Paper Info
Title
The Applied Research Of Kalman In The Dynamic Travel Time Prediction
Abstract
The dynamic travel time prediction is important contents of The intelligent Transportation System. Dynamic travel time is updating the travel time by the prediction model on the same path or segment of the journey. Different forecasting models are corresponded to different methods, and different methods are corresponded to different prediction accuracy. Contrast to the existing methods, such as historical trends method, non-parametric regression, time series, neural networks, travel time prediction method, The Kalman filtering is best in dynamic information forecasts. Due to the unpredictability of traffic Factor impacting the travel time, the change of dynamic travel time have not strict laws. while the Kalman filter can full use of travel time variation to reflect changes. The all parameters of the kalman filter are carefully analyzed in this float car test, the adjacent speed of the float car is selected as the state vector, the observation vector is obtained by converting coordinates and observation time because the speed of floating cars is instantaneous speed whose error is big, and other parameters are introduced in the article. Then the method of establishing the Kalman filter equation is proposed by the long interval GPS datum in the urban. For the continuous driving vehicles, the rear travel time can be constantly updated when the observation value of the travel time is obtain in certain segments due to continuation of time and space. Then the updated travel time will be closer to the real value. So the parameter regression smoothing approach is proposed based on the result of kalman. For Verifying accuracy of Kalman filter, three periods are selected for time of the GPS data. From 5 error targets that included the mean relative error, the mean absolute relative error, the maximum absolute relative error, the equal coefficients and the mean square root of the relative error squared, the error of the kalman filter is researched. So the results are shown that the maximum value is the maximum absolute relative error of the 14:15-14:30 time period, but its value is also only 0.015. It is shown that the forecast model has highly predicting accuracy in all periods. To further prove this conclusion, the all time periods are also studied. It can be seen from the all periods that the predicted result are lower than the statistics travel time. The main reason is excluding the data that the speed is zero in the raw data, and it is always assumed that the vehicle is running when the speed of the float car is calculated. Therefore the results of Kalman filter need to be carried out the necessary smoothing process. As the known data of three segments is used in the second improved model, the data processing should be started form the fourth segment. The predicting result shows that the accuracy of improved model is higher than the kalman.
Year
DOI
Venue
2010
10.1109/GEOINFORMATICS.2010.5567722
2010 18TH INTERNATIONAL CONFERENCE ON GEOINFORMATICS
Keywords
Field
DocType
Kalman filter, Kalman filter, ITS, GPS datum, dynamic travel time prediction, observation white noise
Data mining,State vector,Computer science,Regression analysis,Algorithm,Kalman filter,Vehicle dynamics,Smoothing,Global Positioning System,Invariant extended Kalman filter,Statistics,Approximation error
Conference
Citations 
PageRank 
References 
8
0.81
0
Authors
4
Name
Order
Citations
PageRank
Huifeng Ji181.14
Aigong Xu2116.44
Xin Sui334031.49
Lanyong Li481.48