JP2011113547A - Traffic information estimation device, computer program for estimating traffic information, and traffic information estimation method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To properly estimate traffic information of a road link to be estimated by using an estimation parameter. <P>SOLUTION: The traffic information estimation device estimates traffic information of a road link to be estimated, based on traffic information of other road links. The device includes: an estimation system 11 which estimates the traffic information of the road link by use of the traffic information of other road links and the estimation parameter; and a learning system 12 which accumulates a combination of traffic information based on a measurement value in a possible road link and the traffic information of other road links used for estimating traffic information of the road link as learning data, and optimizes the estimation parameter by use of the accumulated learning data. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a traffic information estimation device for estimating traffic information, a computer program for traffic information estimation, and a traffic information estimation method.

As a technology for providing road traffic information to drivers, VICS (Vehicle Information and Communication System: “VICS” is a registered trademark of the foundation) is widely known.
This VICS tabulates traffic information including traffic congestion and link travel time on each route based on fixed point observation information consisting of the number of vehicles, vehicle speed, etc. collected from various roadside sensors (vehicle detectors, loop coils, etc.). The traffic information is provided to the driver by wide-area communication such as narrow-area communication using beacons or FM broadcasting.

As another technique for providing road traffic information to a driver, a traffic information estimation system using a probe car (hereinafter referred to as a probe system) is also known.
For example, as shown in Patent Documents 1 and 2, this probe system uses a vehicle (probe vehicle) that actually travels on a road as a moving body sensor, and wirelessly communicates probe information such as the current vehicle position and time. Are collected from each probe vehicle to generate road traffic information.

Japanese Patent Laid-Open No. 5-151696 Japanese Patent Laying-Open No. 2005-4467

The VICS information can be obtained only with respect to some roads such as main trunk roads where roadside sensors are installed.
On the other hand, in the probe system, since a vehicle (probe vehicle) that actually travels on the road is used as a moving body sensor, traffic information can be acquired even on a road where no roadside sensor is installed.

  However, at present, the number of probe cars is very small, and traffic information cannot be obtained for road links for which no data is obtained from either the VICS or the probe system.

  Therefore, for road links for which data is not obtained from the (VICS and probe) system, it is conceivable to estimate the traffic information based on the traffic information of another road link. For example, the traffic information of a certain road link can be correlated with the traffic information of other related road links such as a road link connected to the road link. If such a correlation is used, the traffic information of the road link to be estimated can be supplemented using the traffic information of another road link.

However, if estimation is performed as described above, estimation parameters that determine how much the traffic information of each other road link is reflected as the traffic information of the road link to be estimated are appropriately set. It will be necessary.
In order to set the estimation parameters appropriately, it is conceivable that the correlation between traffic information of road links is actually measured and investigated to identify, but such a combination of a large number of road links It is practically impossible to determine the correlation appropriately by human means.

  Therefore, an object of the present invention is to make it possible to appropriately estimate traffic information of an estimation target road link using an estimation parameter.

(1) The present invention is a traffic information estimation device that estimates traffic information of an estimation target road link based on traffic information of another road link, and uses the traffic information and estimation parameters of the other road link. , An estimation unit for estimating traffic information of the estimation target road link, traffic information based on measured values in the road link that can be the estimation target road link, and the other used for estimating the traffic information of the estimation target road link A learning unit that accumulates a combination with traffic information of the road link as learning data and optimizes the estimation parameter using the accumulated learning data.

  According to the present invention, traffic information is estimated for road links for which traffic information could not be acquired. For road links for which traffic information based on measured values can be acquired, a combination of traffic information based on the measured values and traffic information of other road links used to estimate the traffic information is stored as learning data Since the estimation parameters are optimized using the accumulated learning data, it is possible to update the estimation parameters to appropriate values as traffic information is acquired. Even if it is not set manually, an appropriate estimation parameter can be obtained.

In the traffic information estimation device, it is assumed that a road link with traffic information based on actual measurement values is acquired. The road corresponding to this road link has a plurality of lanes. For example, when the lane on the right side is a main line where the vehicle continues to run at substantially the same speed, the traffic information based on the actually obtained values is compared. However, if the left lane is a lane that leads to a branched side road and the vehicle is stagnant, the traffic information based on the measured values obtained is relatively slow. It becomes the information which shows.
Thus, even if traffic information of a certain road link is obtained based on actual measurement values, two types of traffic information may be obtained. Even if the traffic information of the related estimation target road link is estimated or the process for optimizing the estimation parameter is performed, there is a possibility that the information is not accurate.

(2) Therefore, the learning unit uses the learning data classified by a plurality of branch road links existing after branching from the road link that can be the estimation target road link, and uses the learning data. The estimation parameter may be optimized separately, and the estimation unit may be configured to estimate the traffic information of the estimation target road link using the estimation parameter optimized for each branch road link.
According to this traffic information estimation device, as described above, even if the road link is branched ahead and there is a possibility that multiple types of traffic information may be acquired from the same road link, the learning unit Since the estimation parameters are optimized for each branch road link, and the estimation unit uses the estimation parameters optimized for each branch road link, it is possible to accurately estimate the traffic information of the estimation target road link.

(3) In the case of (2), the learning unit may be configured to classify and accumulate the learning data for each of the plurality of branch road links.
That is, the learning unit classifies and accumulates the learning data according to a plurality of branch road links existing after branching after the road link that can be the estimation target road link, and the stored branch road Using the learning data for each link, the estimation parameter is optimized for each branch road link. However, if the learning data can be classified at the time of calculating the estimation parameter from the traffic information (or its attached information), it may be accumulated as a whole without being classified and accumulated according to conditions. Good.

(4) The learning unit optimizes an estimation parameter for each condition using the learning data classified according to at least one of a calendar, a time zone, and weather, and the estimation unit includes: It is preferable that the traffic information of the estimation target road link is estimated using the estimation parameters optimized according to the conditions.
Even for the same road link (same road), traffic information based on measured values with different tendencies may be obtained due to different calendars, time zones, and weather conditions. Even if the traffic information of the estimation target road link is estimated or the estimation parameter is optimized, there is a possibility that the traffic information may not be accurate. However, according to the traffic information estimation device of (4), the learning unit However, the estimation parameter is optimized according to at least one of the calendar, time zone, and weather conditions, and the estimation unit uses the estimation parameter optimized according to the condition. Traffic information can be estimated accurately.

(5) And, in the case of (4), the learning unit may be configured to store the learning data classified and stored according to at least one of calendar, time zone, and weather conditions. .
That is, the learning unit classifies and accumulates the learning data according to at least one of the calendar, time zone, and weather, and uses the accumulated learning data according to the accumulated condition, The estimation parameter is optimized for each condition. However, if the learning data can be classified at the time of calculating the estimation parameter from the traffic information (or its attached information), it may be accumulated as a whole without being classified and accumulated according to conditions. Good.

As described in (1) above, the learning unit is configured to estimate traffic information based on actual measurement values of road links that can be estimation target road links and traffic of other road links used for estimating traffic information of the estimation target road links. Since the combination with the information is accumulated as learning data, when the traffic information based on the actual measurement value is obtained, the learning data can be newly accumulated.
As described above, even when learning data is newly added to the learning database, the original learning data accumulated from old times still remains in the learning database. Therefore, if the learning process of the learning target road link is performed using all the learning data stored in the learning database, the learning accuracy may be prevented from being improved due to the influence of the original old learning data.
Therefore, it is effective to discard the learning data stored in the learning database with an old time and perform learning using the newly stored learning data. In this case, the learning data with the oldest accumulated time may be simply discarded, but it may be preferable to select the learning data to be discarded.

  (6) Therefore, the learning unit classifies and accumulates the learning data for each predetermined category, and prioritizes learning data of a category having a large number of accumulated learning data among the categories. It is preferable to optimize the estimation parameter by using the learning data that is discarded and remains without being discarded. With this configuration, it is possible to narrow down the learning data that is the basis of the learning process.

  In each of the traffic information estimation devices, for example, it is detected that the link travel speed is 100 km / h according to the traffic information of a certain road link, and the link travel speed is 60 km according to the traffic information of other road links. / H is detected. In this case, there is a difference in speed, but this is only due to the difference in the speed limit of the road, and it can be considered that the degree of congestion (traffic volume) of the road is almost the same (free).

(7) Therefore, the traffic information estimation apparatus further converts the link travel speed indicating the speed of the vehicle on the other road link into speed information that takes a value corresponding to the magnitude of the link travel speed. The conversion unit may be configured to change the conversion information according to the type of road link or the traffic information of the road link and convert it into the speed information.
In this case, since the link travel speed is converted into speed information that takes a value corresponding to the magnitude of the link travel speed, the road situation can be expressed more appropriately by the speed information.
Furthermore, according to the road link type or the road link traffic information, the conversion information is changed and converted into speed information. Therefore, the speed information suitable for the road link type or road link traffic information is converted. Can do. The types of road links include, for example, highways and general roads, and the road link traffic information includes, for example, vacant information and crowded information.

(8) In addition, the learning unit may determine the number of the learning data to be accumulated for optimization of the estimation parameter, the number of other road links in which traffic information is used in the estimation process by the estimation unit, or It is preferable to set according to the number of estimation parameters used in the estimation process by the estimation unit.
In order for the learning unit to effectively perform learning for optimizing the estimation parameter, it may be preferable that the number of learning data is large to some extent, but traffic information is used in the estimation process by the estimation unit. When the number of road links or the number of estimation parameters is small, the number of learning data may be small accordingly. According to the learning unit, the number of learning data can be reduced. The amount of information stored can be reduced.

(9) Further, the estimation unit determines a calculation method used when estimating the traffic information of the estimation target road link according to the type of the estimation target road link or the accuracy of the estimation result of the traffic information by the estimation unit. It is preferable to change and estimate traffic information.
In this case, when the correlation between the road corresponding to the estimation target road link and its surrounding road is simple, for example, when the estimation target road link is a highway, the calculation method used when estimating traffic information What is necessary is just to employ | adopt a linear model and can shorten processing time by this. On the other hand, when the correlation between the road corresponding to the estimation target road link and its surrounding roads is complicated, for example, the estimation target road link is a general road with many inflows and outflows of vehicles in the middle of the link In this case, a nonlinear model can be employed as a calculation method used when estimating traffic information.

