JP3901685B2 - LINK TRAVEL TIME PREDICTION METHOD AND DEVICE, LINK TRAVEL TIME PREDICTION PROGRAM, AND ITS RECORDING MEDIUM - Google Patents

Description

  The present invention relates to a link travel time prediction method, a device, a link travel time prediction program, and a recording medium thereof for predicting a future link travel time using current and accumulated past traffic information, where A “link” is a unit constituting a road. For example, in a VICS (registered trademark) system, a line segment connecting an intersection is defined as a “VICS link”. In the present invention, “link” represents a “VICS link” handled in the VICS system.

  Conventionally, there are the following methods as link travel time prediction methods. First, there is a method of virtualizing a road and a vehicle traveling on it using a fluid model and predicting future traffic conditions (such as traffic congestion and link travel time) by simulation. For example, one example is an input / output method (see, for example, Non-Patent Document 1).

This is because the movement of the entire vehicle on the road is treated as one fluid phenomenon, and the traffic flow confluence, branch point, bottleneck, and inflow end are used as key points in the road target section where future conditions are to be predicted. Numbering. This is a method for predicting the future traffic situation by simulating the change of traffic flow caused by passing vehicles as fluid motion on the road network with each key point as the basic unit of the traffic fluid network.
Hideki Otsuka, “Prediction of traffic congestion at Metropolitan Expressway”, Traffic Engineering vol32 special issue, p22-33,1997.

  Further, as another approach, it is an attempt to catch the change in the link travel time at each link as time series data with respect to the time variable T and to predict this by a statistical method. For example, there is a document relating to prediction of traffic congestion, which is another important traffic information along with the link travel time (see Non-Patent Document 2, for example).

  In the literature on this prediction, as one of the methods, based on the hypothesis that the traffic situation of a short time ahead is almost the same as the current traffic situation, the predicted traffic congestion situation of the prediction time ahead is expressed as the prediction processing time point, that is, A continuous prediction method presented as a traffic jam situation at the current time is described.

  Similarly, the congestion status of each link is compiled into a statistical mode value (or average value) for each day of the week or day of the week, and the congestion status that matches the time attribute such as the day of the week or the time of the forecast destination time. A method of presenting statistical values (hereinafter referred to as mode output type statistical prediction or average value output type statistical prediction) is also described.

These continuous prediction and statistical prediction can be used not only for traffic jam prediction but also for link travel time prediction in exactly the same way. There are other methods for predicting future link travel times using a neural network and methods based on case-based reasoning.
Hitoshi Mohri, Riki Horikoshi et al., “Traffic Prediction Method Using Case Base” IEICE Transactions Vol.J82-B No11.p1993-2001,1999

  However, these conventional link travel time prediction methods have the following problems when practical use is considered. For example, in a method of modeling a road network and a car traveling on the road and predicting the link travel time by simulating the movement of the car on the model, the inflow and outflow locations of the car such as an expressway are limited. However, if relatively detailed data can be observed for the movement of the car above it, it is possible to predict with relatively high accuracy, but in a road network where the inflow and outflow of vehicles such as ordinary roads are complicated, The modeling becomes very complicated, and in practice it is very difficult to model.

  Although the statistical method is a realistic prediction method, the continuous prediction method that is one of them has good prediction accuracy for a short time ahead, but the prediction destination time width becomes longer in the prediction for a long time, Prediction accuracy deteriorates.

  In addition, in a state where there is no traffic congestion or in a state where traffic congestion is severe and noronoro driving continues, the link travel time does not change much, so a relatively good prediction accuracy can be expected, but the traffic congestion situation changes every moment Then, there is a problem that good prediction accuracy cannot be expected.

  The mode output type statistical prediction and the average value output type statistical prediction method can predict with almost constant prediction accuracy from short time to long time, but sudden traffic accidents and sudden weather There is a problem that it is difficult to incorporate the influence of traffic regulations due to deterioration into the prediction process, and as a result, it is possible to predict traffic conditions only with relatively low prediction accuracy.

  As described above, each prediction method has advantages and disadvantages. In particular, according to the statistical method, the prediction accuracy varies greatly according to the prediction destination time width and the traffic congestion situation.

  An object of the present invention is to provide a link travel time prediction method and apparatus, a link travel time prediction program, and a recording medium for the link travel time prediction, which have solved the problem that the prediction accuracy depends on the predicted destination time width and traffic congestion situation. It is in.

  In order to solve the above-mentioned problem, the present invention uses the past traffic information accumulated for a specified period, and the prediction accuracy of two or more different link travel time prediction methods is calculated for each prediction target link. And a procedure for selecting an optimal link travel time prediction method in each prediction situation by calculating for each prediction destination time width and each prediction execution time link congestion degree or each prediction destination time link congestion degree. Further, as a procedure for predicting the link travel time in the next period, the prediction execution time link congestion degree or the prediction destination time link congestion degree of the prediction target link is obtained, the link, the prediction destination time width, the prediction execution time link congestion degree. Alternatively, a procedure for outputting the link travel time calculated using the prediction method selected as the optimal prediction procedure in the predicted destination time link congestion level as a predicted value is provided.

  In this way, multiple link travel time prediction methods are prepared, the situation where each method is effective is obtained by learning using past traffic information, and this is applied to actual prediction, so that the predicted destination time width and Highly accurate link travel time prediction is possible, with prediction accuracy being unaffected by traffic congestion.

