Information processing device, information processing program, and information processing method

Abstract

This information processing device recursively repeats processing in which each congested section indicated by training data including the congestion start position and the congestion end position is classified in each congestion direction and adjacent congested sections are connected such that the congested sections are formed as a single integrated congested section, and thereby determines which of the congested sections is the last of the the congested sections forming the integrated congested section, classifies, for each integrated congested section indicating the same congestion, the integrated congested section according to the determined demarcation, and sets, for each integrated congested section indicating the same congestion and for each determined demarcation, the extension scale of said integrated congested section.

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, an information processing program, and an information processing method for predicting the extension length of traffic jams that occur suddenly due to changes in demand.

Background Art

[0002] There is known a traffic jam prediction service that predicts the location, occurrence time, and traffic jam length of a road traffic jam and provides prediction information to a user (for example, Non-Patent Document 1).

[0003] When the user inputs a date and time into the traffic jam prediction service, for example, a road map as shown in FIG. 19 is displayed on the screen, and the section where a traffic jam is predicted to occur at the input date and time is displayed on the road map by, for example, an arrow 50.

[0004] Further, when the user selects a detailed icon 51 displayed corresponding to the arrow 50, for example, a traffic jam detailed screen 52 as shown in FIG. 20 is displayed. The traffic jam detailed screen 52 displays a traffic jam section, a traffic jam occurrence time zone, a location that is a bottleneck of the traffic jam, a traffic jam length at peak time, and a time required to pass through the traffic jam.

[0005] <Non-Patent Document 1> The East Nippon Expressway Company Limited, Zenrin DataCom Co., Ltd., "DriveTraffic" <URL: https: / / www.drivetraffic.jp / .>

Summary of the Invention

Problems to be Solved by the Invention

[0006] These traffic congestion prediction services predict the occurrence of congestion based on historically observed, consistent traffic volume increases (e.g., over a year or more). Therefore, for congestion that has newly arisen due to changes in behavior patterns caused by the COVID-19 pandemic, such as an increase in the number of drive-through services or a rise in demand for visits to home improvement stores, there is a problem in that the number of historical traffic volume increases that can be used for prediction is limited.

[0007] Therefore, with conventional traffic congestion forecasting services, when new congestion occurs in areas where congestion has not previously occurred regularly due to changes in demand (hereinafter referred to as "sudden congestion"), it is difficult to predict how far it will extend once it begins to occur.

[0008] On the other hand, it is possible to obtain information on sudden traffic jams from traffic congestion images taken by cameras installed on roads, but since it is impossible to predict in advance where sudden traffic jams will occur, it is necessary to install cameras in various locations, which leads to an increase in traffic congestion prediction costs. Moreover, if sudden traffic congestion is obtained from traffic congestion images, it will inevitably be the situation after the congestion has occurred, so the timing of providing information to users will be delayed compared to traffic congestion prediction information that notifies the possibility of congestion occurring in advance. For this reason, it is preferable to predict the occurrence of sudden traffic jams from past observational data on sudden traffic jams (hereinafter referred to as "training data") rather than relying on traffic congestion images. In the verification described herein, the information obtained from the Japan Road Traffic Information Center (JARTIC) was used as this observational data on sudden traffic jams.

[0009] In light of the above, the objective is to provide an information processing device, an information processing program, and an information processing method that can predict the extension of sudden traffic congestion even when such sudden traffic congestion occurs for which sufficient training data has not been obtained for use in conventional traffic congestion prediction services. [Means for solving the problem]

[0010] A first aspect of this disclosure is an information processing device comprising: a classification unit that classifies each congested section, represented by learning data including the starting position and ending position of the congestion, according to the congestion direction represented by the starting position and the ending position; and a process that connects adjacent congested sections, whose starting position and ending position are within a predetermined range, as a continuous integrated congested section, for each congested section classified by the congestion direction by the classification unit, until the starting position or ending position of the adjacent congested sections no longer falls within the predetermined range. The system includes a determination unit that recursively repeats up to a certain point to determine which congested sections constitute the integrated congested section, and a setting unit that acquires the integrated congested sections determined by the determination unit for each integrated congested section representing the same congestion, and for each integrated congested section representing the same congestion, classifies the integrated congested section according to determination categories that divide the timing of congestion by day of the week, by work attribute indicating whether it is a weekday or a holiday, and by time of day, and sets the extension scale of the integrated congested section for each integrated congested section representing the same congestion and for each determination category.

[0011] A second aspect of the present disclosure is an information processing device comprising: a selection unit that uses the extension length of consecutive identical congestion sections, each of which is classified into determination categories based on the timing of occurrence by day of the week, by work attribute indicating whether it is a weekday or a holiday, and by time of day, to determine whether there is a bias in the extension scale of a designated congestion section for each determination category, and the extension scale for each determination category, and if there is a bias in the extension scale of the designated congestion section within the determination category, selects the determination category to be used to predict the extension length of the congestion section according to the combination of determination categories in which the bias in the extension scale exists; and a prediction unit that uses the extension length of the designated congestion section included in the determination category selected by the selection unit to predict the extension length of the designated congestion section at a specified date and time.

[0012] A third aspect of this disclosure is an information processing program that causes a computer to function as a component of an information processing device.

[0013] A fourth aspect of this disclosure is an information processing method in an information processing device including a classification unit, a determination unit, a setting unit, a selection unit and a prediction unit, comprising: a classification step in which the classification unit classifies each congestion section represented by learning data including the starting position and the ending position of congestion according to the congestion direction represented by the starting position and the ending position; a determination step in which the determination unit, for each congestion section classified according to the congestion direction, connects adjacent congestion sections such that the starting position and the ending position are within a predetermined range as a continuous integrated congestion section, by recursively repeating this process until there are no more starting positions or ending positions of adjacent congestion sections within the predetermined range, thereby determining which congestion sections constitute the integrated congestion section; and a setting unit, for each integrated congestion section representing the same congestion, The system includes: a setting step of classifying integrated congestion sections according to each determination category which is divided by the day of the week, by work attribute indicating whether the congestion occurs on a weekday or holiday, and by time of day, and setting the extension scale of the integrated congestion sections for each integrated congestion section representing the same congestion and for each determination category; a selection step in which the selection unit uses the extension scale to identify whether there is a bias in the extension scale of a designated congestion section within the determination category, and if there is a bias in the extension scale of a designated congestion section within the determination category, selects the determination category to be used to predict the extension length of the congestion section according to the combination of determination categories in which the bias in extension scale exists; and a prediction step in which the prediction unit uses the extension length of the designated congestion section included in the selected determination category to predict the extension length of the designated congestion section at a specified date and time. [Effects of the Invention]

[0014] The information processing device, information processing program, and information processing method disclosed herein have the effect of being able to predict the extension of sudden traffic congestion even if a sudden traffic congestion occurs for which sufficient training data has not been obtained for use in conventional traffic congestion prediction services. [Brief explanation of the drawing]

[0015] [Figure 1]It is a diagram showing an example of the functional configuration of an information processing device. [Figure 2] It is a diagram showing an example of a congestion section. [Figure 3] It is a diagram showing an example of the connection process of congestion sections. [Figure 4] It is a diagram showing an example of a branched congestion section. [Figure 5] It is a diagram showing an example of a day-of-the-week time zone classification table. [Figure 6] It is a diagram showing an example of an employment time zone classification table. [Figure 7] It is a diagram showing an example of an extension scale determination table. [Figure 8] It is a diagram showing an example of the main part configuration of the electrical system of a computer. [Figure 9] It is a flowchart showing an example of the flow of prediction processing. [Figure 10] It is a flowchart showing an example of the flow of congestion section determination processing. [Figure 11] It is a flowchart showing an example of the flow of extension length prediction processing. [Figure 12] It is a diagram showing an example of the bias of the extension tendency of congestion sections classified by congestion direction. [Figure 13] It is a diagram showing an example of RMSE for each combination of determination categories when the usage ratio of learning data is changed. [Figure 14] It is a diagram showing an example of an analysis regarding a congestion section with the lowest prediction accuracy of the extension length. [Figure 15] It is a diagram showing an example of an analysis regarding a congestion section with the highest prediction accuracy of the extension length. [Figure 16] It is a diagram showing an example of an analysis regarding a congestion section where the extension length is not "0" but the prediction accuracy of the extension length is relatively high. [Figure 17] It is a diagram showing an example of a congestion section in an area with a complex road network. [Figure 18] It is a diagram showing an example of RMSE for each combination of determination categories when the usage ratio of learning data is changed for an area with a complex road network. [Figure 19]This diagram shows an example of traffic congestion display in an existing traffic congestion prediction service. [Figure 20] This figure shows an example of a traffic congestion details screen in an existing traffic congestion prediction service. [Modes for carrying out the invention]

[0016] This embodiment will be described below with reference to the drawings. The same reference numerals are used throughout the drawings for the same components and processes, and redundant explanations are omitted.