(10) In addition, the learning unit performs the optimization process according to the number of the learning data for each road link accumulated or according to the accuracy of the traffic information estimation result by the estimation unit. It is preferable to change the frequency of performing.
Since the process of optimizing the estimation parameter by the learning unit takes time, if the number of accumulated learning data increases, the optimization process is not performed every time the learning data is accumulated. The frequency can be reduced. In addition, when the accuracy of the traffic information estimation result by the estimation unit is not preferable, the frequency of performing the optimization process is increased (for example, every time learning data is accumulated), and the traffic information estimation result by the estimation unit When the accuracy is preferable, the frequency of performing the optimization process can be reduced.

(11) Further, the estimation unit obtains the reliability of the traffic information of the estimation target road link estimated using the traffic information of other road links and the estimation parameter, and the estimation unit is based on the reliability. The traffic information of the estimation target road link is preferably estimated by changing to an estimation process different from the estimation process.
In this case, when the reliability of the traffic information estimated using the traffic information of other road links and the estimation parameters is low, the traffic information can be estimated by changing to another estimation process. In addition, as another estimation process, there exists a process performed using statistical information, for example.

(12) The estimation unit includes a first estimation processing function for estimating traffic information of an estimation target road link using traffic information of other road links and estimation parameters, and traffic information of road links acquired in the past. A second estimation processing function for estimating traffic information of the estimation target road link using statistical information obtained by statistically calculating the traffic information based on the actual measurement value in the road link. , Obtaining an error between the traffic information based on the actual measurement value and the traffic information estimated by each of the first estimation processing function and the second estimation processing function, and based on the comparison of the error, It is preferable to determine which of the first estimation processing function and the second estimation processing function is used.
In this case, when the traffic information based on the actual measurement value in the road link is obtained, the error becomes smaller with respect to the traffic information based on the actual measurement value in the first estimation processing function and the second estimation processing function. In the road link, the traffic information of the estimation target road link can be estimated.

  (13) Moreover, this invention is a computer program for functioning a computer as a traffic information estimation apparatus of any one of said (1)-(12).

  (14) The present invention is also a traffic information estimation method for estimating traffic information of an estimation target road link based on traffic information of another road link, using the traffic information and estimation parameters of the other road link. The estimation step for estimating the traffic information of the estimation target road link, the traffic information based on the actual measurement value of the road link that can be the estimation target road link, and the traffic information of the estimation target road link are used for estimating the traffic information. A traffic information estimation method comprising: a learning step of accumulating combinations with traffic information of other road links as learning data, and optimizing the estimation parameters using the accumulated learning data .

  According to the present invention, it is possible to update an estimation parameter to an appropriate value as traffic information is acquired, and an appropriate estimation parameter can be obtained without artificially setting the estimation parameter. Therefore, the traffic information of the estimation target road link can be appropriately estimated using the estimation parameter and the traffic information of the other road links.

1 is an overall configuration diagram of a traffic information system. It is a block diagram of a traffic information estimation apparatus. It is a block diagram of an input information processing part. It is a figure which shows conversion information. It is a flowchart which shows the traffic information estimation process procedure. It is a block diagram of a neural network. It is a figure which shows the example of a road link. It is a figure which shows the initial state of an estimation database. It is a figure which shows the initial state of a weight database. It is a figure which shows the VICS information and probe information which were acquired as input information. It is a figure which shows the estimation database after the 1st update. It is a figure which shows the estimation database after the 2nd update. It is a figure which shows the estimation database after the 3rd update. It is a figure which shows the part extracted as a snapshot. It is a figure which shows a learning database. It is a flowchart which shows a learning process. It is explanatory drawing of the road which consists of two lanes. It is explanatory drawing of the estimation database which rewrote a part of FIG. It is explanatory drawing of the weight database which rewrote a part of FIG. It is explanatory drawing of a learning database and a weight database. It is a flowchart explaining the change of the model by an estimation system. It is a flowchart explaining the change of the model by an estimation system. It is a flowchart explaining the selection process by an estimation system. It is a flowchart explaining the selection process by an estimation system. It is a figure which shows the conversion information different from FIG. It is a data structure figure of the data for learning with category information. It is a figure which shows the accumulation | storage process to the learning database of the data for learning at the time of classifying. It is a flowchart explaining the process for setting a learning cycle. It is a flowchart explaining the process for setting a learning cycle.

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.
[1. overall structure]
FIG. 1 is an overall configuration diagram of a traffic information system including a traffic information estimation device (central device) 1 according to the present invention. In addition to the traffic information estimation device 1, this traffic information system includes a probe vehicle 3 equipped with an in-vehicle device 2, a roadside communication device 4 that wirelessly communicates with the in-vehicle device 2, a roadside sensor 5, and the like.

  The traffic information estimation device 1 indicates one of various functions in the central device, for example, and the traffic information estimation device 1 acquires traffic information (observation information) such as VICS information and probe information, and It has a function of generating traffic information for provision for provision. The traffic information estimation device 1 may perform various traffic processing such as traffic signal control and traffic control based on the traffic information.

  The traffic information estimation device (central device) 1 is constituted by a computer having a processing device (CPU) and a storage device, and a computer program for causing the computer to function as the traffic information estimation device 1 is stored in the storage device. It is remembered. This computer program is executed by the processing device, and the processing device performs functions as the traffic information estimation device 1 by inputting and outputting to the storage device and the like. Note that the functions of the traffic information estimation apparatus (central apparatus) 1 described below are realized by the computer program unless otherwise specified.

The in-vehicle device 2 generates probe information as observation information of the probe vehicle 3 and wirelessly transmits it to the roadside communication device 4. Alternatively, the in-vehicle device 2 may transmit probe information via the mobile phone 50 connected to the in-vehicle device 2.
The probe information is traffic information including the position of the probe vehicle, the passage time of the position, the vehicle ID of the probe vehicle, and the like. The probe information may include other information such as the passing speed of the probe vehicle. A link travel time for each road link is obtained based on the position and time information of the probe information. The position of the probe vehicle is calculated based on the GPS signal received by the GPS receiver included in the in-vehicle device 2.

The roadside communication device 4 transmits / receives information to / from the in-vehicle device 2 by wireless communication. Specifically, the roadside communication device 4 receives probe information as observation information transmitted by the in-vehicle device 2 and transfers the probe information to the traffic information estimation device (central device) 1. Further, the roadside communication device 4 can acquire traffic information for provision to the vehicle from the traffic information estimation device (central device) 1 and transmit the traffic information to the in-vehicle device 2. The roadside communication device 4 and the traffic information estimation device 1 are connected by a communication line.
As described above, the probe information may be transmitted via the mobile phone 50. In this case, when the probe information is generated by the in-vehicle device 2, the probe information is transmitted to the mobile phone 50, and the mobile phone 50 50 transmits probe information to the base station 49. Then, the base station 49 transfers the probe information to the traffic information estimation device (central device) 1. In addition, the mobile phone 50 can acquire traffic information for provision to the vehicle from the traffic information estimation device (central device) 1 via the base station 51 and send the traffic information to the in-vehicle device 2.

  The roadside sensor 5 is for detecting traffic information as observation information, for example, a vehicle detector for ultrasonically detecting a vehicle passing directly below, a loop coil for detecting the vehicle by inductance change, or It consists of an image sensor that measures the traffic volume and vehicle speed by image processing of the camera video, and is installed on expressways and major arterial roads for the purpose of measuring the number of vehicles flowing into intersections and vehicle speed. Yes.

  The observation information detected by the roadside sensor 5 is transmitted to the VICS center server 6 via the communication line, and the VICS center server 6 generates VICS information based on the observation information of the roadside sensor 5. This VICS information is transmitted to the traffic information estimation device 1 via a communication line.

The VICS information is traffic information including traffic jams and link travel times on each road link. Since the VICS information is updated by acquiring observation information from the roadside sensor 5 every 5 minutes, high-density information can be obtained in time. However, the roadside sensor 5 is not installed on all roads (20% or less on main roads), and the area coverage rate is low.
On the other hand, since the probe information is acquired from the probe vehicle 3 traveling on the road, the area coverage rate can be increased. However, since the penetration rate of the in-vehicle device 2 for becoming the probe vehicle 3 is still low, only low-density data can be obtained in terms of time.

  That is, when a road link is set on a road in a predetermined area, traffic information (VICS information) is always obtained for each VICS information update unit time for a road link from which VICS information is obtained. On the other hand, as for road links for which VICS information cannot be obtained, there are road links for which probe information can be obtained as traffic information, and road links for which probe information cannot be obtained for each VICS information update unit time. become.

[2. Details of Traffic Information Estimator]
[2.1 Overall configuration of estimation device]
The traffic information estimation device 1 according to the present embodiment estimates the traffic information of a road link for which neither VICS information nor probe information is obtained, and complements the traffic information of the road link, so that all road links in the predetermined area Get traffic information. That is, when the traffic information estimation device 1 according to the present embodiment obtains traffic information for all road links in a predetermined area and acquires traffic information for some of the road links, the remaining road links Estimate traffic information.

In this way, by supplementing road link traffic information for which neither VICS information nor probe information can be obtained, more accurate traffic information can be provided to the vehicle (navigation system), or traffic control can be performed with high accuracy. Is possible.
In the following, a road link for which VICS information or probe information is not obtained and traffic information needs to be estimated and complemented is referred to as “estimation target road link”.

  As shown in FIG. 2, the traffic information estimation device 1 learns an estimation system (estimation unit) 11 for estimating traffic information of an estimation target road link and a parameter (estimation parameter) used for estimation. And a learning system 12.

  The traffic information estimation device 1 accumulates data used for learning in the learning system 12 among data in the estimation database (traffic information database) 13 and the estimation database 13 for accumulating traffic information estimated by the estimation system 11. A learning database 14 for estimation, and a weight database (estimation parameter database) 15 for accumulating estimation parameters (in this embodiment, “weights”).

[2.2 Input information processing section]
The traffic information estimation device 1 according to the present embodiment is an input information processing unit that performs processing on VICS information and probe information (hereinafter, collectively referred to as “input information”) acquired by the traffic information estimation device 1. 16 is provided.
The input information processing unit 16 performs processing for generating “speed information” of each link from the input information. The generated “speed information” is given to the estimation database 13 and used for updating the estimation database 13. It is done.

As shown in FIG. 3, the input information processing unit 16 includes a VICS information processing unit 17 that generates “speed information” from the acquired VICS information, and a probe information processing unit that generates “speed information” from the acquired probe information. 18.
The VICS information processing unit 17 converts the link travel speed of the road link from the link travel time obtained based on the VICS information, and converts the link travel speed into “speed information” of the link. Speed information conversion unit 17b.

