JP6420647B2 - Congestion prediction device, congestion prediction system, and congestion prediction program - Google Patents

Description

  The present invention relates to a congestion prediction device, a congestion prediction system, and a congestion prediction method.

  A mobile communication carrier (so-called network operator) has position information of various mobile bodies (for example, mobile phone terminals) in order to provide wireless communication services such as data communication and voice communication. A technique for estimating the population existing in the observation area using the above positional information has been proposed (for example, Patent Document 1).

  Population estimation is applied in various ways. For example, the degree of congestion for each area (for example, a regional mesh) is predicted by short-term population estimation, which is useful for determining a user's behavior (for example, determining whether to go out). In addition, long-term population projections predict future population trends by area, which can be used for city planning and store opening plans.

International Publication No. 2012/105516

  When forecasting the future population based on analysis of time series data using past population estimation results, it is common to select multiple forecast models and then select the appropriate forecast model, and then select the forecast model A method of performing an actual prediction process using the is adopted.

  In the above prediction, the calculation load when generating the prediction model is particularly high. Therefore, when estimating a short-term population (degree of congestion) that is required to produce a prediction result within a shorter time, it is difficult to make all areas a prediction target.

  With regard to the above problems, it is also possible to target an area where there are many residents and an area where a POI (Point Of Interest) such as a tourist spot exists. However, when such static area selection is employed, an area that should be predicted (an area that is highly likely to be crowded) may leak from the prediction target.

  In view of the above circumstances, an object of the present invention is to reduce the amount of calculation while maintaining the accuracy of the congestion prediction by appropriately selecting the prediction target area that is the target of the congestion prediction.

  The congestion prediction device according to the present invention includes a congestion degree acquisition unit that acquires a past congestion degree for each area and for each sub period, and a congestion degree that classifies each of the acquired congestion degrees into at least one of two or more categories. A classification unit, a congestion degree totaling unit that aggregates the classified congestion degree for each area and for each category and obtains a congestion degree aggregation value, and the congestion degree aggregation value that is aggregated by the congestion degree aggregation unit Based on this, a model that generates a congestion prediction model for each of the area to be predicted that is an area in which the degree of congestion should be predicted in the period to be predicted, and the area to be predicted selected by the area selection unit. A generator.

  The congestion prediction system according to the present invention includes a congestion degree acquisition unit that acquires a past congestion degree for each area and for each sub period, and a congestion degree that classifies each of the acquired congestion degrees into at least one of two or more categories. A classification unit, a congestion degree totaling unit that aggregates the classified congestion degree for each area and for each category and obtains a congestion degree aggregation value, and the congestion degree aggregation value that is aggregated by the congestion degree aggregation unit Based on this, a model that generates a congestion prediction model for each of the area to be predicted that is an area in which the degree of congestion should be predicted in the period to be predicted, and the area to be predicted selected by the area selection unit. A generator.

  The congestion prediction method of the present invention is classified into obtaining a past congestion degree for each area and for each sub-period, and classifying each of the acquired congestion degrees into at least one of two or more categories. The congestion degree is aggregated for each area and for each category to obtain a congestion degree aggregate value, and based on the aggregated congestion degree aggregate value, congestion degree prediction should be performed in the prediction target period Selecting a prediction target area, which is an area, and generating a congestion prediction model for each of the selected prediction target areas.

  According to the present invention, a prediction target area that is a target for performing congestion prediction is appropriately selected. As a result, the calculation amount is reduced while maintaining the accuracy of congestion prediction.

It is a figure which shows the congestion prediction server which concerns on embodiment, and the computer apparatus relevant to a congestion prediction server. It is a physical configuration block diagram of the server of the embodiment. It is a physical configuration block diagram of the client terminal of the embodiment. It is a logic block diagram of the congestion prediction server of an embodiment. It is a flowchart which shows the prediction object area selection operation | movement of embodiment. It is a figure which shows an example of the congestion degree classified by time zone of embodiment. It is a figure which shows an example of the daily average congestion degree of embodiment. It is a figure which shows an example of the congestion degree total value of embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

1. Embodiment 1 (1). Configuration Outline Referring to FIG. 1, a congestion prediction server PS according to an embodiment of the present invention and a computer device related to the congestion prediction server PS are schematically shown. The computer apparatus according to the present embodiment executes communication according to a predetermined protocol (TCP / IP or the like).

