Passenger number estimation device and passenger number estimation method

Abstract

To provide a new method for estimating the number of passengers on a train, based on the premise that each passenger's train selection tendencies are not the same. [Solution] From the trains included in the train timetable 310, multiple candidate trains corresponding to each passenger based on passenger demand data 312 are set. For each of the multiple candidate trains corresponding to each passenger based on passenger demand data 312, the following are calculated: 1) the probability of selection of the candidate train based on a simultaneous selection model, and 2) the probability of selection of the candidate train based on a sequential selection model. Based on the selection probabilities of 1) and 2), the probability of adoption of the candidate train is calculated, and the estimated number of passengers for each train in the train timetable 310 is estimated based on the aggregated adoption probabilities for each passenger and each candidate train.

Description

Technical Field

[0001] The present invention relates to a passenger number estimation device and the like.

Background Art

[0002] Especially for toll express trains such as the Shinkansen, temporary trains may be operated according to the passenger demand that varies depending on the day and time zone. As a method for supporting the formulation of an effective temporary train plan, a method of estimating the number of passengers on temporary trains and regular trains before and after them from the allocation ratio of passenger demand calculated based on past performance data to each train is known (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, as a practical problem, not all passengers fully understand the train schedule of the day including temporary trains and regular trains before and after them, and the train selection tendencies of each individual are not the same. Therefore, there is often a bias in the boarding rate, such that the boarding rate of a temporary train operated according to passenger demand is low and the boarding rate of regular trains before and after it is high. Thus, there is a need for a method of estimating the number of passengers on a train that can appropriately reflect the train selection tendencies of each passenger, which are not uniform.

[0005] The problem to be solved by the present invention is to provide a technique for a new method of estimating the number of passengers on a train on the premise that the train selection tendencies of each passenger are not the same.

Means for Solving the Problems

[0006] The first invention for solving the above problems is A passenger number estimation device that estimates the number of passengers for each train related to a given train schedule based on given passenger demand data, A storage means (for example, the storage unit 300 in Figure 13) that stores data indicating a first selection model (for example, a simultaneous selection model in the embodiment) that determines the selection probability for each candidate train when a given number of candidate trains are compared in parallel based on a first parameter set based on passenger properties which include at least the attributes of passengers, and a second selection model (for example, a sequential selection model in the embodiment) that determines the selection probability for each candidate train by sequentially determining the selection probability of each train from the number of candidate trains based on a second parameter set based on the passenger properties, A candidate train setting means (for example, the candidate train setting unit 204 in Figure 13) sets multiple candidate trains corresponding to each passenger based on the passenger demand data from among the trains included in the aforementioned train timetable, For each of the multiple candidate trains corresponding to each passenger based on the passenger demand data, the selection probability calculation means (for example, the selection probability calculation unit 206 in Figure 13) calculates 1) the selection probability of the candidate train based on the first selection model and 2) the selection probability of the candidate train based on the second selection model, and calculates the selection probability of the candidate train based on the selection probabilities of 1) and 2), Based on the aggregated results of the adoption probabilities for each candidate train for each passenger calculated by the adoption probability calculation means, an estimation means (for example, the estimation unit 208 in Figure 13) estimates the estimated number of passengers for each train related to the train schedule, This is a passenger number estimation device equipped with [specific features / equipment].

[0007] Other inventions include, A passenger number estimation method in which a computer system estimates the number of passengers for each train related to a given train schedule based on given passenger demand data, The computer system stores data representing a first selection model that calculates the selection probability for each candidate train when comparing a given number of candidate trains in parallel based on a first parameter set based on passenger properties, which include at least the attributes of the passengers, and a second selection model that calculates the selection probability for each candidate train by sequentially calculating the selection probability for each train from the number of candidate trains based on a second parameter set based on the passenger properties. From among the trains included in the aforementioned train schedule, multiple candidate trains corresponding to each passenger based on the aforementioned passenger demand data are selected. For each of the multiple candidate trains corresponding to each passenger based on the passenger demand data, the candidate train is treated as the candidate train, and 1) the selection probability of the candidate train based on the first selection model and 2) the selection probability of the candidate train based on the second selection model are calculated, and the probability of the candidate train being adopted is calculated based on the selection probabilities of 1) and 2). Based on the results of aggregating the aforementioned probability of each passenger being selected for each candidate train, the estimated number of passengers for each train in the aforementioned train schedule is estimated, A passenger number estimation method that includes this may be constructed.

[0008] According to the first invention, it becomes possible to estimate the number of passengers on a train, assuming that each passenger's train selection tendencies are not the same. Specifically, two selection models are prepared to show the typical train selection tendencies that passengers may take: a first selection model that calculates the selection probability for each candidate train when multiple candidate trains are compared in parallel, and a second selection model that calculates the selection probability for each candidate train by sequentially calculating the selection probability for each train from among multiple candidate trains. Using these two types of selection models, the number of passengers on each train is estimated by probabilistically calculating and aggregating the selection probability for each candidate train, which is which train a passenger will choose. This realizes the estimation of the number of passengers on a train, assuming that each passenger's train selection tendencies are not the same.

