JP6295769B2 - Person dynamic calculation device, human dynamic calculation system, and program - Google Patents

Description

  The present invention relates to a human activity calculation device, a human activity calculation system, and a program, and more particularly to a human activity calculation device, a human activity calculation system, and a program that calculate a person's action sequence and destination sequence.

  Conventionally, in a person trip survey or the like, there is no real-time property and the number of samples may be insufficient.

  In addition, a method is known in which a user's attribute and situation are estimated from information obtained by combining the user's GPS history and attribute information such as a store at the place of stay, and the movement of a person in the city is reproduced (patent) Reference 1).

JP 2012-164227 A

  In the prior art, in order to secure a sufficient number of samples in real time, there is a problem that a large number of investigators are required and the monetary cost is enormous.

  Moreover, in the technique of patent document 1, it must estimate with respect to all the persons in a city, and all the persons carry the device with a GPS function at present, and GPS is turned ON. However, there is a problem that estimation cannot be made for all people. In addition, there is a problem that it is necessary to perform estimation for each person, and enormous computational resources are required.

  The present invention has been made to solve the above-described problems, and is based on statistical data and is capable of calculating a human action sequence and a destination sequence with high accuracy. And to provide a program.

  In order to achieve the above object, the person dynamics calculation device according to the first invention is based on statistical data representing population data by attribute of each area, and a person having the attribute for each attribute for each area. An agent generating means for generating an agent representing a person living in the area in a number corresponding to the population data of the attribute of the area, and each of a plurality of action pairs learned in advance for each attribute An action sequence generating means for generating an action sequence consisting of actions of each time period for each of the agents generated by the agent generating means based on a transition probability from one action of the pair to the other action; For each of the agents generated by the agent generating means, a pair of areas learned in advance for each of the plurality of actions. The purpose of each time zone is based on the destination selection probability of selecting the other area from one area of the pair as a destination and the action sequence generated for the agent by the action sequence generating means. For each of the agent generated by the agent generating unit and the agent sequence generated by the agent generating unit, the action sequence generated for the agent by the action sequence generating unit and the purpose Output means for outputting the destination series generated for the agent by the ground series generation means.

  According to the first invention, the agent generation means generates, for each area, an agent representing a person living in the area for each attribute in the number corresponding to the population data of the area attribute, and the action sequence generation means Based on the transition probability from one action of the pair to the other action for each of a plurality of action pairs learned in advance by attribute, an action sequence consisting of actions in each time zone is generated for each agent Then, the destination selection probability of selecting the other area as the destination from one area of each pair of areas learned in advance for each of the plurality of actions for each of the agents by the destination sequence generation means And a destination sequence consisting of areas that are destinations for each time zone based on It means by, for each of the agents, and outputs the action sequences and the destination sequence generated for the agent.

  In this way, an agent is generated, and an action sequence composed of actions of each time period is determined for each of the agents based on the transition probability from one action of the pair to the other action for each of the plurality of action pairs. Generated for each agent, the destination selection probability of selecting the other area as the destination from one area of the pair for each of the pair of areas, learned in advance for each of the plurality of actions, and generated for the agent Based on the action sequence, a destination sequence consisting of areas that are the destinations of each time zone is generated, and by outputting the action sequence and destination sequence generated for the agent, all the people in the city The movement can be accurately reproduced.

  Moreover, in the person dynamics calculation apparatus of 1st invention, the said action was performed in the said time slot | zone among the persons with the said attribute prepared beforehand for every combination of the said attribute, the said action, and the said time slot | zone. Actor-specific data by time period representing the proportion of persons, and 1 representing the proportion of persons who performed the act in one day among the persons having the attribute prepared in advance for each combination of the attribute and the act A transition probability learning means for learning the transition probability for each attribute based on the daily actor rate data may be further included.

  In the human dynamics calculation device according to the first aspect of the invention, facility distribution data representing physical quantities of facilities related to each action in each area prepared in advance and generated for each of the agents by the action series generation means. Based on the action series, the area and the combination of the actions are required. Based on the number of persons who perform the action among the persons living in the area, the action is determined for each of the plurality of actions. It may further include destination selection probability learning means for learning the destination selection probability using a model in which the number of persons who have moved from one of the pair of areas to the other is modeled.

  In the human dynamics calculation device according to the first aspect of the invention, facility distribution data representing physical quantities of facilities related to each action in each area prepared in advance and generated for each of the agents by the action series generation means. Based on the action series, the area and the combination of the actions are required. Based on the number of persons who perform the action among the persons living in the area, the action is determined for each of the plurality of actions. Using a model that models the number of people who have moved from one of the pair of areas to the other, learning each of the destination selection probabilities when applying each value of the model parameter, the parameter of the model Of the destination selection probabilities when each value is applied, each of the combinations of the time zone and the area prepared in advance Compatible with the time zone area population by data representative of the population of may further include a destination selection probability learning means for selecting the destination selection probability.

  A human activity calculation system according to a second invention is the human activity calculation device according to the first invention, and is prepared in advance for each combination of the attribute, the action, and the time zone. , The actor rate data for each time zone representing the percentage of the person who performed the act in the time zone, and prepared in advance for each combination of the attribute and the act, the day of the person having the attribute And a probability learning device including transition probability learning means for learning the transition probability for each attribute based on the daily actor rate data representing the percentage of persons who have acted.

  A human activity calculation system according to a third invention is the personal activity calculation device according to the first invention, the facility distribution data representing the physical quantity of the facility related to each activity in each area, and the activity sequence generation Based on the number of persons performing the action among the persons living in the area, which are obtained for the combination of the area and the action based on the action sequence generated for each of the agents by means. Probability including destination selection probability learning means for learning the destination selection probability using a model that models the number of persons who perform the action from one of the pair of areas to the other for each of the acts And a learning device.

  A human activity calculation system according to a fourth invention is the personal activity calculation device according to the first invention, the facility distribution data representing the physical quantity of the facility related to each activity in each area, and the activity sequence generation Based on the number of persons performing the action among the persons living in the area, which are obtained for the combination of the area and the action based on the action sequence generated for each of the agents by means. For each of the acts, the destination selection probability when each value of the model parameter is applied using a model that models the number of people who perform the act moves from one of the pair of areas to the other Of the destination selection probability when each value of the parameter of the model is applied, the time zone and A probability learning device including destination selection probability learning means for selecting the destination selection probability, which is matched with time-specific area-specific population data representing the population of each area combination. .

  Moreover, the program of this invention is a program for functioning a computer as each means which comprises said person dynamics calculation apparatus.

  As described above, according to the human activity calculation device, the human activity calculation system, and the program of the present invention, an agent is generated based on statistical data representing population data for each attribute of each area, and learning is performed in advance for each attribute. For each agent, a sequence of actions consisting of actions for each time period is generated for each agent, and for each agent, a destination selection probability learned in advance for each of a plurality of actions and for each agent are generated. Based on the statistical data by generating a destination sequence consisting of areas that are destinations in each time zone based on the generated action sequence and outputting the generated action sequence and destination sequence for each of the agents. Thus, it is possible to accurately calculate a person's action series and destination series.