[0017] Figure 1 is a diagram showing an example of the functional configuration of the information processing device 10 according to this disclosure. The information processing device 10 is a device that predicts the future extension of the congested section 3 from past congestion information recorded in chronological order.

[0018] Traffic congestion information is information that records the status of each traffic congestion occurring at a predetermined interval, such as every 5 minutes, over a predetermined period, in chronological order. Each traffic congestion information entry includes at least chronological information representing the time series of the traffic congestion, such as the date and time the congestion occurred, the starting point of the congestion, and the ending point of the congestion. In other words, while information recording the status of traffic congestion may also include other information such as the length of the congestion, in this embodiment, the traffic congestion information only needs to include chronological information, the starting point of the congestion, and the ending point of the congestion.

[0019] Traffic congestion information recorded at the same time contains the same time-series information. The starting and ending locations of congestion are represented, for example, by two-dimensional coordinate values ​​using latitude and longitude.

[0020] Since the information processing device 10 uses this traffic congestion information to predict the future extension of the congested section 3, each piece of traffic congestion information in chronological order will be referred to as "training data" from now on.

[0021] In this embodiment, the extension of the congested section 3 refers to the absolute increase or decrease between the congestion length of congested section 3 represented by specific training data and the congestion length of congested section 3 represented by the immediately preceding training data recorded at the same congestion point adjacent to the training data in the time series. In other words, if the training data is recorded at 5-minute intervals, the extension of congested section 3 will be the absolute increase or decrease in the congestion length of congested section 3 compared to congested section 3 5 minutes prior.

[0022] As shown in Figure 1, the information processing device 10 includes a traffic congestion classification unit 11, a traffic congestion identity determination unit 12, a congestion scale setting unit 13, a classification selection unit 14, and a congestion length prediction unit 15, as well as a storage device 16.

[0023] The congestion classification unit 11 refers to the starting and ending positions of congestion included in the training data and classifies each congestion section 3 represented by the training data according to the direction of congestion. In other words, the congestion classification unit 11 is an example of a classification unit that classifies each training data according to the direction of congestion.

[0024] There are no restrictions on the direction of congestion to be classified, but in this embodiment, as an example, the congested section 3 represented by the training data is classified into four directions: west to east, east to west, south to north, and north to south. Therefore, just as there are uphill and downhill lanes on the same road, congested section 3 occurring between opposite lanes are classified into different directions, such as west to east and east to west.

[0025] The traffic congestion classification unit 11 stores the learning data classified by traffic congestion direction as direction-specific learning data information 16A in the storage device 16.

[0026] On the other hand, when congestion occurs intermittently, a congested section 3 that forms a single congestion area as a whole may be perceived as multiple fragmented congested sections 3. In such cases, it is preferable to treat each congested section 3 not as an independent congested section 3, but as a fragment of a continuous congested section 3.

[0027] Therefore, the congestion identity determination unit 12 acquires learning data classified by congestion direction from the direction-specific learning data information 16A and determines which congestion sections 3 constitute a continuous congestion section 3. A continuous congestion section 3 represented by multiple congestion sections 3 in this way is called an "integrated congestion section 5". Hereafter, the integrated congestion section 5 will be referred to as "congested section 5". The congestion identity determination unit 12, which determines the range of congestion section 5, is an example of a determination unit.

[0028] The traffic congestion identity determination unit 12 acquires learning data classified by traffic congestion direction from direction-specific learning data information 16A, and converts the starting and ending positions of the traffic congestion included in each learning data into a geohash 1.

[0029] A geohash 1 is an example of a predetermined range, referring to each divided region of the Earth based on latitude and longitude. Each geohash 1 is represented by a different multi-digit symbol, and as the number of digits representing the geohash 1 increases, the range of the region represented by the geohash 1 becomes smaller, and the precision of the geohash 1 increases.

[0030] Figure 2 shows an example of a map display showing the start and end points of a traffic jam represented by traffic jam section 3. In the example in Figure 2, the endpoints of the traffic jam included in the training data, i.e., the start and end points of the traffic jam, are converted into geohash 1A, represented by the symbol "xn76q41", and geohash 1B, represented by the symbol "xn76q2c", respectively. When explaining each geohash 1, such as geohash 1A and geohash 1B, an alphabet letter is added to the end of the geohash 1 to distinguish them.

[0031] The congestion identity determination unit 12 selects training data for which it has not yet been determined whether or not it is a congestion section 3 that constitutes congestion section 5. The congestion identity determination unit 12 then performs a process to connect another congestion section 3 to the selected congestion section 3, such that the geohash 1 corresponding to the starting position of the congestion section 3 represented by the selected training data ("selected congestion section 3") ("starting geohash 1") is the geohash 1 corresponding to the ending position of the congestion ("ending geohash 1"). In other words, the congestion identity determination unit 12 performs a process to connect another congestion section 3 to the selected congestion section 3, such that the starting geohash 1 of the selected congestion section 3 is the ending geohash 1. Then, the congestion identity determination unit 12 recursively repeats the process of connecting congestion sections 3, treating the connected congestion section 3 as the newly selected congestion section 3, until there are no more congestion sections 3 that have the starting geohash 1 of the selected congestion section 3 as their ending geohash 1.

[0032] Furthermore, the congestion identity determination unit 12 uses the endpoint geohash 1 of the selected congestion section 3 as the starting point and performs a process to connect another congestion section 3 that has the endpoint geohash 1 of the selected congestion section 3 as its starting geohash to the selected congestion section 3. Then, the congestion identity determination unit 12 uses the connected congestion section 3 as the newly selected congestion section 3 and recursively repeats the process of connecting congestion sections 3 until there are no more congestion sections 3 that have the endpoint geohash 1 of the selected congestion section 3 as their starting geohash 1.

[0033] By performing the above processing on each training data set classified in the same direction of congestion, congestion section 5 is generated.

[0034] Figure 3 is a diagram showing an example of the connection process of congested sections 3 displayed on a map. In the example in Figure 3, congested section 3A is connected to congested section 3B, which has geohash 1C represented by the symbol "xn76v23", which is the endpoint geohash 1 of congested section 3A, as its starting geohash 1; congested section 3C, which has geohash 1D represented by the symbol "xn76v0p", which is the endpoint geohash 1 of congested section 3B, as its starting geohash 1; and congested section 3D, which has geohash 1F represented by the symbol "xn76v27", which is the starting geohash 1 of congested section 3A, as its ending geohash 1. This process generates congested section 5, which has geohash 1G represented by the symbol "xn76v8b" as its starting geohash 1 and geohash 1E represented by the symbol "xn76tpv", which is its ending geohash 1.

[0035] In the example shown in Figure 3, the starting point of the congestion represented by one of the congested sections 3 to be connected overlaps with the ending point of the congestion represented by the other congested section 3. In contrast, even if the starting point of the congestion represented by one congested section 3 and the ending point represented by the other congested section 3 are far apart, it goes without saying that adjacent congested sections 3 can be connected as long as their respective starting and ending points are included in the same geohash 1. Furthermore, even if the starting point of the congestion represented by a newly connected congested section 3 is not included in the same geohash 1 as the ending point of the congestion represented by the other congested section 3, the congested sections 3 can be connected as long as it is included in the geohash 1 of any of the congested sections 3 that were connected as a continuous congested section 3 in the previous connection operation.