  The congestion prediction server PS selects the prediction target area TA, which is the area A where congestion degree prediction is to be performed, by analyzing the position information LI of the user (mobile terminal) provided from the position information server LS in time series. Then, the congestion degree prediction is executed for each of the selected prediction target areas TA. The congestion prediction server PS presents the result of the congestion degree prediction to the client terminal CL.

  The time unit of congestion degree prediction in the present embodiment is a sub-period (for example, one day or one hour) included in a prediction target period (for example, one week after the next day). The prediction target period may be predetermined as a default value, or may be dynamically set based on a request from the client terminal CL. The geographical unit of the congestion degree prediction is area A. In the present embodiment, the area A is a rectangular area (mesh) divided into a lattice shape based on latitude and longitude. The mesh described above may be a standard area mesh (for example, a quarter area mesh) provided by the Statistics Bureau of the Ministry of Internal Affairs and Communications in Japan, or may be a mesh defined by another method.

  The event information server ES stores event information EI related to various events held by designating the date and time, and provides the stored event information EI to the congestion prediction server PS. The event information EI includes at least an event name, an event location (latitude / longitude), and an event date. The event information EI can be used for selecting the prediction target area TA.

  The blog server BS receives and stores post information PI from a user terminal (for example, a smartphone), and provides the stored post information PI to the congestion prediction server PS. In the present embodiment, post information PI is used to extract event information EI.

Hereinafter, examples of the posting information PI are enumerated without limitation.
-Short postings (eg, tweets [registered trademark]) of 140 characters or less in a microblog service (such as Twitter [registered trademark])
-Posting diaries and reviews on blog (web log) services-Status posting on social networking services (SNS)

1 (2). Physical configuration 1 (2) -1. Configuration of Server Device FIG. 2 is a block diagram showing a physical configuration of each server device of this embodiment including the congestion prediction server PS. Each server device (BS, ES, LS, PS) includes a network interface 10, an input unit 12, an output unit 14, a CPU (Central Processing Unit) 16, a RAM (Random Access Memory) 18, a ROM (Read Only Memory) 20, An HDD (Hard Disk Drive) 22 is provided.

  The network interface 10 communicates with other computer devices via a network. The input unit 12 receives an input signal from an input device such as a keyboard. The output unit 14 transmits an output signal to an output device such as a display. The CPU 16 performs various controls and calculations by executing programs stored in the RAM 18 and the ROM 20 which are main storage devices. The HDD 22 is an auxiliary storage device that stores programs and data that can be expanded on the RAM 18. A storage medium such as an SSD (Solid State Drive) may be adopted instead of or in addition to the HDD 22.

  As a person skilled in the art understands naturally, one server may be configured by a plurality of computer devices, or one computer device may include a plurality of virtualized servers.

1 (2) -2. Configuration of Client Terminal FIG. 3 is a block diagram showing a physical configuration of the client terminal CL of the present embodiment. The client terminal CL includes a network interface 30, an input unit 32, an output unit 34, a CPU 36, a RAM 38, a ROM 40, and an HDD 42. Each of the above elements has the same configuration as the element described with reference to FIG. The client terminal may be a portable terminal that provides position information to the position information server LS, or may be a general personal computer.

1 (3). Logical configuration 1 (3) -1. Configuration of Congestion Prediction Server FIG. 4 is a block diagram showing a logical configuration of the congestion prediction server PS. The congestion prediction server PS includes a congestion level acquisition unit 90, a congestion level classification unit 100, a congestion level totaling unit 110, an area selection unit 120, a model generation unit 130, and a predicted value calculation unit 140. Further, the congestion prediction server PS includes various logical tables described later.

  The congestion degree acquisition unit 90 acquires the past congestion degree for each area A and for each sub period based on the position information LI provided from the position information server LS. The unit of the sub period is arbitrary, for example, one day (calendar day) or one hour. When “day” is adopted as the unit of the sub-period, the daytime average population or the daytime maximum population may be adopted as the congestion degree in view of the fact that the majority of people are at home at night and early morning.

  The congestion level classification unit 100 classifies each acquired congestion level into at least one of two or more categories. As a category for classification, when the unit of the sub-period is “day”, weekdays / holidays, rainy days / non-rainy days (days including sunny days and cloudy days), singular days (extremely for some reason) (Congested day) / normal day, event date / non-event day, etc.

  The category can be set for each area A. Depending on the category type, there may be areas A belonging to different categories in the same sub-period. For example, on the same day, an area A that is a rainy day and another area A that is a non-rainy day can coexist.