It is a block diagram which shows the functional structure of the probability learning apparatus which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the structure of the person dynamics calculation apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the example of movement data in a city. It is a figure which shows the probability learning process routine in the probability learning apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the calculation processing routine of the hourly action transition probability matrix in the probability learning apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the calculation process routine of the destination selection probability matrix according to action in the probability learning apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the person dynamics calculation process routine in the human dynamics calculation apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the action series production | generation processing routine in the person dynamics calculation apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the destination series production | generation routine in the person dynamics calculation apparatus which concerns on the 1st Embodiment of this invention. It is a block diagram which shows the functional structure of the probability learning apparatus which concerns on the 2nd Embodiment of this invention. It is a figure which shows the probability learning process routine in the probability learning apparatus which concerns on the 2nd Embodiment of this invention. It is a figure which shows the calculation process routine of the destination selection probability matrix according to action in the probability learning apparatus which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows the functional structure in the person dynamics calculation apparatus which concerns on the 3rd Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the present embodiment, the T-time division number, the type of action to K, the number of areas of ^{L, y (0) ... y} (T-1) the time zone actors constant (K element) , A ⁽⁰⁾ ... A ^(T-2) is an action transition probability matrix by time zone (K × K elements), and α _k ^(t) is the k-th row vector (K element) of the action transition probability matrix by time zone. Or the transition probability of the action at time t when the action at time t-1 is k, P ^k is the destination selection probability matrix (K × K element) of action k, and p _l ^k is the destination selection probability matrix of action k The l-th row vector (L element) or the destination selection probability at time t when the location at time t−1 is l, k ⁽⁰⁾ ... K ^(T−1) is the action sequence, l ⁽⁰⁾ . l ^(T-1) is defined as the destination series.

<Configuration of Human Dynamics Calculation System According to First Embodiment>
First, the configuration of the human behavior calculation system according to the first embodiment of the present invention will be described. The human behavior calculation system according to the first embodiment of the present invention includes a probability learning device 100 and a personal behavior calculation device 200 described later.

<Configuration of Probability Learning Device According to First Embodiment>
Next, the configuration of the probability learning device 100 according to the first embodiment of the present invention will be described. As shown in FIG. 1, a probability learning device 100 according to an embodiment of the present invention includes a CPU, a RAM, and a ROM that stores a program and various data for executing a probability learning processing routine described later. Can be configured with a computer. Functionally, the probability learning apparatus 100 includes an input unit 10, a calculation unit 20, and an output unit 90 as shown in FIG.

The input unit 10 is a daily actor rate data indicating the percentage of persons who acted by attribute per day, a parameter β _i, and an actor by time zone representing the percentage of persons who acted by attribute and by time zone The rate data, the generation amount by action of each area obtained from the action sequence generated in the previous process in the personal behavior calculation device 200 described later, and the facility distribution data representing the distribution of the number of facilities are accepted.

  The computing unit 20 includes a daily actor rate data storage unit 24, an hourly actor rate data storage unit 26, an act-based generation amount storage unit 28, a facility distribution data storage unit 30, and an hourly activity transition. A probability matrix learning unit 40, a time-dependent action transition probability matrix storage unit 42, an action-specific destination selection probability matrix learning unit 44, and an action-specific destination selection probability matrix storage unit 46 are configured.

  The daily actor rate data storage unit 24 stores the daily actor rate data received by the input unit 10.

  The hourly actor rate data storage unit 26 stores hourly actor rate data received by the input unit 10. In addition, as the actor rate data by time zone, the basic survey data on social life of the Ministry of Internal Affairs and Communications may be used.

  The action generation amount storage unit 28 stores the action generation amount of each area received by the input unit 10. Here, the generated amount by action is obtained from the action series for each of the agents generated in the person dynamics calculation device 200, and for each area, among the agents whose residence mesh is the area, This represents the number of agents that perform the action.

  The facility distribution data storage unit 30 stores the facility distribution data received by the input unit 10. The facility distribution data represents the number of facilities related to each action in each area. The facility distribution data may represent the number of facilities related to each action in each area. In addition, the facility distribution data may represent the sales floor area, sales, etc. of the facility related to each action in each area. The number of facilities is an example of the physical quantity of the facility.

The action transition probability matrix learning unit for each time zone 40 includes the daily actor rate data stored in the daily actor rate data storage unit 24 and the time stored in the actor rate data storage unit 26 for each time zone. a band by actors rate data, on the basis of the parameter beta _i accepted in the input unit 10, a time zone for each demographic Acts transition probability matrix a ^{(0) ...} learn a ^(T-2), time zone While storing in the action transition probability matrix memory | storage part 42, it outputs to the output part 90. FIG. In addition, the action transition probability matrix classified by time zone represents the transition probability from one action of the pair to the other action for each of a plurality of action pairs.

The action transition probability matrix for each time zone is learned for each attribute by solving a quadratic programming problem expressed by the following equation (1). Specifically, the actor rate data y ⁽⁰⁾ ... Y ^(T-1) by time zone in the attribute stored in the hourly actor rate data storage unit 26 and the parameters received by the input unit 10 based on the beta _i, using the following (1) to solve the quadratic programming problem in the expression, the resulting time zone Acts transition probability matrix ^{a (0) ... a (T} -2), human pharmacokinetics Similar to the calculation device 200, a simulation is performed. Then, the simulation results are aggregated to calculate the daily actor rate for the person having the attribute, and the daily actor rate for the attribute and the daily actor rate data storage unit 24 are stored. The daily actor rate data in the attribute is compared. When the error is within a predetermined range as a result of the comparison, the action transition probability matrix for each time zone in the attribute obtained by the processing is converted into the action transition probability matrix for each time zone A ^{(0) in the} attribute. ... A ^(T-2) is confirmed. On the other hand, if the error is outside the predetermined range as a result of the comparison, the value of the parameter β _i changed to an appropriate value is received by the input unit 10 and the quadratic programming problem shown in the following equation (1) The process of solving the process, the process of performing the simulation, the process of calculating the daily actor rate, and the comparison process are repeatedly executed.

  The action transition probability matrix storage unit for each time zone stores the action transition probability matrix for each time zone learned for each attribute by the action transition probability matrix learning unit for each time zone.

  The action-specific destination selection probability matrix learning unit 44 performs an action based on the action-based generation amount stored in the action-specific generation amount storage unit 28 and the facility distribution data stored in the facility distribution data storage unit 30. Each destination selection probability matrix is learned, and the learned result is stored as an action-specific destination selection probability matrix in the action-specific destination selection probability matrix storage unit 46 and output to the output unit 90. The action-specific destination selection probability matrix represents a destination selection probability for selecting one area of each pair of areas as the destination from the other area.

