JP2004288178A - Travel time prediction method and program - Google Patents

Abstract

An object of the present invention is to accurately predict a travel time of a certain road section and a certain time zone on a prediction target day using a travel time statistical value.
A travel time statistic Mb for the same road section and the same time zone over a predetermined period B (three months) before the present is obtained, and a predetermined time period A (one year) before the prediction target date is calculated. The corresponding day of the same day of the week is determined, the travel time statistical value Yt of the same road section and the same time zone on the corresponding day is determined, and the predetermined period C (three months ), The travel time statistics Mc of the same road section and the same time zone are obtained, and the travel time Y is predicted by correcting the Mb based on the relationship between Yt and Mc.
[Selection] Fig. 2

Description

  The present invention relates to a method for predicting a travel time in a specific time zone on a future prediction target day for a prediction target road section.

With the recent increase in traffic demand and the development of the advanced information society, the needs for traffic control systems are becoming more complex and diverse. Traffic information that is particularly needed for information provision is the time required for a departure vehicle to reach a destination, that is, travel time, and a method for predicting this travel time has been conventionally developed.
In predicting travel time, the difference between the actual measured travel time value of the road section to be predicted at the present time and the current time zone value of the travel time statistical value for the same date or the same day of the week is predicted for the time zone to be predicted. Was done by adding to the travel time statistics.
"Travel Time Prediction Method and Experiment" IEICE Technical Meeting, Road Traffic Study Group RTA-92-18, September 24, 1992

However, in the above-mentioned method, since the travel time measurement information obtained at the present time is used, the travel time can be accurately predicted only several hours ahead where the current traffic condition is considered to continue. There is a demand to predict the travel time further.
Accordingly, an object of the present invention is to provide a travel time prediction method and a program that can accurately predict travel time using only travel time statistics.

  Further, the present invention combines the travel time predicted using the travel time statistical value and the travel time predicted using the measured travel time value, regardless of the length of the predicted time, the travel time can be accurately calculated. It is an object of the present invention to provide a travel time prediction method and a program capable of predicting travel time.

  The travel time prediction method according to the present invention obtains a travel time statistical value Mb in the same road section and the same time zone over the past predetermined period B from the present, and calculates a travel time statistical value Mb before the predetermined period A (A> B) from the prediction target date. One or more corresponding days of the same type as the prediction target day are obtained, and travel time statistics Yt of the same road section and the same time zone on the corresponding day are obtained. The travel time statistics Mc for the same road section and the same time zone over the past predetermined period C (assuming C <A) are obtained, and the travel time is corrected based on the relationship between Yt and Mc, thereby obtaining travel time. This is a method of estimating the time Y.

The corresponding date may be a single day or a plurality of days.
According to the above-described method, in order to predict the travel time in a certain road section and a certain time zone on the prediction target day, referring to the travel time statistical value, the date included in the predetermined period B before the current time point is calculated. The average Mb of travel times in the same time zone of the road section to be predicted is used. This is based on the expectation that the average Mb of travel times over the last predetermined period B will take a value close to the travel time of the forecast target day.

However, if any relationship can be found between the average Mb of the travel time of the predetermined period B and the travel time of the prediction target day, it is preferable to correct Mb using the relationship.
Therefore, one or more corresponding days before the predetermined period A from the prediction target date are selected, and the travel time statistical value Yt of the corresponding time zone in the prediction target road section in the same time zone and the corresponding date before the corresponding date are selected. Mb is corrected using the relationship with the average Mc of travel time in the same road section and the same time zone in the predetermined period C.

Accordingly, the relationship between the travel time statistical value Yt on the corresponding day before the predetermined period A and the average Mc of the travel time in the predetermined period C before the corresponding day can be slid to the present and used for prediction. it can. Thereby, it is possible to perform the correction in consideration of the periodic fluctuation factor over the predetermined period A, and it is possible to improve the accuracy of the prediction.
The types of the prediction target days may be classified into weekdays, Saturdays, and Sundays. This is because the tendency of travel time is common every weekday, Saturday, and Sunday. Further, the classification may be performed from Monday to Sunday. This is effective when the tendency of travel time is common for each day of the week.