More specifically, the present invention has the following configuration. First, the link travel time prediction method of the present invention is a link travel time prediction method for predicting a future link travel time using current and accumulated past traffic information,
A procedure to calculate the degree of link congestion (hereinafter referred to as link congestion degree) using the congestion information and congestion length information included in the traffic information,
For each prediction target link, for each time from the prediction execution time to the prediction destination time (hereinafter referred to as the prediction destination time width), and for each link congestion degree at the prediction execution time (hereinafter referred to as the prediction execution time link congestion degree). In addition,
(1) two or more different link travel time prediction procedures;
(2) For each link travel time prediction procedure, a procedure for calculating the prediction accuracy of each link travel time procedure in the period by comparing the predicted link travel time in the specified period and the actual link travel time;
(3) A procedure for selecting an optimal link travel time prediction procedure from the calculated prediction accuracy is provided.
To predict the link travel time for the next period,
(4) Calculate the prediction execution time link congestion degree of the prediction target link,
(5) The link travel time calculated using the prediction procedure selected as the optimal prediction procedure in the link, the prediction destination time width, and the prediction execution time link congestion degree according to the selection procedure in (3) above. And a procedure for outputting as a predicted value.

Secondly, the link travel time prediction method of the present invention has a method for predicting a traffic congestion state of a prediction target link (hereinafter referred to as a traffic congestion prediction method) inside or outside, and the procedure (1) described above, (2) and (3) are performed for each prediction target link, each prediction destination time width, and each link congestion degree at the prediction destination time (hereinafter referred to as prediction destination time link congestion degree),
To predict the link travel time for the next period,
The forecast destination time link traffic congestion degree of the target link is obtained from the traffic congestion prediction method,
The link travel time calculated by using the prediction procedure selected as the optimal prediction procedure for the link, the prediction destination time width, and the prediction destination time link congestion degree by the selection procedure of (3) above is calculated as a predicted value. And a procedure for outputting as a feature.

  3rdly, the link travel time prediction method of this invention is a link travel time in the prediction destination time of the said link as a prediction method among one of several link travel time prediction methods. And a procedure for outputting as a feature.

4thly, the link travel time prediction method of this invention is included in traffic information from the past traffic information by which the prediction object link was accumulate | stored as one prediction method among the multiple link travel time prediction methods. Procedures to calculate link travel time averages and mode statistics for each time attribute such as day of the week and time,
And a procedure for outputting statistical information of a time attribute that matches the time attribute of the prediction destination time as a link travel time at the prediction destination time.

Fifth, the link travel time predicting device of the present invention is a link travel time predicting device that predicts a future link travel time using current and accumulated past traffic information,
A means for calculating the degree of link congestion (hereinafter referred to as link congestion degree) using the congestion information and congestion length information included in the traffic information,
For each prediction target link, for each time from the prediction execution time to the prediction destination time (hereinafter referred to as the prediction destination time width), and for each link congestion degree at the prediction execution time (hereinafter referred to as the prediction execution time link congestion degree). In addition,
(A) two or more different link travel time prediction means;
(B) For each link travel time prediction means, a means for calculating the prediction accuracy of each link travel time manual method in the period by comparing the predicted link travel time in the specified period and the actual link travel time;
(C) a means for selecting an optimal link travel time prediction procedure from the calculated prediction accuracy;
As a way to predict the link travel time for the next period,
(D) Calculate the predicted execution time link congestion level of the target link,
(E) The link travel time calculated by using the prediction means selected as the optimal prediction means in the link, the prediction destination time width, and the prediction execution time link congestion degree by the selection means in (c) above. And means for outputting as a predicted value.

Sixth, the link travel time predicting device of the present invention has a device for predicting a traffic jam situation of a prediction target link (hereinafter referred to as a traffic jam predicting device) inside or outside, and the means (b) described above, (B) and (c) are performed for each prediction target link, each prediction destination time width, and each link congestion degree at the prediction destination time (hereinafter referred to as prediction destination time link congestion degree)
As a way to predict the link travel time for the next period,
Obtain the predicted destination link congestion level of the target link from the congestion prediction method or device,
The link travel time calculated by the prediction means selected as the optimal prediction means in the link, the prediction destination time width, and the prediction destination time link congestion degree by the selection means of (c) is calculated as a predicted value. As an output means.

  7thly, the link travel time prediction apparatus of this invention is a link travel time in prediction execution time of the said link travel time in one prediction apparatus as link prediction time of the said link travel time as one prediction apparatus among multiple link travel time prediction apparatuses. As an output means.

Eighth, the link travel time prediction side device of the present invention is included in the traffic information from the past traffic information accumulated in the prediction target link as one prediction device among a plurality of link travel time prediction devices. A means of calculating statistical information such as the average value and mode of link travel time for each time attribute such as day of the week and time;
Means for outputting statistical information of a time attribute that matches the time attribute of the prediction destination time as a link travel time at the prediction destination time.

Ninth, the present invention is a link travel time prediction program for predicting a future link travel time using current and accumulated past traffic information,
The step of predicting the future link travel time includes the steps of executing the link travel time prediction procedures described in the first to fourth aspects.
10thly, the link travel time prediction program of this invention was recorded on the computer-readable recording medium.