To learn the action-specific destination selection probability matrix, first, for each action k, X _ij ^(k) is calculated using the gravity model shown in the following equation (2). Here, X _ij ^(k) is the number of persons moving between meshes ij of the person who performs act k, A _i ^(k) and B _j ^(k) are coefficients, and O _i ^(k) is the mesh is the number of people living in i and performing act k, D _j ^(k) is the number of facilities related to act k in mesh j, and dist _ij is a predetermined inter-ij distance (or Required time), and γ is a distance coefficient. The coefficient A _i ^(k) is expressed by the following expression (3), and the coefficient B _j ^(k) is expressed by the following expression (4). It is assumed that 1 is given as an initial value of the coefficients A _i ^(k) and B _j ^(k) . Further, it is assumed that the distance coefficient γ is given an appropriate value (γ> 0). In addition, the gravity model represents a model in which the number of persons who perform an action moves from one of the pair of areas to the other.

Specifically, first, the number O _i ^{(k) of} people who live in the residence mesh i and perform the action k, which is obtained from the action-specific generation amount stored in the action-specific generation amount storage unit 28, is acquired. . Next, for each action, the acquired O _i ^(k) , D _j ^(k) obtained from the facility distribution data, each of the distances dist _ij between _ij , and distances set in advance with appropriate values Based on the coefficient γ and the coefficients A _i ^(k) and B _j ^(k) , the coefficients A _i ^(k) and B _j ^(k) are alternately switched according to the above expressions (3) and (4). Repeat the calculation to calculate the coefficients A _i ^(k) and B _j ^(k) . For each action, the converged coefficients A _i ^(k) and B _j ^(k) , O _i ^(k) , D _j ^(k) obtained from the facility distribution data, and each distance dist _ij between _ij X _ij ^(k) is calculated according to the above equation (2) based on each of the above and the distance coefficient γ.

Next, using X _ij ^(k) calculated using the gravity model for each action k, the destination selection probability P _ij ^(k) for each action k is set in areas i and j according to the following equation (5). It calculates about each of a pair and it is set as the learning result of the destination selection probability matrix for every action. Here, P _ij ^(k) is the probability that the agent of the current location i and the action k will select the destination j.

  The action-specific destination selection probability matrix storage unit 46 stores the action-specific destination selection probability matrix learned by the action-specific destination selection probability matrix learning unit 44.

  The output unit 90 includes an action transition probability matrix for each time zone learned by the action transition probability matrix learning unit for each time zone 40 and an action destination selection probability matrix for each action learned by the action destination selection probability matrix learning unit 44. Output.

<Configuration of Person Dynamic Calculation Device According to First Embodiment>
Next, the configuration of the human behavior calculation device 200 according to the first embodiment of the present invention will be described. As shown in FIG. 2, a human activity calculation apparatus 200 according to an embodiment of the present invention includes a CPU, a RAM, a ROM that stores a program and various data for executing a personal activity calculation processing routine described later, Can be configured with a computer including Functionally, the person dynamics calculation apparatus 200 includes an input unit 210, a calculation unit 220, and an output unit 290 as shown in FIG.

  The input unit 210 includes area-specific population data, which is statistical data representing an attribute-specific population for each area, an attribute-specific time-dependent action transition probability matrix and an action-specific destination selection probability learned by the probability learning device 100 described above. It accepts a matrix and hourly actor rate data consisting of probabilities of actors by attribute and by time zone.

  The calculation unit 220 includes an area-specific population data storage unit 222, a time-zone-specific action transition probability matrix storage unit 224, a time-zone-specific actor rate data storage unit 226, an action-specific destination selection probability matrix storage unit 228, An agent generation unit 230 and an in-city movement data storage unit 260 are included.

  The area-specific population data storage unit 222 stores area-specific population data received by the input unit 210. As census data by area, national census data or the like may be used.

  The time zone action transition probability matrix storage unit 224 stores a time zone action transition probability matrix for each attribute received by the input unit 210.

  The hourly actor rate data storage unit 226 stores the hourly actor rate data received by the input unit 210.

  The action-specific destination selection probability matrix storage unit 228 stores the action-specific destination selection probability matrix received by the input unit 210.

  The agent generation unit 230 includes the area-specific population data stored in the area-specific population data storage unit 222, the attribute-specific time-zone action transition probability matrix stored in the time-zone-specific action transition probability matrix storage unit 224, Based on the hourly actor rate data stored in the hourly actor rate data storage unit 226 and the action destination selection probability matrix stored in the action destination selection probability matrix storage unit 228. The action sequence and the destination sequence for each of the agents are generated, stored in the in-city movement data storage unit 260, and output to the output unit 290. The agent generation unit 230 includes an attribute generation unit 232, an action sequence generation unit 234, and a destination sequence generation unit 236.

  Based on the population data for each area stored in the population data storage unit for each area 222, the attribute generation unit 232 assigns an agent representing a person having the attribute to the area and the area for each attribute and area. Generate as many as the population of the attribute. Specifically, for example, a combination of age and gender included in the population data by area is defined as an attribute as shown in the C column of FIG. 3, and the residence of the person having the attribute is shown in the B column of FIG. As such a residence mesh, the attribute and the residence mesh are combined and created as one agent. Note that IDs as shown in the column A of FIG. 3 are given in the order of creation. In addition, the residence mesh is assumed to be defined either in advance which residence belongs to which residents mesh. The attribute generation unit is an example of an agent generation unit.

  The action sequence generation unit 234 includes each of the agents created in the attribute generation unit 232, the action transition probability matrix for each time zone stored in the action transition probability matrix storage unit for each time zone 224, and the time zone Based on the actor rate data by time zone stored in the actor rate data storage unit 226, an action sequence as shown in columns D1 to DT of FIG. 3 is generated for each agent, and for each agent, Data that combines the ID, the residence mesh, the attribute, and the action sequence is created and output to the destination sequence generation unit 236 and also to the outside of the person dynamics calculation device 200.

Specifically, first, for each agent, the first time zone for the same attribute as the agent attribute of the hourly actor rate data stored in the hourly actor rate data storage unit 226. The action k ⁽⁰⁾ at time t = 0 is determined based on the actor rate y ⁽⁰⁾ at.

Then, for each agent, the action k ^(t) at time t (t> 0) is based on the action transition probability matrix for each time zone by attribute and the action k ^{(t−1) at} time t−1. To decide. That is, it is the k ^(t−1) -th row vector of the action transition probability matrix by time zone.

The action k ^(t) at the time t is generated with the probability according to the above. For example, the action k ⁽⁰⁾ at the determined time t = 0 and the k ⁽⁰⁾ -th row vector of the attribute of the agent in the action transition probability matrix for each attribute by time zone.

The action sequence k ⁽⁰⁾ ... K ^{(T-1) of the} agent is determined by determining the action k ⁽¹⁾ based on the probability of obeying and repeating the determination process until time t = T-1. Is generated.

  The destination series generation unit 236 is based on the action series of each agent input by the action series generation unit 234 and the action-specific destination selection probability matrix stored in the action-specific destination selection probability matrix storage unit 228. For each of the agents, a destination series of E1 column to ET column shown in FIG. 3 is generated, and created for each agent as in-city movement data combined as shown in column A to ET column of FIG. The data is stored in the city movement data storage unit 260 and output to the output unit 290.