The link travel speed calculation unit 17a of the VICS information processing unit 17 divides the link length of the road link for which the link travel speed is to be calculated by the link travel time of the road link, so that the link travel speed [km / h] Is calculated. The link lengths of all road links in the target area are stored in advance in the storage device of the device 1.
The speed information conversion unit 17b of the VICS information processing unit 17 converts the link travel speed obtained by the link travel speed calculation unit 17a into an index value (speed index information) called “speed information” (standardized speed). This conversion is performed by the speed information conversion unit 17 b referring to the “speed-speed information” conversion information 19 preset in the apparatus 1. The “speed-speed information” conversion information 19 is set as shown in FIG. 4 and will be described later in detail.

The link travel speed calculation unit 18a of the probe information processing unit 18 calculates the link travel speed of the road link based on the position and time included in the probe information.
The speed information conversion unit 18b of the probe information processing unit 18 refers to the “speed-speed information” conversion information 19 to obtain the link travel speed obtained by the link travel speed calculation unit 18a as “speed information” (standardized speed). ) To an index value (speed index information). The processing content is the same as that of the speed information conversion unit 17b of the VICS information processing unit 17.

[2.3 Speed information]
“Speed information” is an index value that takes a value according to the magnitude of the link travel speed. In this embodiment, as shown in FIG. 4, the speed information takes a value from 0 to 1, and the link travel speed. When the link travel speed is 0 [km / h], the value is “1”.
The “speed information” in FIG. 4 is not simply inversely proportional to the link travel speed. That is, the link travel speed-speed information relational expression is different depending on a plurality of speed areas of the link travel speed.
More specifically, the plurality of speed regions are a first speed region (near zero speed region) having a link travel speed of 0 [km / h] to about 10 (more strictly 10.53) [km / h]. ), A second speed region in which the link travel speed is about 10 (more precisely 10.5) [km / h] to about 60 (more precisely 57.9) [km / h], the link travel speed is about These are the three speed regions of the third speed region (free travel speed region) exceeding 60 (more precisely, 57.9) [km / h].

  The first speed region is set as a near-zero speed region where the vehicle speed is 0 or a value close thereto, and such a speed region value is detected when the road is congested. The third speed region is set as a speed region that is equal to or higher than the speed limit (60 km / h) of a general road. When the road is relatively smooth and the vehicle can be in a free-running state, such speed region is set. The value of is detected. The second speed area is set as an intermediate area between the first and third speed areas.

  If the value of the horizontal axis (link travel speed) in FIG. 4 is x and the value of the vertical axis (speed information) is y, the relational expression of link travel speed-speed information in the first speed region is y = 1− (x / 1000), the relational expression of the link travel speed-speed information in the second speed region is set to y = 1.2- (x / 50), and the link travel speed-speed information of the third speed region is The relational expression is set to y = 0.1− (x / 1000).

As described above, the link travel speed-speed information relationship line is focused on the third speed area and the second speed area. In the third speed area (inclination = 1/1000), the second speed area (inclination = 1/1000). 50), the negative slope is smaller.
That is, in the third region, which is a free speed region in which the vehicle can be in a free-running state, the rate of change in the speed information (y) with respect to the change in the link travel speed (x) (as compared to the lower second speed region) (Negative slope) is smaller.

In addition, the link travel speed-speed information relational line is focused on the first speed area and the second speed area. In the first speed area (inclination = 1/1000), the second speed area (inclination = 1/50). The inclination is smaller than that.
That is, in the first region where the vehicle speed is near zero, the rate of change in the speed information (y) with respect to the change in the link travel speed (x) (negative slope) compared to the higher-speed second speed region. Is getting smaller.

  According to the conversion information 19 set as described above, in addition to an element of simply “speed”, the road information is busy and traffic jams, or the road is empty and the vehicle is in good condition. It is also possible to have an element of the degree of traffic congestion (the degree of traffic flow).

For example, if the road travel speed is 100 km / h on a road link with a speed limit of 100 km / and the road travel speed is 60 km / h on a road link with a speed limit of 60 km / h, the speed difference is 40 km / h. Although there are vehicles on all road links, all are considered to be free-running.
On the other hand, when the link travel speed of one road link is 60 km / h and the link travel speed of the other road link is 20 km / h, the difference in speed is 40 km / h as described above. However, although the vehicle is flowing smoothly on the former road link, the latter road link is considered to be congested.

In this way, when the link travel speed reaches the free travel speed range, the relationship between the number of vehicles on the road (congestion) and the link travel speed becomes relatively low, which may be caused by other factors such as the road speed limit. The link speed is likely to change.
For this reason, for example, assuming a case where traffic information (link travel speed) of an estimation target road link is to be estimated from traffic information (link travel speed) of other related road links, the third of other road links The influence of the difference in the link travel speed in the speed region (free travel speed region) on the estimated value of the link travel speed of the estimation target road link is smaller than the difference in the link epidemic speed in the slower second speed region.

In the first speed region, as in the third speed region, when the speed is 0 to 10 [km / h], the road is considered to be considerably congested at any speed. The difference in the link travel speed in the area is less meaningful than the difference in the link epidemic speed in the high-speed second speed area.
The speed information of this embodiment can estimate the traffic information of the estimation target road link with high accuracy by reducing the slopes of the first and third speed regions.
Further, as a result of reducing the slopes of the first and third speed regions, the slope of the second speed region is increased, and the difference in the link travel speed in the second speed region can be more emphasized in the index of speed information. .

[2.4 Processing by Traffic Information Estimator]
[2.4.1 Overview of estimation process]
FIG. 5 shows a traffic information estimation method by the traffic information estimation apparatus 1 (processing apparatus thereof).
First, initial values are set in the estimation database (traffic information database) 13 and the weight database 15 (step S1).

  And if the traffic information estimation apparatus 1 acquires VICS information and probe information, the input information processing part 16 will produce | generate the speed information of each road link from VICS information and probe information (step S2). However, the road links for which speed information can be acquired in step S2 are a part of all road links in the target area, and there are road links for which speed information cannot be obtained in step S2.

  Subsequently, the speed information obtained in step S2 is set in the estimation database 13, and the estimation database DB4 is updated (step S3). Then, the estimation system 11 estimates speed information about the road link for which speed information is not obtained based on the contents of the estimation database 13 updated in step S3, and sets the estimated speed information in the estimation database 13. Then, the estimation database 13 is updated (step S4). By this step S4, speed information (a mixture of measured values and estimated values) for all road links in the target area is obtained.

The speed information for all road links obtained in step S4 is output to the outside of the device (step S5). Specifically, it is displayed on the display of the device 1 or output as information to be provided to the in-vehicle device 2.
A part of the content of the estimation database 13 updated in step S4 is accumulated as a snapshot in the learning database 14 and used as learning data by the learning system 12 (step S6).

  The processes in steps S2 to S6 are repeatedly executed every time speed information is generated. The input information processing unit 16 acquires the VICS information and the probe information in accordance with the update period (for example, 5 minutes) of the VICS information, and generates the speed information. It is repeatedly executed in accordance with the update cycle (for example, 5 minutes).

[2.4.2 Model for estimating velocity information]
FIG. 6 shows a neural network for the estimation system 11 to estimate speed information of the estimation target road link. The illustrated neural network is configured as a simple perceptron that generates an output value y by multiplying each of N input signals x _i (i: 1 to N) having values of 0 to 1 by weights w _i . .
Here, the input signal x _i is speed information of a road link other than the estimation target road link (in this embodiment, a road link connected to the estimation target road link), and the output value y is the estimation target road link. Speed information.

The weight x _i are the respective velocity information plurality of road links than the estimated target road link, a value of how to reflect what proportion of correlation high road link between the estimated target road link (e.g., A road link that constitutes the same road as the estimation target road link and is adjacent to the estimation target road link) should be set to a larger value, and a road link having a lower correlation should be set to a smaller value.

However, in the case of FIG. 6, unlike the general simple perceptron, an independent parameter w ₀ that is added to the node without being multiplied by the input signal x _i is provided. This independent parameter w ₀ expresses a factor (for example, a difference in speed limit between road links) that a factor other than speed information in a road link other than the estimation target road link gives to the speed information of the estimation target road link. Speed information can be estimated with high accuracy. Further, when learning the estimation parameter (weight), the estimation parameter is easily converged to the optimum value, and the learning process can be performed easily or at high speed.

  As described above, in order for the estimation system 11 to obtain the speed information of a certain estimation target road link, the speed information of the other road link that is slightly correlated with the estimation target road link and the other road It is only necessary to obtain a weight indicating how much the link speed information is reflected in the speed information of the estimation target road link. The speed information of the other road links is accumulated in the estimation database (traffic information database) 13, the weight is accumulated in the weight database 15, and the estimation system 11 obtains necessary information from both the databases 13 and 15. get.

  Since the above-described neural network is provided for each road link, the estimation system 11 uses the speed information about each of the plurality of road links, the speed information about the other road links, and the estimation parameters indicating the degree of contribution thereof. It is configured to be able to be estimated using (weight).

[2.4.3 Details of estimation processing]
Here, a road link connected as shown in FIG. 7 is assumed. In FIG. 7, Vx indicates road link speed information (value of 0 to 1). The subscript x in Vx indicates the link number of the road link and takes a value of 1 to 24. That is, in FIG. 7, there are 24 road links with link numbers 1 to 24. Moreover, the arrow direction of each link shows the traveling direction of the vehicle.

In FIG. 7, road links (link numbers x = 1 to 10) indicated by solid arrows are links from which VICS information can be acquired, for example, road links corresponding to expressways and main trunk roads. Further, road links (link numbers x = 11 to 24) indicated by dotted arrows are links for which VICS information cannot be acquired, and are mainly general roads other than highways and main trunk roads, for example. A road link indicated by a dotted arrow may be an estimation target road link of speed information. That is, the road link indicated by the dotted arrow does not become the estimation target road link when the probe information is obtained, and becomes the estimation target road link when the probe information is not obtained. The estimation process is executed.
Even if the VICS information can be acquired, if the VICS information cannot be acquired for some reason, it is treated as an estimation target road link.

Hereinafter, the speed information estimation procedure will be described in detail with the road link of FIG. 7 as a premise and with reference to FIG. 5 again.
First, initial values are input to the estimation database 13 and the weight database 15 by the device administrator (step S1).

  As shown in FIG. 8, the estimation database 13 has data items of “check” 31, “speed information” 32, “related road link information” 33, “link length” 34, and “learning?” 35. The value of each data item can be stored for each road link (link number).

Among the data items, “check” 31 is an item indicating whether or not the speed information of each road link is updated and the order of update. “Speed information” 32 is an item in which speed information of each road link is set.
The “related road link information” 33 indicates a road link (related road link) correlated with each road link, and here indicates a road link (link number) connected to each road link. .