  As described above, according to the present invention, according to batch processing using accumulated past traffic information, each prediction target link, each prediction destination time width, prediction execution time link congestion degree or prediction destination time link congestion degree By learning the optimal method from a plurality of link travel time prediction methods with different effective conditions for each prediction situation, and applying the learning result to the real-time processing system, It is possible to make a stable and highly accurate link travel time prediction that is not affected by prediction conditions such as traffic conditions and traffic congestion. In addition, although it takes a certain amount of processing time to create the prediction accuracy evaluation result, once it is created, it is only necessary to refer to it later, so the real-time processing system can perform high-speed processing.

  Embodiments of the present invention will be described below in detail with reference to the drawings, and the outline of each embodiment will be described. First, the first embodiment is a method of selecting an optimal link travel time prediction procedure for each prediction target link, each prediction destination time width, and each prediction execution time link congestion degree.

  In the second embodiment, a traffic jam prediction method or device is held internally or externally, and an optimum link travel time prediction procedure is selected for each prediction target link, each prediction destination time width, and each prediction destination time link congestion level. It is a method to do.

  The prediction destination time link congestion degree can be obtained by the congestion prediction method or apparatus, and using this may improve the accuracy compared to the first embodiment.

  Furthermore, the third embodiment is a method of selecting an optimal link travel time prediction procedure for each prediction target link, each prediction destination time width, and each prediction execution time link congestion level, as in the first embodiment. This is a case where the traffic jam information predicted by the traffic jam prediction method or apparatus is used for the link travel time prediction.

  In the current VICS system, travel time information is not provided for about half of the links, and in order to predict travel time for all links, the travel time is reduced by converting traffic information, which is almost 100% provided. It is necessary to calculate.

  In the third embodiment, since the traffic jam information predicted by the traffic jam prediction method or apparatus is used for link travel time prediction, it is possible to perform travel time prediction even for links for which travel time is not provided.

Furthermore, the fourth embodiment is a method of selecting an optimal link travel time prediction procedure for each prediction target link, each prediction destination time width, and each prediction destination time link congestion level. This is a case where the traffic jam information predicted by the prediction method or apparatus is used. Since the optimum procedure is selected based on the predicted destination time link congestion degree as in the second embodiment, there is a possibility of improving accuracy. The first embodiment will be described below.
[First Embodiment]
The first embodiment is a method of selecting an optimal link travel time prediction procedure for each prediction target link, each prediction destination time width, and each prediction execution time link congestion level.

  FIG. 1A shows an apparatus configuration diagram of the first embodiment, and FIG. 1B shows a processing flowchart.

  In FIG. 1-a, 100 is a traffic information center that distributes traffic information. The processing is roughly divided into two, one is a batch processing system 1000, and the other is a real-time processing system 2000. Is done.

  The batch processing system 1000 includes traffic information 120 accumulated for a specified period by the traffic information collecting means 110 described later, two or more different batch processing system link travel time prediction means 130, prediction results 140 of each link travel time prediction means, The link congestion degree calculation means 150 for calculating the link congestion degree at each prediction execution time from the traffic information 120 accumulated for the specified period, the prediction execution time link congestion degree 160 obtained as a result of the calculation means 150, the traffic information 120 and the prediction result 140, the link travel time prediction accuracy calculating means 170 for calculating the link travel time prediction accuracy, the link travel time prediction accuracy 180 obtained as a result of the calculation means 170, and the prediction accuracy of each link travel time prediction means are predicted. Prediction to be evaluated for each target link, each predicted destination time width, and each predicted execution time link congestion level Degree evaluation unit 190, the prediction accuracy evaluation result 200 obtained as a result of the evaluation unit 190 is present.

  The real-time processing system 2000 includes a traffic information collection unit 110 that collects traffic information distributed from the traffic information center 100, a real-time traffic information 210, two or more different real-time system travel time prediction units 220, and a link travel time prediction. By referring to the result 230, the link congestion degree calculation means 240 for obtaining the link congestion degree using the real-time traffic information, the calculated prediction execution time link congestion degree 250, and the prediction accuracy evaluation result 200, the prediction target link and the Predicted link travel time output switching means 260 for outputting the link travel time calculated by the prediction procedure selected as the optimal prediction procedure as the predicted value for each predicted destination time width and the predicted execution time link congestion level, the predicted link A travel time output value 270 exists.

  The traffic information center 100 is a place that regularly distributes real-time traffic information such as the degree of congestion, link travel time, and failure status, for example, every 5 minutes. In the first embodiment, the traffic information center 100 distributes VICS information. An example of the traffic information center 100 is a road traffic information communication system center.

  The traffic information collecting unit 110 receives real-time VICS traffic information sent from the traffic information center 100.

  The traffic information 120 accumulated for the specified period is the traffic information accumulated for any specified period obtained by extracting necessary information from the real-time VICS traffic information. In the first embodiment, the date, time, traffic jam information, and link travel time information are extracted from the real-time VICS traffic information, but this is not restrictive.

  Further, in the first embodiment, the traffic information accumulated for 14 months from February 1, 2001 0: 0: 0 to March 31, 2002 23:59:59 is accumulated for the specified period. The traffic information 120 is an example.

  The batch processing system link travel time prediction means 130 is composed of a plurality of link travel time prediction means 131, 132,. The batch processing system link travel time predicting means 130 designates a part of the accumulation target period of the traffic information 120 accumulated for the designated period as the forecast data generation period, and the link ahead of m minutes at each time of the forecast data generation period. Predict travel time.