  The congestion level totaling unit 110 totals the classified congestion levels for each area A and for each category, and acquires a congestion level total value. Any value can be adopted as the congestion degree total value. For example, when “weekday / holiday” is adopted as the category, the average congestion degree or maximum congestion degree on weekdays and the average congestion degree or maximum congestion degree on holidays can be adopted as the congestion degree total value.

  The area sorting unit 120 sorts the prediction target area TA based on the congestion level total value counted by the congestion level totaling unit 110. The prediction target area TA is an area A in which congestion degree prediction is to be executed in the prediction target period (that is, an area A in which a congestion prediction model is to be generated). Details of the area selection process will be described later.

  The model generation unit 130 generates a congestion prediction model for each prediction target area TA selected by the area selection unit 120. In the above generation process, an arbitrary congestion prediction model can be adopted.

  For example, autoregressive (AR) model, moving average (MA) model, autoregressive moving average (ARMA) model, autoregressive integrated moving average model (autoregressive integrated) Known models such as a moving average (ARIMA) model, a seasonal autoregressive integrated moving average (SARIMA) model, and a SARIMAX model (SARIMA model with exogenous variables) may be employed. It is preferable that a plurality of congestion prediction models are generated for one prediction target area TA.

  The predicted value calculation unit 140 calculates a predicted value of the congestion degree using the congestion prediction model generated by the model generation unit 130. When a plurality of congestion prediction models are generated, it is preferable to calculate a prediction value by selecting a model predicted to have higher accuracy.

  The above functional blocks are realized by the CPU 16 executing a computer program stored in the main storage device of the congestion prediction server PS. The logical table is a component of a relational database in which a plurality of tables are linked based on a predetermined relation (relation). The relational database may be implemented as an on-disk database in which data is stored in the HDD 22 that is a non-volatile memory, or may be implemented as an in-memory database in which data is stored in the RAM 18 that is a volatile memory.

1 (4). Prediction target area selection operation An example of the prediction target area selection operation of the present embodiment will be described below. FIG. 5 is a flowchart showing the prediction target area selection operation. In this example, “day (calendar day)” is adopted as a unit of the sub-period, and “holiday” and “weekday” are adopted as categories (first category date and second category date) for classifying the congestion degree.

  The congestion level acquisition unit 90 acquires the congestion level by time zone for each area A based on the position information LI provided from the position information server LS (S1010). FIG. 6 is a diagram illustrating an example of the acquired congestion degree by time zone. The degree of congestion by time zone in a certain area A is indicated by, for example, the total number of mobile terminals that have existed in that area A during that time zone. In addition, the shortest stay time (for example, 10 minutes) required in order to be treated as the calculation object of the time zone congestion degree may be set. In the present embodiment, the period for which the degree of congestion for each time zone is acquired is a predetermined period (for example, four weeks immediately before) that comes immediately before the prediction target period.

  The congestion degree acquisition unit 90 acquires the daily average congestion degree for each area A (that is, the past congestion degree for each area A and for each sub-period) based on the acquired hourly congestion degree (S1020). . FIG. 7 is a diagram illustrating an example of the acquired daily average congestion degree. In addition, instead of the daily average congestion degree, the daily maximum congestion degree may be acquired. Moreover, the daytime average daytime congestion degree or the daytime maximum daytime congestion degree may be acquired.

  The congestion level classification unit 100 classifies each of the daily average congestion levels for each area A into categories (S1030). That is, the congestion degree classification unit 100 assigns category information (in this example, “holiday” or “weekday”) indicating the category to be classified to each of the acquired daily average congestion degrees for each area A. . The category information is given by an arbitrary method. For example, it is preferable that the congestion degree classification unit 100 refers to a table (calendar table) in the congestion prediction server PS in which dates and category information are associated with each other and assigns category information to each daily average congestion degree. is there.

  The congestion level totaling unit 110 totals the daily average congestion levels classified by the congestion level classification unit 100 for each area A and for each category, and acquires the average congestion level (congestion level total value) (S1040). FIG. 8 is a diagram illustrating an example of the aggregated congestion level aggregate value. More specifically, the congestion degree totaling unit 110 averages the average daily congestion degree classified as “holiday” for each area A to obtain the average holiday congestion degree (holiday congestion degree total value), Average the daily average congestion degree classified as “weekdays” to obtain the weekday average congestion degree (weekday congestion degree total value).