Specifically, first, the resident mesh l _{rsd of the} agent included in the data input from the action sequence generation unit 234 and the action-specific destination selection probability stored in the action-specific destination selection probability matrix storage unit 228. Corresponds to the first action k ⁽⁰⁾ of the agent in the queue

Based on the above, the destination l ⁽⁰⁾ at t = 0 is determined. Then, the destination l ^(t) at time t (t> 0) is determined based on the action-specific destination selection probability matrix and the destination l ^{(t−1) at} time t−1. That is, it is the l ^(t−1) -th row vector of the action-specific destination selection probability matrix corresponding to action k (t).

The destination l ^(t) at time t is determined by the probability of following. For example, the destination l ^{(0) at} the time t = 0 of the determined agent and the l ⁽⁰⁾ -th row vector of the action-specific destination selection probability matrix corresponding to the action k (1) of the agent.

The destination l ⁽¹⁾ is determined based on the probability according to the above. By repeating the process until time t = T−1, the destination series l ⁽⁰⁾ ... L ^(T−1) of the agent is generated. The destination transition is performed only when the action changes from the previous time. Also, under the assumption that there are no multiple workplaces or schools, if the act is work or school, it is necessary to reproduce the destination that has been given so far. If action is taken, recreate the destination.

  The intra-city movement data storage unit 260 stores intra-city movement data such as columns A to ET shown in FIG. 3 input from the destination series generation unit 236.

  The output unit 290 visualizes the city movement data shown in FIG.

<Operation of the probability learning device according to the first embodiment>
Next, the operation of the probability learning device 100 according to the first embodiment will be described. First, the input unit 10 accepts the daily actor rate data, the parameter β _i , the hourly actor rate data, the act-related generation amount, and the facility distribution data. Then, the probability learning processing routine shown in FIG.

  First, in step S100, the daily actor rate data stored in the daily actor rate data storage unit 24 is read.

  Next, in step S102, the actor rate data classified by time zone memorize | stored in the actor rate data memory | storage part 26 classified by time zone is read.

  Next, in step S <b> 104, the action-specific generation amount stored in the action-specific generation amount storage unit 28 is read.

  Next, in step S106, the facility distribution data stored in the facility distribution data storage unit 30 is read.

Next, in step S108, for each attribute, based on the daily actor rate data acquired in step S100, the hourly actor rate data acquired in step S102, and the parameter β _i received in the input unit 10. Then, the action transition probability matrix A ⁽⁰⁾ ... A ^(T-2) for each time zone is learned and stored in the action transition probability matrix storage section 42 for each time zone.

  Next, in step S110, a destination selection probability matrix is learned for each action based on the action-generated amount acquired in step S104 and the facility distribution data acquired in step S106. Is stored in the action-specific destination selection probability matrix storage unit 46.

  Next, in step S112, the action transition probability matrix classified by time zone for each attribute acquired in step S108 and the destination selection probability matrix classified by action acquired in step S110 are output from the output unit 90, and a probability learning process routine is performed. Exit.

  The calculation of the action transition probability matrix for each time zone in step S108 will be described in detail in the calculation processing routine for the action transition probability matrix for each time zone shown in FIG.

In step S200 of FIG. 5, the hourly actor rate y ⁽⁰⁾ ... Y ^(T−1) is acquired for the attribute to be processed based on the hourly actor rate data acquired in step S102. .

Next, in step S202, based on the time zone actor rate for the attribute to be processed acquired in step S200 and the parameter β _i received in the input unit 10, the second order is performed according to the above equation (1). Solve planning problems.

Next, in step S204, a simulation similar to that of the person dynamics calculation apparatus 200 is performed using the time zone action transition probability matrix A ⁽⁰⁾ ... A ^(T-2) for the attribute to be processed acquired in step S202. Do.

  Next, in step S206, based on the simulation result of the attribute to be processed acquired in step S204, the daily actor rate for the person having the attribute to be processed is calculated.

  Next, in step S208, the daily actor rate in the attribute to be processed acquired based on the daily actor rate data acquired in step S100, and the attribute to be processed acquired in step S206. Compare the daily actor rate of the person you have.

  Next, in step S210, based on the comparison result acquired in step S208, it is determined whether or not the error is within a predetermined range. If the error is within a predetermined range, the process proceeds to step S212. If the error is outside the predetermined range, the process proceeds to step S211.

In step S211, the input unit 10 receives an input for β _i , proceeds to step S202, and repeats the processes in steps S202 to S210.

  In step S212, the action transition probability matrix for each time zone acquired in step S202 is determined as the action transition probability matrix for each time zone of the attribute to be processed.

  Next, in step S214, it is determined whether or not processing has been completed for all attributes. When processing has been completed for all attributes, the processing routine for calculating the action transition probability by time zone is terminated, and when processing has not been completed for all attributes, the attribute to be processed is changed. And it transfers to step S200 and repeats the process of step S200-step S214.

  The calculation of the action-specific destination selection probability matrix in step S110 will be described in detail in the action-specific destination selection probability matrix calculation processing routine shown in FIG.

In step S300 of FIG. 6, the generated amount representing the number of persons who live in each mesh i and perform the act k to be processed for the action k to be processed from the generated amount by action acquired in step S104. Obtain O _i ^(k) .

Next, in step S302, the coefficient A _i ^(k) and the coefficient B _j ^(k) are initialized, and each value is set to 1.

Next, in step S304, for the action k to be processed, mesh j acquired based on the coefficient B _j ^(k) acquired in step S302 or the previous step S306 and the facility distribution data acquired in step S106. Based on the number of facilities D _j ^(k) related to the action k to be processed in, the predetermined distance dist _ij between the mesh i and the mesh j, and the predetermined distance coefficient γ The coefficient A _i ^(k) is calculated according to the equation (3).

Next, in step S306, for the action k to be processed, the coefficient A _i ^(k) acquired in step 304, the generation amount O _i ^(k) acquired in step S300, and a predetermined mesh i Based on the distance dist _{ij from the} mesh j and a predetermined distance coefficient γ, the coefficient B _j ^(k) is calculated according to the above equation (4).

Next, in step S308, it is determined whether or not the coefficient A _i ^(k) acquired in step S304 and the coefficient B _j ^(k) acquired in step S306 have converged for the action k to be processed. If it has converged, the process proceeds to step S310. If it has not converged, the process proceeds to step S304, and the processes in steps S304 to S308 are repeated.

Next, in step S310, for the action k to be processed, the coefficient A _i ^(k) acquired in step S304, the coefficient B _j ^(k) acquired in step S306, and the generation amount O _i acquired in step S300. ^(K) , the number of facilities D _j ^(k) related to the action k to be processed in the mesh j, acquired based on the facility distribution data acquired in step S106, and a predetermined mesh i and mesh Based on the distance dist _{ij from} j and a predetermined distance coefficient γ, X _ij ^(k) is calculated for the action k to be processed using the gravity model shown in the above equation (2). To do.