The predicted link travel time is determined by the following three variables:
Variable i = 1,2 ..., n indicating which link travel time prediction means
Variable t _exe representing the predicted execution time
A variable m indicating the predicted destination time width indicating the number of minutes ahead
Present in each, and represent this _{TTp (i, t exe, m} ) and.

  In the first embodiment, the batch processing system link travel time prediction means 130 includes link travel time prediction means 1 (sustained prediction), link travel time prediction means 2 (mode value output type statistical prediction), link travel time prediction means 3. (Mean value output type statistical prediction)

  For example, the following time series data prediction method can be considered as the link travel time prediction means 4.

  (1) A travel time frequency histogram (predicted execution time travel time histogram) at the predicted execution time in the past year is created.

  (2) Create a travel time frequency histogram (predicted time travel time histogram) at the predicted time of the past year.

  (3) In the prediction execution time histogram, the cumulative degree S represented by the travel time value at the actual prediction execution time is calculated.

  (4) In the prediction destination time histogram, a value that is the cumulative degree S calculated in (3) is obtained, and the value is output as a predicted value.

  In the first embodiment, the prediction data generation period is one month from February 1, 2002 0: 0: 0 to February 28, 2002 23:59:59. 8 predicted destination time spans at 5 minute intervals (5 minutes, 10 minutes, 20 minutes, 30 minutes, 60 minutes, 90 minutes, 120 minutes, 240 minutes ahead) in the data generation period Travel time will be predicted.

  The continuous prediction in the first embodiment is a method of outputting the link travel time at the prediction execution time as it is as the predicted link travel time.

  The mode output type statistical prediction and the average value output type statistical prediction in the first embodiment are performed on February 1, 2001, 0 hours, 0 minutes, 0 seconds to 2002 years 1 in 12 months immediately before the prediction data generation period. Based on the traffic information from 23:59:59 on March 31, the link travel time mode value and link travel time average value for each day of the week / time zone is accumulated, and the day / Outputs time zone statistics.

  The day of the week is divided into seven types: Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, and Sunday, the weekday (Monday to Friday), Saturday, and Sunday, and the weekday, Saturday and Sunday. Alternatively, there is a method of collecting all the days of the week into one type, but in the first embodiment, it is divided into seven types: Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, and Sunday.

  There are various ways of dividing time zones, such as every 5 minutes, every 30 minutes, and every hour, but in the first embodiment, one day is divided into 288 time zones every 5 minutes. .

  The link travel time prediction result 140 in the designated period is the predicted link travel time output by the batch processing system link travel time prediction means 130, and is composed of the link travel time prediction results 141, 142,.

For example prediction target time width m = 10 min, for 30 minutes, the link travel time prediction result 1 for a specified period (141), the prediction data generated period t ₀ ~t period the link travel time prediction unit 1 to _e (131 ) Predicted link travel time information TTp (1, t ₀ , 10), TTp (1, t ₁ , 10), ..., TTp (1, t _e , 10) and 30 minutes ahead prediction link travel time information _{TTp (1, t 0, 30} ), TTp (1, t 1, 30), ..., representing the _{TTp (1, t e, 30} ).

Similarly, the link travel time prediction result 2 (142) in the designated period is TTp (2, t ₀ , 10), TTp (2, t ₁ , 10),... Predicted by the link travel time prediction means 2 (132). TTp (2, t _e , 10) and predicted link travel time information TTp (2, t ₀ , 30), TTp (2, t ₁ , 30), ... TTp (2, t _e , 30) 30 minutes ahead is there.

In the first embodiment, TTp (i, t _exe , m) is
i = 1,2,3
t _exe : Time every 5 minutes from February 1, 2002 _0:00:00 to February 28, 2002 23:59:59 m = 10, 20, 30, 60, 90, 120, 180, 240 [minutes]
However, this is not the case.

  The link congestion degree calculation means 150 calculates the link congestion degree of the prediction target link at each prediction execution time using the traffic information 120 accumulated for the specified period. Here, a method of calculating the link congestion degree from the VICS data used in the first embodiment will be described with reference to FIG.

  For each link, the traffic information in the VICS data includes the top coordinates of sublinks that are further divided into links (indicated by the distance from the link end / 10 [m]), and the congestion level at each traffic location (0: unknown, 1: no traffic jam, 2: congestion, 3: traffic jam). An example is shown in FIG.

  In FIG. 2, 151 is a link start point, and 152 is a link end point. The link in FIG. 2 is divided into five sublinks, and the congestion degree of each sublink is 1, 1, 2, 3, 1 from the link start point side. The top coordinates of each sublink are 10, 8, 6, 3, 1. The link congestion degree is calculated by the following formula (A).

  Therefore, the link congestion degree in the example of FIG.

  The link congestion degree 160 is the link congestion degree of the prediction target link at each prediction execution time calculated by the link congestion degree calculation unit 150.

  The congestion level in the VICS data is a quantized value of 0, 1, 2, 3, but the link congestion level is a continuous analog value. The link congestion degree can be handled as an analog value, but the link congestion degree can be divided into an arbitrary number according to an appropriate threshold value.