[0009] The second invention relates to the above invention, The aforementioned passenger demand data is data in which passengers are arranged chronologically in order of their desired boarding time or desired train. A time zone setting means (for example, the time zone setting unit 202 in Figure 13) that sets multiple time zones separated along the time series of the aforementioned passenger demand data, Furthermore, The aforementioned candidate train setting means sets a plurality of candidate trains corresponding to the passengers included in each time period. This is a passenger number estimation device.

[0010] According to the second invention, it becomes possible to appropriately and easily set as candidate trains that passengers are likely to choose.

[0011] The third invention is, in the above invention, The first parameter set includes a subset of parameters for each of the first classes, which classifies the passenger into a plurality of first classes based on the passenger properties. The second parameter set includes a subset of parameters for each of the two classes, which classifies the passengers into multiple second classes based on the passenger properties. The aforementioned means for calculating the probability of adoption is: Using the Class 1 parameter subset and the Class 1 belonging probability determined by the passenger properties defined for the passenger, the selection probability of the candidate train for each Class 1 is calculated, and the selection probability of 1) is calculated based on the calculation result. Using the aforementioned Class 2 parameter subset and the Class 2 belonging probability determined by the passenger properties defined for the passenger, the selection probability for the Class 2 of the candidate train is calculated, and the selection probability of 2) above is calculated based on the calculation results. This is a passenger number estimation device.

[0012] According to the third invention, for both the first and second selection models, it becomes possible to calculate the selection probability of each candidate train based on the passenger's class, which is determined by classifying passengers based on their passenger properties.

[0013] The fourth invention is in the above invention, the first parameter set and the second parameter set each include parameters based on the train type, the adoption probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the train type of the adoption candidate train, and it is a passenger number estimation device.

[0014] According to the fourth invention, it becomes possible to variably calculate the selection probability according to the train type of the adoption candidate train.

[0015] The fifth invention is in the above invention, the first parameter set and the second parameter set each include parameters based on the usage fee of the train, the adoption probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the usage fee of the adoption candidate train, and it is a passenger number estimation device.

[0016] According to the fifth invention, it becomes possible to variably calculate the selection probability based on the usage fee of the adoption candidate train.

[0017] The sixth invention is in the above invention, the first parameter set and the second parameter set each include parameters based on the seat availability of the train, in the process of calculating the adoption probability of each of the plurality of adoption candidate trains corresponding to each passenger based on the passenger demand data, the adoption probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the seat availability of the adoption candidate train based on the calculated adoption probability, and it is a passenger number estimation device.

[0018] According to the sixth invention, it becomes possible to variably calculate the selection probability based on the seat availability of the adoption candidate train.

[0019] The seventh invention is, in the above invention, The candidate train setting means selects an estimated optimal candidate train corresponding to the passenger from among the trains included in the train timetable, and sets a plurality of candidate trains by selecting a predetermined number of optimal candidate trains in order of proximity to the estimated optimal candidate train in the sequence of trains. The first parameter set and the second parameter set each include parameters based on the relative time difference of the train to the estimated optimal candidate train, The selection probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the relative time difference of the candidate train to be selected with respect to the estimated optimal candidate train corresponding to the passenger. This is a passenger number estimation device.

[0020] According to the seventh invention, it is possible to variably calculate the selection probability of each passenger based on the relative time difference between the estimated optimal candidate train and the candidate train that the passenger chooses. [Brief explanation of the drawing]

[0021] [Figure 1] Summary of passenger number estimates. [Figure 2] Explanation of passenger demand data. [Figure 3] An example of a model selection probability table. [Figure 4] An explanation of the probability of passenger trains being selected. [Figure 5] An example of a parameter set for a degree of belonging model used in a simultaneous selection model. [Figure 6] An example of a parameter set for a degree of belonging model used in a sequential selection model. [Figure 7] The probability of belonging to each class in a simultaneous selection model. [Figure 8] The probability of belonging to each class in a sequential selection model. [Figure 9] An example of a parameter set for a simultaneous selection model. [Figure 10] Selection probability of each candidate train by class in the simultaneous selection model. [Figure 11]An example of a parameter set for a sequential selection model. [Figure 12] Selection probability for each candidate train in each class of the sequential selection model. [Figure 13] Example of a passenger count estimation device's functional configuration. [Figure 14] An example of passenger demand data. [Figure 15] An example of recruitment probability data by passenger type. [Figure 16] An example of estimated passenger count data by time of day. [Figure 17] An example of estimated passenger data. [Figure 18] Flowchart for passenger number estimation process. [Modes for carrying out the invention]

[0022] Preferred embodiments of the present invention will be described below with reference to the drawings. However, the applicable forms of the present invention are not limited to the following embodiments. Furthermore, in the drawings, the same elements are denoted by the same reference numerals.

[0023] This embodiment estimates the number of passengers for each train in a given train schedule based on given passenger demand data.