[0036] In other words, the congestion identity determination unit 12 determines which congestion sections 3 represent a continuous congestion section 3 by recursively repeating the process of connecting adjacent congestion sections 3 whose starting and ending points are within a predetermined range, for each congestion section 3 classified according to the direction of congestion, until there are no more adjacent congestion sections 3 whose starting or ending points are within the predetermined range.

[0037] On the other hand, Figure 4 shows an example of a congested section 5 displayed on a map. For example, in densely populated urban areas, as shown in Figure 4, the congested section 5 may not be represented by a single line, but rather by multiple lines, with the congested section 5 branching off at point P3, which corresponds to a road junction. The existence of a road junction means that one of the roads branching off from the junction is managed as a separate road from the other road.

[0038] Therefore, it is preferable to treat the branching congested sections 5 as different congested sections 5 with respect to the branching point.

[0039] Specifically, in the case of congested section 5 shown in Figure 4, it is preferable to treat it as two separate sections: congested section 5A connecting point P1 and point P4, and congested section 5B connecting point P3 and point P5.

[0040] To this end, the congestion identity determination unit 12 shown in Figure 1 includes a division unit 12A. The division unit 12A divides the congestion section 5 using the azimuth angles of adjacent congestion sections 3 that constitute the congestion section 5. The azimuth angle of a congestion section 3 is the value representing the direction of travel of the congestion section 3 as an angle, with north being 0 degrees, east being 90 degrees, south being 180 degrees, and west being 270 degrees.

[0041] The division unit 12A calculates the difference in azimuth angles between each of the adjacent congested sections 3 that make up the congested section 5, i.e., the azimuth angle difference. If the azimuth angle difference between the congested sections 3 is greater than or equal to a predetermined angle, the division unit 12A determines that the adjacent congested sections 3 are congested sections 3 that make up different congested sections 5, and divides the congested section 5 at the connection point between the congested sections 3.

[0042] In the congested section 5 shown in Figure 4, if the azimuth angle of one adjacent congested section 3 at point P2 is 28.54 degrees and the azimuth angle of the other congested section 3 is 58.49 degrees, the difference in azimuth angles between the congested sections 3 is 29.95 degrees. Also, if the azimuth angle of one adjacent congested section 3 at point P3 is 18.07 degrees and the azimuth angle of the other congested section 3 is 81.79 degrees, the difference in azimuth angles between the congested sections 3 is 63.72 degrees.

[0043] Therefore, if the threshold for the azimuth difference of congested section 3 is set to, for example, 40 degrees, it can be determined that adjacent congested sections 3 at point P2 constitute a single congested section 5, while adjacent congested sections 3 at point P3 constitute different congested sections 5. In this way, the division unit 12A divides congested section 5 into congested section 5A and congested section 5B, respectively. Note that the threshold for the azimuth difference of congested section 3 is just an example and will be set according to road conditions.

[0044] In other words, if the difference in azimuth angles between adjacent congested sections 3 that constitute the congested section 5 is greater than or equal to a predetermined angle, the division section 12A divides the congested section 5 into two different congested sections 5 at the connection point of adjacent congested sections 3 where the difference in azimuth angles is greater than or equal to a predetermined angle.

[0045] In this way, the congestion identity determination unit 12 works in conjunction with the division unit 12A to determine the range of the congestion section 5 based on the learning data classified for each congestion direction, and calculates the extension length for each congestion section 5 from the changes in the congestion section 5 over time. The extension length of the congestion section 5 is expressed in a predetermined unit, for example, in units of 10m.

[0046] The congestion identity determination unit 12 applies the determined congestion section 5 to the learning data in each time series, which is classified according to the direction of congestion. Then, the congestion identity determination unit 12 classifies the congestion section 5 into those that can be considered the same congestion section 5, and stores the information that associates the time series information and the extension length for each congestion section 5 as congestion section information 16B in the storage device 16. Whether or not they are the same congestion can be determined, for example, by whether or not they are congestion sections 5 connected starting from the same geohash 1. Hereafter, a set of congestion sections 5 connected starting from the same geohash 1 will be referred to as "the same congestion section 5".

[0047] The extension scale setting unit 13 acquires the congested section 5 from the congested section information 16B for each congested section 5.

[0048] The extension scale setting unit 13 classifies each congested section 5 into predetermined judgment categories according to the timing of congestion, in order to obtain the extension trend in each congested section 5. A judgment category is a category established to determine whether or not there is a significant difference, i.e., a bias, in the extension trend of the congested sections 5 belonging to that category, and is set in advance by the user, for example.

[0049] In this embodiment, as an example, the congested section 5 is classified into determination categories based on the day of the week, work schedule, and time of day. However, it may also be classified into other determination categories, such as classifying days that are multiples of 5 (e.g., the 5th and the 10th) from other days. It may also be classified into determination categories based on the month, the season (spring, summer, autumn, winter), and the year. By classifying the congested section 5 by season, it becomes easier to predict the extension of seasonal congested sections 5, such as the sudden congestion caused by cherry blossom viewers that began occurring only in spring two years ago.

[0050] The day of the week classification indicates which of the seven days of the week (Sunday through Saturday) each traffic congestion occurred on. The time of day classification indicates which of the 24 time periods (divided into one-hour intervals) each traffic congestion occurred on. The work schedule classification indicates whether each traffic congestion occurred on a weekday or a holiday.

[0051] Therefore, the extension scale setting unit 13 uses the day-of-the-week and time-of-the-week classification table 19A, which is divided by a combination of day-of-the-week and time-of-the-week classifications, to classify each congested section 5 into one of the 168 classifications in the day-of-the-week and time-of-the-week classification table 19A.

[0052] Figure 5 shows an example of the day-of-the-week time-of-day classification table 19A. For example, a traffic congestion section 5 that occurred between 7:00 and 8:00 on a Friday would be classified under the category represented by "F7".

[0053] By referring to the day-of-the-week and time-of-the-week classification table 19A, we can obtain the congested sections 5 classified by day of the week and time of day. Note that while the day-of-the-week and time-of-the-week classification table 19A shown in Figure 5 divides the day into one-hour units, it goes without saying that it could also be divided into other units, such as two-hour units.

[0054] Furthermore, the extension scale setting unit 13 uses the work time classification table 19B, which is divided by combinations of work categories and time zones, to classify each congested section 5 into one of the 48 categories in the work time classification table 19B.

[0055] Figure 6 shows an example of the work time classification table 19B. For example, a traffic congestion section 5 that occurs between 14:00 and 15:00 on a holiday would be classified under the category represented by "H14". Holidays vary from person to person, but in this embodiment, Saturdays, Sundays, and public holidays are defined as holidays, and days other than holidays are defined as weekdays.

[0056] By referring to the work time classification table 19B, we can obtain the congested sections 5 classified by time zone and by work category. Note that the congested sections 5 classified by time zone can also be obtained from the day of the week time zone classification table 19A, so it is not always necessary to classify work categories by time zone, as shown in the work time classification table 19B in Figure 6; it is sufficient to simply classify them into two categories: weekdays and holidays.

[0057] In this way, the extension scale setting unit 13 stores information classifying each congested section 5 into a determination category as category-specific congested section information 16C in the storage device 16 for each congested section 5. That is, category-specific congested section information 16C exists for each congested section 5. For the sake of explanation, the day of the week time zone classification table 19A and the working time zone classification table 19B are sometimes collectively referred to as "classification table 19".

[0058] Meanwhile, the extension scale setting unit 13, in each determination category, sorts the extension lengths of the congested sections 5 in ascending or descending order for each congested section 5 that has a continuous time series, and identifies the extension length located in the middle of the sort, that is, the median extension length. A congested section 5 with a continuous time series refers to a congested section 5 that is included in the range from when the congestion occurs until it disappears, and a congested section 5 with a continuous time series will be referred to as a "continuous congested section 5" from now on.