  In the first embodiment, the threshold value processing is performed on the link congestion degree with threshold values 1, 25, 50, 75, and 99, and the link congestion degree is divided into 6 levels of 0 to 5. That is, link congestion degree 0, link congestion degree 1 (link congestion degree 1 to 24), link congestion degree 2 (link congestion degree 25 to 49), link congestion degree 3 (link congestion degree 50 to 74), link congestion degree 4 (Link congestion degree 75 to 99) and link congestion degree 5 (link congestion degree 100) are divided into six stages.

  The link travel time prediction accuracy calculation means 170 is composed of a plurality of prediction accuracy calculation means 171-172,. The prediction accuracy calculation means 170 reads the predicted link travel time information from the link travel time prediction result 140 in the specified period, reads the actual link travel time from the traffic information 120 accumulated for the specified period, and compares the two to compare each prediction. The prediction accuracy for each prediction destination time width of the means is calculated by the following formulas (B), (C), and (D).

  Although many prediction accuracy is considered according to the objectives, such as said formula, in 1st Embodiment, prediction accuracy is calculated on the basis of said formula (B). The value calculated by the equation (B) indicates that the closer to “0”, the better the accuracy.

  In the first embodiment, for each prediction target link, the prediction accuracy is calculated for each link travel time prediction means, for each prediction destination time width, and for each predicted execution time link congestion level.

  Therefore, within the prediction data generation period, the samples used for the prediction accuracy evaluation are divided into six sets for each prediction execution time link congestion degree divided into six stages, and the prediction accuracy is calculated using the formula (B) for each set. calculate.

  The link travel time prediction accuracy 180 is the prediction accuracy calculated by the link travel time prediction accuracy calculation means 170. The link travel time prediction accuracy 180 exists for each of the following three variables.

Variable i = 1,2, ... n indicating which link travel time prediction means
Variable prediction destination time width m indicating the number of minutes ahead
Variable j = 0,1,2,3,4,5 indicating the predicted execution time link congestion level
The above variable is represented as P (i, m, j).

  Based on the link travel time prediction accuracy 180, the prediction accuracy evaluation unit 190 selects an optimal link travel time prediction unit for each prediction target link for each prediction destination time width and for each prediction execution time link congestion degree. That is, for each prediction target link, for each m and j of the link travel time prediction accuracy P (i, m, j), i for which P (i, m, j) is the lowest value is obtained. A processing flow of the prediction accuracy evaluation unit 190 will be described with reference to FIG.

  In step S191 in FIG. 3, j, which is one of the parameters of P (i, m, j), is fixed. In the first embodiment, it starts from j = 0. In step S192, m which is one of the parameters of P (i, m, j) is fixed. In the first embodiment, the process starts from m = 10.

  In step S193, i that is the lowest value is obtained from P (i, m.j) in which m and j are fixed. Here, P, which is the lowest value, is obtained from the three values P (1,10,0) and P (2,10,0) and P (3,10,0), and the value of i is output. In step S194, it is determined whether the processing has been completed for all m.

  If there is an unprocessed m, the process returns to 192 to proceed with the process for the unprocessed m. In step S195, it is determined whether the processing has been completed for all j. If there is an unprocessed j, the process returns to step S191, and the process proceeds for the unprocessed j.

  The process shown in FIG. 3 is performed on all the prediction target links, and the prediction accuracy evaluation result 200 is output. The prediction accuracy evaluation result 200 is output as a result of the prediction accuracy evaluation means 190. An explanatory diagram of an example of the prediction accuracy evaluation result 200 is shown in FIG.

  As shown in FIG. 4, the number of the link travel time prediction means showing the best accuracy is described for each prediction destination time width and each prediction execution time link congestion degree. Taking FIG. 4 as an example, when the predicted destination time width m is 10 minutes and the predicted execution time link congestion degree j is “0”, the optimal link travel time prediction means is “1”, that is, continuous prediction. Is shown.

  The prediction accuracy evaluation result 200 shown in FIG. 4 is created for all prediction target links. The real-time traffic information 210 is current real-time VICS traffic information obtained from the traffic information collection unit 110. In the first embodiment, the date, time, traffic jam information, and link travel time information are extracted from the real-time VICS traffic information, but this is not restrictive.

  The real-time processing system link travel time predicting means 220 includes a plurality of link travel time predicting means 221, 222,. The real-time processing system link travel time predicting means 220 predicts the link travel time of the requested predicted destination time width m based on the real-time traffic information 210. The predicted link travel time exists for each of the following three variables.

Variable i = 1,2, ..., n indicating which link travel time prediction means
Variable t _exe representing the predicted execution time
Prediction time span variable m indicating how many minutes ahead
And to represent the variable _{TTpnow (i, t exe, m} ) and.

  The link travel time prediction result 230 is a predicted link travel time output by the real-time processing system link travel time prediction means 220, and includes link travel time prediction results 231, 232,.

  The link congestion degree calculation means 240 calculates the link congestion degree of the prediction target link at the prediction execution time based on the real-time traffic information 210. The method of calculating the link congestion level in the first embodiment is the same as that of the link congestion level calculation means 150.

  The link congestion degree 250 is the link congestion degree of the prediction target link calculated by the link congestion degree calculation unit 240 at the prediction execution time. Similar to the link congestion degree 160, the link congestion degree 250 is also subjected to threshold processing with threshold values 1, 25, 50, 75, and 99, and the link congestion degree is divided into 6 levels of 0 to 5.