The “related road link information” 33 is divided into five types of “reverse”, “A order”, “A reverse”, “B order”, and “B reverse”.
“Reverse” indicates a link number of a road link opposite to an arbitrary road link. For example, for a road link with link number 1, the road link with link number 8 becomes a “reverse” road link.
“A order” is a road link connected in the forward direction behind a certain road link. For example, for the road link with link number 1, the road links with link numbers 12, 7, and 15 are “A order”. It becomes the road link.
“A reverse” is a road link connected in the reverse direction behind a certain road link. For example, for the road link with link number 1, the road links with link numbers 11, 15, and 13 are “A reverse”. It becomes the road link.

“B order” is a road link connected in the forward direction in front of a certain road link. For example, for the road link with link number 1, the road links with link numbers 4 and 9 are “B order”. It becomes a link.
“B reverse” is a road link connected in the reverse direction in front of a certain road link. For example, for the road link with link number 1, the road links with link numbers 3 and 6 are “B reverse”. It becomes a link.

  Of the data items, “link length” 34 indicates the link length of each road link. “Learning?” 35 indicates whether or not the speed information of each road link is data (learning data) for learning by the learning unit 20 (learning system 12). Or takes the value of 1. “0” indicates that the speed information of the road link is not the learning data, and “1” indicates that the speed information of the road link is the learning data.

  The initial value input in step S1 is performed when the operation of the apparatus 1 is started or reset. Regarding the estimation database 11, among the above data items, “speed information” 32, “related road link information” 33, “link length”. An initial value is set for 34. As an initial value of the “speed information” 32, for example, 0 may be set for all road links. The initial values of “related road link information” 33 and “link length” 34 are set according to the road configuration of the target area. The initial values of “related road link information” 33 and “link length” 34 are not updated even if traffic information (speed information) including VICS information and probe information is acquired.

On the other hand, the value of “speed information” 34 is updated for all road links every time traffic information (speed information) composed of VICS information and probe information is acquired (steps S3 and S4).
The “check” 31 is initialized to 0 each time the traffic information (speed information) is acquired (step S2). “Learning?” 35 is also set to 0 or 1 each time traffic information (speed information) is acquired (step S2), depending on whether or not the speed information of each road link should be used as learning data. .

As shown in FIG. 9, in the weight database 15, a weight wi used when an estimated value of speed information for each road link is obtained from speed information of other road links is set as an initial value. The weight wi is set to “number of related road links + 1” for each road link. In FIG. 9, w ₀ is an independent parameter, and w _{1 and} subsequent weights are weights to be multiplied by the speed information of the related road links.

  Since the weight wi that is an estimation parameter is automatically updated to a more appropriate value by learning by the learning system 12, an appropriate value may be set as an initial value. Therefore, initial setting is easy.

  Now, after performing the above initialization, it is assumed that VICS information and probe information (traffic information) as shown in FIG. 10 have been obtained at a certain point in time for the road link shown in FIG. In FIG. 10, VICS information (link travel time) for road links with link numbers 1 to 10 is obtained, and probe information for road links with link number 20 is obtained. Neither VICS information nor probe information is obtained for other road links. In FIG. 10, the probe information about the road link with the link number 20 is indicated by “link travel time” in the same manner as the VICS information for easy notation.

  In the information of FIG. 10, a value “1” indicating that the road link (link number 20) where the probe information exists is set as a learning target is set, and the road link (link number) where the VICS information exists. As for 1 to 10), a value “0” indicating that it is not a learning target is set.

When the input information as shown in FIG. 10 is acquired, the “travel time” in this information is converted into “speed information” by the input information processing unit 16 (step S2), and the correspondence of the estimation database 13 is determined by the speed information. The “speed information” 32 of the road link to be updated is updated (first update; step S3). For the updated road link, “1” indicating that the speed information has been updated in the first update is set in “check” 31. For the road link whose “learning target” of the input information in FIG. 11 is “1” (here, the road link of link number 20), “learning?” 35 is set to “1”.
FIG. 11 shows the contents of the estimation database 13 after the first update (set of input information) is performed as described above.

Subsequently, the speed information about the road links (link numbers 11 to 19 and 21 to 24) whose speed information is not updated is estimated, and the estimation database 13 is updated with the estimated value (step S4).
Specifically, first, among the related road links (adjacent road links) of road links (link numbers 1 to 10, 20) in which “1” indicating the previous update is set in the “check” 31 item, A road link (check = 0) for which speed information has not yet been updated based on the current input information (FIG. 10) is extracted. Here, 11 road links of link numbers 12, 15, 11, 13, 19, 18, 17, 21, 21, 24, 14 and 16 are extracted (see FIG. 12). These 11 road links serve as estimation target road links here.

  Then, the estimation system 11 reads the related road links for each of the 11 estimation target road links from the estimation database 13 and uses the weights (estimation parameters) set for each of the 11 estimation target road links as the weights. Reading from the database 15 and using these, the estimated value of the speed information of each estimation target road link is calculated. The calculated estimated value of the speed information is set in “speed information” of the estimation database 13, and the second update of the speed information is performed. For the updated road link, “2” indicating that the speed information has been updated in the second update is set in “check” 31.

Among the estimation target road links, for example, an arithmetic expression for obtaining an estimated value (V ₁₂ , V ₁₅ , V ₁₁ , V ₁₃ ) of speed information of road links of link numbers ₁₂ , ₁₅ , ₁₁ , and ₁₃ is as follows. It is as follows.
V ₁₂ = w ₀ + w ₁ V ₁₁ + w ₂ V ₁ + w ₃ V ₅ + w ₄ V ₁₃ + w ₅ V ₈ + w ₆ V ₇ + w ₇ V ₁₅
V ₁₅ = w ₀ + w ₁ V ₁₃ + w ₂ V ₂₀ + w ₃ V ₁₆ + w ₄ V ₁₇ + w ₅ V ₁₄ + w ₆ V ₁ + w ₇ V ₁₁
+ W ₈ V ₅ + w ₉ V ₈ + w ₁₀ V ₁₂ + w ₁₁ V ₅
V ₁₁ = w ₀ + w ₁ V ₁₂ + w ₂ V ₈ + w ₃ V ₁₅ + w ₄ V ₇ + w ₅ V ₁ + w ₆ V ₁₁ + w ₇ V ₅
V ₁₃ = w ₀ + w ₁ V ₁₅ + w ₂ V ₈ + w ₃ V ₁₂ + w ₄ V ₇ + w ₅ V ₁ + w ₆ V ₁₁ + w ₇ V ₅
+ W ₈ V ₁₄ + w ₉ V ₁₇ + w ₁₀ V ₁₆ + w ₁₁ V ₂₀

  And among the related road links (adjacent road links) of the road links (link numbers 11 to 19, 21, 24) in which “2” indicating the previous update is set in the “check” 31 item, this input A road link (check = 0) for which speed information has not yet been updated based on the information (FIG. 10) is extracted. Here, two road links with link numbers 22 and 23 are extracted (see FIG. 13). These two road links become the next estimation target road links.

  Then, the estimation system 11 reads the related road links for each of the two estimation target road links from the estimation database 13 and uses the weights (estimation parameters) set for each of the two estimation target road links as the weights. Reading from the database 15 and using these, the estimated value of the speed information of each estimation target road link is calculated. The calculated estimated value of the speed information is set in “speed information” of the estimation database 13, and the third update of the speed information is performed. As for the updated road link, “3” indicating that the speed information has been updated in the third update is set in “check” 31.

  The above process is repeated until the speed information for all road links is estimated (step S4). Here, since all speed information has been complemented by three updates, the estimation process is terminated.

Then, the contents of the estimation database are output (step S5).
Furthermore, among the road link data of the estimation database 13 for which the estimation process has been completed based on the current input information (FIG. 10), the road link of the link number 20 with “1” set to “Learning?” 35 is displayed. A snapshot (see FIG. 14) is added to the learning database 14. Here, the snapshot includes “speed information” of a road link in which “1” is set in “learning?” 35 and speed information of a road link in which “1” is set in “learning?” 35. A combination of speed information of one or more other road links (related road links) used for estimation is stored. Here, the speed information of the road link in which “1” is set in “Learning?” 35 is a measured value because it is a value obtained from the probe information. In addition, in the “speed information” of the related road link of the road link, an actual measurement value and an estimated value obtained from VICS information are mixed.

A snapshot is generated each time input information is acquired and all road links in the estimation database are updated (steps S3 and S4), and stored as learning data. This snapshot is generated for a road link for which probe information has been acquired.
Therefore, in the above example, the snapshot is generated only for the road link with the link number 20, but if all the road links in the estimation database are updated many times (steps S3 and S4), the other road links are also snapped. Shots are accumulated. In addition, when a sufficient time has passed so that the input information is repeatedly generated, a plurality of snapshots are accumulated for one road link.

[2.4.4 Details of learning process]
FIG. 15 shows the contents of the learning database 14 in which a plurality of snapshots are accumulated for a plurality of road links.
Based on the learning database 14 in which a large number of snapshots are accumulated as learning data in this way, the learning system 12 learns (optimizes) the weights (estimation parameters) stored in the weight database 15. Thus, the contents of the weight database 15 are updated.

FIG. 16 shows the procedure of learning processing by the learning system 12. First, the learning system 12 sets parameters for learning such as an allowable error for error determination, the number of intermediate layers constituting the neural network (FIG. 6), and a learning count (step S11).
The learning system 12 is configured as a neuro engine that optimizes (learns), for example, a neural network as shown in FIG. That is, the learning system 12 uses the speed information that is an actual measurement value among the snapshots stored in the learning database 14 as a teacher signal, and from the speed information about the related road link of the road link having the speed information that becomes the teacher signal. Reconstruct an appropriate neural network to output this teacher signal. That is, the learning system 12 calculates the optimum value of the weight from the learning data including the speed information that is the actually measured value and the speed information about the related road link having the speed information. The calculation of the optimum value of the weight can be performed by, for example, the least square method.

The optimum weight value is calculated until the speed information calculated from the speed information about the related road link approaches the teacher signal and the error from the teacher signal is less than the set allowable error (step S13). ).
When the weights converge and the learning process ends, the obtained weights are reflected in the weight database 15 and the weight database 15 is updated.

  When input information including VICS information and probe information is generated after the weight database 15 is updated, speed information is estimated by the estimation system 11 using new weights with higher accuracy. Thus, in the traffic information estimation apparatus 1 of this embodiment, the estimation accuracy of traffic information (speed information) naturally improves by continuing operation.