[0059] Then, the extension scale setting unit 13 calculates the average of the median extension lengths over all sections and the standard deviation of the median extension lengths for each judgment category, and generates an extension scale determination table 4 for each judgment category. The extension scale determination table 4 is a table for determining the extension scale of the congested sections 5 classified into the corresponding judgment category.

[0060] Figure 7 shows an example of Extension Scale Determination Table 4. In Extension Scale Determination Table 4, criteria are defined for each extension scale. In the example of Extension Scale Determination Table 4 shown in Figure 7, if the extension length of congested section 5 is greater than or equal to (average of all sections of median extension length + standard deviation), the extension scale is set to "large". If the extension length of congested section 5 is greater than or equal to (average of all sections of median extension length - standard deviation) but less than (average of all sections of median extension length + standard deviation), the extension scale is set to "medium". Also, if the extension length of congested section 5 is greater than or equal to 0 and less than (average of all sections of median extension length - standard deviation), the extension scale is set to "small". Note that "average of medians" in Figure 7 refers to the average of all sections of median extension length. In this embodiment, the extension scale of congested section 5 is set to three levels: "large", "medium", and "small", but it goes without saying that other levels of granularity may be set, such as five levels or two levels.

[0061] Since the extension length of the congested section 5 is expressed in units of 10m, the extension scale setting unit 13 rounds up the last digit of each value representing (overall average of extension length - standard deviation) and (overall average of extension length + standard deviation) so that the extension length used as the judgment criterion is expressed in units of 10m. In addition, if the value of (overall average of extension length - standard deviation) becomes 0 or less, the value of (overall average of extension length - standard deviation) is set to "10".

[0062] The extension scale setting unit 13 compares the extension length of each congested section 5 classified into a determination category with the extension scale determination table 4 corresponding to that determination category to identify the extension scale for each congested section 5, and sets the extension scale with the highest number of occurrences as the extension scale for the congested section 5 in that determination category.

[0063] Specifically, the extension scale setting unit 13 classifies the extension scale of each congested section 5 that occurred on each day of the week and time of day into "large," "medium," or "small" categories by referring to the extension scale determination table 4, and then aggregates the number of classifications for each category. Finally, it determines the extension scale for that determination category to be "large," "medium," or "small" by majority vote of the number of classifications. There may be two extension scales with the most classifications, but for example, if the number of cases classified as "small" and "medium" for the congested section 5 on Monday are tied for first place with 10 cases each, the extension scale for congested section 5 on Monday will be "medium." If the number of cases classified as "small" and "large" are tied for first place, the extension scale for congested section 5 on Monday will be "medium," and if the number of cases classified as "medium" and "large" are tied for first place, the extension scale for congested section 5 on Monday will be "large."

[0064] In other words, the extension scale setting unit 13 sets the extension scale for each congested section 5 and for each determination category. The extension scale setting unit 13 stores the extension scale for each congested section 5 and for each determination category as extension scale information 16D in the storage device 16.

[0065] Thus, the extension scale setting unit 13, which sets the extension scale of the congested section 5 for each congested section 5 and for each judgment category, is an example of a setting unit.

[0066] When the classification selection unit 14 receives prediction request information including the congested section 5 and date and time to be predicted for extension length, it uses the extension scale information 16D set by the extension scale setting unit 13 to determine whether there is a bias within the judgment category in the extension scale of the congested section 5 specified by the prediction request information, i.e., the congested section 5 to be predicted for extension length. Specifically, the classification selection unit 14 determines whether there is a bias in at least one of the judgment categories for each day of the week, time of day, and work attribute with respect to the extension scale of the specified congested section 5.

[0067] If a bias exists in at least one judgment category, the category selection unit 14 selects a judgment category to be used to predict the extension length of the congested section 5, according to the combination of judgment categories in which the bias in the extension scale exists.

[0068] For example, if there is a bias in the day of the week classification, the classification selection unit 14 selects the day of the week classification; if there is a bias in the work classification, the classification selection unit 14 selects the work classification; and if there is a bias in the time of day classification, the classification selection unit 14 selects the time of day classification. Also, for example, if there is a bias in both the day of the week classification and the time of day classification, the classification selection unit 14 selects both the day of the week classification and the time of day classification; and if there is a bias in both the work classification and the time of day classification, the classification selection unit 14 selects both the work classification and the time of day classification. Since both the day of the week classification and the work classification are classifications based on the day of the week, if there is a bias in both the day of the week classification and the work classification, it is sufficient to consider that the bias exists only in the classification with the stronger degree of bias. If the bias is the same in both the day of the week classification and the work classification, the work classification is given priority. This makes it possible to increase the amount of data used for prediction and improve the prediction accuracy of the extension length of the congested section 5.

[0069] As will be described later, the determination category selected by the category selection unit 14 is used to predict the extension length of the congestion section 5 specified by the prediction request information.

[0070] In other words, the classification selection unit 14 uses the extension scale for each congested section and each classification, obtained from the overall average and standard deviation of the median extension length for each consecutive identical congested section 5 classified by day of the week, work attribute, and time of day, to determine whether there is a bias in the extension scale of the designated congested section 5 within the classification. If there is a bias in the extension scale of the designated congested section 5 within the classification, the unit selects the classification to be used to predict the extension length of the congested section 5 according to the combination of classifications in which the bias in extension scale exists.

[0071] Thus, the classification selection unit 14, which selects a judgment category to be used in predicting the extension length of the congested section 5, is an example of a selection unit.

[0072] The extension length prediction unit 15 obtains the extension length for each designated congestion section 5, which is classified according to the determination category selected by the category selection unit 14, from the category-specific congestion section information 16C, and predicts the extension length of the designated congestion section 5 at the specified date and time.

[0073] For example, suppose the classification selected by the classification selection unit 14 is a day of the week classification, and the date and time specified by the prediction request information is Monday. In this case, the extension length prediction unit 15 refers to the day of the week and time zone classification table 19A of the specified congested section 5 that constitute the classification-specific congestion section information 16C, and outputs the average of the extension lengths of all congested section 5 classified as Monday as the predicted extension length for the specified congested section 5. In the example of the day of the week and time zone classification table 19A shown in Figure 5, the average of the extension lengths of the congested section 5 classified into categories from "M0" to "M23" becomes the predicted extension length for the congested section 5 specified by the prediction request information.

[0074] Furthermore, suppose the judgment category selected by the category selection unit 14 is a time zone category, and the date and time specified by the prediction request information corresponds to the time zone between 18:00 and 19:00. In this case, the extension length prediction unit 15 refers to the day-of-the-week time zone classification table 19A of the specified congestion section 5 that constitutes the category-specific congestion section information 16C, and outputs the average extension length of all congestion sections 5 classified in the time zone between 18:00 and 19:00 as the predicted extension length for the specified congestion section 5. In the example of the day-of-the-week time zone classification table 19A shown in Figure 5, the average extension lengths of the congestion sections 5 classified into the categories "M18", "Tu18", "W18", "Th18", "F18", "Sa18", and "Su18" become the predicted extension length for the congestion section 5 specified by the prediction request information.

[0075] Furthermore, suppose the judgment category selected by the category selection unit 14 is a day of the week category and a time zone category, and the date and time specified by the prediction request information corresponds to the time zone from 18:00 onwards to before 19:00 on Monday. In this case, the extension length prediction unit 15 refers to the day of the week and time zone category table 19A of the specified congested section 5 that constitutes the category-specific congestion section information 16C, and outputs the average extension length of all congested sections 5 classified in the time zone from 18:00 onwards to before 19:00 on Monday as the predicted extension length for the specified congested section 5. In the example of the day of the week and time zone category table 19A shown in Figure 5, the average extension length of the congested section 5 classified in the "M18" category becomes the predicted extension length for the congested section 5 specified by the prediction request information.

[0076] Thus, the extension length prediction unit 15, which predicts the extension length of the congested section 5 specified by the prediction request information, is an example of a prediction unit.