  That is, link congestion degree 0, link congestion degree 1 (link congestion degree 1 to 24), link congestion degree 2 (link congestion degree 25 to 49), link congestion degree 3 (link congestion degree 50 to 74), link congestion degree 4 (link congestion degree 75 to 99) and link congestion degree 5 (link congestion degree 100).

  The predicted link travel time output switching means 260 uses the prediction accuracy evaluation result 200 and the predicted execution time link congestion degree 250 created by the batch processing, and finally outputs the predicted link from the link travel time prediction result 230. This is a process for obtaining travel time.

  If the explanatory diagram of the prediction accuracy evaluation result shown in FIG. 4 is an example of the prediction accuracy evaluation result 200 of the prediction target link, when the prediction destination time width is 20 minutes and the prediction execution time link congestion degree is “0”, The predicted link travel time finally output according to FIG. 4 is the link travel time prediction result 2 (232).

  At this time, if the predicted execution time link congestion degree is 2, the link travel time prediction result 1 (231) is output.

The predicted link travel time output value 270 is the final predicted link travel time output as a result of the predicted link travel time output switching means 260.
[Second Embodiment]
The second embodiment of the present invention is a method for selecting a link travel time prediction procedure that is optimal for each prediction target link, each prediction destination time width, and each prediction destination time link congestion degree by holding a traffic jam prediction method or apparatus internally or externally. It is.

  FIG. 5-a shows a device configuration diagram of the second embodiment, and FIG. 5-b shows a processing flowchart.

  Differences between the first embodiment and the second embodiment are the following ten points, and the other points are the same as those in the first embodiment, and thus the description thereof is omitted.

"Link congestion level calculation means 150"
In the second embodiment, the link congestion degree calculation means 150 calculates the link congestion degree of the prediction target link at each prediction destination time using the traffic information 120 accumulated for the specified period. The calculation method is the same as in the first embodiment.

"Link traffic congestion degree 160"
In the second embodiment, the link congestion degree 160 is a link congestion degree of the prediction target link at each prediction destination time calculated by the link congestion degree calculation unit 150. The method for dividing the link congestion degree into six stages is the same as in the first embodiment.

"Link travel time prediction accuracy calculation means 170"
In the second embodiment, the prediction accuracy is calculated for each prediction target link for each link travel time prediction means, for each prediction destination time width, and for each prediction destination time link congestion degree. Accordingly, within the prediction data generation period, the samples used for prediction accuracy evaluation are divided into six sets for each prediction destination time link congestion degree divided into six stages, and the prediction accuracy is calculated for each set. The calculation formula for the prediction accuracy is the formula (B) as in the first embodiment.

"Link travel time prediction accuracy 180"
In the second embodiment, the variable j of the link travel time prediction accuracy P (i, m, j) represents the predicted destination time link congestion degree.

"Prediction accuracy evaluation means 190"
In the second embodiment, based on the link travel time prediction accuracy 180, an optimal link travel time prediction means is selected for each prediction target link for each prediction destination time width and each prediction destination time link congestion degree. The processing flowchart is the same as in the first embodiment, and j in FIG. 3 is the predicted destination time link congestion degree.

"Prediction accuracy evaluation result 200"
In the second embodiment, the prediction accuracy evaluation result 200 describes the number of the link travel time prediction procedure showing the best accuracy for each prediction destination time width and for each prediction destination time link congestion degree. In FIG. 4, j is the predicted destination link congestion level.

"Congestion prediction method or device 3000"
The traffic jam prediction method or device 3000 is a method or device for predicting the traffic jam status of the prediction target link. The congestion prediction method or apparatus exists inside or outside the present invention.

  Non-patent documents 1 and 2 cited in <Background Art> described above exist as methods for predicting a traffic jam situation.

"Link traffic congestion degree calculation means 240"
In the second embodiment, the link congestion degree calculation unit 240 calculates the link congestion degree of the prediction target link at the prediction destination time based on the prediction congestion information output from the congestion prediction method or the device 3000. The method of calculating the link congestion degree is the same as that of the link congestion degree calculation means 150.

"Link traffic congestion degree 250"
In the second embodiment, the link congestion degree 250 is the link congestion degree of the prediction target link at the prediction destination time calculated by the link congestion degree calculation unit 240. The method for dividing the link congestion degree into six stages is the same as in the first embodiment.

"Predictive link travel time output switching means 260"
In the second embodiment, the predicted link travel time output switching means 260 uses the prediction accuracy evaluation result 200 and the predicted destination time link congestion degree 250 created by the batch processing, and from the link travel time prediction result 230, This is a process for obtaining the predicted link travel time to be finally output. The processing flowchart is the same as in the first embodiment.
[Third Embodiment]
The third embodiment is a method of selecting an optimal link travel time prediction procedure for each prediction target link, each prediction destination time width, and each prediction execution time link congestion level. Use the traffic information predicted by the device.

  FIG. 6A shows an apparatus configuration diagram of the third embodiment, and FIG. 6B shows a processing flowchart.

  The differences between the first embodiment and the third embodiment are the following three points, and the other points are the same as those of the first embodiment, and thus the description thereof is omitted.

"Batch processing link travel time prediction means 130"
In the third embodiment, it is possible to introduce a link travel time prediction means for estimating the link travel time from the traffic jam situation at the predicted destination time. For links for which link travel time is not provided, neither continuous prediction nor statistical prediction can be applied. Therefore, when such a link is targeted for prediction, the link travel time is obtained from the traffic jam situation.