  The area sorting unit 120 sorts the prediction target area TA based on the congestion level total value counted by the congestion level totaling unit 110 (S1050). For example, when “prevention” (first category day) is included in the prediction target period, the area selection unit 120 compares the holiday congestion degree total value with the weekday congestion degree total value (holiday congestion degree total value / weekday congestion degree totalization). It is preferable to select an area A whose value) exceeds the first threshold as the prediction target area TA. By the above sorting process, the area A where holidays are more crowded than weekdays is sorted as the prediction target area TA.

  Instead of performing the above sorting process using the first threshold value in the above step S1050, the area selecting unit 120 compares the ratio of the holiday congestion degree total value to the weekday congestion degree total value (holiday congestion degree total value / weekday congestion degree). The area A (area A up to the upper first order) that is higher than or equal to the first order when the (total value) is sorted in descending order may be selected as the prediction target area TA.

  In step S1050, “weekdays” can be handled as the first category date, and the same processing can be performed. Further, when both the “holiday” and the “weekday” are included in the prediction target period, it is possible to perform the sorting process in step S1050 for each of the “holiday” and the “weekday”. That is, both the area A where the holiday is more crowded than the weekday and the area A where the weekday is crowded than the holiday may be selected as the prediction target area TA. The sorting process based on the first threshold value or the first order may be performed nationwide (that is, for all areas A) or may be performed for each prefecture.

  In step S1050, when either the holiday congestion level total value or the weekday congestion level total value is small, the ratio value increases in spite of the fact that it is not so crowded and is selected as the prediction target area TA. The case where it ends up is assumed. In order to avoid the above cases, the area selection unit 120 may determine that the holiday congestion level total value (that is, the first category day congestion level total value) and the weekday congestion level total value (that is, the second category day congestion level total value). It is preferable to select an area A in which at least one of ()) exceeds the cut-off threshold (cut-off threshold) as a target of the selection process of the prediction target area TA in step S1050.

  The above-described prediction target area selection operation is preferably performed, for example, at sparse intervals (for example, once a week or once a day). Further, it is preferable that the model generation unit 130 generates a congestion prediction model following the above prediction target area selection operation.

  Note that it is preferable that the predicted value calculation unit 140 calculates the predicted value at an interval (for example, once per hour) closer than the prediction target area selection operation. This is because the calculation value of the predicted value has a lower calculation load than the model generation, and more timely congestion prediction information is provided to the user by calculating the predicted value with high frequency.

1 (5). Effects of the present embodiment According to the configuration of the present embodiment described above, the prediction target area TA that is a target for which congestion prediction is performed is appropriately selected. That is, a prediction target area TA that is a target for generating a prediction model with a high calculation load is selected based on a congestion degree total value indicating a past congestion degree. Accordingly, the calculation amount necessary for generating the prediction model is reduced while maintaining the accuracy of the congestion prediction by area selection based on the congestion degree total value.

2. Modifications The above embodiment can be variously modified. Specific modifications are exemplified below. Two or more aspects arbitrarily selected from the above embodiments and the following examples can be appropriately combined as long as they do not contradict each other.

2 (1). Modification 1
In the above embodiment, “holiday” and “weekday” are adopted as categories (first category date and second category date) for classifying the congestion degree. Similarly, “rainy day” and “non-rainy day” can be adopted as categories. In the above case, the congestion degree classification unit 100 refers to the rainfall record table in which the date, the area A, and the rainfall record are associated with each other in the category information (“rainy day” in step S1030) Or “non-rainy day”).

  In this example, whether or not “rainy day” or “non-rainy day” is included in the prediction target period is determined based on, for example, a rainfall information table in which date, area A, and rainfall information are associated with each other. Is done. The rainfall information table is a table that stores and acquires rainfall information from an external server.

2 (2). Modification 2
In the above embodiment, the prediction target area TA is selected based on the ratio between the congestion degree total value of the first category date (for example, “holiday”) and the congestion degree total value of the second category day (for example, “weekday”). (S1050). Instead of the above sorting process, the prediction target area TA may be sorted based on only one of the congestion degree total values.

  For example, when the first category date is included in the prediction target period, the area selection unit 120 may select the area A in which the congestion degree total value of the first category date exceeds the second threshold as the prediction target area TA. In order to detect an extreme increase in the degree of congestion, it is preferable that the second threshold value is N times (for example, 5 times) the standard deviation of the congestion degree total value of the first category day.

  In addition, when the first category date is included in the prediction target period, the area selection unit 120 selects the area A that becomes the second or higher rank as the prediction target area TA when the congestion degree total value of the first category date is sorted. May be.