[0077] In this embodiment, a configuration in which the information processing device 10 includes a storage device 16 has been described, but the information processing device 10 does not necessarily have to include a storage device 16. In this case, the information processing device 10 may store and acquire various types of information with the storage device 16 connected to a communication line, for example, via a communication unit 27, which will be described later. Alternatively, the information processing device 10 may be separated into a first device including a congestion classification unit 11, a congestion identity determination unit 12, and an extension scale setting unit 13, and a second device including a classification selection unit 14 and an extension length prediction unit 15. The first device is an example of an information processing device 10 that takes learning data as input and outputs classification congestion section information 16C and extension scale information 16D, and the second device is an example of an information processing device 10 that takes classification congestion section information 16C, extension scale information 16D, and prediction request information as input and outputs a predicted value of the extension length in the congestion section 5 specified by the prediction request information.

[0078] In the example shown above, the extension length of the congestion in section 5 was predicted, but this method can also be applied to predict the length of the congestion in section 5 itself. By predicting the length of the congestion, it becomes possible to notify drivers of the occurrence of congestion in section 5 without having to refer to past congestion information.

[0079] An information processing device 10 having such functions is configured, for example, using a computer 20. Figure 8 is a diagram showing an example of the main components of the electrical system of the computer 20 applied to the information processing device 10.

[0080] Computer 20 includes a CPU (Central Processing Unit) 21 responsible for processing in each functional unit of the information processing device 10 shown in Figure 1. Computer 20 also includes a ROM (Read Only Memory) 22 for storing information processing programs that enable computer 20 to function as the information processing device 10, and a RAM (Random Access Memory) 23 used as a temporary workspace for the CPU 21. Furthermore, computer 20 includes non-volatile memory 24 and an input / output interface (I / O) 25. The CPU 21, ROM 22, RAM 23, non-volatile memory 24, and I / O 25 are connected by a bus 26.

[0081] The non-volatile memory 24 is an example of a storage device 16 that maintains stored information even when the power supplied to the non-volatile memory 24 is cut off. For example, semiconductor memory is used, but a hard disk may also be used. The non-volatile memory 24 does not necessarily have to be included in the computer 20; for example, a portable non-volatile memory 24 that can be attached to and detached from the computer 20 may be used.

[0082] The non-volatile memory 24 stores, for example, direction-specific learning data information 16A, congestion section information 16B, category-specific congestion section information 16C, and extension scale information 16D.

[0083] For example, a communication unit 27, an input unit 28, and a display unit 29 are connected to I / O 25.

[0084] The communication unit 27 is connected to a communication line such as the Internet and a LAN (Local Area Network), and is equipped with a communication protocol for data communication with external devices (not shown) also connected to the communication line. The CPU 21 receives, for example, training data from external devices through the communication line connected to the communication unit 27. The communication line connected to the communication unit 27 may be either wired or wireless.

[0085] The input unit 28 is a device that receives user instructions and notifies the CPU 21 of the content of the received instructions. Examples of input units include buttons, touch panels, keyboards, and mice.

[0086] The display unit 29 is an example of a device that visually displays information processed by the CPU 21, and may include, for example, a liquid crystal display or an organic EL (Electro-Luminescence) display.

[0087] When the information processing device 10 receives user instructions from an external device via a communication line and transmits the processed information to the external device via the communication line according to the instructions, the input unit 28 and the display unit 29 do not necessarily need to be connected to the I / O 25.

[0088] Next, the operation of the information processing device 10 of this disclosure will be described.

[0089] Figure 9 is a flowchart showing an example of the flow of prediction processing performed by the CPU 21 of the information processing device 10.

[0090] The information processing program that defines the prediction process is pre-stored in, for example, the ROM 22 of the information processing device 10. The CPU 21 of the information processing device 10 reads the information processing program stored in the ROM 22 and executes the prediction process. It is assumed that training data is pre-stored in the non-volatile memory 24.

[0091] In step S10 of Figure 9, the CPU 21 executes a congestion determination process that determines congestion section 3 from the training data and generates congestion section information 16B.

[0092] Next, in step S20, the CPU 21 uses the congestion section information 16B generated in step S10 to perform an extension length prediction process to predict the extension length of the congestion section 3 specified by the prediction request information at the specified date and time, and then terminates the prediction process shown in Figure 9.

[0093] Figure 10 is a flowchart showing an example of the flow of the congestion section determination process performed in step S10 of the prediction process shown in Figure 9.

[0094] In step S100 of Figure 10, the CPU 21 acquires training data from the non-volatile memory 24 and classifies each acquired training data according to the direction of congestion. This generates direction-specific training data information 16A.

[0095] In step S110, the CPU 21 selects one of the congestion directions that will be used to classify the training data. The congestion direction selected in step S110 is called the selected congestion direction.

[0096] In step S120, the CPU 21 selects the training data containing the oldest time-series information from the direction-specific training data information 16A among the unselected training data classified as the selected congestion direction. If there are multiple training data containing time-series information representing the same date and time, multiple training data will be selected. The training data selected in step S120 is called selected training data.

[0097] In step S130, the CPU 21 converts the start and end locations of the congestion included in the selected training data into geohash 1.

[0098] In step S140, the CPU 21 uses the geohash 1 converted in step S130 to determine which of the congested sections 3 represented by each selected learning data constitute a continuous congested section 5, and generates the congested section 5.

[0099] In step S150, the CPU 21 calculates the length of each congested section 5, classifies the congested sections 5 into those that can be considered to be the same congested section 5, and calculates the extension length of each congested section 5 included in the same congested section 5 by comparing the length of each congested section 5 with the length of the congested section 5 observed in the previous time series. For each congested section 5, the CPU 21 associates the time series information included in the selected learning data with the calculated extension length.

[0100] The length of the congested section 5 can be obtained, for example, by referring to map data stored in the non-volatile memory 24 beforehand and calculating the distance along the road with the starting and ending points of the congested section 5 as endpoints. Furthermore, if there is no congested section 5 observed in the immediately preceding time series for the generated congested section 5, the CPU 21 can simply set the extension length to "0".

[0101] In step S160, the CPU 21 determines whether there is any unselected training data among the training data classified in the selected congestion direction. If there is unselected training data, the process proceeds to step S120 and steps S120 to S160 are repeated. As a result, the training data classified in the selected congestion direction is selected in chronological order, and the chronological information and extension length are associated for each congestion section 5 in the selected congestion direction.

[0102] On the other hand, if the judgment process in step S160 determines that there is no unselected training data, the process proceeds to step S170.

[0103] In step S170, the CPU 21 determines whether there are any unselected congestion directions among the congestion directions classified by the learning class. If there are unselected congestion directions, the process proceeds to step S110, and steps S110 to S170 are repeatedly executed. As a result, time-series information and extension length are associated for each congestion section 5 of all congestion directions, and congestion section information 16B is generated.

[0104] The congestion section determination process shown in Figure 10 is now complete.

[0105] On the other hand, Figure 11 is a flowchart showing an example of the flow of the extension length prediction process performed in step S20 of the prediction process shown in Figure 9.

[0106] In step S200 of Figure 11, the CPU 21 selects the same congested section 5 from the congested section information 16B. The congested section 5 selected in step S200 is called selected congested section 5.

[0107] In step S210, the CPU 21 classifies the selected congested section 5 according to the classification table 19 and generates congested section information 16C by classification.

[0108] In step S220, the CPU 21 generates an extension scale determination table 4 for each determination category using the average of the median extension lengths across all sections and the standard deviation of the median extension lengths, which are calculated from the extension lengths of the congested sections 5 classified into the determination category.