  Regarding the method of estimating the link travel time from the traffic jam situation, for example, the link destination traffic congestion degree is calculated by, for example, calculating the link travel time by multiplying the value by a constant, or the correspondence table between the link traffic congestion degree and the travel time is previously stored. A method of generating travel time based on the table and the predicted destination time link congestion level can be considered.

"Congestion prediction method or device 3000"
The traffic jam clearing prediction method or apparatus 3000 is the same as the traffic jam prediction method or apparatus 3000 described in the second embodiment.

"Real-time processing system link travel time prediction means 220"
In the third embodiment, the real-time processing system link travel time predicting means 220 uses the predicted traffic jam information output from the traffic jam prediction method or the device 3000 in addition to the real-time traffic information 210 to request the predicted destination time width m. Estimate link travel time for. By using the predicted traffic jam information, it is possible to calculate the predicted link travel time even for links for which the link travel time is not provided.
[Fourth Embodiment]
The fourth embodiment is a method of selecting an optimal link travel time prediction procedure for each prediction target link, for each prediction destination time link, and for each prediction destination time link congestion level. Use the traffic information predicted by the device.

  FIG. 7A shows an apparatus configuration diagram showing the fourth embodiment, and FIG. 7B shows a processing flowchart. Differences between the first embodiment and the fourth embodiment are the following 12 points, and the other points are the same as those in the first embodiment, and thus the description thereof is omitted.

"Batch processing link travel time prediction means 130"
In the fourth embodiment, as in the third embodiment, it is possible to introduce link travel time prediction means for estimating the link travel time from the traffic jam situation at the prediction destination time. The batch processing system link travel time prediction means 130 of the fourth embodiment is the same as that of the third embodiment, and a description thereof will be omitted.

"Link congestion level calculation means 150"
The link congestion degree calculation means 150 in the fourth embodiment is the same as that in the second embodiment.

"Link traffic congestion degree 160"
The link congestion degree 160 in the fourth embodiment is the same as that in the second embodiment.

"Link travel time prediction accuracy calculation means 170"
The link travel time prediction accuracy calculating means 170 in the fourth embodiment is the same as in the second embodiment.

"Link travel time prediction accuracy 180"
The link travel time prediction accuracy 180 in the fourth embodiment is the same as that in the second embodiment.

"Prediction accuracy evaluation means 190"
The prediction accuracy evaluation means 190 in the fourth embodiment is the same as that in the second embodiment.

"Prediction accuracy evaluation result 200"
The prediction accuracy evaluation result 200 in the fourth embodiment is the same as that in the second embodiment.

"Congestion prediction method or device 3000"
The traffic jam prediction method or device 3000 is the same as the traffic jam prediction method or device 3000 described in the second embodiment.

"Real-time processing system link travel time prediction means 220"
The real-time processing system link travel time predicting means 220 in the fourth embodiment is the same as that in the third embodiment.

"Link traffic congestion degree calculation means 240"
In the fourth embodiment, the link congestion degree calculation unit 240 calculates the link congestion degree of the prediction target link at the prediction destination time based on the prediction congestion information output from the congestion prediction method or the device 3000. The method of calculating the link congestion degree is the same as that of the link congestion degree calculation means 150.

"Link traffic congestion degree 250"
In the fourth embodiment, the link congestion degree 250 is the link congestion degree of the prediction target link calculated by the link congestion degree calculation unit 240 at the prediction destination time. The method for dividing the link congestion degree into six stages is the same as in the first embodiment.

"Predictive link travel time output switching means 260"
The predicted link travel time output switching means 260 in the fourth embodiment is the same as in the second embodiment.

  The processing procedure or apparatus means of the method of the present invention can also be realized by a computer and a program, and the program can be read by a computer-readable recording medium such as a floppy (registered trademark) disk, MO, ROM, CD, or DVD. , Can be provided by being recorded on a memory card, a removable disk, etc., and can also be provided through a network.

The apparatus block diagram which shows the 1st Embodiment of this invention. The process flowchart of 1st Embodiment of this invention. Explanatory drawing which shows the example of a calculation method of a link congestion degree. The processing flowchart of a prediction accuracy evaluation means. Explanatory drawing of a prediction accuracy evaluation result. The apparatus block diagram which shows the 2nd Embodiment of this invention. The process flowchart of 2nd Embodiment of this invention. The apparatus block diagram which shows the 3rd Embodiment of this invention. The process flowchart of 3rd Embodiment of this invention. The apparatus block diagram which shows the 4th Embodiment of this invention. The processing flowchart of 4th Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Traffic information center 110 ... Traffic information collection means 120 ... Traffic information accumulated for the designated period 130 ... Batch processing system link travel time prediction means 140 ... Link travel time prediction result 150 ... Link congestion degree calculation means 160 ... Link congestion degree 170 Link travel time prediction accuracy calculation means 180 ... Link travel time prediction accuracy 190 ... Prediction accuracy evaluation means 200 ... Prediction accuracy evaluation result 210 ... Real time traffic information 220 ... Real time processing system link travel time prediction means 230 ... Link travel time prediction result 240 ... Link traffic level calculation means 250 ... Link traffic level 260 ... Predicted link travel time output switching means 270 ... Predicted link travel time output value 1000 ... Batch processing system 2000 ... Real time processing system

Claims (10)