[2.5 Functions further provided by the traffic information estimation device]
[2.5.1 Estimation process]
In the embodiment, as shown in FIG. 10, the VICS information is acquired for the road links with the link numbers 1 to 10, the probe information about the road link with the link number 20 is acquired, and the speed information of the remaining road links is Estimated. And the case where the speed information based on the actual measurement value of the road link of the link number 20 was used and the learning process was performed was demonstrated.
Here, the road corresponding to the road link with the link number 20 is composed of a plurality of lanes, and the road link with the link number 20 is branched on the downstream side as shown in FIG. Will be described.

  For example, as shown in FIG. 17, the road of the road link R20 with the link number 20 is composed of two lanes L1 and L2, and the right lane L1 is substantially the same with the road link R14 with the adjacent link number 14 ahead. This is the main line on which the vehicle continues to run at speed, and the left lane L2 is a lane that continues to the branched side road (link number 15). In addition, in this side road and the lane L2 that is downstream thereof, Is congested. In addition, as a road containing such a road link R20, it is an exit of a highway, for example.

In this case, even on the same road R20, the speed information (probe information) V20-1 acquired from the probe vehicle 3 traveling on the right lane L1 is a relatively high speed (for example, 0.1 at the standardized speed). The speed information (probe information) V20-2 acquired from the probe vehicle 3 traveling in the lane L2 is a value indicating a relatively slow speed (for example, 0.42 in the normalized speed). It is possible to become.
As described above, even if the speed information of the road link R20 is obtained based on the actual measurement value, two types of speed information V20-1 and V20-2 may be acquired. Even if the estimation system 11 estimates the speed information of the road link related to the road link R20 (for example, the branch road links R14 and R15 ahead of the road link R20), the learning system 12 also optimizes the speed information. Doing so may not be accurate.

Therefore, in the traffic information estimation apparatus 1 (estimation system 11) of the present embodiment, the learning system 12 uses the learning data classified for each of the plurality of branch road links, and optimizes the weight for each branch road link. When the weight optimization process is completed, the weight database is updated with the new weight value obtained by the process.
Then, the estimation system 11 estimates the traffic information of the estimation target road link using the weight optimized (updated) for each branch road link. This configuration is applied to roads where traffic information may differ greatly in each lane, for example, because there are multiple lanes on the road before branching in addition to the exit of the expressway, and those include a right-turn exclusive lane. It is effective to do. Some such roads can be registered in the system in advance. Hereinafter, this configuration will be described.

In the present embodiment, as shown in FIG. 2, a learning database 14 that accumulates learning data (snapshot) in the learning system 12, an estimation database 13 that accumulates speed information of each road link, and the like A weight database 15 for accumulating parameters (weights) is provided. The learning database 14 is provided with a selection unit 51 for classifying and accumulating learning data by a predetermined method. The selection unit 51 also has a function of classifying and storing estimation parameters (weights) in the weight database 15 by a predetermined method.
In this embodiment (FIG. 15), the selection unit 51 branches the learning data of the road link R20 with the link number 20 after the road link R20 into two branch road links (link numbers 20 ( → 15) and link number 20 (→ 14)), and the case where they are stored in the learning database 14 is described.

In the learning database 14, learning data is stored by being classified according to branch road links (link numbers 14 and 15). When learning data is stored in this manner, the learning system 12 stores the learning data for each branch road link. Using the classified learning data, a process of optimizing the weight for each branch road link (link numbers 14 and 15) is performed. When the weight is optimized for each branch road link, the weight database is updated with the new weight value (see FIG. 19). FIG. 19 is an explanatory diagram of the weight database 15 in which a part of FIG. 9 is rewritten. In FIG. 19, the optimized (updated) weights are stored for each branch road link (link numbers 14 and 15) for the road link R20 with the link number 20.
As described above (see FIGS. 15 and 19), the learning data (speed information) and the weights are classified and accumulated by destination according to the same road link R20.

In this way, in order to classify the information according to the destination of the probe vehicle 3 and accumulate the information in the databases 14 and 15, it is necessary to determine the destination after the probe vehicle 3 passes the road link R20. For this purpose, the estimation system 11 includes a determination unit 52 (see FIG. 2) that determines the destination of the probe vehicle 3.
Here, the in-vehicle device 2 of the probe vehicle 3 is configured to collectively transmit probe information of a plurality of continuous road links (for example, to transmit to the roadside communication device 4 every 5 minutes). The probe information includes information related to the ID, position, and time of the probe vehicle 3. For this reason, the discrimination | determination part 5 can discriminate | determine the destination (traveling locus | trajectory) of the probe vehicle 3 by acquiring such probe information.

  Therefore, if the sorting unit 51 obtains speed information of each road link and information for determining the destination from the obtained probe information, the speed information based on the actually measured value of the link number 20 is obtained from the branch road link (link number 14). 15) It is possible to classify and accumulate separately. That is, if the speed information on the road link of the link number 20 acquired from a certain probe vehicle 3 travels on the road link R15 of the link number 15, the speed information V20-2 of the link number 20 is 0. .42 is accumulated (the portion surrounded by the broken line in FIG. 15).

Specifically, according to the probe information, when it is determined that the speed information based on the actual measurement value in the road link with the link number 20 travels to the branch road link with the link number 14, in FIG. The speed information V20-1 when traveling to the branch road link R14 of the link number 14 is accumulated in the data item of “speed information” of the road link of the number 20 (→ 14), and this link number 20 is combined with this. The speed information of other road links used for estimating the speed information of the other road link is accumulated as learning data.
On the other hand, according to the probe information, if it is determined that the speed information based on the actually measured value in the road link with the link number 20 travels to the branch road link with the link number 15, the link number 20 (→ 15 The speed information V20-2 when traveling to the branch road link R15 with the link number 15 is accumulated in the “speed information” data item of the road link), and combined with this, the speed of the road link with the link number 20 The speed information of other road links used for estimating the information is accumulated as learning data.

Further, in FIG. 19, the estimation parameters (weights) for the road link with the link number 20 that are the related road links of the road links with the link numbers 14 and 15 are represented by two branch road links (link numbers 14 and 15). They are classified separately and stored in the weight database 15.
That is, regarding the branch road link R14 with the link number 14, the weight in the case of proceeding to the branch road link with the link number 14 is accumulated in each weight (weight w _{3 in the} illustrated example), and further distinguished from this, The weight when traveling to the branch road link of link number 15 is accumulated. Then, regarding the branch road link R15 with the link number 15, each weight (weight w _{2 in the} illustrated example) is accumulated with the weight when proceeding to the branch road link with the link number 14, and further distinguished from this, The weight when traveling to the branch road link of link number 15 is accumulated.

A case will be described in which the estimation system 11 estimates speed information using one branch road link of the two branch road links (link numbers 14 and 15) as an estimation target road link.
The weight corresponding to the one branch road link is selected from the weights classified by the two branch road links (link numbers 14 and 15), that is, classified by destination and stored in the weight database 15 (FIG. 19). Use the selected weight.

That is, for example, when estimating the speed information of the branch road link R15 with the link number 15, as described above, the estimated value of the speed information of the branch road link R15 can be obtained by the following arithmetic expression.
V ₁₅ = w ₀ + w ₁ V ₁₃ + w ₂ V ₂₀ + w ₃ V ₁₆ + w ₄ V ₁₇ + w ₅ V ₁₄ + w ₆ V ₁ + w ₇ V ₁₁
+ W ₈ V ₅ + w ₉ V ₈ + w ₁₀ V ₁₂ + w ₁₁ V ₅
At this time, the speed information of the link number 20 that is the related road link of the branch road link R15 is selected from the estimation database, and in FIG. 19, as the weight w ₂ corresponding to the speed information of the road link of the link number 20, 20 → A value of 0.7, which is 15, is selected.
Then, in the same manner as in the above embodiment, the estimated value of the speed information of the branch road link R15 is calculated by the above calculation formula.

  In the above description, the speed information of the link number 20 that is the related road link of the branch road link R15 is described as being selected from the estimation database (for example, FIG. 11). However, when the speed information of the link number 20 is estimated by the estimation system 11, that is, the estimation system 11 is a road link (link) in front of the two-branch road link (link numbers 14 and 15). The case where the speed information is estimated using the number 20) as the estimation target road link will be described.

In the estimation processing by the estimation system 11, the following arithmetic expression (V ₂₀ =...) Is used, but each weight (see FIG. 19) classified according to branch road links (link numbers 14 and 15) is used. Thus, speed information for each branch road link is estimated. That is, two types of speed information are estimated.
V ₂₀ = w ₀ + w ₁ V ₁₇ + w ₂ V ₂₁ + w ₃ V ₂₄ + w ₄ V ₁₄ + w ₅ V ₁₅ + w ₆ V ₁₆ + w ₇ V ₁₃
As described above, the speed information of the road link with the link number 20 can be estimated by the above-described arithmetic expression (V ₂₀ =...). The values from the weights w ₀ to w ₇ used in this case are as follows. When going to the branch road link with link number 14 (upper link → 14 of FIG. 19) and when going to the branch road link with link number 15 (lower link → 15 of FIG. 19), Two types are distinguished and used.
In this case, as shown in FIG. 18, speed information V20-1 toward the link number 20 → 14 and speed information V20-2 toward the link number 20 → 15 are estimated. When estimating one of the speed information V20-1 and V20-2, the other traffic information may be used for the estimation.

As described above, the estimation system 11 estimates the speed information using the road link with the link number 20 before branching as the estimation target road link, and when this is accumulated in the estimation database, for example, the branch with the link number 15 In order to estimate the speed information of the road link R15, as described above, the estimated value of the speed information of the branch road link R15 can be obtained by the arithmetic expression (V ₁₅ =...). as the speed information V ₂₀ toward number 20 → 15, 0.42 is chosen is V20-2.
In the case of estimating the speed information of the branch road links R15 of the link number 15, as the speed information V _20, it is sufficient that at least V20-2 is used, in addition to V20-2, the V20-1 It may be used. In this case, the weight of V20-2 is set to a high value, but the weight of V20-1 may be set to a small value.

  According to this estimation system 11, even if there is a possibility that a road link is branched ahead and multiple types of speed information may be acquired from the same road link, an appropriate weight is selected. Link speed information can be accurately estimated, and estimation parameters can be optimized correctly.

[2.5.2 Estimation process]
Further, as another function, the selecting unit 51 (see FIG. 2) classifies road link speed information in the learning database 14 using at least one of the calendar, time zone, and weather as a classification criterion. And has a function of accumulating. Furthermore, the selection unit 51 has a function of causing the weight database 15 to accumulate and accumulate estimation parameters (weights) using at least one of the calendar, time zone, and weather as classification criteria. . In FIG. 20, a database for each of the calendar, time zone, and weather is prepared for each of the learning database 14 and the weight database 15.