[0109] In step S230, the CPU 21 compares the extension length of each consecutive congested section 5 classified into each judgment category with the extension scale judgment table 4 in the judgment category to which the consecutive congested section 5 is classified, and determines which extension scale the extension length of the consecutive congested section 5 belongs. Then, the CPU 21 sets the extension scale that has the most extension lengths classified as the extension scale of the congested section 5 in that judgment category. If there are multiple extension scales that have the most extension lengths classified as such, the CPU 21 sets the extension scale of the congested section 5 in the judgment category based on predetermined rules. For example, if the number of extension scale classifications is tied for first place between "small" and "medium," or between "medium" and "large," the larger extension scale is adopted. If the number of extension scale classifications is tied for first place between "large" and "small," "medium," which is the extension scale between "large" and "small," is adopted. In other words, the CPU 21 sets the scale of the congestion in the selected congestion section 5 for each determination category.

[0110] In step S240, the CPU 21 determines whether there are any unselected congested sections 5 in the congested section information 16B. If there are unselected congested sections 5, the process proceeds to step S200, and steps S200 to S240 are repeatedly executed. This allows for obtaining the extension scale of congested sections 5 for each congested section 5 and for each determination category, based on the learned data.

[0111] In step S250, the CPU 21 determines, for each determination category, whether or not there is a bias in the scale of extension of the congested section 5, that is, the congested section 6, which is the target of extension length prediction.

[0112] Specifically, CPU 21 determines that there is no bias if the extension scale of all predicted congestion sections 6 in the judgment category is the same, and determines that there is a bias if at least one extension scale differs from the others. It goes without saying that the determination of whether or not there is a bias in the judgment category is not limited to this method, and other methods may also be used.

[0113] In step S260, the CPU 21 selects a combination of judgment categories that were determined to have a bias in step S250. If no judgment categories have a bias, the CPU 21 does not select any judgment categories. In this case, the CPU 21 will predict the extension length of the congested section 5 by referring to the congested section 5 for all days of the week and all time periods.

[0114] The CPU 21 predicts the length of congestion in the predicted congestion section 6 at the specified date and time by taking the average length of all congestion sections 5 that are classified into the category corresponding to the specified date and time of the predicted congestion section 6 within the selected judgment category. Then, the CPU 21 outputs the predicted length of congestion.

[0115] The CPU 21 may output the predicted extension length in any form, as long as the user can recognize the predicted extension length from the information processing device 10. For example, the CPU 21 may display the predicted extension length on the display unit 29, or transmit the predicted extension length to an external device connected to a communication line via the communication unit 27. The CPU 21 may also print the predicted extension length on paper, or store the predicted extension length in the non-volatile memory 24. If the predicted extension length exceeds a certain level, it may be used to change the guidance route in the road guidance navigation system to a detour route that bypasses the congested section 5.

[0116] With the above steps completed, the extension length prediction process shown in Figure 11 is terminated, and at the same time, the prediction process shown in Figure 9 is terminated.

[0117] <Verification Results> Next, the verification results of the predicted extension length of the congested section 5 using the information processing device 10 according to this embodiment are shown.

[0118] To conduct this verification, training data from October 2019 to February 2021 in a predetermined verification area A was collected from traffic congestion information provided by the Japan Road Traffic Information Center (JARTIC). The predicted extension length of congested section 5, generated using each extension scale information 16D created by varying the amount of training data, was compared with the actual extension length of congested section 5. The congested section 5 shown on the map in Figures 14, 15, 16, and 17 represents the training results obtained using traffic congestion information acquired from the Japan Road Traffic Information Center (JARTIC).

[0119] Specifically, the RMSE of the predicted extension length of the congested section 5, i.e., the predicted extension length of the congested section 5, and the actual extension length of the congested section 5, i.e., the actual extension length of the congested section 5, were calculated using the extension scale information 16D to evaluate the prediction accuracy of the congested section 5 in the information processing device 10.

[0120] RMSE is an abbreviation for "Root Mean Squared Error". RMSE is an example of an accuracy evaluation index that is expressed as the root mean square of the difference between the predicted extension of congested section 5 and the actual extension of congested section 5, and is calculated by equation (1).

[0121]

number

[0122] In equation (1), “n” is the predicted extension length of congested section 5, “k” is the index of the predicted extension length, and “f” k " is the k-th predicted extension, and "y k " represents the k-th actual extension.

[0123] As can be seen from equation (1), RMSE is an accuracy evaluation index that indicates higher prediction accuracy as the value approaches 0.

[0124] The information processing device 10 classified the learning data representing traffic congestion information in verification area A by direction of congestion and used an 8-digit geohash 1 to determine the congestion sections 5, resulting in 63 classifications of congestion sections 5.

[0125] Figure 12 summarizes the 63 congested sections 5 by direction of congestion and shows an example of the number of congested sections 5 with a bias in extension trends for each judgment category.

[0126] In Figure 12, the denominator of each column represents the number of congestion sections 5 classified according to the congestion direction corresponding to the row direction, and the numerator represents the number of congestion sections 5 in each congestion direction that show a bias in the judgment category corresponding to the column direction.

[0127] Figure 13 shows an example of the RMSE for each combination of judgment categories that takes into account the bias in the extension trend, when predicting the extension length of congested section 5 while changing the usage ratio of training data in 10% increments from 10% to 100%. The RMSE when all extension lengths of congested section 5 were set to "0" was "163.7245". The unit of RMSE is meters.

[0128] As shown in the RMSE example in Figure 13, when the proportion of training data used to predict the extension length of congested section 5 ranges from 10% to 40% of the total training data, predicting the extension length of congested section 5 while considering the bias in the extension trend tends to result in higher prediction accuracy than predicting the extension length of congested section 5 without considering the bias in the extension trend. In other words, as the amount of training data used to predict the extension length of congested section 5 decreases, predicting the extension length of congested section 5 while considering the bias in the extension trend results in higher prediction accuracy.

[0129] Next, we will analyze the characteristics of the congested sections 5 with relatively low RMSE (Remote Mean Squared) scores, i.e., congested sections 5 with low prediction accuracy, and the characteristics of the congested sections 5 with relatively high RMSE scores, i.e., congested sections 5 with high prediction accuracy, out of the 63 classified congested sections 5.

[0130] Figure 14 shows an example of analysis for the congested section 5 that had the lowest prediction accuracy for extension length among the 63 classified congested sections 5.

[0131] The congested section 5 analyzed in Figure 14 is a congested section 5 running from east to west, and is included in geohash 1 of “xn76s76g”. Furthermore, the number of congestion occurrences by extension scale in the congested section 5 analyzed in Figure 14 was 50 for “small”, 24 for “medium”, and 105 for “large”, resulting in an RMSE of “399.4”.

[0132] In the congested section 5 analyzed in Figure 14, the extension length exceeding 1000m occurred between 18:00 and 21:00. This likely resulted in a relatively larger error in the predicted extension length for congested section 5, considering the bias in extension trends by day of the week and time of day, leading to a higher RMSE. In other words, congested section 5, which is more prone to congestion with extension lengths exceeding 1000m compared to other congested sections 5, tends to show a decrease in the accuracy of extension length predictions.

[0133] Figure 15 shows an example of analysis for the congested section 5 that had the highest prediction accuracy for extension length among the 63 classified congested sections 5.

[0134] The congested section 5 analyzed in Figure 15 is a congested section 5 running from north to south, and is included in geohash 1 of “xn76svef”. Furthermore, the number of congestion occurrences by extension scale in the congested section 5 analyzed in Figure 15 was 89 for “small”, 0 for “medium”, and 0 for “large”, resulting in an RMSE of “0.0”.

[0135] In the congested section 5 analyzed in Figure 15, there were only 89 congestion incidents, which is less than half the number of congestion incidents in the congested section 5 analyzed in Figure 14. Moreover, the extension length of the congested section 5 analyzed in Figure 15 was "0" in all cases, and the scale of the extension of the congested section 5 was classified as "small". In other words, there is a tendency for the accuracy of extension length prediction to increase as the number of congestion incidents decreases and the extension length of the congested section 5 approaches "0".

[0136] On the other hand, among the congested sections 5, there are some where the extension length of the congested section 5 is not "0", but the prediction accuracy of the extension length is higher compared to other congested sections 5.