[0024] Figure 1 is a schematic diagram of the estimation of the estimated number of passengers in this embodiment. As shown in Figure 1, given a train schedule 310 and passenger demand data 312 for a certain target day, the estimated number of passengers for each train is estimated so as to allocate each passenger determined by the passenger demand data 312 to each train determined by the train schedule 310.

[0025] Specifically, for each passenger defined in passenger demand data 312, the probability of selecting each candidate train, which is a subset of the trains defined in train schedule 310, is calculated. Next, the selection probabilities for each candidate train for each passenger are aggregated to calculate the selection probability for each train defined in train schedule 310, and this selection probability is multiplied by the number of passengers defined in passenger demand data 312 to calculate the estimated number of passengers for each train.

[0026] The probability of a passenger selecting each candidate train is calculated as follows: Based on the passenger's preferred boarding time or preferred train, multiple candidate trains are selected from the trains specified in train timetable 310. Then, based on passenger properties that include at least the attributes specified for the passenger, the probability of each candidate train being selected is calculated using a train selection model. There are two types of train selection models: a simultaneous selection model and a sequential selection model, each corresponding to two representative train selection patterns that a passenger may take. The probability of each candidate train being selected is calculated using each of these two types of train selection models, and based on these, the probability of each candidate train being selected is calculated.

[0027] Figure 2 shows an example of passenger demand data 312. Passenger demand data 312 is data that arranges passengers (passenger demand) who wish to ride between target stations (stations from departure station to arrival station) on a target day in chronological order in order of desired boarding time or desired train. Passenger demand data 312 is given as a passenger demand wave, for example, as shown in the upper part of Figure 2, with the horizontal axis being time and the vertical axis being passenger demand (number of passengers), representing the relationship between the desired boarding time of a train or the departure time of a desired train and the passenger demand.

[0028] As shown in the lower part of Figure 2, this passenger demand fluctuation is discretized by dividing the horizontal axis, which represents time, into multiple time zones σ (for example, every 10-15 minutes) along the time series. Then, based on the midpoint time tσ of each time zone σ, candidate trains are set for passengers included in that time zone σ. That is, the same candidate train is set for passengers whose desired boarding time or departure time falls within the same time zone. Specifically, predetermined time ranges before and after the midpoint time tσ of a given time zone σ are set (for example, the pre-range from one hour before the midpoint time tσ to the midpoint time tσ, and the post-range from the midpoint time tσ to one hour after that). Then, for each pre- and post-range, a maximum of a predetermined number of trains (for example, a maximum of 3) are selected from the trains within that range in order of their departure time being closest to the midpoint time tσ.

[0029] Then, the probability P that passengers in time period σ choose candidate train i is multiplied by the passenger demand W (number of passengers) in that time period σ to calculate the estimated number of passengers for candidate train i in that time period σ. Therefore, the estimated number of passengers for train i is the sum of the estimated number of passengers for train i in each time period σ, and is given by equation (1) below.

number

[0030] The simultaneous selection model is a train selection model that reproduces the train selection pattern in which passengers compare multiple candidate trains in parallel and make a selection. The sequential selection model is a train selection model that reproduces the train selection pattern in which passengers decide whether or not to select one train at a time from among multiple candidate trains. Furthermore, the model equations for the simultaneous selection model and the sequential selection model can be expressed by applying, for example, a latent class logit model, and are defined as model equations for determining the selection probability of each candidate train for passengers.

[0031] The choice of train selection model for passenger n is probabilistically determined based on the attributes (passenger properties) of passenger n, according to the model selection probability table 316, an example of which is shown in Figure 3. The attributes (passenger properties) of passenger n include information about the ticket purchase location (ticket purchase channel), which indicates where the ticket is purchased. Based on this ticket purchase location information, the train selection model to be adopted is determined. In Figure 3, the model selection probability table 316 associates the selection probability for each ticket purchase location for both the simultaneous selection model and the sequential selection model, which are the train selection models.

[0032] Since the sum of the probabilities R that passenger n adopts both the simultaneous selection model and the sequential selection model is 1, for example, the probability R that passenger n adopts the simultaneous selection model is... n d As shown in equation (2a), is given as the weighted average of the selection probabilities for each ticket purchase location, and the adoption probability R of the successive selection model is given. n cAs shown in equation (2b), the adoption probability R of the simultaneous selection model is n d It is given by.

number

[0033] Furthermore, whether a passenger adopts a simultaneous selection model or a sequential selection model as their train selection model may be based not only on the ticket purchase location, but also on other factors included in the passenger's attributes (such as age, gender, and purpose of travel), or on a combination of multiple factors. In that case, the model selection probability table 316 will be a table showing the selection probability of the simultaneous selection model and the sequential selection model for each of those factors or combinations of factors.