Then, the learning system 12 can perform the process of optimizing the estimation parameter for each condition using the learning data classified according to the conditions such as the history, and when the weight optimization process is completed. The weight database is updated with the new weight value obtained by the processing.
And the estimation system 11 estimates the traffic information of an estimation object road link using the weight optimized (updated) according to the said conditions. Hereinafter, this configuration will be described.

  For example, the sorting unit 51 has a calendar function to classify history as a condition, and accumulates learning data and weights for each day of the week and each holiday. That is, a learning database 14 for each day of the week and a weight database 15 for each day of the week are set. Therefore, the learning system 12 performs a learning process for each learning data accumulated separately for each day of the week or each holiday. In the estimation process by the estimation system 11, the weight database 15 corresponding to the day of the week on which the process is performed is used, and the holiday weight database 15 is used for a holiday.

Similarly, the selection unit 51 has a clock function, and accumulates learning data and weights for each time zone. That is, for example, the learning database 14 and the weight database 15 are set for morning, daytime, evening, nighttime, and midnight respectively. Therefore, the learning system 12 performs a learning process for each learning data accumulated separately for each time zone. And in the estimation process by the estimation system 11, the weight database 15 according to the time zone which performs the process is used.
The sorting unit 51 accumulates the learning data and the weight separately for each weather condition (sunny, cloudy, rain, and snow). That is, the learning database 14 and the weight database 15 for each weather condition are set. Therefore, the learning system 12 performs a learning process for each learning data accumulated separately for each weather condition. And the selection part 51 can acquire weather information from the outside, such as a VICS center, for example, and in the estimation process by the estimation system 11, the weight database 15 according to the weather conditions at the time of performing the process is used.

In this way, the individual learning database 14 and the individual weight database 15 are prepared by classifying the calendar, time zone, and weather as classification criteria.
The learning data is stored in the learning database 14 by classifying each condition such as the history as a classification criterion, and the learning system 12 performs optimization for each condition to update the weight for each condition. it can. For this reason, the speed information is accurately estimated by the estimation system 11 performing the estimation process using the weight for each condition.
Then, when estimating the speed information of the estimation target road link, the estimation system 11 corresponds to the condition corresponding to the estimation from the weights classified according to the classification criteria and stored in the weight database 15. The selected weight and use the selected weight. Thus, the speed information is estimated using an arithmetic expression corresponding to the classification at the time of processing.

  As mentioned above, even if it is the same road link, even if speed information of a different tendency may be acquired based on a measured value by a different calendar, time zone, and weather, the learning database 14 and the weight database 15 is classified according to a predetermined classification standard, and the traffic information of the estimation target road link can be accurately estimated by using a database that is suitable at the time of processing, and the estimation parameter can be learned. It can be done correctly.

  In the description of the estimation process, when the learning parameters classified for each branch road link are used to optimize the estimation parameters for each branch road link, and in the calendar, time zone, and weather The case where the estimation parameter is optimized according to the condition using the learning data classified according to at least one condition has been described. For example, when the road link includes a plurality of lanes, the learning data classified according to the lane May be used to optimize the estimation parameter for each lane, and the traffic information of the estimation target road link may be estimated using the weight optimized for each lane.

[2.5.3 Model selection for velocity information estimation]
In the above embodiment, the case where the estimation system 11 uses a single-layer perceptron as shown in FIG. 6 as a neural network for estimating velocity information is not limited to this. A multilayer perceptron having an output layer may be used. Accordingly, the estimation system 11 has a function of estimating speed information by changing a mathematical model (calculation method) used when estimating the traffic information of the estimation target road link according to the road type of the estimation target road link. Have.

  FIG. 21 is a flowchart for explaining the model change by the estimation system 11. For example, when the correlation between a road corresponding to a certain estimation target road link and its surrounding road is simple like a highway (Yes in step S21), as a model used when estimating traffic information, A linear model such as a single-layer perceptron as shown in FIG. 6 is employed (step S22). If a linear model is adopted, the processing time can be shortened. Since the road link is set on the central device (traffic information estimation device) 1 side, the traffic information estimation device 1 can grasp the road type of each road link.

  On the other hand, when the correlation between the road corresponding to the estimation target road link and its surrounding roads is complicated, for example, the estimation target road link is a general road with many inflows and outflows of vehicles in the middle of the link In this case (in the case of No in step S21), a nonlinear model such as a multilayer perceptron is adopted as a calculation method used when estimating traffic information (step S23). If the estimation model is configured as a multi-layer perceptron, the speed information of the estimation target road link can be appropriately obtained even when the speed information has a non-linear relationship between the estimation target road link and another road. .

  In the case of FIG. 21, the case where the mathematical model is changed according to the road type such as a highway or a general road and the speed information is estimated has been described. However, according to the accuracy of the estimation result of the speed information by the estimation system 11, The mathematical model (calculation method) may be changed. FIG. 22 is a flowchart for explaining model change by the estimation system 11.

  The estimation system 11 estimates the speed information of each road link using a linear model (step S25) and also obtains an error of the estimation result (step S26). This error is obtained based on the difference between the speed information acquired as an actual measurement value for a certain road link and the speed information acquired as an estimated value for the certain road link. If this error is larger than a preset threshold value (Yes in step S27), the road information is estimated using a non-linear model instead of a linear model thereafter. (Step S28).

If the error is equal to or smaller than the threshold value (No in step S27), the road link is set so as to estimate the speed information using the linear model (step S29).
According to this estimation system 11, in principle, the linear model is used to shorten the processing time. However, the linear model is used only when the estimation accuracy is low, and the overall estimation accuracy is ensured. be able to.

[2.5.4 Selection of speed information estimation method]
In the estimation system 11 of the above-described embodiment, for road links from which VICS information and probe information are not acquired (estimation target road links), traffic information and estimation parameters (weights) of other road links are used. The case where the speed information is estimated by the neural network shown in FIG.
By adopting a mathematical model based on this neural network, it is possible to accurately estimate the speed information of the estimation target road link. Moreover, since the traffic information estimation device 1 includes the learning system 12, actual measurement values are obtained. It is possible to further improve the estimation accuracy each time the speed information based on is obtained.

However, depending on the type of road link, various factors may overlap, and even if estimation processing using a neural network is performed, the estimated value may vary. When this variation is large, it is considered that the reliability of the estimation result (estimated value) is low.
Therefore, the estimation system 11 of the present embodiment has a function of estimating the traffic information of the estimation target road link by changing to an estimation process different from the estimation process using the neural network based on the reliability of the estimated value. Have. For this purpose, the estimation system 11 has a reliability determination unit 53 (see FIG. 2) that calculates the reliability of the estimation processing of the speed information of the estimation target road link by the neural network. Note that the reliability in this embodiment is determined based on variations in estimated values obtained by a neural network. That is, the reliability is determined by comparing the standard deviation of the estimated value with the threshold value. The reliability may be determined based on an error between the speed information based on the estimated value and the speed information based on the actually measured value.

  FIG. 23 is a flowchart for explaining the selection process by the estimation system 11. For example, since the neural network shown in FIG. 6 is set for each road link, the estimation system 11 uses the speed information on each road link and the speed information and weight (estimation parameter) regarding the other road links. (Step S31). Then, the estimated speed information on the road link is accumulated in the estimation database 13 and then updated. However, the estimated speed information is not used for the estimation process but is accumulated. (Step S32).

  In this way, when a plurality of pieces of speed information based on estimated values are accumulated for a certain road link, the reliability determination unit 53 calculates a standard deviation for the speed information based on the plurality of estimated values (step S33). ). Then, the reliability determination unit 53 compares the standard deviation with a preset threshold value. When the standard deviation exceeds the threshold value (Yes in step S34), the reliability determination unit 53 As described above, it is determined that the reliability of the estimated value obtained using the neural network is low. When this determination is made, the certain road link is thereafter changed to an estimation process different from the estimation process using the neural network (step S35), and the traffic information of the estimation target road link is obtained by the estimation system 11. Presumed.

  On the other hand, when the standard deviation is equal to or less than the threshold (No in step S34), the reliability determination unit 53 determines that the reliability of the estimation process using the neural network is high. In this case, an estimation process using a neural network is performed on the certain road link as it is (step S36).

  The “another estimation method” for estimating the speed information of the estimation target road link is not an estimation method for obtaining speed information by calculation using a mathematical model as shown in FIG. There is a method of estimating traffic information (speed information) of an estimation target road link using statistical information obtained by statistically calculating traffic information (speed information) of other road links. For this purpose, the estimation system 11 has a statistical database (not shown) that accumulates road link traffic information (speed information) acquired in the past from probe information or the like for each road link.

A method of estimating speed information using this statistical information will be further described. This method is a method of estimating the traffic information of the estimation target road link using the traffic information about the actual measurement values acquired and accumulated in the past.
That is, for each road link, an average value of traffic information based on actual measurement values collected by acquiring VICS information and probe information is accumulated in the database as statistical information. In addition, the average value of this traffic information is accumulate | stored according to a day type (for example, a day of the week, a holiday) and a time slot | zone (for example, 15 minutes width).
When the traffic information estimation process is performed, the statistical information (the average value of the traffic information) corresponding to the day type (for example, day of the week, holiday) and the time zone (for example, 15 minutes) for which the estimation process is performed. ) As an estimated value. Note that the second estimation processing function described later can also be used in this method.

[2.5.5 Selection of speed information estimation method]
Since the traffic information estimation device 1 includes the learning system 12, when speed information based on actual measurement values is acquired, an optimum value of the weight is obtained, and as a result, the estimation system 11 performed using this weight. It becomes possible to improve the estimation accuracy of speed information by. However, depending on the type of road link, various factors may overlap, and even if an estimation process using a neural network is performed, the error between the actual measurement value and the estimated value may remain large.

Therefore, the estimation system 11 of the present embodiment has a first estimation processing function and a second estimation processing function as functions for estimating speed information of the estimation target road link. The first estimation processing function is a function for performing the estimation processing described above, and the estimation target road is obtained by a neural network (for example, see FIG. 6) using speed information and estimation parameters (weights) of other road links. Estimate link speed information.
On the other hand, the second estimation processing function is not an estimation method for calculating speed information by calculation using a mathematical model as shown in FIG. 6, but is a statistic obtained by statistically calculating speed information of other road links acquired in the past. This function uses information to estimate speed information of the estimation target road link. For this second estimation processing function, the estimation system 11 has a statistical database (not shown) for accumulating road link speed information acquired in the past based on probe information and the like for each road link. ing.
And the estimation system 11 is based on the difference | error of the speed information based on a measured value, and the speed information estimated by each of the 1st estimation processing function and the 2nd estimation processing function, and the said 1st estimation processing function and the said 2nd It has a function of determining which of the estimation processing functions to use.