  When there are a plurality of response dates, the date may include the direct response date before the predetermined period A from the prediction target date and the response dates before and after that. If the range of correspondence days is too wide, the peculiar relationship between the prediction target day and the direct correspondence day before the predetermined period A will be reduced by averaging. On the other hand, if the range of the correspondence day is narrowed (for example, a single day), there is a problem that if the correspondence day represents a sudden travel time, the suddenness is dragged. From this viewpoint, the range of the response date may be selected to be an appropriate range based on the system operation results.

Further, a plurality of the predetermined periods A may be selected, for example, one year ago, two years ago, and so on. This has the same effect as expanding the range of corresponding days in order to eliminate suddenness, and has the effect of not diminishing the specificity of the relationship between the prediction target date and the corresponding day before the predetermined period A.
It is desirable that the travel time statistical value Mb be taken for the same type of day in which the tendency of travel time is similar in the predetermined period B. It is desirable to take about the same kind of day with similar tendency.

The method of correcting the Mb is arbitrary, but is performed using, for example, a ratio operation or a difference operation.
Further, a weighted average value of the corrected Mb and the uncorrected Mb may be obtained. If the weight of the corrected Mb is increased, the periodic fluctuation factors over the predetermined period A will be emphasized. If the weight of the Mb before correction is increased, the recent fluctuation factors will be emphasized.
A weighted average of the travel time Y predicted by the travel time prediction method and the travel time Z predicted by another prediction method using real-time measured travel time values may be taken. The travel time Y is suitable for predicting the travel time ahead of a relatively long time, and the travel time Z predicted using the real-time measured travel time is suitable for predicting the travel time ahead of a relatively short time. Therefore, the travel time can be predicted by performing a weighted averaging process of these travel times Y and Z.

  When the predicted time t is short, the other prediction method using the real time travel time measurement value has higher accuracy, and when the predicted time t is long, the travel time predicted by the travel time prediction method of the present invention is more accurate. Therefore, if the weight for the travel time Y is increased and the weight for the travel time Z is reduced as the predicted time t increases, a continuous and accurate travel can be performed regardless of the length of the predicted time t. Time is gained.

When the predicted time t is equal to or less than a predetermined lower limit, the weight for the travel time Y may be set to 0, and when the predicted time t is equal to or more than the predetermined upper limit, the weight for the travel time Z may be set to 0.
Further, the travel time prediction method of the present invention predicts a travel time Y ahead by a predicted time t on the prediction target day based on past travel time statistics of the same road section, and performs real-time travel on the same prediction target day. This method predicts the travel time Z by predicting the travel time Z ahead by the predicted time t on the prediction target day using the measured time value, and performs a weighted averaging process on the travel time Y and the travel time Z. .

In this method, the travel time Y predicted based on the past travel time statistics is suitable for predicting the travel time ahead of a relatively long time, and the travel time Z predicted using the real-time measured travel time value. Is suitable for predicting the travel time ahead of a relatively short time, the travel time is predicted by performing a weighted averaging process of these travel times Y and Z.
It is desirable that the weight when the weighted average of the travel time Y and the travel time Z is changed according to the length of the predicted time t. If these travel times are weighted and averaged according to the length of the predicted time t, a continuous and accurate travel time can be obtained regardless of the length of the predicted time t.

When the predicted time t is short, the travel time Z predicted based on the real-time measured travel time is more accurate, and when the predicted time t is long, the travel predicted based on the past travel time statistics is used. Since the time Y is considered to be more accurate, the weight for the travel time Y should be heavier and the weight for the travel time Z should be lighter as the predicted time t becomes longer.
When the predicted time t is equal to or less than a predetermined lower limit, the weight for the travel time Y may be set to 0, and when the predicted time t is equal to or more than the predetermined upper limit, the weight for the travel time Z may be set to 0.

  Further, a travel time prediction program of the present invention is a program for realizing the travel time prediction method, which is stored and used in a computer.