[0034] Figure 4 illustrates the calculation of the probability of passenger n adopting a candidate train i using a train selection model. The probability P that passenger n adopts a candidate train i is calculated as a weighted average of the probability P of train i when the simultaneous selection model is adopted and the probability P of train i when the sequential selection model is adopted, with the adoption probabilities R of the simultaneous selection model and sequential selection model respectively being used as weights, as shown in equation (3).

number

[0035] The selection probability P of candidate train i when using the simultaneous selection model and the sequential selection model, respectively, can be calculated as a weighted average of the selection probabilities P of candidate trains belonging to each latent class, with the probability Q of belonging to each latent class as the weight, as shown in equation (4).

number

[0036] Latent classes are classifications based on passengers' train selection tendencies, such as which factors they prioritize when choosing a train. The probability Q that passenger n belongs to a latent class s in both the simultaneous selection model and the sequential selection model is determined probabilistically based on passenger n's attributes (passenger properties) according to the degree of belonging model shown in equation (5). This degree of belonging model equation (5) is a model equation to which a multinomial logit model is applied.

number

[0037] In equation (5), the explanatory variable z for the membership function Φ consists of items related to the individual passenger, such as gender, age, ticket purchase location, and purpose of travel (business or private). Furthermore, this explanatory variable z is a binary variable with a value of either 1 or 0, and is determined by the attributes (passenger properties) of passenger n.

[0038] The first type of latent class of the simultaneous selection model and the second type of latent class of the sequential selection model are different classes. Therefore, the parameter sets that define the type of explanatory variable z in the membership function Φ in equation (5), the parameter α for the explanatory variable z, and the constant term α are prepared for the simultaneous selection model and the sequential selection model, respectively, as shown in examples in Figures 5 and 6.

[0039] In the example in Figure 5, the number of latent classes of the simultaneous selection model, which are classes of type I, is N, and the parameter set 324 of the degree of belonging model for the simultaneous selection model has N parameter subsets for each of the classes of type I. Similarly, in the example in Figure 6, the number of latent classes of the sequential selection model, which are classes of type II, is M, and the parameter set 326 of the degree of belonging model for the sequential selection model has M parameter subsets for each of the classes of type II.

[0040] As shown in Figure 7, the probability Q that passenger n belongs to a latent class (Type 1 class) s in the simultaneous selection model is determined using the degree of belonging model shown in equation (5), based on the parameter subset of the latent class (Type 1 class) s in the parameter set 324 of the degree of belonging model for the simultaneous selection model, and the attributes (passenger properties) of passenger n. Also, as shown in Figure 8, the probability Q that passenger n belongs to a latent class (Type 2 class) s in the sequential selection model is determined using the degree of belonging model shown in equation (5), based on the parameter subset of the latent class (Type 2 class) s in the parameter set 326 of the degree of belonging model for the sequential selection model, and the attributes (passenger properties) of passenger n.

[0041] The simultaneous selection model is a model equation that applies a conditional logit model and is given by equation (6) below. Equation (6) allows us to calculate the probability P that a passenger n belonging to latent class (class 1) s at time σ will select train i.

number

[0042] In equation (6), the explanatory variable x of the utility function V is an item related to train transportation services, such as the type of train, the train fare, seat availability, the passenger's desired boarding time, or the departure time of the desired train (or the relative time difference with respect to the midpoint tσ of the time zone σ). The value of this explanatory variable x is a binary variable, either 1 or 0, and is determined by the train i.

[0043] The parameter set 334 that determines the type of explanatory variable x in the utility function in equation (6), the parameter β for the explanatory variable x, and the constant term β includes N parameter subsets for each latent class (first class) of the simultaneous selection model, as shown in an example in Figure 9. Then, as shown in Figure 10, the selection probability P of passenger n who chooses train i in a time period σ belonging to the latent class (first class) s of the simultaneous selection model is determined using the simultaneous selection model shown in equation (6), based on the parameter subset of that latent class (first class) s in the parameter set 334 of the simultaneous selection model and the transport service of train i.

[0044] The model equation for the successive selection model is the model equation obtained by applying the binomial logit model, and is given by equation (7) below.

number

[0045] Specifically, the probability P of selecting the first train i (m=1) from the candidate trains set for passenger n is calculated by equation (8a), and the probability P of selecting the second or subsequent train i (m≧2) is calculated by equation (8b).

number

[0046] In equation (7), the explanatory variable x of the utility function V is an item related to train transportation services, such as the type of train, the train fare, seat availability, the passenger's desired boarding time, or the departure time of the desired train (or the relative time difference with respect to the midpoint tσ of the time zone σ). The value of this explanatory variable x is a binary variable, either 1 or 0, and is determined by the train i.

[0047] The parameter set 338 that determines the type of explanatory variable x in the utility function in equation (7), the parameter β for the explanatory variable x, and the constant term β includes M parameter subsets for each latent class (Type II class) of the sequential selection model, as shown in Figure 11 as an example. Then, as shown in Figure 12, the selection probability P that a passenger n belonging to the latent class (Type II class) s of the sequential selection model selects train i during the time period σ is calculated using the sequential selection model shown in equation (7), based on the parameter subset of that latent class (Type II class) in the parameter set 338 of the sequential selection model and the transport service of train i.