[0137] Figure 16 shows an example of an analysis of such a congested section 5. The congested section 5 analyzed in Figure 16 is a congested section 5 running from west to east and is included in geohash 1 of “xn76ubgy”. Furthermore, the number of congestion occurrences by extension scale in the congested section 5 analyzed in Figure 16 was 496 for “small”, 1608 for “medium”, and 1143 for “large”, and the RMSE was “79.4”.

[0138] In the congested section 5 analyzed in Figure 16, congestion occurred between 8:00 and 18:00 on weekdays, and the number of congestion occurrences was higher than in the congested sections 5 analyzed in Figures 14 and 15. This indicates a tendency for congestion to extend, with the most congested section 5 being classified as "medium" in scale. In other words, congestion classified as "medium" in scale is more likely to occur, and there is a bias in the time of day and day of the week when extensions occur. Therefore, it is thought that the accuracy of predicting the extension length tended to be higher in congested section 5.

[0139] In addition, while it is preferable for the congested section 5 shown in Figure 16 to be displayed along the dotted line 7 representing the road, due to errors in the start and end positions of the congested section 3 included in the training data, the congested section 5 is displayed shifted from the dotted line 7.

[0140] Up to this point, we have explained the verification results for the predicted extension length of congestion section 5 in verification area A. Next, we will explain the verification results for the predicted extension length of congestion section 5 in verification area B, which is different from verification area A.

[0141] The data collection period for training data in test area B was the same as that for test area A, but the amount of training data collected in test area B during the same period was approximately 6.7 times that of test area A. Furthermore, test area B has a more complex network of roads than test area A.

[0142] The information processing device 10 classified the learning data representing traffic congestion information in verification area B by direction of congestion and used an 8-digit geohash 1 to determine the congestion sections 5, resulting in 204 congestion sections 5. However, of the 204 classified congestion sections 5, 25 had branching points.

[0143] Figure 17 shows an example of a congested section 5 in verification area B. The congested section 5 shown in Figure 17 includes points P6, P7, and P8, which correspond to the branching points.

[0144] At point P6, if the azimuth angle of one adjacent congested section 3 is 15.21 degrees and the azimuth angle of the other congested section 3 is 74.76 degrees, the difference in azimuth angles between the congested sections 3 is 59.55 degrees. At point P7, if the azimuth angle of one adjacent congested section 3 is 62.28 degrees and the azimuth angle of the other congested section 3 is 6.64 degrees, the difference in azimuth angles between the congested sections 3 is 55.64 degrees. Also, at point P8, if the azimuth angle of one adjacent congested section 3 is 6.64 degrees and the azimuth angle of the other congested section 3 is 80.06 degrees, the difference in azimuth angles between the congested sections 3 is 73.42 degrees.

[0145] Therefore, if the threshold for the azimuth angle difference in congested section 3 is set to, for example, 40 degrees, the congested section 5 shown in Figure 17 will be divided into three congested sections 5.

[0146] The threshold for the azimuth angle difference in congested section 3 can be set to an angle smaller than the minimum azimuth angle difference among the azimuth angle differences between different congested sections 3 that make up congested section 5.

[0147] By dividing the congested section 5, the congested section 5 in verification area B was classified into 235 congested section 5.

[0148] Figure 18 shows an example of the RMSE for each combination of judgment categories that takes into account the bias in the extension trend when predicting the extension length of congested section 5 while changing the usage ratio of the training data from 10% to 100% in 10% increments from the oldest data. Note that the RMSE when the extension length of congested section 5 was set to "0" for all cases was "179.8048".

[0149] As shown in the RMSE example in Figure 18, when the proportion of training data used to predict the extension length of congested section 5 is 10% of the oldest training data, it is shown that predicting the extension length of congested section 5 while considering the bias in the extension trend results in higher prediction accuracy than predicting the extension length of congested section 5 without considering the bias in the extension trend.

[0150] Furthermore, in the RMSE example in verification area A shown in Figure 13, the proportion of training data used to predict the extension length of congested section 5 ranged from 10% to 40% of the total training data, and it was found that predicting the extension length of congested section 5 while considering the bias in the extension trend tended to result in higher prediction accuracy than predicting the extension length of congested section 5 without considering the bias in the extension trend. This phenomenon stems from the fact that the number of training data collected in verification area B is greater than the number of training data collected in verification area A.

[0151] In other words, the same trend observed in verification area B as in verification area A—that the fewer the amount of training data used to predict the extension length of congested section 5, the higher the accuracy of the extension length prediction when considering the bias in the extension trend—is also evident in the verification results for verification area B.

[0152] Although one embodiment of the information processing device 10 has been described above, the disclosed embodiment of the information processing device 10 is merely an example, and the embodiment of the information processing device 10 is not limited to the scope described in this embodiment. Various modifications or improvements can be made to this embodiment without departing from the gist of this disclosure, and such modified or improved embodiments are also included within the technical scope of the disclosure. For example, the order of the prediction processing, including the congestion section determination processing shown in Figure 10 and the extension length prediction processing shown in Figure 11, may be changed without departing from the gist of this disclosure.

[0153] Furthermore, this disclosure describes a method of implementing prediction processing in software as an example. However, processing equivalent to the flowcharts shown in Figures 10 and 11 may also be implemented in hardware, for example, using an ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), or PLD (Programmable Logic Device). In this case, processing speed can be increased compared to implementing prediction processing in software.

[0154] Thus, the CPU 21 of the information processing device 10 may be replaced with a dedicated processor specialized for specific processing, such as an ASIC, FPGA, PLD, GPU (Graphics Processing Unit), and FPU (Floating Point Unit).

[0155] The prediction process may be implemented by a single CPU 21, or by a combination of two or more processors of the same or different types, such as multiple CPUs 21, or a combination of a CPU 21 and an FPGA.

[0156] Furthermore, the prediction process may be implemented, for example, through the collaboration of processors located in physically separate locations connected via the Internet.

[0157] Furthermore, although this embodiment describes an example in which the information processing program is stored in the ROM 22 of the information processing device 10, the storage location of the information processing program is not limited to the ROM 22. The information processing program of this disclosure can also be provided in a form recorded on a storage medium readable by the computer 20. For example, the information processing program may be provided in a form recorded on an optical disc such as a CD-ROM (Compact Disk Read Only Memory) or DVD-ROM (Digital Versatile Disk Read Only Memory). Alternatively, the information processing program may be provided in a form recorded on a portable semiconductor memory such as a USB (Universal Serial Bus) memory or a memory card.

[0158] ROM22, non-volatile memory24, CD-ROM, DVD-ROM, USB, and memory cards are examples of non-transitory storage media.

[0159] Furthermore, the information processing device 10 may download an information processing program from an external device via the communication unit 27 and store the downloaded information processing program in, for example, the non-volatile memory 24. In this case, the information processing device 10 reads the information processing program downloaded from the external device and performs prediction processing.

[0160] All documents, patent applications, and technical standards described herein are incorporated by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted to be incorporated by reference.

[0161] With regard to the embodiments described above, the following additional information is disclosed.

[0162] (Additional note 1) Memory and At least one processor connected to the memory, Includes, The aforementioned processor, Each congested section, represented by learning data including the starting and ending points of the congestion, is classified according to the congestion direction represented by the starting and ending points. For each congested section classified by direction of congestion, the process of connecting adjacent congested sections whose starting and ending points are within a predetermined range as a single integrated congested section is repeated recursively until there are no more starting or ending points of adjacent congested sections within the predetermined range, thereby determining which congested sections constitute the integrated congested section. An information processing device configured to acquire the aforementioned integrated congestion sections for each integrated congestion section representing the same congestion, classify each integrated congestion section according to determination categories that divide the congestion occurrence time by day of the week, by work attribute indicating whether it is a weekday or a holiday, and by time of day, and set the extension scale of the integrated congestion section for each integrated congestion section representing the same congestion and for each determination category.