  As described above, according to the present invention, the relationship between the travel time statistical value Yt of the corresponding day before the predetermined period A before the prediction target date and the average Mc of the travel time of the predetermined period C before the corresponding day is calculated. , Can be slid as it is to the average Mb of the travel time of the predetermined period B before the present, and used for the prediction of the travel time on the prediction target day. Thereby, the travel time can be corrected in consideration of the periodic fluctuation factors over the predetermined period A, and the prediction accuracy can be improved.

  Further, according to the present invention, the travel time Y predicted based on the past travel time statistics is suitable for predicting the travel time ahead of a relatively long time, and is predicted using the real-time measured travel time. Since the travel time Z is suitable for predicting a travel time ahead of a relatively short time, the travel time is predicted by performing a weighted averaging process of the travel times Y and Z, thereby obtaining the predicted time t. Precise prediction can be made regardless of the length.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
FIG. 1 is a block diagram for explaining a function of a travel time prediction device 1 that executes a method of determining a prediction target date and a prediction time zone and predicting a travel time Y for a certain prediction target road section. . As shown in the figure, the travel time prediction device 1 performs a travel time prediction method of the present invention and a conventionally known travel time prediction method in parallel, and performs a weighted averaging process on each of the obtained predicted values. Finally, the predicted travel time Y is predicted.

The travel time prediction device 1 stores travel time statistics data in a storage device 2 such as a hard disk. In addition, travel time measurement data of each road section obtained by a roadside beacon or a vehicle sensor is received in real time through a communication line.
All or a part of the functions of the travel time prediction device 1 that realizes this travel time prediction method are realized by the computer of the travel time prediction device 1 executing a program recorded on a predetermined medium such as a CD-ROM or a hard disk. Is achieved.

In the travel time prediction method of the present invention, data of travel time statistics is used.
Tables 1 and 2 show data examples of the travel time statistics.

  The data in Table 1 is obtained by classifying days into weekdays, Saturdays, and Sundays for a certain road section, and recording statistical travel time data for each classification for each time zone of 5 minutes. For example, in the time zone from 0:00 to 0:05, the travel time on weekdays is 310 seconds, and the travel time on Saturday is 303 seconds. Such data is recorded on weekdays, Saturdays and Sundays, and all time periods.

  The data in Table 2 is obtained by classifying days from Monday to Sunday for a certain road section, and recording statistical travel time data for each classification for each time zone of 5 minutes. For example, in the time zone from 0:00 to 0:05, the travel time on Monday is 302 seconds and the travel time on Tuesday is 315 seconds. Such data is recorded all day and all time.

The outline of the travel time prediction method of the present invention will be described with reference to FIG. The horizontal axis in FIG. 2 represents time. The current time point is represented by a mark, and the time point at which prediction is desired (predicted time point) is represented by a triangle. The day to which the prediction time belongs is a “prediction target date”.
The corresponding date that is a predetermined period A earlier than the predicted time ▽ is represented by Ａ. The concept of “response date” will be described later. Further, a predetermined period B before the current time point △ and the predicted time point 、 and a predetermined time period C before the corresponding date 日 are also shown.

In the example of FIG. 2, the predetermined period A is one year, and the predetermined periods B and C are three months. However, the present invention is not limited to this numerical value, and may have an arbitrary period length. However, it is necessary to satisfy B <A and C <A. Further, it is preferable that C = B.
Further, the predetermined period A is one period, but may be a plurality of periods. For example, a period of one year or two years may be set. Hereinafter, the case where the predetermined period A is a single period will be described, and then the case where the predetermined period A is plural will be described.

Here, the "response date" will be described. The "correspondence date" refers to one or more days selected from the same type of days as the prediction target date in the past. The “correspondence date” includes a “direct response date” and a “corresponding date” as described later.
Experience shows that travel times vary by day of the week during the same time period. For example, on roads such as industrial estates, travel time tends to be longer on Monday to Friday than on Saturday and Sunday during commuting hours, and on touristy roads most times on Monday to Friday. Also, Saturday and Sunday tend to travel longer. In addition, even when comparing Saturday and Sunday, there may be a difference in the tendency of travel time.