[0048] Figure 13 shows an example of the functional configuration of the passenger number estimation device 1. According to Figure 13, the passenger number estimation device 1 is configured with an operation unit 102, a display unit 104, a communication unit 106, a processing unit 200, and a storage unit 300, and is implemented as a type of computer system. The passenger number estimation device 1 may be implemented with a single computer, or it may be configured by connecting multiple computers.

[0049] The operation unit 102 is implemented by an input device such as a keyboard, mouse, touch panel, or various switches, and outputs an operation signal to the processing unit 200 according to the operation performed. The display unit 104 is implemented by a display device such as a liquid crystal display or touch panel, and displays various information based on the display signal from the processing unit 200. The communication unit 106 is a communication device implemented by a wireless communication module, router, modem, jack or control circuit for wired communication cables, etc., and connects to a given communication network to perform data communication with external devices.

[0050] The processing unit 200 is a processor implemented using arithmetic devices and circuits such as a CPU (Central Processing Unit) or FPGA (Field Programmable Gate Array), and it performs overall control of the passenger number estimation device 1 based on programs and data stored in the memory unit 300, input data from the operation unit 102 and the communication unit 106, etc.

[0051] Furthermore, the processing unit 200 performs a passenger number estimation process (see Figure 18) by executing a passenger number estimation program 302 to estimate the number of passengers for each train related to a given train schedule based on given passenger demand data. Functional processing blocks for this purpose include a time zone setting unit 202, a candidate train setting unit 204, a selection probability calculation unit 206, and an estimation unit 208. Each of these functional units in the processing unit 200 can be implemented either through software execution by the processing unit 200 or through dedicated calculation circuits. In this embodiment, the former, software implementation, will be described.

[0052] The time zone setting unit 202 sets multiple time zones σ by dividing the time axis included in the passenger demand data 312 along the time series (see Figure 2).

[0053] Passenger demand data 312 is data showing the number of passengers (passenger demand) who wish to travel between the target stations (stations from departure station to arrival station) on the target day, arranged chronologically in order of desired departure time or desired train.

[0054] Figure 14 shows an example of passenger demand data 312. As shown in Figure 14, the passenger demand data 312 stores, for each passenger, a desired boarding time or the departure time of the desired train, and passenger properties, associated with a passenger ID, which is an identification number. Passenger properties include at least attributes. Attributes include not only the place of ticket purchase (ticket purchase channel), but also other elements such as age, gender, purpose of travel, presence or absence of companions, and degree of railway usage experience. Based on the attributes included in the passenger properties, the probability R that a passenger adopts the simultaneous selection model and the sequential selection model, respectively, as train selection models, and the probability Q that a passenger belongs to each latent class (Type 1 class and Type 2 class) of the simultaneous selection model and the sequential selection model, respectively, are determined.

[0055] The candidate train setting unit 204 sets multiple candidate trains corresponding to each passenger in a given time period, based on passenger demand data 312, from among the trains included in the train timetable 310. Alternatively, it may set multiple candidate trains by selecting an estimated optimal candidate train corresponding to the passenger from among the trains included in the train timetable 310, and then selecting a predetermined number of optimal candidate trains in order of proximity to the estimated optimal candidate train.

[0056] Specifically, for example, predetermined time ranges are set before and after the central time tσ of a time period σ (for example, the pre-range from one hour before the central time tσ to the central time tσ, and the post-range from the central time tσ to one hour after that). Then, for each of the pre- and post-ranges, a maximum of a predetermined number of trains (for example, a maximum of 3) are selected from among the trains in that range in order of their departure time being closest to the central time tσ, and these are selected as candidate trains corresponding to the passengers included in that time period σ (see Figure 2). Alternatively, the train whose departure time is closest to the passenger n's desired train or desired boarding time may be used as the estimated optimal candidate train, and candidate trains may be set in order of their proximity to this estimated optimal candidate train.

[0057] The adoption probability calculation unit 206 calculates, for each of the multiple candidate trains corresponding to each passenger based on the passenger demand data 312, 1) the selection probability of the candidate train based on the first selection model and 2) the selection probability of the candidate train based on the second selection model, and then calculates the adoption probability of the candidate train based on the selection probabilities of 1) and 2).

[0058] Specifically, the selection probability of the candidate train for each Class 1 is calculated using a Class 1 parameter subset and the Class 1 belonging probability determined by the passenger properties defined for the passenger. Then, based on the calculation results, 1) the selection probability of the candidate train based on the first selection model is calculated. Furthermore, the selection probability of the candidate train for each Class 2 is calculated using a Class 2 parameter subset and the Class 2 belonging probability determined by the passenger properties defined for the passenger. Then, based on the calculation results, 2) the selection probability of the candidate train based on the second selection model is calculated.

[0059] Furthermore, based on the type of train being selected, the fare for the selected train, the seat availability of the selected train, and the relative time difference of the selected train to the estimated optimal train, we calculate 1) the selection probability of the selected train based on the first selection model and 2) the selection probability of the selected train based on the second selection model.