FIG. 24 is a flowchart for explaining the selection process by the estimation system 11. It is assumed that speed information based on an actual measurement value on a certain road link is obtained by probe information or the like (step S41).
For example, since the neural network shown in FIG. 6 is set for each road link, according to the first estimation processing function, the speed information regarding the certain road link is the speed information and the weight of the other road link. (Estimation parameter) is used for estimation (step S42-1).
Moreover, since the speed information of the road link acquired in the past about each road link is accumulated in the statistical database, according to the second estimation processing function, the speed information regarding the certain road link can be estimated. Yes (step S42-2).

  Since the speed information based on the actual measurement value in the certain road link is obtained (step S41), the first error between the speed information based on the actual measurement value and the speed information estimated by the first estimation processing function. (Step S43-1), and second error between the speed information based on the actually measured value and the speed information estimated by the second estimation processing function is obtained (step S43-2).

And these 1st error and 2nd error are calculated | required several times during a fixed period (for example, 1 week), and the average value of the said fixed period is calculated | required.
The estimation system 11 compares the first error (average value) and the second error (average value), and if the first error is equal to or less than the second error (Yes in step S44), the certain road link Thereafter, when the speed information based on the actual measurement value cannot be obtained, the first estimation processing function is adopted to estimate the speed information (step S45).
When the first error exceeds the second error (in the case of No in step S44), when the speed information based on the actual measurement value is not obtained for the certain road link, the second estimation processing function is adopted. Speed information is estimated (step S46).

  Thus, by obtaining speed information based on actual measurement values for a certain road link, whether to adopt the first estimation processing function or the second estimation processing function is fixed for the certain road link. It is performed for each period, and it is set as to which function is adopted in the certain road link. Thereby, in each road link, the speed information of the estimation target road link can be estimated by the estimation processing function in which the error is smaller than the other.

[2.5.6 Select speed information]
As described in FIGS. 3 and 4, the speed information conversion units 17 b and 18 b convert the link travel speed indicating the vehicle speed on the road link based on the “speed-speed information” conversion information 19 to the link travel speed. It is converted into an index value called “speed information” (standardized speed) that takes a value according to the size of.
In the embodiment, the “speed-speed information” conversion information 19 is set as shown in FIG. 4, and the first conversion information 19 is set for a general road, for example. is there.
As the “speed-speed information” conversion information, in addition to the relationship shown in FIG. 4, “speed-speed information” conversion information 59 shown in FIG. 25 is also set, and these conversion information 19 and 59 are input information. It is stored in the processing unit 16. The second conversion information 59 is set for a highway, for example.

In the present embodiment, each of the speed information conversion units 17b and 18b changes the “speed-speed information” conversion information according to the road type of the road link to change the link travel speed to “speed information” (standardized). Speed).
For example, when the link travel speed acquired from the VICS information and the probe information is acquired from a vehicle traveling on a general road, the link travel speed is “speed-speed information” shown in FIG. The conversion information 19 is used and converted into “speed information” (standardized speed). Since the road link is set on the central device (traffic information estimation device) 1 side, the input information processing unit 16 can grasp the road type of each road link.

On the other hand, when the link travel speed is obtained from a vehicle traveling on a highway, the link travel speed is used by the “speed-speed information” conversion information 59 shown in FIG. And converted into “speed information” (standardized speed).
In this way, conversion information having different characteristics can be switched according to the types of roads having different speed limits such as general roads and highways, so that “speed information” ( Normalization speed) can be obtained.

In addition to changing the “speed-speed information” conversion information according to the type of road of the road link, each of the speed information conversion units 17b and 18b is configured according to traffic information indicating, for example, the congestion status of the road link. The “speed-speed information” conversion information may be changed and converted into “speed information” (standardized speed).
That is, even if it is the same general road, according to the traffic information of a certain road link, when it is determined that the road is vacant, the “speed-speed information” conversion information 59 shown in FIG. 25 is used. On the contrary, according to the traffic information of a certain road link, when it is determined that the road is congested, the “speed-speed information” conversion information 59 shown in FIG. 4 is used.

  For example, the traffic information indicating the congestion status is obtained by calculating “speed information” (standardized speed) using the “speed-speed information” conversion information 19 shown in FIG. To determine the congestion situation. For example, when “speed information” (standardized speed) is 0.9 or more, it is determined that the traffic is congested. Then, it is determined which of the “speed-speed information” conversion information 19 and 59 is to be used according to the determined congestion situation. For example, when it is determined that the traffic is congested, the “speed-speed information” conversion information 19 is determined to be adopted. Then, “speed information” (standardized speed) is obtained again based on the determined conversion information, and this “speed information” is used by the estimation system 11 and the learning system 12.

[2.5.7 Accumulation of learning data in the learning system]
As described above, the combination of the speed information based on the actual measurement value of the road link that can be the estimation target road link and the speed information of the other road link used for estimating the traffic information of the estimation target road link is used for learning. As data is accumulated in the learning database 14 and speed information based on actual measurement values is obtained, learning data is newly accumulated.
As described above, even when new learning data is newly added to the learning database 14, the original learning data that has been accumulated for a long time remains in the learning database 14. Therefore, if the learning process for the learning target road link is performed using all of the learning data stored in the learning database 14, the learning accuracy may be prevented from being improved due to the influence of the original old learning data. .
Therefore, it is effective to discard the learning data stored in the learning database 14 with an old time and perform learning using the newly stored learning data.

Therefore, the learning system 12 can adopt a criterion of discarding learning data that exceeds a reference accumulation period (for example, six months) as a discarding criterion for discarding old learning data. In this case, learning data for the past six months is left in the learning database 14, and learning data whose accumulation period has passed six months is sequentially discarded.
However, if no traffic jam has occurred on a road link in the past six months, the learning database 14 does not have any learning data when the traffic jam has occurred on the road link. In this case, when traffic jam occurs every year on a specific day of the year, the learning data for the specific day last year does not remain in the learning database 14. For this reason, learning in consideration of traffic jam on the specific day is not performed, and there is a possibility that the estimation (complementation) of speed information on the specific day of the year cannot be performed appropriately.

Therefore, even if the old learning data is discarded, it is preferable to select the learning data to be discarded instead of simply discarding the oldest learning data.
Specifically, as shown in FIG. 26, each learning data is composed of a learning data body B and category information C indicating the category to which each learning data belongs, and the category of each learning data Make it possible to classify. The learning data body B is a combination of the speed information of the learning target road link (speed information based on the probe information) and the speed information of the other road links used for estimating the speed information of the learning target road link. Consists of.

  In FIG. 26, as category information C, speed category information categorized based on the speed information of the learning target road link in the learning data main body B is added. As the speed category, the speed information of the learning target road link can be classified into three categories of “large”, “medium”, and “small” according to the size. The speed category information C added to the learning data body B is information indicating whether the speed information of the learning target road link in the learning data body B belongs to “large”, “medium”, or “small”. .

  FIG. 27 shows an accumulation process when learning data with category information C is added to the learning database 14 by snapshot. This accumulation process is executed by the management unit 55 (see FIG. 2) of the learning system 12.

When the learning data is to be stored in the learning database 14, first, it is determined whether or not the learning data already stored in the learning database 14 needs to be discarded according to the discard criteria as described above. (Step S51). If there is no need for discarding, the learning data is stored in the learning database 14.
When the discarding is necessary, the category having the maximum number of learning data already accumulated in the learning database 14 is determined from the three categories (step S52).

  Then, the oldest learning data in the category determined to have the largest number of learning data is deleted from the learning database 14 (step S53). Then, the learning data to be accumulated is accumulated in the learning database 14 (step S54). That is, among the three categories, the learning data of the category having a large number of accumulated learning data is discarded with priority.

For example, it is assumed that the storage process of FIG. 27 is performed when the discarding standard of discarding 300 pieces of learning data in 6 months and discarding the learning data that has passed 6 months after the storage is adopted. Then, it is assumed that the number of learning data belonging to each category of learning data accumulated in the learning database 14 is large: 10 medium: 50, small: 240.
In this case, when the 301st learning data is obtained, it is necessary to delete one learning data from the learning database 14, but according to the accumulation process of FIG. Category is selected as the category to be deleted, and the oldest learning data in the [small] category is deleted.

When the accumulation process of FIG. 27 is repeated, the learning data of the category with a small number of accumulations is less likely to be deleted even if it is old, so it is likely to remain without being deleted. As a result, the number of learning data in each category is almost equal.
Then, the learning system 12 optimizes the weight using the learning data that remains without being discarded.
In this way, it is possible to narrow down the learning data that is the basis of the learning process by accumulating speed information divided into categories and preferentially discarding learning data in categories with a large number of learning data. The learning data to be discarded can be selected.

  Note that categorization is not limited to speed information, and any categorization can be adopted. For example, it may be categorized according to the degree of congestion of road links. Further, the learning data may be categorized according to the day (special day such as a day of the week or a holiday) / time zone when the learning data is obtained, the weather of the day, and the like.

[2.5.8 Data amount of learning data in learning system]
As described above, the learning system 12 calculates the optimum weight value from the learning data by, for example, the least square method. For this reason, in order for the learning system 12 to effectively perform learning for optimizing the weights, it is considered preferable that the number of learning data is large to some extent.
However, since weight optimization should be performed for each road link, if the number of learning data is excessively increased for all road links, the amount of data accumulated in the learning database 14 becomes enormous. Become.

Here, since the learning system 12 calculates the optimum value of the weight (estimation parameter) from the learning data by, for example, the least square method or the like, the optimization of the weight for the road link with a small weight has a large weight. The number of learning data may be smaller than that of road links.
Therefore, the learning system 12 of the present embodiment has a function of setting the number of learning data to be accumulated for the optimization of weights for each learning target road link, and the number of learning data is It is set according to the number of weights used in the estimation process by the estimation system 11.

That is, for example, in the weight database 15 shown in FIG. 9, in order to estimate the speed information of the road link with the link number 15, twelve weights from W _{0 to} W ₁₁ are used. On the other hand, in order to estimate the speed information of the road link with the link number 21, six weights from W _{0 to} W ₅ may be used.
Therefore, the learning system 12 sets the number of learning data to be accumulated for optimization of weights (estimation parameters), with the upper limit being an integer multiple (for example, five times) the number of weights. For example, in the case of the road link with the link number 15, the learning data (snapshot) is accumulated in the learning database 14 with “60” being five times the number of weights “12” as the upper limit. In the case of the road link with the link number 21, the learning data (snapshot) may be accumulated in the learning database 14 with “30” being five times the number of related road links “6” as the upper limit.