  Therefore, days are classified into weekdays, Saturdays, and Sundays. Further, a day corresponding to the prediction target day before the predetermined period A is referred to as a “corresponding day”. "Applicable date" is determined according to the calendar. For example, if the predetermined period A is six months, and if the prediction target date is February 28, the corresponding date will be August 28, six months earlier. If the predetermined period A is one year, if the prediction target date is February 28, the corresponding date is February 28 one year ago. However, when the prediction target date is February 29 and there is no corresponding date one year ago, the last day of the month is set as the relevant date.

In the “correspondence date”, a day directly corresponding to the forecast target date is referred to as a “direct correspondence date”. This is a single day. Days of the same type before and after the "direct response date" are referred to as "front and rear response dates." There are multiple days before and after.
For example, if the “direct correspondence day” is a certain Wednesday on a weekday, the following Tuesday, Monday, last Friday, etc., such as Thursday, Friday, the next Monday, etc., will be the “front and back correspondence days”. If the “direct correspondence date” is a Sunday, the next Sunday, the next Sunday, the previous Sunday, the previous previous Sunday, etc. will be the “front and rear correspondence days”.

  Table 3 shows the relationship between the prediction target date, the relevant date, and the direct correspondence date.

According to Table 3, if the prediction target day is a weekday and the corresponding day is also a weekday, the corresponding day becomes the direct corresponding day. If the prediction target day is a weekday and the corresponding day is Saturday, the previous Friday, which is the closest weekday to the relevant day, is set as the direct corresponding day. If the forecast target day is a weekday and the corresponding day is Sunday, the next Monday which is the closest weekday to the corresponding day is set as the direct corresponding day.
If the prediction target day is Saturday and the corresponding day is a weekday, the closest Saturday to the corresponding day is set as the direct corresponding day. If the prediction target day is Saturday and the corresponding day is also Saturday, the corresponding day is directly used as the corresponding day. If the prediction target day is Saturday and the corresponding day is Sunday, the Saturday before the corresponding day is set as the direct corresponding day.

If the forecast target day is Sunday and the corresponding day is a weekday, the closest Sunday to the corresponding day is set as the direct corresponding day. If the prediction target day is Sunday and the corresponding day is Saturday, the next day of Sunday of the corresponding day is set as the direct corresponding day. If the prediction target day is Sunday and the corresponding day is Sunday, the corresponding day is directly used as the corresponding day.
In the above description, the days are classified into weekdays, Saturdays, and Sundays. However, the days can be classified into seven from Monday to Sunday. This classification is effective when it is desired to make a difference in travel time for each day of the week even on weekdays.

In this case, if the “direct correspondence date” is a Monday, the next Monday, the next next Monday, the previous Monday, the previous previous Monday, and the like are the “front and rear correspondence days”.
Table 4 shows the relationship between the prediction target date, the relevant date, and the direct correspondence date.

  According to Table 4, if the prediction target day is Monday and the corresponding day is also Monday, the corresponding day becomes the direct corresponding day. If the prediction target day is Monday and the corresponding day is Tuesday, the Monday one day before the corresponding day is set as the direct corresponding day. If the applicable date is Wednesday, the Monday two days before the applicable date is set as the direct correspondence date. If the relevant date is Thursday, the Monday three days before the relevant date will be the direct corresponding date. If the relevant day is Friday, the Monday three days after the relevant day will be the direct corresponding day. If the relevant day is Saturday, the Monday two days after the relevant day is set as the direct corresponding day. If the relevant day is Sunday, the Monday one day after the relevant day is set as the direct corresponding day. Hereinafter, the same applies when the prediction target day is Tuesday to Sunday.

Next, a travel time prediction method according to the present invention will be described.
First, referring to the travel time statistics, an average Mb of travel times in the same time zone of the target road section is calculated for a plurality of days included in the predetermined period B before the current time point.
For example, if the current time point is a certain day in February and the predetermined period B is three months, data for the past three months, November, December, and January, will be targeted.