[0060] Specifically, first, the probability R that each passenger n adopts either the first selection model, the simultaneous selection model, or the second selection model, the sequential selection model, is calculated according to equation (2), based on the model selection probability table 316 and the attributes (passenger properties) of the passenger n.

[0061] Next, the probability Q that each passenger n belongs to each of the latent classes (Type 1 classes) s of the simultaneous selection model is calculated using the belongingness model shown in equation (5), based on the parameter subset of the latent class (Type 1 class) s in the parameter set 334 of the belongingness model for the simultaneous selection model and the attributes (passenger properties) of the passenger n. Subsequently, the selection probability P of each candidate train i for each latent class (Type 1 class) of the simultaneous selection model for each passenger n is calculated using the simultaneous selection model shown in equation (6), based on the parameter set 334 of the simultaneous selection model and the transportation service of the candidate train i. Then, the selection probability P of each candidate train i when each passenger n adopts the simultaneous selection model is calculated according to equation (4), based on the calculated selection probability P for each latent class (Type 1 class).

[0062] Furthermore, the probability Q that each passenger n belongs to each latent class (Type 2 class) s of the sequential selection model is calculated using the belongingness model shown in equation (5), based on the parameter subset of the latent class (Type 2 class) s in the parameter set 326 of the belongingness model for the sequential selection model and the attributes (passenger properties) of the passenger n. Subsequently, the selection probability P of each candidate train i for each latent class (Type 2 class) of the sequential selection model for each passenger n is calculated using the sequential selection model shown in equation (7), based on the parameter set 338 of the sequential selection model and the transportation service of the train i. Then, the selection probability P of each candidate train i when each passenger n chooses the sequential selection model is calculated according to equation (4), based on the calculated selection probability P for each latent class (Type 2 class).

[0063] Next, the probability P that each passenger n adopts each candidate train i is calculated according to equation (3), based on the selection probability P of each candidate train i when the passenger n adopts both the simultaneous selection model and the sequential selection model, respectively (see Figure 4).

[0064] The data relating to the degree of belonging model is stored as the degree of belonging model data 320. The degree of belonging model data 320 includes the model equation 322 of the degree of belonging model shown in equation (5), the parameter set 324 for the degree of belonging model for the simultaneous selection model, and the parameter set 326 for the degree of belonging model for the sequential selection model.

[0065] The data relating to the first selection model, the simultaneous selection model, and the second selection model, the sequential selection model, are stored as train selection model data 330. The train selection model data 330 includes the model equation 332 for the simultaneous selection model shown in equation (6), the parameter set 334 for the simultaneous selection model, the model equation 336 for the sequential selection model shown in equation (7), and the parameter set 338 for the sequential selection model.

[0066] The transportation services for each train are stored as transportation service data 314. For each train specified in the train schedule 310, transportation service data 314 stores the train's transportation services, such as type and fare (usage fee), associated with the train number.

[0067] The calculated probability P of each candidate train i for each passenger n is stored as passenger-specific probability data 340. Figure 15 shows an example of passenger-specific probability data 340. As shown in Figure 15, the passenger-specific probability data 340 stores the time period σ and the probability P of each candidate train in association with each passenger n.

[0068] The estimation unit 208 estimates the number of passengers for each train in the train schedule based on the aggregated results of the adoption probability calculation unit 206 for each candidate train for each passenger.

[0069] Specifically, based on passenger-specific adoption probability data 340, the selection probability of each candidate train i in each time period σ is calculated as the average of the adoption probabilities of each candidate train i for each passenger n included in that time period σ. Next, the estimated number of passengers on each candidate train i in each time period σ is calculated by multiplying the number of passengers in that time period σ by the selection probability of each candidate train i in that time period σ. Finally, the estimated number of passengers on each train i is calculated as the sum of the estimated number of passengers on that train i in each time period σ.

[0070] The estimated number of passengers for each candidate train i within each calculated time period σ is stored as time-period estimated passenger data 342. Figure 16 shows an example of time-period estimated passenger data 342. As shown in Figure 16, the time-period estimated passenger data 342 stores the number of passengers, the selection probability and estimated number of passengers for each candidate train, in association with each time period σ.

[0071] Furthermore, the estimated number of passengers for each train i is stored as estimated passenger data 344. Figure 17 shows an example of estimated passenger data 344. As shown in Figure 17, the estimated passenger data 344 stores the estimated number of passengers associated with each train i.

[0072] By performing the calculation of the probability of each passenger being selected for each candidate train by the selection probability calculation unit 206 and the estimation of the estimated number of passengers by the estimation unit 208 in a time-series order with time period σ as the unit, it is possible to calculate the selection probability P for each candidate train based on both the simultaneous selection model and the sequential selection model, respectively, which more appropriately reflects the seat availability status of the candidate trains.

[0073] In other words, by allocating the estimated number of passengers for each candidate train in each time period σ calculated by the estimation unit 208 (estimated passenger data by time period 342; see Figure 16) as passengers for the candidate train, the number of passengers on each train up to that time period σ can be calculated. As a result, when the adoption probability calculation unit 206 calculates the adoption probability P for each candidate train for each passenger in time period σ, it can obtain the availability of seats on the train from the estimated number of passengers on each train in the preceding time periods.