Next, in step S312, based on X _ij ^(k) calculated using the gravity model for the action k to be processed acquired in step S310 for the action k to be processed, the above equation (5) Accordingly, the destination selection probability P _ij ^(k) of the action k to be processed is calculated for each of the pairs of areas i and j, and the result is the learning result of the destination selection probability matrix of the action k.

  Next, in step S314, it is determined whether or not processing has been completed for all actions. When processing has been completed for all actions, the routine for calculating the destination selection probability by action is terminated, and when processing has not been completed for all actions, the action k to be processed is changed. And it transfers to step S300 and repeats the process of step S300-step S314.

<Operation of Human Dynamics Calculation Device According to First Embodiment>
Next, the operation of the human behavior calculation apparatus 200 according to the first embodiment will be described. First, the population data for each area in the input unit 210, the action transition probability matrix for each time zone learned in step S108 in the probability learning device 100 described above, and the action-specific destination selection probability matrix learned in step S110. , And actor rate data by time zone, which is composed of the probabilities of actors by time zone. Then, the human activity calculation processing routine shown in FIG.

  First, in step S400, the population data by area stored in the population data storage unit by area 222 is read.

  Next, in step S402, the action transition probability matrix according to time zone classified by attribute memorize | stored in the action transition probability matrix memory | storage part 224 classified by time zone is read.

  Next, in step S404, the actor rate data for each time zone stored in the hourly actor rate data storage unit 226 is read.

  Next, in step S406, the action destination selection probability matrix stored in the action destination selection probability matrix storage unit 228 is read.

  Next, in step S408, based on the area-specific population data acquired in step S400, an agent representing the person having the attribute with the area as a residence mesh is displayed as the population of the area and the attribute for each attribute and area. Generate as many as appropriate.

  Next, in step S410, acquired in step S408 based on the attribute-specific action transition probability matrix for each attribute acquired in step S402 and the attribute-specific time zone actor rate data acquired in step S404. An action sequence is generated for each agent.

  Next, in step S412, for each of the agents acquired in step S408, based on the action-specific destination selection probability matrix acquired in step S406 and each of the agent action sequences acquired in step S410. Is generated as city movement data.

  Next, in step S414, the in-city movement data acquired in step S412 is visualized as a result in the output unit 290, and the human activity calculation processing routine is terminated.

  The action sequence generation in step S410 will be described in detail in the action sequence generation processing routine shown in FIG.

In step S500 of FIG. 8, for the agent to be processed, the action k ^{(0) at} time t = 0 based on the attribute of the agent and the actor rate data classified by time period for each attribute acquired in step S404. To decide.

  Next, in step S502, 1 is set to the value of the variable t.

Next, in step S504, for the agent to be processed, the action transition probability matrix for each attribute by time period acquired in step S402 and the action k ^{(0) of the} agent acquired in step S500, or the previous step S504. The agent's action k ^(t) is determined based on the agent's action k ^(t-1) acquired in step ⁽¹⁾ .

  Next, in step S506, t + 1 is set as the value of the variable t.

  Next, in step S508, it is determined whether t = T. If t = T, the process proceeds to step S510. If t = T, the process proceeds to step S504.

  Next, in step S510, it is determined whether or not processing has been completed for all agents generated in step S408. If the processing has been completed for all agents, the action sequence generation processing routine is terminated. If the processing has not been completed for all agents, the agent to be processed is changed, and the process proceeds to step S500. Then, the processing from step S500 to step S510 is repeated.

  The generation of the destination series in step S412 will be described in detail in the destination generation processing routine shown in FIG.

  In step S600 of FIG. 9, 0 is set as the value of the variable t.

Next, in step S602, for the agent to be processed, the value of the resident mesh of the agent to be processed is set as the value of l _rsd based on the agent's action sequence acquired in step S410, and l _off and l A value of -1 is set to the value of _sch .

Next, in step S604, for the agent to be processed, based on the action sequence acquired in step S410, it is determined whether or not the agent action k ^(t) is at home. If the agent action k ^(t) is at home, the process proceeds to step S606. If the agent action k ^(t) is not at home, the process proceeds to step S612.

Next, in step S606, the destination l ^(t) of the agent to be processed is a mesh represented by the value of l _rsd acquired in step S602.

  Next, in step S608, the value of the variable t is set to t + 1.

  Next, in step S610, it is determined whether the value of the variable t is equal to the value of T. If the value of the variable t is equal to the value of T, the process proceeds to step S650. If the value of the variable t is not equal to the value of T, the process proceeds to step S604, and the processes of steps S604 to S610 are repeated. .

  In step S612, it is determined whether the value of the variable t is not zero. If the value of the variable t is not 0, the process proceeds to step S614. If the value of the variable t is 0, the process proceeds to step S630.

Next, in step S614, it is determined whether or not the agent action k ^(t) to be processed is work. If the agent action k ^(t) is work, the process proceeds to step S616. If the agent action k ^(t) is not work, the process proceeds to step S620.

Next, in step S616, it is determined whether or not the value of the variable l _off is -1. If the value of the variable _{l off} is not -1, the process proceeds to step S618, if the value of the variable _{l off} is -1, the process proceeds to step S620.

Next, in step S618, the destination l ^(t) of the agent to be processed is set as a mesh represented by the value of l _off acquired in step S636.

In step S620, it is determined whether or not the agent action k ^(t) to be processed is academic. If the agent action k ^(t) is academic, the process proceeds to step S622. If the agent action k ^(t) is not academic, the process proceeds to step S626.

Next, in step S622, it is determined whether or not the value of the variable _lsch is -1. If the value of the variable l _sch is not -1, the process proceeds to step S624. If the value of the variable l _sch is -1, the process proceeds to step S626.

Next, in step S624, the destination l ^(t) of the agent to be processed is a mesh represented by the value of l _sch acquired in step S640.

In step S626, it is determined whether or not the action k ^{(t) of the} agent to be processed is the same as the action k ^{(t-1) of the} agent one hour before. When the action ^(t) and the action ^(t-1) are the same, the process proceeds to step S628, and when the action ^(t) and the action ^(t-1) are not the same, the process proceeds to step S632. .

In step S628, the destination ^l of agents to be processed to ^(t), 1 time step S606 in the previous processing, step S618, step S624, step S628, step S632, or the destination ^{l (t} acquired in step S630 ^The mesh represented by the value of ^-1) .

In step S632, it is the l ^(t−1) -th row vector of the action-specific destination selection probability matrix acquired in step S406.

Accordingly, the destination l ^(t) of the agent to be processed is determined using a random number.

Next, in step S634, for the agent to be processed, it is determined whether or not the agent's action k ^(t) acquired based on the agent's action sequence acquired in step S410 is work. If the agent action k ^(t) is work, the process proceeds to step S636. If the agent action k ^(t) is not work, the process proceeds to step S638.

Next, in step S636, a value representing the mesh that is the destination determined in step S632 or step S630 is set as the value of the variable l _off .

