURE TRAVEL TIMES OF A TARGET VEHICLE
[Technical Field]
[0001]
This invention relates generally to vehicle traffic conditions, and more particularly to predicting future travel times of vehicles on routes.
[Background Art]
[0002]
Various methods are known for predicting future travel times for a target vehicle traveling on a route based on probe data collected by probe vehicles for the same route. The route includes links or sequential road segments. The methods include linear or nonlinear regression, non-parametric regression, neural networks, and support vector machines (SVM). The methods can use locations, current times and date of the target vehicle and compare that data to similar probe data acquired by the probe vehicle. Having a good estimate of the current travel time can significantly improve the prediction of future travel times.
[0003]
It is important to predict future travel times for each link on the route, instead of simply using the current travel time estimates for each link. This is because a particular route can take hours to travel, and the current travel time can be very inaccurate, e.g., one or two hours later, particularly as traffic conditions change. [0004]
On less traveled roads, new probe data from the probe vehicles can be very infrequent and sparse. If there is no recent probe data for a link, then it is difficult to predict the current travel time on the link. For a method that relies on an estimate of the current travel time to make a prediction of future travel times, this is a major problem.
[0005]
U.S. Patent 7,894,980 describes a method for estimating real-time travel times or traffic loads in a transportation network based on limited real-time data. That method estimates traffic flow, i.e., number of vehicles on a link per unit time, from real time data on one link and using that estimate to predict flow on a connected link based on knowledge of splitting percentages, which is the percentage of vehicles that leave the link instead of continuing along the link.
[0006]
U.S. Patent 7,375,649 describes a method for identifying a fastest possible travel route. A traveler data processor collects traffic speed data and associates the traffic speed data with road segments. The road segments collectively represent one or more possible travel routes from a start point to an end point. A forecast engine determines a predicted travel time for each of the road segments based on the traffic speed data for each of the road segments. A routing engine determines the fastest possible travel route from the start point to the end point.
[0007]
U.S. Publication 20120290204 describes a system for predicting a travel time for a traffic route including one or more road segments. A predicted travel time for each of those segments is based on traffic speed data for each road segments. A total travel time is then calculated for the route.
[0008]
U.S. Publication 20120136561 describes a method predicting traffic conditions at future times by using probabilistic techniques for road segments in real-time based on changing current conditions for a network of roads. One or more predictive Bayesian models and corresponding decision trees are created based on based on historical traffic conditions.
[0009]
U.S. Publication 20100063715 describes a method for predicting traffic on a transportation network where real time data points are missing. The missing data are estimated using a calibration model of historical data that can be periodically updated, from select links constituting a relationship vector. The missing data can be estimated off-line for part of the network.
[0010]
Most prior art method use historical data for the prediction. This is a problem because the historical data may not reflect current traffic conditions.
[Summary of Invention]
[0011]
It is desired to predict future travel times for a target vehicle on a route, e.g., sequence of streets, roads and highways, generally "a set of one or more links," using real-time probe data acquired by probe vehicles traveling on a set of one or more alternative links that substantially parallel the route traveled by the target vehicle.
[0012] It is noted that the method uses real-time data, which should be more accurate than the historical data used by conventional methods.
[0013]
The real-time probe data can be acquired by a cell phone in the probe vehicle, a vehicle navigation system, or other means. The locations and times for the probe vehicles can be sampled frequently, e.g., multiple times each minute.
[0014]
The embodiments of the invention provide a method for predicting the future travel times when there is no recent estimate of the travel time for any of the links on the route traveled by the target vehicle. This is a fundamental problem when relying on the probe data acquired by the probe vehicles.
[0015]
Specifically, future travel times of a target vehicle traveling on a route from a starting point to a destination are predicted by first acquiring, by a probe vehicle, realtime probe data for the set of one or more alternate links. A current travel time is estimated for each link. Then, the future travel time for the route is predicted using a set of regression functions.
[Brief Description of the Drawings]
[0016]
[Fig. 1]
Fig. 1 is a flow diagram of a method for predicting travel times according to embodiments of the invention; and
[Fig. 2] Fig. 2 is a schematic of possible routes and an optimal route according to embodiments of the invention.
[Description of Embodiments]
[0017]
Prediction Method and Route Optimization
As shown in Fig. 1, the embodiments of our invention provide a method for predicting future travels for a target vehicle 10 traveling on a set of one or more links (L) of a route from a starting point to a destination 102. The starting point and destination can be input by an operator of the target vehicle.
[0018]
Real-time Probe data 101 are acquired by probe vehicles 20 traveling on a set of one or more alternative links from the starting point to the destination. As defined herein, an alternative link is a nearby road segment that is traveled by the probe vehicles. The probe data includes times (f) for the probe vehicles along the set of one or more alternative links.
[0019]
A current travel time τ ' 111 for each alternative link at the current time t is estimated 110 from current or very recent probe data. The current travel time is used to predict 120 a future travel time Ti+ for each linkL along the route at a time t + Δί in the future. The future travel times can then be input to a routing procedure 130 to determine an optimal route 104, which can be output to the operator of the target vehicle. Optimality can be based on minimum travel time, minimum distance, or combinations thereof. [0020]
The steps of the method can be performed in a processor connected to memory and input/output interface as known in the art. The processor 15 can be part of a navigation system arranged in the target vehicle.
[0021]
Fig. 2 shows the set of one or more links (solid lines) from the starting point (S) to the destination (D), and the set of one or more alternative links (dashed lines).
[0022]
Geospatial Inference
To solve the prediction problem, we use the idea of geospatial inference. The idea is to predict travel times on a link using estimates of recent travel times for the alternative links.
[0023]
Our assumption is that when a road is congested, then nearby roads are likely to be congested as well. Therefore, the travel times for a link used by the target vehicle are predicted using estimates of current travel times for the alternative links, when accurate travel times for the route used by the target vehicle are not available.
[0024]
Regression Function
To use this idea, we use a regression functions for each alternative linkLN. Each regression function predicts future travel times given an estimate of the current travel time for an alternative link along with the current time and the number of minutes ( Δ t) in the future for the prediction. [0025]
For each alternative linkL2, we use the regression function
N
fiiL2 t, T L2 t Lt = ^ Wi G i t r T^ At; 0 έ )
i=l
[0026] where t is the current time, T " is the current travel time estimate of the alternative linkZ,2, Δ t is the future time for the prediction, with respect to the current time, Wi is a weight, G is a real-valued functions with corresponding parameters ' , e.g., a Gaussian function. The regression function is learned using any number of known techniques.
[0027]
To use this in practice, a travel time is predicted using each alternative link for which there is recent probe data. The various predictions are averaged, possibly using a weighted average, to obtain the estimated travel time for the link.
[0028]
Application
A practical system provides optimal routes between an arbitrary starting point and destination using geospatial inference. First, during an off-line training, prior to use, the set of regression functions fu L i is trained for each link/,; and the alternative link L j using historical probe vehicle data.
[0029] Then, during real-time operation, the system in the target vehicle periodically acquires real-time probe data for the alternative links to estimate current travel times for the route.
[0030]
When the routing procedure is given the starting point and destination, the procedure uses the predicted travel times for the links provided by the set of trained regression functions fun- Only regression functions f L1Lj , for which a recently updated current travel time on alternative link L } is available, are used.