A link travel time prediction method that predicts future link travel time using current and accumulated past traffic information,
A procedure to calculate the degree of link congestion (hereinafter referred to as link congestion degree) using the congestion information and congestion length information included in the traffic information,
For each prediction target link, for each time from the prediction execution time to the prediction destination time (hereinafter referred to as the prediction destination time width), and for each link congestion degree at the prediction execution time (hereinafter referred to as the prediction execution time link congestion degree). In addition,
(1) two or more different link travel time prediction procedures;
(2) For each link travel time prediction procedure, a procedure for calculating the prediction accuracy of each link travel time procedure in the period by comparing the predicted link travel time in the specified period and the actual link travel time;
(3) A procedure for selecting an optimal link travel time prediction procedure from the calculated prediction accuracy is provided.
To predict the link travel time for the next period,
(4) Calculate the prediction execution time link congestion degree of the prediction target link,
(5) The link travel time calculated using the prediction procedure selected as the optimal prediction procedure in the link, the prediction destination time width, and the prediction execution time link congestion degree according to the selection procedure in (3) above. , Procedure to output as predicted value,
A link travel time prediction method characterized by comprising: The link travel time prediction method according to claim 1,
The method (1), (2), (3) according to claim 1, wherein a method for predicting a traffic congestion state of a prediction target link (hereinafter referred to as a traffic congestion prediction method) is held internally or externally, Every time, every forecast destination time span and every link congestion degree at the forecast destination time (hereinafter referred to as forecast destination time link congestion degree)
To predict the link travel time for the next period,
The forecast destination time link traffic congestion degree of the target link is obtained from the traffic congestion prediction method,
The link travel time calculated by using the prediction procedure selected as the optimal prediction procedure for the link, the prediction destination time width, and the prediction destination time link congestion degree by the selection procedure of (3) above is calculated as a predicted value. As output procedure,
A link travel time prediction method characterized by comprising: In the link travel time prediction method according to claim 1 or 2,
Among the multiple link travel time prediction methods, as one prediction method, a procedure for outputting the link travel time at the prediction execution time as the link travel time at the predicted destination time of the link,
A link travel time prediction method characterized by comprising: In the link travel time prediction method according to claim 1 or 2,
Among the multiple link travel time prediction methods, as one prediction method, the average value of the link travel time for each time attribute such as day of the week and time included in the traffic information from the past traffic information accumulated in the prediction target link To calculate statistical information such as
A procedure for outputting the time attribute statistical information that matches the time attribute of the predicted time as the link travel time at the predicted time,
A link travel time prediction method characterized by comprising: A link travel time prediction device that predicts future link travel time using current and accumulated past traffic information,
A means for calculating the degree of link congestion (hereinafter referred to as link congestion degree) using the congestion information and congestion length information included in the traffic information,
For each prediction target link, for each time from the prediction execution time to the prediction destination time (hereinafter referred to as the prediction destination time width), and for each link congestion degree at the prediction execution time (hereinafter referred to as the prediction execution time link congestion degree). In addition,
(A) two or more different link travel time prediction means;
(B) a means for calculating the prediction accuracy of each link travel time procedure in the period by comparing the predicted link travel time in the specified period and the measured link travel time for each link travel time prediction means;
(C) a means for selecting an optimum link travel time prediction means from the calculated prediction accuracy;
As a way to predict the link travel time for the next period,
(D) Calculate the predicted execution time link congestion level of the target link,
(E) The link travel time calculated by using the prediction means selected as the optimal prediction means in the link, the prediction destination time width, and the prediction execution time link congestion degree by the selection means in (c) above. , Means for outputting as a predicted value,
A link travel time prediction apparatus comprising: In the link travel time prediction device according to claim 5,
A device for predicting a traffic jam condition of a prediction target link (hereinafter referred to as a traffic jam prediction device) is held internally or externally, and means (a), (b), (c) according to claim 5 is used as the prediction target link. Every time, every forecast destination time span and every link congestion degree at the forecast destination time (hereinafter referred to as forecast destination time link congestion degree)
As a way to predict the link travel time for the next period,
Obtain the forecast destination time link congestion degree of the prediction target link from the congestion prediction device,
The link travel time calculated by the prediction means selected as the optimal prediction means in the link, the prediction destination time width, and the prediction destination time link congestion degree by the selection means of (c) is calculated as a predicted value. Means to output as
A link travel time prediction apparatus comprising: In the link travel time prediction device according to claim 5 or 6,
A means for outputting a link travel time at a prediction execution time as a link travel time at a prediction destination time of the link as one prediction device among a plurality of link travel time prediction devices,
A link travel time prediction apparatus comprising: In the link travel time prediction device according to claim 5 or 6,
Among the multiple link travel time prediction devices, as one prediction device, the average value of the link travel time for each time attribute such as day of the week and time included in the traffic information from the past traffic information accumulated in the prediction target link A means of calculating statistical information such as
Means for outputting statistical information of a time attribute that matches the time attribute of the predicted destination time as a link travel time at the predicted destination time;
A link travel time prediction apparatus comprising: A link travel time prediction program that predicts future link travel time using current and accumulated past traffic information,
The link travel time prediction program comprising the step of executing each link travel time prediction procedure according to claim 1 as a step of predicting a future link travel time. A recording medium, wherein the link travel time prediction program according to claim 9 is recorded on a computer-readable recording medium.

Images