In step S638, for the agent to be processed, it is determined whether or not the agent's action k ^(t) acquired based on the action-specific generation amount acquired in step S410 is academic. If the agent action k ^(t) is academic, the process proceeds to step S640. If the agent action k ^(t) is not academic, the process proceeds to step S608.

Next, in step S640, it sets the value indicating the mesh is determined destination at a value in step S632 or step S630 of variables _{l sch.}

In step S630, it is obtained from l _rsd acquired in step S602 and the action-specific destination selection probability matrix acquired in step S406.

Then, the destination l ^(t) of the agent to be processed is determined using a random number.

  In step S650, it is determined whether or not processing has been completed for all agents to be processed. If the processing is completed for all agents, the processing of the target sequence generation processing routine is terminated. If the processing is not completed for all agents, the agent to be processed is changed, and steps S600 to S600 are performed. The process up to S650 is repeated.

  As described above, according to the human dynamics calculation system according to the first embodiment of the present invention, agents are generated based on statistical data representing population data by attribute of each area, and learning is performed in advance for each attribute. For each agent, a sequence of actions consisting of actions for each time period is generated for each agent, and for each agent, a destination selection probability learned in advance for each of a plurality of actions and for each agent are generated. Based on the statistical data by generating a destination sequence consisting of areas that are destinations in each time zone based on the generated action sequence and outputting the generated action sequence and destination sequence for each of the agents. Thus, it is possible to accurately calculate a person's action series and destination series.

  Moreover, since the attributes of all people in the city and the purpose of movement can be estimated from the aggregated data, it is possible to obtain real-time and sufficient sample data at a very low cost. The cost of work resulting from the activity is greatly reduced.

  In addition, although the statistical data is considered to have high accuracy as the aggregate value, although the characteristics that represent the individual are lost, the agent follows a probability model that reproduces statistical data composed of movement action sequence estimation and destination selection Can be estimated from information (attributes, purpose of movement) given to the agent.

  Further, when the number of agents is equal to the population in the city, the movements of all the people in the city can be reproduced on the computer at a low cost and in a short time.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

For example, in the first embodiment, D _j ^(k) is described as the number of facilities related to the action k in the mesh j. However, the present invention is not limited to this, and the mesh concentration is not limited thereto. A value other than the number of facilities may be used as long as the value is proportional to the amount (the number of trips arriving in a certain area). Specifically, if act = work, the number of employees may be used, and if act = study, the number of students may be used. Further, the facility distribution data may represent a physical quantity of the facility other than the number of facilities as long as the amount is proportional to the number of customers.

  Next, a human behavior calculation system according to the second embodiment will be described.

  In the second embodiment, the first embodiment is that the destination selection probability matrix for each action is learned from the generation amount for each action, the facility distribution data, and the population data for each area for each time zone. And different. In addition, about the structure and effect | action similar to the person dynamics calculation system which concerns on 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

<Configuration of Person Dynamics Calculation System According to Second Embodiment>
First, the configuration of the human behavior calculation system according to the second embodiment of the present invention will be described. The human activity calculation system according to the second embodiment of the present invention includes a probability learning device 300 and a human activity calculation device 200 described later.

<Configuration of Probability Learning Device According to Second Embodiment>
Next, the configuration of the probability learning device 300 according to the second embodiment of the present invention will be described. As shown in FIG. 10, a probability learning apparatus 300 according to an embodiment of the present invention includes a CPU, a RAM, and a ROM that stores a program and various data for executing a probability learning processing routine described later. Can be configured with a computer. Functionally, the probability learning apparatus 300 includes an input unit 310, a calculation unit 320, and an output unit 390 as shown in FIG.

The input unit 310 is a daily actor rate data indicating the percentage of persons who acted by attribute per day, a parameter β _i, and an actor by time zone representing the percentage of persons who acted by attribute and by time zone Rate data, the amount generated by each action obtained from the action sequence generated in the previous process in the personal dynamics calculation apparatus 200 to be described later, the facility distribution data representing the distribution of the number of facilities, and the population by area by time zone And data.

  The calculation unit 320 includes a daily actor rate data storage unit 24, an hourly actor rate data storage unit 26, an act-based generation amount storage unit 28, a facility distribution data storage unit 30, and a time zone-specific area. Population data storage unit 322, action transition probability matrix learning unit by time zone 40, action transition probability matrix storage unit by time zone, destination selection probability matrix learning unit by action 344, destination selection probability matrix by action And a storage unit 346.

  The time-specific area-specific population data storage unit 322 stores the time-specific area-specific population data received by the input unit 310.

  The action-specific destination selection probability matrix learning unit 344 includes an action-specific generation amount stored in the action-specific generation amount storage unit 28, facility distribution data stored in the facility distribution data storage unit 30, and a time zone area. A destination selection probability matrix is learned for each action based on the time-specific area-specific population data stored in the separate population data storage unit 322, and the learned result is used as an action-specific destination selection probability matrix. The data is stored in the separate destination selection probability matrix storage unit 346 and output to the output unit 390.

To learn the action-specific destination selection probability matrix, first, the initial value, increment, and end value of the distance coefficient γ accepted by the input unit 310 are set. Next, for each action k, the distance coefficient γ is applied, and, similarly to the action-specific destination selection probability matrix learning unit 44 in the probability learning device according to the first embodiment, X _ij ^{( k)} is calculated. Next, similar to the action-specific destination selection probability matrix learning unit 44 in the probability learning device according to the first embodiment, using X _ij ^(k) calculated for each action k using a gravity model. In addition, the destination selection probability P _ij ^(k) is calculated for each of the pairs of areas i and j to obtain a destination selection probability matrix for each action. Next, for each action, based on the calculated destination selection probability matrix, a simulation similar to the person dynamics calculation device 200 is performed, and based on the simulation results, the mesh population by time zone is totaled, The fitness with the population data according to different area is evaluated using the fitness evaluation formula shown in the following formula (6). The process is applied while increasing the value of γ up to a predetermined end value of γ, and processing of gravity model calculation, destination selection probability matrix calculation, and fitness calculation are performed. Run repeatedly.

Here, n _l ^(t) is the population of area l and time zone t (simulated value), and the value of ^ n _l ^(t) is the population of area l and time zone t (population by area by time zone) Actual data based on the data).

  Then, for each action, the destination selection probability matrix calculated using γ having the highest fitness obtained by the above equation (6) is determined as the action-specific destination selection probability matrix for the action.

  The action destination selection probability matrix storage unit 346 stores the action destination selection probability matrix learned by the action destination selection probability matrix learning unit 344.

<Operation of the probability learning device according to the second embodiment>
Next, the operation of the probability learning device 300 according to the first embodiment will be described. First, the input unit 310 accepts the daily actor rate data, the parameter β _i , the hourly actor rate data, the behavior-specific generation amount, the facility distribution data, and the time zone-specific population data. Then, the probability learning processing routine shown in FIG. 11 is executed by the probability learning device 300.

  In step S700, the time-specific area-specific population data stored in the time-specific area-specific population data storage unit 322 is read.