When the prediction target day is distinguished between weekdays, Saturdays, and Sundays, if the prediction target day is a weekday, the average of the travel time of the three-month weekday is taken. If the prediction target date is Saturday, the average travel time on Saturday for three months is calculated. If the prediction target date is Sunday, the average of three months of travel time on Sunday is taken.
When the prediction target day is distinguished from Monday to Sunday, for example, if the prediction target day is Monday, the average of three months of travel time on Monday is taken. If the prediction target day is Tuesday, the average of the travel time on Tuesday for three months is calculated.

Next, with reference to the travel time statistics, traffic information of the target road section in the same time zone on the corresponding day before the predetermined period A from the prediction target date is obtained. The travel time of the corresponding day is represented by Y (k). The subscript k represents a direct correspondence date or a before and after correspondence date. k = 0 indicates the direct corresponding date, k = ± 1 indicates the closest preceding and following corresponding date, k = ± 2 indicates the second closest corresponding preceding and following date, and the like.
The following average calculation is performed to determine the travel time on the corresponding day.

Yt = ΣY (k) / (k + 1)
The sum Σ takes k from −N to N. Here, N is an integer of 0 or more. If N is 0,
Yt = Y (0)
, Which is the travel time on the day of the corresponding day. If N is 1,
Yt = [Y (-1) + Y (0) + Y (+1)] / 3
And the average of the travel time on the direct response day and the travel time on the nearest preceding and following response day.

If N is set to be too large, the range of the correspondence day is widened, and a peculiar relationship between the prediction target day and the direct correspondence day before the predetermined period A cannot be obtained. For example, when the predetermined period A is one year, if N = 0, the specificity of the data on the same day one year ago can be used as it is, whereas if N is greater than 1, the same day on the same day one year ago. The specificity of the data is diminished by averaging.
On the other hand, if N is made small, there is also a problem that if the corresponding date accidentally represents a sudden travel time, the suddenness is dragged. From this viewpoint, it is better that N is large to some extent.

Therefore, it is most preferable to determine the value of N by actually implementing the method of the present invention and accumulating data.
Until now, it has been described that the predetermined period A is a single period. However, a plurality of periods may be used. In this case, a corresponding date is determined for each predetermined period A. For example, when the predetermined period A is one year and two years, the corresponding day one year ago is determined and its travel time is calculated, and the corresponding day two years ago is determined and its travel time is calculated. Then, a simple average or a weighted average of the two calculated travel times is calculated and set as the travel time Yt of the corresponding day. The weight in the case of weighted averaging may be increased as the predetermined period A is shorter, and lighter as the predetermined period A is longer.

Next, referring to the travel time statistics, an average Mc of travel times in the same time zone of the target road section is calculated for a plurality of days included in the predetermined period C before the direct correspondence date.
For example, if the current time point is a certain day in February, the predetermined period A is one year, and the predetermined period C is three months, data for three months of November, December, and January about one year ago is targeted. .
When the prediction target day is distinguished between weekdays, Saturdays, and Sundays, if the prediction target day is a weekday, the average of the travel time of the three-month weekday is taken. If the prediction target date is Saturday, the average travel time on Saturday for three months is calculated. If the prediction target date is Sunday, the average of three months of travel time on Sunday is taken.

When the prediction target day is distinguished from Monday to Sunday, for example, if the prediction target day is Wednesday, the average of travel time on Wednesday for three months is taken. If the prediction target date is Saturday, the average travel time on Saturday for three months is calculated. If the prediction target date is Sunday, the average of three months of travel time on Sunday is taken.
Next, the ratio Yt / Mc of the travel time Yt, which is obtained as described above, a predetermined period A before the prediction target date, and the average Mc of the predetermined period C, which is before the direct corresponding date, is obtained. Then, using the ratio Yt / Mc, the travel time Y is predicted by correcting the average Mb of the travel time in the predetermined period B before the current time.

Y = Mb (Yt / Mc) (1)
Alternatively, the travel time Mb may be corrected by using the difference Yt-Mc between Yt and Mc, and the travel time Mb may be corrected to predict the travel time Y.
Y = Mb + (Yt−Mc) (2)
The meaning of the above formula (1) or (2) can be explained as the following (A) or (B).