[0074] The memory unit 300 is implemented using IC (Integrated Circuit) memory such as ROM (Read Only Memory) and RAM (Random Access Memory), storage devices such as hard disks, and external storage devices built in a cloud environment. It stores programs and data for the processing unit 200 to comprehensively control the passenger number estimation device 1, and is also used as a workspace for the processing unit 200. Calculation results performed by the processing unit 200 and input data from the operation unit 102 and communication unit 106 are temporarily stored there.

[0075] In this embodiment, the storage unit 300 stores a passenger number estimation program 302, a train schedule 310, passenger demand data 312, transportation service data 314, a model selection probability table 316, belongingness model data 320, train selection model data 330, passenger adoption probability data 340, estimated passenger number data by time of day 342, and estimated passenger number data 344.

[0076] Figure 18 is a flowchart illustrating the passenger number estimation process. This process is performed on a train schedule 310 and passenger demand data 312 for a specific station on a specific day.

[0077] First, the time zone setting unit 202 sets multiple time zones σ by dividing the time axis of the passenger demand data 312 along the time series (step S1). Next, it selects one of the set time zones σ in chronological order as the target time zone σ (step S3), and performs the following processing (steps S5 to S25) on the selected time zone σ.

[0078] Specifically, by referring to the passenger demand data 312, one passenger n included in the target time period σ is selected as the target passenger n (step S5), and the following processing (steps S5 to S19) is performed on the selected passenger n.

[0079] First, the candidate train setting unit 204 sets multiple candidate trains for the target passenger n (step S7). Next, the adoption probability calculation unit 206 calculates the probability R that the target passenger n adopts the simultaneous selection model and the sequential selection model, respectively, as train selection models (step S9). Subsequently, the probability Q that the target passenger n belongs to each latent class (Type 1 class) of the simultaneous selection model is calculated (step S11). Then, the selection probability of each candidate train i is calculated for each latent class (Type 1 class) of the simultaneous selection model in which the target passenger n belongs. Finally, based on the calculated selection probability P of each candidate train i for each latent class (Type 1 class), the probability that the target passenger n selects each candidate train i is calculated (step S13).

[0080] Next, the probability Q that the target passenger n belongs to each latent class (Type 2 class) of the sequential selection model is calculated (Step S15). Then, the selection probability P of each candidate train i is calculated for each latent class (Type 2 class) that the target passenger n belongs to. Then, based on the calculated selection probability P of each candidate train i for each latent class (Type 2 class), the probability P that the target passenger n selects each candidate train i is calculated (Step S17). Subsequently, the adoption probability P of each candidate train i for passenger n is calculated based on the selection probability P of each candidate train i when using the simultaneous selection model and the sequential selection model, respectively (Step S19).

[0081] Next, it is determined whether all passengers n in the target time period σ have already been selected as target passengers. If they have not been selected (step S21: NO), the process returns to step S5 to select the next passenger. If all passengers have been selected (step S21: YES), the selection probability P for each candidate train in the target time period σ is then calculated based on the selection probability P for each candidate train for each passenger n in the target time period σ (step S23). Subsequently, the estimation unit 208 calculates the estimated number of passengers on each candidate train in the target time period σ by multiplying the number of passengers in the target time period σ by the selection probability P for that candidate train (step S25).

[0082] Next, it is determined whether all time periods σ have been selected as the target time periods σ. If they have not been selected (step S27: NO), the process returns to step S3 to select the next time period σ. If all time periods σ have been selected (step S27: YES), the estimation unit 208 calculates the estimated number of passengers for each train by summing the estimated number of passengers for each time period σ of that train (step S29). After completing the above process, the passenger number estimation process is finished.

[0083] According to this embodiment, it becomes possible to estimate the number of passengers on a train, assuming that each passenger's train selection tendencies are not the same. Specifically, as selection models that show the typical train selection tendencies that passengers may take, a simultaneous selection model is prepared that calculates the selection probability of each candidate train when multiple candidate trains are compared in parallel, and a sequential selection model is prepared that calculates the selection probability of each candidate train by sequentially calculating the selection probability of each candidate train from among multiple candidate trains. Using these two types of train selection models, the number of passengers on each train is estimated by probabilistically calculating and aggregating the selection probability of each candidate train to determine which train a passenger will choose. This realizes the estimation of the number of passengers on a train, assuming that each passenger's train selection tendencies are not the same.