2 (3). Modification 3
When there is an area A where congestion has occurred on a specific date in the past (specific past date), a response date (for example, a day one year after the specific past date) that responds to the specific past date is a prediction target When included in the period, the area selecting unit 120 may select the area A as the prediction target area TA. Whether or not the congestion has occurred in the area A on the specific past date is preferably determined based on whether or not the congestion level of the area A on the specific past date exceeds a threshold (third threshold).

  This modification assumes a large-scale event such as a fireworks display held every year or a small-scale event such as a special sale day held every month. Therefore, the reception day can be variously set according to the nature of each event held on a specific past day. For example, the anniversary day may be “the day of the same day of the week as the specific past day closest to the day one year after the specific past day” or “the day after the specific past day (eg, October 29)” The same day (for example, November 29) ".

2 (4). Modification 4
In the above embodiment, a configuration in which the unit of the sub period is other than “day” can be employed. For example, as described above, the unit of the sub period may be “hour (1 hour)”. When the unit of the sub-period is other than “day”, the degree of congestion is also classified into categories in units other than “day”. For example, the degree of congestion for each sub-period is classified into one of the categories of weekday period (period included in weekdays) / holiday period (period included in holidays). In addition, categories corresponding to sub-periods include rainy periods / non-rainy periods, peculiar periods (periods that are extremely crowded for some reason) / normal periods, event periods (periods where events are held) / non-event periods, Etc. are exemplified.

  Based on this modification, the categories “first category date” and “second category date” in the embodiment correspond to “first category corresponding to the first type sub-period” and “second type sub-period”. It can be generally expressed as “second category”. The “first type sub period” is, for example, a holiday period, and the “second type sub period” is, for example, a weekday period.

  Note that in the configuration of the present modification in which the congestion level is classified into categories other than “day”, for example, in units of “hours”, it is preferable to eliminate the influence of the congestion level variation that occurs depending on the time zone. For example, when the congestion level totaling unit 110 counts the congestion level, the calculation is performed on the same time section (for example, from 13:00 to 15:00) included in the prediction target period (that is, by layer), and the congestion level It is preferable to obtain the total value. In addition, prior to the congestion level totaling unit 110 counting the congestion level, seasonal adjustment may be performed to eliminate the influence of fluctuations in the congestion level.

2 (5). Modification 5
When there is an area A where an event is held in the prediction target period, the area selection unit 120 may select (add) the area A as the prediction target area TA. The above operation is preferably performed based on the event information EI provided from the event information server ES. As described above, the event information EI includes at least the event name, the venue (latitude / longitude), and the date and time of the event.

  Further, the event information EI may be extracted from the posting information PI posted on the blog server BS. The method for extracting the event information EI is arbitrary. For example, event information EI can be extracted from words (morphemes) obtained by performing morphological analysis on a plurality of posted information PIs. For the analysis of the acquired word (morpheme), it is preferable to use machine learning such as a conditional random field (CRF) or a support vector machine (SVM).

  If an event with the same name and location as the event held in the prediction target period has been held in the past, the effect of raising the congestion level due to the presence or absence of the event can be achieved by the same method as in the previous embodiment. If it is determined whether there is a push-up effect, the area A corresponding to the event may be selected (added) as the prediction target area TA. According to the above configuration, it is possible to exclude the area A where an event having a small influence on the congestion degree is held from the prediction target area TA.

2 (6). Modification 6
The area sorting unit 120 can sort the prediction target area TA by adding an arbitrary condition in addition to the conditions of the above embodiment. For example, only the area A including the POI may be a target of the selection process of the prediction target area TA.

2 (7). Modification 7
The area sorting unit 120 may sort the prediction target area TA by combining a plurality of conditions described in the above embodiments and modifications. For example, for each area A, an index calculated based on each condition is multiplied by a weighting coefficient, and a combined linear combination value is compared with a threshold value to determine whether or not to be a prediction target area TA. Good.

2 (8). Modification 8
In the above embodiment, the congestion prediction server PS, which is one device, executes the prediction target area selection operation and the subsequent operation (model generation or the like). However, a congestion prediction system constituted by a plurality of devices may execute the selection operation and the subsequent operation of the prediction target area. For example, the congestion level acquisition unit 90, the congestion level classification unit 100, the congestion level totaling unit 110, the area selection unit 120, the model generation unit 130, and the predicted value calculation unit 140 are implemented by a system including two or more server devices. Also good.

2 (9). Modification 9
Each function executed by the CPU in the computer apparatus, particularly the congestion prediction server PS, may be executed by a programmable logic device such as an FPGA (Field Programmable Gate Array) or a DSP (Digital Signal Processor).