  (A) To predict the travel time of the prediction target day, referring to the travel time statistic, for a day of the same type as the prediction target day included in the predetermined period B before the current time point, the target road section , The average Mb of travel times in the same time zone is used. This is based on the expectation that the average Mb of travel times for the same type of day as the forecast target day will take a value close to the travel time of the forecast target day.

However, if there is a past or some relationship between the average Mb of the travel time of the predetermined period B and the travel time of the prediction target day, it is preferable to correct Mb using the relationship.
Therefore, Mb is corrected using the relationship between the travel time statistic data Yt on the corresponding day before the predetermined period A from the prediction target date and the average Mc of the travel time in the predetermined period C before the corresponding day.
(B) Travel time statistical value data is used to predict the travel time of the prediction target day on the corresponding day before the predetermined period A from the prediction target date. This is based on the expectation that the travel time statistics on the corresponding day will take values close to the travel times on the forecast target day.

  However, the prediction target date and the corresponding date are separated by a predetermined period A. Therefore, for the same type of day included in the predetermined period C before the corresponding date, the average Mc of the travel time in the same time zone and the same time for the same type of day included in the predetermined period B before the current time point. The long-term fluctuation appearing between the prediction target date and the corresponding date is absorbed by using the relationship between the travel time of the belt and the average Mb. (B) Above.

Although the description of (A) or (B) is different, the result is the same since the same formula is used.
Incidentally, a weighted average value of the value of the expression (1) or (2) obtained by the above correction and Mb before correction is obtained, and the final value is obtained as in the expression (3) or (4). The travel time prediction value Y can also be obtained.

Y = αMb (Yt / Mc) + (1−α) Mb (3)
Y = α [Mb + (Yt−Mc)] + (1−α) Mb (4)
Here, α is a weighting coefficient for the fluctuation tendency, and 0 ≦ α ≦ 1. If α is close to 1, the prediction will be based on the periodic fluctuation of the predetermined period A over the recent fluctuation tendency, and if α is close to 0, the prediction will be based on the recent fluctuation tendency rather than the periodic fluctuation of the predetermined period A.

  The travel time prediction method of the present invention described above uses travel time statistical value data and does not consider the current real-time traffic situation. This is because it is assumed that the travel time prediction method of the present invention is used for prediction several hours or more ahead of the present. However, for prediction within a few hours from now, a conventionally known travel time prediction method that uses actually measured travel time information is more accurate.

Here, a method of acquiring travel time information in real time will be described.
FIG. 3 is a road map showing a road section L in one direction from an intersection to an intersection. The vehicle detector and the roadside beacon are provided at any position in the road section L, for example, at one end. The vehicle detector is a sensor that emits ultrasonic waves or light pulses from above the road and measures the time required to return to detect the presence of the vehicle. A roadside beacon is a communication device that identifies a vehicle by performing two-way communication with an in-vehicle communication device, and acquires information on a roadside beacon that the vehicle has passed last time and information on the time of the passage.

FIG. 4 is a schematic diagram of a travel time acquisition device. The travel time acquisition device includes a processing device 11 including a computer and the like, a storage device 12 for storing measured travel time values and the like, and a signal for taking in signals from a vehicle detector or a road beacon installed at each point on a road under jurisdiction. An interface 13 and an input / output device 14 such as a display device and a keyboard are provided.
Several methods by which the processing device 11 collects travel time data will be described. The processing device 11 calculates a traffic amount q (the number of vehicles passing per unit time (for example, 5 minutes)) based on the detection signal of the vehicle sensor. Since the vehicle detector is provided for each road section, the traffic q is also obtained for each road section. Further, the processing device 11 detects the occupation time O (total Σtk of the time tk at which each vehicle k crosses the vehicle sensor within a unit time (for example, 5 minutes)). The processing device 11 calculates the average speed V of the vehicle by the formula V = I · q / O using the traffic volume q, the occupation time O, and the average vehicle length (constant value) I. Using the length L, the travel time T is calculated by the equation T = L / V.