  In step S702, the destination selection probability matrix for each action based on the action-generated amount acquired in step S104, the facility distribution data acquired in step S106, and the time-specific area-specific population data acquired in step S700. Is stored in the action-specific destination selection probability matrix storage unit 346 as an action-specific destination selection probability matrix.

  The calculation of the action-specific destination selection probability matrix in step S702 will be described in detail in the action-specific destination selection probability matrix calculation processing routine shown in FIG.

  In step S800 of FIG. 12, the initial value, increment, and end value of the distance coefficient γ, to which a constant value is given in advance, are set.

  Next, in step S802, the value of the distance coefficient γ, which is given a constant value in advance, is set to the initial value set in step S800.

  In step S804, a simulation is performed using the destination selection probability matrix acquired in step S312 for the action k to be processed.

  Next, in step S806, for the action k to be processed, the mesh population by time zone is acquired based on the simulation result acquired in step S804.

  Next, in step S808, for the action k to be processed, based on the time-specific area population data acquired in step S700 and the time-based mesh population acquired in step S806, according to the above equation (6). The fitness is calculated and stored in a memory (not shown) in association with the distance coefficient γ.

  Next, in step S810, it is determined whether or not the value of the distance coefficient γ matches the end value of the distance coefficient γ set in step S800. If the value of the distance coefficient γ is the end value, the process proceeds to step S813. If the value of the distance coefficient γ is not the end value, the process proceeds to step S812.

  In step S812, the value of the distance coefficient γ is incremented and added in step S800, the process proceeds to step S300, and the processes in steps S300 to S810 are repeated.

  In step S813, the distance coefficient γ when the fitness obtained in step S808 is the highest is selected, and the destination selection probability matrix acquired in step S312 by applying the selected distance coefficient γ is set as the processing target. It is determined as a destination selection probability matrix of the act k.

  As described above, according to the human dynamics calculation system according to the second embodiment of the present invention, agents are generated based on statistical data representing population data by attribute of each area, and learning is performed in advance for each attribute. For each agent, a sequence of actions consisting of actions for each time period is generated for each agent, and for each agent, a destination selection probability learned in advance for each of a plurality of actions and for each agent are generated. Based on the statistical data by generating a destination sequence consisting of areas that are destinations in each time zone based on the generated action sequence and outputting the generated action sequence and destination sequence for each of the agents. Thus, it is possible to accurately calculate a person's action series and destination series.

  Next, a human behavior calculation apparatus according to a third embodiment will be described.

  The third embodiment is different from the first embodiment in that the person dynamics calculation apparatus learns the action transition probability matrix for each time zone and the destination selection probability matrix for each action. In addition, about the structure and effect | action similar to the person dynamics calculation system which concerns on 1st Embodiment, the same code | symbol is attached | subjected and description is abbreviate | omitted.

<Configuration of Person Dynamic Calculation Device According to Third Embodiment>
First, the configuration of the human behavior calculation device 400 according to the third embodiment of the present invention will be described. A human activity calculation apparatus 400 according to an embodiment of the present invention is configured by a computer including a CPU, a RAM, and a ROM that stores a program for executing a personal activity calculation processing routine described later and various data. I can do it. Functionally, the person dynamics calculation apparatus 400 includes an input unit 410, a calculation unit 420, and an output unit 290 as shown in FIG.

The input unit 410 includes one-day actor rate data indicating the percentage of persons who acted on a daily basis by attribute, parameter β _i , and hourly actors representing the percentage of persons who acted by attribute and time period It accepts rate data, population data by area that is statistical data representing the population by attribute for each area, and facility distribution data that represents the distribution of the number of facilities.

  The calculation unit 420 includes an area-specific population data storage unit 222, a daily actor rate data storage unit 24, a time zone actor rate data storage unit 26, an action-specific generation amount storage unit 428, and a facility distribution data storage. Unit 30, action transition probability matrix learning unit by time zone 40, action transition probability matrix storage unit by time zone, destination selection probability matrix learning unit by action 444, and destination selection probability matrix storage unit by action 46 And an agent generation unit 430 and an in-city movement data storage unit 260.

  The agent generation unit 430 includes the area-specific population data stored in the area-specific population data storage unit 222, the attribute-specific time zone action transition probability matrix stored in the time zone-specific action transition probability matrix storage unit 42, Based on the hourly actor rate data stored in the hourly actor rate data storage unit 26 and the action destination selection probability matrix stored in the action destination selection probability matrix storage unit 46. The action sequence and the destination sequence for each of the agents are generated, stored in the in-city movement data storage unit 260, and output to the output unit 290. The agent generation unit 430 includes an attribute generation unit 232, an action sequence generation unit 434, and a destination sequence generation unit 236.

  The action series generation unit 434 includes the agents created in the attribute generation unit 232, the action transition probability matrix for each time zone stored in the action transition probability matrix storage unit for each time zone, and the time zone Based on the actor rate data for each time zone stored in the actor rate data storage unit 26, an action sequence as shown in columns D1 to DT of FIG. 3 is generated for each agent, and for each agent, Data that combines the ID, the residence mesh, the attribute, and the action series is created and output to the destination series generation unit 236 and stored in the action-specific generation amount storage unit 428.

  The action-based generation amount storage unit 428 stores the action-specific generation amount obtained from the data input from the action series generation unit 434. Here, the generated amount by action is obtained from the action series for each of the agents generated in the action series generation unit 424, and for each area, among the agents whose residence mesh is the area, This represents the number of agents that perform the action.

  The action-specific destination selection probability matrix learning unit 444 performs an action based on the action-specific generation amount stored in the action-specific generation amount storage unit 428 and the facility distribution data stored in the facility distribution data storage unit 30. Each destination selection probability matrix is learned, and the learned result is stored in the action-specific destination selection probability matrix storage unit 46 as an action-specific destination selection probability matrix. The action-specific destination selection probability matrix represents a destination selection probability for selecting one area of each pair of areas as the destination from the other area.

<Operation of Human Dynamics Calculation Device According to Third Embodiment>
Next, the operation of the human activity calculation device 400 according to the third embodiment will be described. First, the input unit 410 receives area-specific population data, daily actor rate data, parameter β _i , time zone actor rate data, and facility distribution data.

  Then, in the human behavior calculation device 400 according to the third embodiment, a processing routine similar to the calculation processing routine for the hourly action transition probability matrix shown in FIG. 5 is executed, and the action transition probability matrix for each time zone is learned. Is done.

  Further, in the human behavior calculation apparatus 400 according to the third embodiment, a processing routine similar to the calculation processing routine of the action-specific destination selection probability matrix shown in FIG. 6 is executed, and the action-specific destination selection probability matrix is To be learned. It should be noted that the action-specific generation amount stored in the action-specific generation amount storage unit 428 is used as the action-specific generation amount used in the learning.