DESCRIPTION OF SYMBOLS 10 ... Network interface, 12 ... Input part, 14 ... Output part, 16 ... CPU, 18 ... RAM, 20 ... ROM, 22 ... HDD, 30 ... Network interface, 32 ... Input part, 34 …… Output unit, 36 …… CPU, 38 …… RAM, 40 …… ROM, 42 …… HDD, 90 …… Congestion degree acquisition unit, 100 …… Congestion degree classification unit, 110 …… Congestion degree totaling unit, 120 …… Area selection unit, 130 …… Model generation unit, 140 …… Predicted value calculation unit, A …… Area, BS …… Blog server, CL …… Client terminal, EI …… Event information, ES …… Event information Server, LI ... position information, LS ... position information server, PI ... posted information, PS ... congestion prediction server, TA ... predicted area.

Claims (10)

A congestion degree acquisition unit for acquiring a past congestion degree for each area and for each sub period;
A congestion degree classifying unit that classifies each acquired degree of congestion into at least one of two or more categories related to the sub-period ;
A congestion degree totaling unit that totalizes the classified degree of congestion for each area and for each category to obtain a congestion degree total value; and
An area selection unit that selects a prediction target area that is an area in which the congestion degree prediction is to be performed in the prediction target period, based on the congestion degree total value calculated by the congestion degree calculation unit;
A congestion prediction device comprising: a model generation unit that generates a congestion prediction model for each of the prediction target areas selected by the area selection unit. The category includes a first category corresponding to the first type sub-period and a second category corresponding to the second type sub-period,
The area selection unit
When the first target sub-period is included in the prediction target period, an area where the ratio of the congestion degree total value of the first category and the congestion degree total value of the second category exceeds a first threshold is the prediction target area. The congestion prediction device according to claim 1. The category includes a first category corresponding to the first type sub-period and a second category corresponding to the second type sub-period,
The area selection unit
When the first type sub-period is included in the prediction target period, an area that is higher than the first rank when the ratio of the congestion degree aggregation value of the first category and the congestion degree aggregation value of the second category is sorted. The congestion prediction device according to claim 1. The area selection unit
The area for which at least one of the congestion level total value of the first category and the congestion level total value of the second category exceeds a cut-off threshold is set as a target of the selection process of the prediction target area. Congestion prediction device. The category includes a first category corresponding to a first type sub-period;
The area selection unit
2. The congestion prediction device according to claim 1, wherein when the first type sub-period is included in the prediction target period, an area in which the congestion degree total value of the first category exceeds a second threshold is selected as the prediction target area. The category includes a first category corresponding to a first type sub-period;
The area selection unit
The congestion according to claim 1, wherein when the prediction target period includes the first type sub-period, an area that is ranked second or higher when the congestion degree aggregation value of the first category is sorted is selected as the prediction target area. Prediction device. The area selection unit
When there is an area whose congestion degree exceeds the third threshold in the specific past period, and the period corresponding to the specific past period is included in the prediction target period, the area is selected as the prediction target area. The congestion prediction device according to any one of claims 1 to 6. The area selection unit
The congestion prediction device according to claim 1, wherein when there is an area where an event is held in the prediction target period, the area is selected as the prediction target area. A congestion degree acquisition unit for acquiring a past congestion degree for each area and for each sub period;
A congestion degree classifying unit that classifies each acquired degree of congestion into at least one of two or more categories related to the sub-period ;
A congestion degree totaling unit that totalizes the classified degree of congestion for each area and for each category to obtain a congestion degree total value; and
An area selection unit that selects a prediction target area that is an area in which the congestion degree prediction is to be performed in the prediction target period, based on the congestion degree total value calculated by the congestion degree calculation unit;
A congestion prediction system comprising: a model generation unit that generates a congestion prediction model for each of the prediction target areas selected by the area selection unit. Computer
A congestion degree acquisition unit for acquiring a past congestion degree for each area and for each sub period;
A congestion degree classifying unit that classifies each acquired degree of congestion into at least one of two or more categories related to the sub-period ;
A congestion degree totaling unit that totalizes the classified degree of congestion for each area and for each category to obtain a congestion degree total value; and
An area selection unit that selects a prediction target area that is an area in which the congestion degree prediction should be performed in the prediction target period, based on the calculated congestion degree total value;
For each sorted the prediction target area, congestion prediction program to function as a model generation for generating a congestion prediction model.

Images