If a road beacon is installed in the road section and the vehicle can be identified by two-way communication with the in-vehicle device, the time when the same vehicle passes through the end of the road section and the time when it passes through the other end of the road section The travel time required for traveling on a road section can also be obtained. If a plurality of vehicles can be identified within a unit time (for example, 5 minutes), the travel time of each vehicle is averaged.
In addition to the above, the vehicle is identified by matching the plate number of the vehicle from the measurement image of the camera, and the time when the same vehicle passes through the end of the road section and the time when the same vehicle passes through the other end of the road section are determined based on the time of the road section. The time T required to drive the vehicle may be determined. Also in this case, when a plurality of vehicles can be identified within a unit time (for example, 5 minutes), an average of travel time of each vehicle is taken.

A weighted average value may be obtained for a plurality of types of travel times obtained as described above, and this may be used as the actual travel time value. As the weight, a value determined empirically may be used.
In this way, real-time travel time information can be obtained.
Next, a conventionally known travel time prediction method based on real-time travel time information will be briefly described. In a conventionally known travel time prediction method, data of the real-time travel time actual measurement value measured by a vehicle sensor or the like is used together with travel time statistical value data.

FIG. 5 is a graph in which the horizontal axis represents the time of the day and the vertical axis represents the travel time, and shows the transition of the travel time statistics collected in a certain period in the past and the measured real-time travel time of the day. The predicted value ahead of the current time can be predicted by directly adding the difference ΔY between the travel time statistical value at the current time and the measured travel time value to the travel time statistical value.
FIG. 6 is a graph comparing the accuracy of the travel time obtained by the conventionally known travel time prediction method with the accuracy of the travel time obtained by the travel time prediction method of the present invention. This indicates that the accuracy of the travel time prediction method of the present invention increases, and the accuracy of the travel time prediction method of the present invention increases as the prediction time increases.

Therefore, the travel time obtained by the conventionally known travel time prediction method and the travel time obtained by the travel time prediction method of the present invention are determined according to the time t (predicted time) from the current time to the predicted time. Decide which to use. If the travel time obtained by the travel time prediction method of the present invention is written as Y and the travel time obtained by the conventionally known travel time prediction method is written as Z, the averaging process mY + (1-m) Z (5)
Go for travel time. The weight coefficient m (0 ≦ m ≦ 1) takes m close to 0 when the predicted time t is shorter than the predetermined reference time T1. This is because the travel time Z obtained by the conventionally known travel time prediction method uses a real-time travel time measurement value. This is because it is considered that the more accurate the travel time Z obtained by the time prediction method, the higher the prediction accuracy.

  If the predicted time t is longer than the predetermined reference time T1, m is set close to 1. This is because the travel time Y obtained by the travel time prediction method of the present invention uses the travel time statistical value in consideration of the periodic fluctuation, so that the current traffic condition is not considered to continue in the distant future. This is because it is considered that the prediction accuracy is improved by giving importance to the travel time Y obtained by the travel time prediction method of the present invention.

As the reference time T1, a time at which the accuracy of the conventionally known travel time prediction method is substantially equal to the accuracy of the travel time prediction method of the present invention is selected.
There are also the following methods. Two reference times T2 and T3 (T2 <T1 <T3) are set. When the predicted time t is shorter than T2, m = 0, and when the predicted time t is longer than T3, m = 1. And When the predicted time t is T2 <t <T3, the above equation mY + (1-m) Z
And take a weighted average. The weight coefficient m is, for example, m = (t−T2) / (T3−T2)
Set to. In this way, a predicted value of the travel time that continuously changes according to the predicted time t is obtained.

FIG. 2 is a block diagram for explaining a function of a travel time prediction device 1 that executes a method of determining a prediction target date and a prediction destination time zone and predicting the travel time Y for a certain prediction target road section. It is an outline explanatory view of a travel time prediction method of the present invention. It is a road map which shows one-way road section L from an intersection to an intersection. It is a schematic diagram of a travel time acquisition device for acquiring a travel time measurement value. In order to explain a conventionally known travel time prediction method, a horizontal axis represents time and a vertical axis represents travel time. 5 is a graph comparing the accuracy of a travel time obtained by a conventionally known travel time prediction method with the accuracy of a travel time obtained by a travel time prediction method of the present invention.