  As described above, the learning database 14 is configured so that the number of learning data related to the road link of the link number 21 is smaller than the number of learning data related to the road link of the link number 15 in accordance with the small number of weights. The learning data may be stored in That is, it is not necessary to accumulate a fixed number of learning data for all road links. As a result, an enormous amount of data to be accumulated in the learning database 14 can be suppressed.

The difference in the number of weights between road links is caused by the fact that adjacent road links are set as related road links, and the number of related road links is different for each road link. In other words, the number of related road links corresponds to the number of weights (as described above, the number of weights is “number of related road links + 1”), and the number of related road links related to the learning target road link is This is because as the number increases, the number of weights also increases.
In addition, the number of learning data has been described as being set according to the number of weights used in the estimation process by the estimation system 11, but the number of related road links corresponds to the number of weights. The number of learning data may be set according to the number of other road links (related road links) for which speed information is used in the estimation process by the estimation system 11.

[2.5.9 Learning cycle by learning system]
The learning system 12 may set the learning cycle so as to optimize the weight each time the learning data is acquired, but in this embodiment, the learning system 12 performs the optimization process. The frequency of performing the optimization process is set according to the number of learning data for each road link accumulated in the learning database 14.
That is, the learning system 12 calculates the optimum value of the weight from the learning data by, for example, the least square method or the like, so that the road link having a large number of learning data is compared with the case where the number of learning data is small. Therefore, it is thought that learning is progressing.

Therefore, the learning system 12 increases the frequency of performing the optimization process when the number of learning data for each road link stored in the learning database 14 is small. FIG. 28 is a flowchart illustrating a process for setting a learning cycle.
In FIG. 15, the number of learning data for the learning target road link with the link number 20 is “3”. Therefore, the learning system 12 acquires the number of learning data for each road link (step S71).

  For road links with less learning data, for example, every time learning data is accumulated in the learning database 14, optimization processing is performed. That is, the learning system 12 compares the number of learning data obtained in step S71 with a preset threshold value, and when the number of learning data is less than the threshold value (in the case of Yes in step S72), The frequency of performing the optimization process is increased (step S73). For example, every time learning data is accumulated in the learning database 14, an optimization process is performed.

The number of learning data obtained in step S71 is compared with a preset threshold value. When the number of learning data is equal to or greater than the threshold value (in the case of No in step S72), the frequency of performing optimization processing Is lowered (step S74). That is, when the number of learning data for each road link stored in the learning database 14 is large, the learning system 12 makes the frequency of optimization processing lower than when the number of learning data is small. Change as follows.
This is because the road link with a lot of learning data is considered to have been sufficiently learned and has reached high learning accuracy. Therefore, even if the learning data is accumulated in the learning database 14, the optimization process is performed each time. For example, the optimization process is performed at a low frequency of once a day or once a week.

  Thus, by changing the frequency of performing the optimization process, the load of the learning process by the learning system 12 can be reduced. That is, since the process of optimizing the weights by the learning system 12 takes time, if the number of accumulated learning data increases, the frequency of performing the optimization process is reduced, thereby reducing the learning system. Twelve loads can be reduced.

  In FIG. 28, the case has been described in which the frequency of performing the optimization process is changed according to the number of learning data for each road link accumulated in the learning database 14. However, the accuracy of the speed information estimation result is described. Depending on the frequency, the frequency of performing the optimization process may be changed.

That is, as shown in FIG. 29, when speed information based on actual measurement values is obtained for a certain road link (step S81), the speed information and weights of other road links (related road links) are used. Speed information of a certain road link can be estimated (step S82). Then, a difference between the estimated value and the actually measured value is obtained, and if the difference is equal to or less than a preset threshold value, it is determined that the accuracy of the speed information estimation result is high (Yes in step S83), The optimization process is performed at a low frequency of once or once a week (step S84).
On the other hand, if the difference between the estimated value and the actually measured value exceeds a preset threshold, it is determined that the accuracy of the speed information estimation result is low (No in step S83). Every time data is accumulated in the learning database 14, the frequency is increased so that optimization processing is performed (step S85).

  In the above description, the difference (error) between the estimated value and the actually measured value is the accuracy of the estimated result of the speed information, but otherwise, the speed information based on the actually measured value cannot be acquired (FIG. 29). Step S81 may be omitted), and the accuracy of the speed information estimation result may be determined from the variation of the estimated value, that is, the standard deviation of the estimated value. That is, when the variation in measured values is larger than a preset threshold value, it is determined that the reliability is low and the accuracy of the estimation result is low (No in step S83). In this case, the frequency of optimization processing is set high (step S85).

  Thus, when changing the frequency of the optimization process according to the accuracy of the speed information estimation result, when the accuracy of the speed information estimation result by the estimation system 11 is not preferable, the frequency of performing the optimization process It is possible to increase the accuracy. On the other hand, when the accuracy of the estimation result of the estimation information by the estimation system 11 is favorable, the frequency of performing the optimization process can be reduced, and the load on the learning system 12 can be reduced.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the meanings described above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

Further, other road links (related road links) used for estimating speed information of the estimation target road link are not limited to those connected to the estimation target road link, but are separated from the estimation target road link. However, it may be a road link (for example, a road link parallel to the estimation target road link) whose speed information changes in time.
In this embodiment, standardized speed information is used as information to be estimated. However, link travel speed itself may be used as speed information without using standardized speed information. .

  Furthermore, the traffic information to be estimated by the estimation system 11 of the traffic information estimation apparatus 1 of the present embodiment is not limited to speed information, but may be other traffic information such as traffic jam information in addition to / in place of the speed information. it can. In this case, the degree of traffic congestion can be calculated by assigning 1 in the speed information to a busy binary value and 0 as a smooth binary value. Further, when the traffic congestion degree is handled as the traffic information to be estimated, the neural network also requires a threshold value for determining one of the two values as the estimation parameter. In this case, the threshold value is also a learning target by the learning unit.

  1: Traffic information estimation device, 11: Estimation system (estimation unit), 12: Learning system (learning unit), 13: Estimation database, 17b: Speed information conversion unit (conversion unit), 18b: Speed information conversion unit (conversion unit) ), 19, 59: Conversion information

Claims (14)

A traffic information estimation device that estimates traffic information of an estimation target road link based on traffic information of another road link,
An estimation unit that estimates traffic information of an estimation target road link using traffic information of other road links and estimation parameters;
Accumulation of traffic information based on measured values on road links that can be estimated road links and traffic information on the other road links used to estimate traffic information on the estimated road links as learning data A learning unit that optimizes the estimation parameter using the accumulated learning data;
A traffic information estimation device comprising: The learning unit uses the learning data classified by a plurality of branch road links existing after branching from a road link that can be the estimation target road link, and sets an estimation parameter for each branch road link. Optimize,
The traffic information estimation apparatus according to claim 1, wherein the estimation unit estimates traffic information of an estimation target road link using the estimation parameter optimized for each branch road link.   The traffic information estimation apparatus according to claim 2, wherein the learning unit classifies and accumulates the learning data by classifying the plurality of branch road links. The learning unit uses the learning data classified according to at least one of the calendar, time zone, and weather, and optimizes the estimation parameter for each condition,
The said estimation part is a traffic information estimation apparatus as described in any one of Claims 1-3 which estimates the traffic information of an estimation object road link using the said parameter for estimation optimized according to the said conditions.   The traffic information estimation apparatus according to claim 4, wherein the learning unit classifies and accumulates the learning data according to at least one of a calendar, a time zone, and a weather.   The learning unit classifies and accumulates learning data for each predetermined category, and among the categories, the learning data of a category having a large number of accumulated learning data is preferentially discarded and discarded. The traffic information estimation apparatus according to any one of claims 1 to 5, wherein the estimation parameter is optimized using remaining learning data. A conversion unit that converts link travel speed indicating the speed of the vehicle on the other road link into speed information that takes a value corresponding to the magnitude of the link travel speed based on predetermined conversion information;
The said conversion part changes the said conversion information according to the road link type or the traffic information of the said road link, and converts into the said speed information, The traffic information estimation apparatus as described in any one of Claims 1-6. .   The learning unit stores the number of the learning data to be accumulated for optimization of the estimation parameter, the number of other road links in which traffic information is used in the estimation process by the estimation unit, or the estimation by the estimation unit The traffic information estimation device according to claim 1, wherein the traffic information estimation device is set according to the number of estimation parameters used in the processing.   The estimation unit changes the calculation method used when estimating the traffic information of the estimation target road link according to the type of the estimation target road link or the accuracy of the estimation result of the traffic information by the estimation unit. The traffic information estimation device according to any one of claims 1 to 8, which estimates information.   The learning unit changes the frequency of performing the optimization processing according to the number of the learning data for each road link accumulated or according to the accuracy of the traffic information estimation result by the estimation unit. The traffic information estimation device according to any one of claims 1 to 9. The estimation unit obtains the reliability of the traffic information of the estimation target road link estimated using the traffic information of other road links and the estimation parameters,
The said estimation part changes to the estimation process different from the said estimation process based on the said reliability, The traffic as described in any one of Claims 1-10 which estimates the traffic information of an estimation object road link. Information estimation device. The estimation unit includes
A first estimation processing function for estimating traffic information of an estimation target road link using traffic information of other road links and parameters for estimation;
A second estimation processing function for estimating the traffic information of the estimation target road link using statistical information obtained by statistically calculating the traffic information of the road link acquired in the past,
The estimation unit includes
When the traffic information based on the actual measurement value in the road link is obtained, an error between the traffic information based on the actual measurement value and the traffic information estimated by each of the first estimation processing function and the second estimation processing function is obtained. The traffic information according to any one of claims 1 to 11, wherein which of the first estimation processing function and the second estimation processing function is used in the road link is determined based on an error comparison. Estimating device.   A computer program for causing a computer to function as the traffic information estimating device according to any one of claims 1 to 12. A traffic information estimation method for estimating traffic information of an estimation target road link based on traffic information of another road link,
An estimation step of estimating traffic information of the estimation target road link using traffic information of other road links and estimation parameters;
Accumulation of traffic information based on measured values on road links that can be estimated road links and traffic information on the other road links used to estimate traffic information on the estimated road links as learning data And a learning step of optimizing the estimation parameter using the accumulated learning data;
A traffic information estimation method comprising:

Images