[0163] (Additional note 2) Memory and At least one processor connected to the memory, Includes, The aforementioned processor, Using the extension length for each congested section obtained from the extension length for each consecutive identical congested section classified into determination categories based on the timing of occurrence by day of the week, by work attribute indicating whether it is a weekday or holiday, and by time of day, and using the extension scale for each determination category, it is determined for each determination category whether or not there is a bias in the extension scale of the designated congested section. If there is a bias in the extension scale of the designated congested section within the determination category, the determination category to be used to predict the extension length of the congested section is selected according to the combination of determination categories in which the bias in extension scale exists. An information processing device configured to predict the extension of the designated congestion section at a specified date and time using the extension of the designated congestion section included in the selected determination category.

[0164] (Additional note 3) Memory and At least one processor connected to the memory, Includes, The aforementioned processor, Each congested section, represented by learning data including the starting and ending points of the congestion, is classified according to the congestion direction represented by the starting and ending points. For each congested section classified by direction of congestion, the process of connecting adjacent congested sections whose starting and ending points are within a predetermined range as a single integrated congested section is repeated recursively until there are no more starting or ending points of adjacent congested sections within the predetermined range, thereby determining which congested sections constitute the integrated congested section. The aforementioned integrated congestion section is acquired for each integrated congestion section representing the same congestion, and for each integrated congestion section representing the same congestion, the integrated congestion section is classified according to each determination category, which is divided by the day of the week, by work attribute indicating whether it is a weekday or a holiday, and by time of day, and the extension scale of the integrated congestion section is set for each integrated congestion section representing the same congestion and for each determination category, Using the extension scale, determine whether there is a bias in the extension scale of the designated congested section within the determination category, and if there is a bias in the extension scale of the designated congested section within the determination category, select the determination category to be used to predict the extension length of the congested section according to the combination of determination categories in which the bias in extension scale exists. An information processing device configured to predict the extension of the designated congestion section at a specified date and time using the extension of the designated congestion section included in the selected determination category.

[0165] (Additional note 4) On the computer, Each congested section, represented by learning data including the starting and ending points of the congestion, is classified according to the congestion direction represented by the starting and ending points. For each congested section classified by direction of congestion, the process of connecting adjacent congested sections whose starting and ending points are within a predetermined range as a single integrated congested section is repeated recursively until there are no more starting or ending points of adjacent congested sections within the predetermined range, thereby determining which congested sections constitute the integrated congested section. An information processing program for performing the following steps: acquiring the aforementioned integrated congestion sections for each integrated congestion section representing the same congestion; classifying each integrated congestion section according to determination categories that divide the congestion occurrence time by day of the week, by work attribute indicating whether it is a weekday or holiday, and by time of day; and setting the extension scale of the integrated congestion section for each integrated congestion section representing the same congestion and for each determination category.

[0166] (Additional note 5) On the computer, Using the extension length for each congested section obtained from the extension length for each consecutive identical congested section classified into determination categories based on the timing of occurrence by day of the week, by work attribute indicating whether it is a weekday or holiday, and by time of day, and using the extension scale for each determination category, it is determined for each determination category whether or not there is a bias in the extension scale of the designated congested section. If there is a bias in the extension scale of the designated congested section within the determination category, the determination category to be used to predict the extension length of the congested section is selected according to the combination of determination categories in which the bias in extension scale exists. An information processing program for performing a process to predict the extension of the designated congested section at a specified date and time, using the extension length of the designated congested section included in the selected determination category.

[0167] (Additional note 6) A non-temporary storage medium that stores a program executable by a computer to perform predictive processing, The aforementioned prediction process, A classification step of classifying each congested section, represented by learning data including the starting and ending points of the congestion, according to the congestion direction represented by the starting and ending points, A determination step to determine which congested sections constitute the integrated congested section, by recursively repeating the process of connecting adjacent congested sections whose starting and ending points are within a predetermined range, for each congested section classified according to the direction of congestion, until there are no more starting or ending points of adjacent congested sections within the predetermined range, thereby determining which congested sections constitute the integrated congested section. A setting step involves classifying each of the aforementioned integrated congestion sections that represent the same congestion according to various determination categories that divide the congestion occurrence time by day of the week, by work attribute indicating whether it is a weekday or a holiday, and by time of day, and setting the extension scale of the aforementioned integrated congestion section for each of the aforementioned integrated congestion sections that represent the same congestion, and for each of the aforementioned determination categories. Using the extension scale, determine whether there is a bias within the determination category in the extension scale of the designated congested section, and if there is a bias within the determination category in the extension scale of the designated congested section, select the determination category to be used to predict the extension length of the congested section according to the combination of determination categories in which the bias in extension scale exists; A prediction step of predicting the extension of the designated congested section at a specified date and time using the extension length of the designated congested section included in the selected determination category, Non-temporary storage media including [this].

Claims

1. A determination unit determines which congested sections constitute the integrated congested section by recursively repeating the process of connecting intermittently occurring congested sections, where the start and end points of the congestion are within a predetermined range, for each congested section classified according to the direction of congestion, until there are no more congested sections with a start or end point within the predetermined range. A setting unit that, for each of the integrated congestion sections representing the same congestion, classifies the integrated congestion section according to each determination category which is divided by the day of the week, by work attribute indicating whether the congestion occurs on a weekday or holiday, and by time of day, and sets the extension scale of the integrated congestion section for each integrated congestion section representing the same congestion and for each determination category, A selection unit that uses the extension scale to determine whether there is a bias within the determination category in the extension scale of the designated congested section, and if there is a bias within the determination category in the extension scale of the designated congested section, selects the determination category to be used to predict the extension length of the congested section according to the combination of determination categories in which the bias in extension scale exists, A prediction unit that predicts the extension length of the designated congested section at a specified date and time using the extension length of the designated congested section included in the selected determination category, Equipped with an information processing device.

2. A selection unit that uses the extension length of consecutive identical congested sections, each classified into determination categories based on the timing of occurrence by day of the week, by work attribute indicating whether it is a weekday or holiday, and by time of day, and the extension scale for each determination category, to determine whether there is a bias in the extension scale of a designated congested section for each determination category, and if there is a bias in the extension scale of the designated congested section within the determination category, selects the determination category to be used to predict the extension length of the congested section according to the combination of determination categories in which the bias in extension scale exists, A prediction unit predicts the extension length of the designated congested section at a specified date and time using the extension length of the designated congested section included in the determination category selected by the selection unit, Equipped with an information processing device.

3. The prediction unit determines that there is a bias in the scale of extension of the designated congested section. Using the extension length of the designated congestion section included in the combination of determination categories corresponding to the specified date and time, among the combinations of categories, the extension length of the designated congestion section at the specified date and time is predicted. The information processing apparatus according to claim 2.

4. The determination unit determines the congested sections that constitute the integrated congested section without limiting the starting point of the integrated congested section to a specific location on the road. The information processing apparatus according to claim 1.

5. An information processing program for causing a computer to function as a component of an information processing device described in any one of claims 1 to 4.

6. An information processing method in an information processing device including a classification unit, a determination unit, a setting unit, a selection unit, and a prediction unit, The classification unit performs a classification step of classifying each congested section, which is represented by learning data including the starting point and ending point of the congestion, according to the congestion direction represented by the starting point and ending point. The determination unit performs a determination step in which it determines which congested sections constitute the integrated congested section by recursively repeating the process of connecting congested sections whose starting and ending points are within a predetermined range, for each congested section classified according to the direction of congestion, until there are no more congested sections whose starting or ending points are within the predetermined range. The setting unit classifies each of the integrated congestion sections representing the same congestion according to the determination categories which are divided by day of the week, by work attribute indicating whether the congestion occurs on a weekday or holiday, and by time of day, and sets the extension scale of the integrated congestion section for each integrated congestion section representing the same congestion and for each determination category, The selection unit uses the extension scale to determine whether there is a bias within the determination category in the extension scale of the designated congested section, and if there is a bias within the determination category in the extension scale of the designated congested section, it selects the determination category to be used to predict the extension length of the congested section according to the combination of determination categories in which the bias in extension scale exists. The prediction step involves the prediction unit predicting the extension length of the designated congested section at a specified date and time using the extension length of the designated congested section included in the selected determination category, Information processing methods including

Images