Explanation of reference numerals

1 travel time prediction device 2 storage device

Claims (19)

A method for predicting a travel time Y in a prediction target time zone of a prediction target day for a prediction target road section,
The travel time statistical value Mb of the same road section and the same time zone over the past predetermined period B from the present is obtained,
One or more corresponding dates of the same type as the prediction target date before a predetermined period A (A> B) before the prediction target date are obtained,
The travel time statistical value Yt of the same road section and the same time zone on the corresponding day is obtained,
A travel time statistic Mc of the same road section and the same time zone over a predetermined period C (C <A) in the past is determined with reference to a period before the predetermined period A from the present time.
A travel time prediction method, wherein the travel time Y is predicted by correcting the Mb based on the relationship between the Yt and the Mc.   2. The travel time prediction method according to claim 1, wherein the types of the prediction target days are classified into weekdays, Saturdays, and Sundays.   The travel time prediction method according to claim 1, wherein the type of the prediction target day is classified from Monday to Sunday.   The travel time prediction method according to any one of claims 1 to 3, wherein the plurality of corresponding days include a direct corresponding day before a predetermined period A from the prediction target date and corresponding days before and after the corresponding day.   The travel time prediction method according to claim 1, wherein a plurality of the predetermined periods A are selected.   The travel time prediction method according to any one of claims 1 to 3, wherein the travel time statistical value Mb is a statistical value obtained for the same type of day as the prediction target day in the predetermined period B.   The travel time prediction method according to any one of claims 1 to 3, wherein the travel time statistical value Mc is a statistical value obtained for the same type of day as the prediction target day in the predetermined period C.   2. The travel time prediction method according to claim 1, wherein the correction of Mb is performed by taking a ratio Yt / Mc of the Yt and Mc and multiplying the ratio by Mb.   2. The travel time prediction method according to claim 1, wherein the correction of Mb is performed by taking a difference Yt-Mc between Yt and Mc and adding the difference to Mb.   10. The travel time prediction method according to claim 1, wherein the travel time Y is predicted by calculating a weighted average value of the corrected Mb and the uncorrected Mb.   2. The travel time prediction method according to claim 1, wherein a weighted average of the travel time Y predicted by the travel time prediction method and the travel time Z predicted by another prediction method using a real-time measured travel time value is calculated.   The travel time prediction method according to claim 11, wherein the weight for the travel time Y is increased and the weight for the travel time Z is reduced as the prediction time t increases.   13. The weight for the travel time Y is set to 0 when the predicted time t is equal to or less than a predetermined lower limit, and the weight for the travel time Z is set to 0 when the predicted time t is equal to or more than a predetermined upper limit. Travel time prediction method.   A travel time prediction program which is a program stored and used in a computer, wherein the travel time prediction method according to any one of claims 1 to 13 is realized. A method for predicting travel time of a prediction target day for a prediction target road section,
Based on the past travel time statistics of the same road section, predict the travel time Y ahead by the predicted time t on the prediction target day,
Using the real-time travel time measurement value of the same forecast target day, predict the travel time Z ahead by the forecast time t of the forecast target day,
A travel time prediction method, wherein a travel time is predicted by performing a weighted averaging process of the travel time Y and the travel time Z.   16. The travel time prediction method according to claim 15, wherein a weight when the weighted average of the travel time Y and the travel time Z is calculated according to the length of the predicted time t.   17. The travel time prediction method according to claim 16, wherein the weight for the travel time Y is increased and the weight for the travel time Z is reduced as the prediction time t increases.   18. The weight for the travel time Y is set to 0 when the predicted time t is equal to or less than a predetermined lower limit value, and set to 0 when the predicted time t is equal to or more than the predetermined upper limit value. Travel time prediction method.   A travel time prediction program which is a program stored and used in a computer, wherein the travel time prediction method according to any one of claims 15 to 18 is realized.

Images