  Further, in the human activity calculation apparatus 400 according to the third embodiment, a processing routine similar to the human activity calculation processing routine shown in FIG. 7 is executed. In addition, the action transition probability matrix classified by time zone memorize | stored in the action transition probability matrix memory | storage part 42 classified by time slot | zone is used for the action transition probability matrix classified by time slot | zone used in the said processing routine. Further, the action-specific destination selection probability matrix used in the processing uses the action-specific destination selection probability matrix stored in the action-specific destination selection probability matrix storage unit 46.

  As described above, according to the human activity calculation device according to the third embodiment of the present invention, agents are generated based on statistical data representing population data for each area attribute, and learning is performed in advance for each attribute. For each agent, a sequence of actions consisting of actions for each time period is generated for each agent, and for each agent, a destination selection probability learned in advance for each of a plurality of actions and for each agent are generated. Based on the statistical data by generating a destination sequence consisting of areas that are destinations in each time zone based on the generated action sequence and outputting the generated action sequence and destination sequence for each of the agents. Thus, it is possible to accurately calculate a person's action series and destination series.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium or provided via a network. Is also possible.

DESCRIPTION OF SYMBOLS 10 Input part 20 Computation part 24 1 day actor rate data storage part 26 Act part rate data storage part 28 according to time zone Occurrence amount storage part 30 Facility distribution data memory part 40 Time zone action transition probability matrix learning part 42 hours Band-specific action transition probability matrix storage unit 44 Action-specific destination selection probability matrix learning unit 46 Action-specific destination selection probability matrix storage unit 90 Output unit 100 Probability learning device 200 Human dynamics calculation device 210 Input unit 220 Calculation unit 222 Population by area Data storage unit 224 Action transition probability matrix storage unit by time zone 226 Actor rate data storage unit by time zone 228 Destination selection probability matrix storage unit by action 230 Agent generation unit 232 Attribute generation unit 234 Action sequence generation unit 236 Destination sequence Generation unit 260 Urban movement data storage unit 290 Output unit 300 Probability learning device 310 Input unit 320 Calculation unit 322 Time Different area population data storage unit 344 Action-specific destination selection probability matrix learning unit 346 Action-specific destination selection probability matrix storage unit 390 Output unit 400 Human dynamics calculation device 410 Input unit 420 Operation unit 424 Action sequence generation unit 428 Generation by action Quantity storage unit 430 Agent generation unit 434 Action sequence generation unit 444 Action-specific destination selection probability matrix learning unit

Claims (8)

Based on statistical data representing population data for each area attribute, for each area, an agent representing the person who has the attribute for each attribute and who resides in the area is represented by the attribute of the area. Agent generation means for generating the number corresponding to the population data of
Based on the transition probability from one action of the pair to the other action for each of a plurality of action pairs learned in advance for each attribute, for each of the agents generated by the agent generating means, An action sequence generating means for generating an action sequence of actions;
For each of the agents generated by the agent generating means, learning in advance for each of the plurality of actions, and selecting a destination from one area of the pair for each of the pair of areas as a destination Based on the probability and the action sequence generated for the agent by the action sequence generating means, a destination sequence generation means for generating a destination sequence consisting of areas that are destinations of each time zone;
For each of the agents generated by the agent generation means, an output for outputting the action sequence generated for the agent by the action sequence generation means and the destination sequence generated for the agent by the destination sequence generation means Means,
Human dynamics calculation device including   Prepared for each combination of the attribute, the action, and the time zone, among the persons having the attribute, the actor rate data by time zone representing the ratio of the person who performed the act in the time zone, Based on the daily actor rate data representing the ratio of the person who performed the act in a day among the persons having the attribute prepared in advance for each combination of the attribute and the action, for each attribute, The person dynamics calculation device according to claim 1, further comprising transition probability learning means for learning a transition probability.   Combination of the area and the action based on the facility distribution data representing the physical quantity of the facility related to each action in each area prepared in advance and the action series generated for each of the agents by the action series generating means For each of the plurality of acts, the person who performed the act moves from one of the pair of areas to the other, based on the number of people residing in the area, The person dynamics calculation device according to claim 1 or 2, further comprising destination selection probability learning means for learning the destination selection probability using a model obtained by modeling the number of people. Combination of the area and the action based on the facility distribution data representing the physical quantity of the facility related to each action in each area prepared in advance and the action series generated for each of the agents by the action series generating means For each of the plurality of acts, the person who performed the act moves from one of the pair of areas to the other, based on the number of people residing in the area, Using each model of the number of people, learning each of the destination selection probabilities when applying each value of the model parameters,
Of the destination selection probabilities when each value of the model parameter is applied, it is matched with the population data by area by time zone representing the population for each combination of the time zone and the area prepared in advance. The person dynamics calculation device according to claim 1, further comprising destination selection probability learning means for selecting the destination selection probability. A human behavior calculation device according to claim 1;
Prepared for each combination of the attribute, the action, and the time zone, among the persons having the attribute, the actor rate data by time zone representing the ratio of the person who performed the act in the time zone, Based on the daily actor rate data representing the ratio of the person who performed the act in a day among the persons having the attribute prepared in advance for each combination of the attribute and the action, for each attribute, A probability learning device including a transition probability learning means for learning a transition probability;
Including human dynamics calculation system. A human behavior calculation device according to claim 1;
Combination of the area and the action based on the facility distribution data representing the physical quantity of the facility related to each action in each area prepared in advance and the action series generated for each of the agents by the action series generating means For each of the plurality of acts, the person who performed the act moves from one of the pair of areas to the other, based on the number of people residing in the area, A probability learning device including a destination selection probability learning means for learning the destination selection probability using a model obtained by modeling the number of people;
Including human dynamics calculation system. A human behavior calculation device according to claim 1;
Combination of the area and the action based on the facility distribution data representing the physical quantity of the facility related to each action in each area prepared in advance and the action series generated for each of the agents by the action series generating means For each of the plurality of acts, the person who performed the act moves from one of the pair of areas to the other, based on the number of people residing in the area, Using each model of the number of people, learning each of the destination selection probabilities when applying each value of the model parameters,
Of the destination selection probabilities when each value of the model parameter is applied, it is matched with the population data by area by time zone representing the population for each combination of the time zone and the area prepared in advance. A probability learning device including destination selection probability learning means for selecting the destination selection probability;
Including human dynamics calculation system. Computer
Based on statistical data representing population data for each area attribute, for each area, an agent representing the person who has the attribute for each attribute and who resides in the area is represented by the attribute of the area. Agent generation means to generate the number according to the population data of
Based on the transition probability from one action of the pair to the other action for each of a plurality of action pairs learned in advance for each attribute, for each of the agents generated by the agent generating means, Action sequence generation means for generating an action sequence of actions;
For each of the agents generated by the agent generating means, learning in advance for each of the plurality of actions, and selecting a destination from one area of the pair for each of the pair of areas as a destination Destination sequence generation means for generating a destination sequence consisting of areas serving as destinations for each time zone based on the probability and the action sequence generated for the agent by the action sequence generation means; and the agent generation Output means for outputting the action sequence generated for the agent by the action sequence generation means and the destination sequence generated for the agent by the destination sequence generation means for each of the agents generated by the means;
Program to function as.