[0084] It should be noted that the applicable embodiments of the present invention are not limited to those described above, and can be modified as appropriate without departing from the spirit of the invention. [Explanation of Symbols]

[0085] 1... Passenger number estimation device 200... Processing Unit 202...Time zone setting section 204…Recruitment candidate train setting department 206...Employment probability calculation unit 208…Estimation Department 300...Storage section 302... Passenger Count Estimation Program 310...Train schedule data 312…Passenger demand data 314…Transportation service data 316…Model Selection Probability Table 320... Belonging Model Data 330...Train selection model data 340… Passenger-specific recruitment probability data 342…Estimated passenger count data by time of day 344…Estimated passenger count data

Claims

1. A passenger number estimation device that estimates the number of passengers for each train related to a given train schedule based on given passenger demand data, A storage means for storing data representing a first selection model that determines the selection probability for each candidate train when a given number of candidate trains are compared in parallel based on a first parameter set based on passenger properties which include at least the attributes of passengers, and a second selection model that determines the selection probability for each candidate train by sequentially determining the selection probability for each train from the number of candidate trains based on a second parameter set based on the passenger properties, A candidate train setting means for setting multiple candidate trains corresponding to each passenger based on the passenger demand data from among the trains included in the aforementioned train timetable, For each of the multiple candidate trains corresponding to each passenger based on the passenger demand data, the candidate train is treated as the candidate train, and 1) the selection probability of the candidate train based on the first selection model and 2) the selection probability of the candidate train based on the second selection model are calculated, and the selection probability of the candidate train is calculated based on the selection probabilities of 1) and 2), the selection probability of the candidate train is calculated, An estimation means for estimating the number of passengers for each train in the train schedule based on the results of aggregating the adoption probabilities for each candidate train for each passenger calculated by the adoption probability calculation means, A passenger number estimation device equipped with [a specific feature].

2. The aforementioned passenger demand data is data in which passengers are arranged chronologically in order of their desired boarding time or desired train. A time zone setting means for setting multiple time zones divided along the time series of the aforementioned passenger demand data, Furthermore, The aforementioned candidate train setting means sets a plurality of candidate trains corresponding to the passengers included in each time period. The passenger number estimation device according to claim 1.

3. The first parameter set includes a subset of parameters for each of the first classes, which classifies the passengers into a plurality of first classes based on the passenger properties. The second parameter set includes a subset of parameters for each of the two classes, which classifies the passengers into a plurality of two classes based on the passenger properties. The aforementioned means for calculating the probability of adoption is: Using the Class 1 parameter subset and the Class 1 belonging probability determined by the passenger properties defined for the passenger, the selection probability of the candidate train for each Class 1 is calculated, and the selection probability of 1) is calculated based on the calculation result. Using the aforementioned subset of parameters for each second class and the probability of belonging to each second class determined by the passenger properties defined for the passenger, the selection probability for each second class of the candidate train is calculated, and the selection probability of 2) above is calculated based on the calculation results. The passenger number estimation device according to claim 1.

4. The first parameter set and the second parameter set each include parameters based on the train type. The aforementioned selection probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the train type of the candidate train. A passenger number estimation device according to any one of claims 1 to 3.

5. The first parameter set and the second parameter set each include parameters based on train fares, The aforementioned selection probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the usage fee of the candidate train. A passenger number estimation device according to any one of claims 1 to 3.

6. The first parameter set and the second parameter set each include parameters based on the availability of seats on the train. The aforementioned probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the availability of seats on the candidate trains corresponding to each passenger based on the calculated probability of adoption, in the process of calculating the probability of adoption for each of the multiple candidate trains corresponding to each passenger based on the passenger demand data. A passenger number estimation device according to any one of claims 1 to 3.

7. The candidate train setting means selects an estimated optimal candidate train corresponding to the passenger from among the trains included in the train timetable, and sets a plurality of candidate trains by selecting a predetermined number of optimal candidate trains in order of proximity to the estimated optimal candidate train in the sequence of trains. The first parameter set and the second parameter set each include parameters based on the relative time difference of the train to the estimated optimal candidate train, The selection probability calculation means calculates the selection probability of 1) and the selection probability of 2) based on the relative time difference of the candidate train to be selected with respect to the estimated optimal candidate train corresponding to the passenger. A passenger number estimation device according to any one of claims 1 to 3.

8. A passenger number estimation method in which a computer system estimates the number of passengers for each train related to a given train schedule based on given passenger demand data, The computer system stores data representing a first selection model that calculates the selection probability for each candidate train when comparing a given number of candidate trains in parallel based on a first parameter set based on passenger properties, which include at least the attributes of the passengers, and a second selection model that calculates the selection probability for each candidate train by sequentially calculating the selection probability for each train from the number of candidate trains based on a second parameter set based on the passenger properties. From among the trains included in the aforementioned train schedule, multiple candidate trains corresponding to each passenger based on the aforementioned passenger demand data are selected. For each of the multiple candidate trains corresponding to each passenger based on the passenger demand data, the candidate train is treated as the candidate train, and 1) the selection probability of the candidate train based on the first selection model and 2) the selection probability of the candidate train based on the second selection model are calculated, and the probability of the candidate train being adopted is calculated based on the selection probabilities of 1) and 2). Based on the results of aggregating the aforementioned probability of each passenger being selected for each candidate train, the estimated number of passengers for each train in the aforementioned train schedule is estimated, A method for estimating the number of passengers, including [specific details omitted].

Images