JP6807822B2 - Human flow predictors, methods, and programs - Google Patents

Description

The present invention relates to a human flow rate predictor, method, and program, and more particularly to a human flow rate predictor, method, and program that predicts the flow rate of a person moving between observation points based on the history of human position information. ..

Human location information obtained using sensors, Wi-Fi (registered trademark), Bluetooth (registered trademark), cameras, lasers, etc. is provided as aggregated data so that individuals cannot be tracked from the viewpoint of privacy. Sometimes. Aggregate data refers to data that aggregates samples at a certain particle size with respect to time and space.

Specifically, at a certain observation time t, the number of people who flowed out from a certain observation point i (outflow amount) and the number of people who flowed into a certain observation point i (inflow amount) are given. As a conventional technique, the number of people moving between observation points is estimated based on aggregate data by expressing the connection between observation points where humans move in a graph and expressing the flow of people on the graph by a stochastic model. A method for predicting future human flow (Collective Flow Diffusion Model) has been proposed (see, for example, Non-Patent Document 1).

Further, in such a conventional technique, "the total number of people who flowed out from an arbitrary observation point at a certain observation time t is equal to the total number of people who flowed into an arbitrary observation point at the next observation time (t + 1)". An algorithm that can learn a stochastic model while estimating the number of people moving between observation points has been proposed by solving an optimization problem that takes into account the constraint of "preserving the number of people moving" (for example, Non-Patent Document 2). reference).

By using the trained probability model, it is possible to predict the inflow to each observation point at the future time t'by giving the outflow amount from each observation point at the observation time (t'-1). is there.

Kumar, D. Sheldon, and B. Srivastava, “Collective Diffusion over Networks: Models and Inference”, In Proceedings of International Conference on Uncertainty in Artificial Intelligence (UAI), pp. 351-360, 2013. Sheldon, T. Sun, A. Kumar, and T. G. Dietterich, “Approximate Inference in Collective Graphical Models”, In Proceedings of International Conference on Machine Learning (ICML), pp. 1004-1012, 2013.

By the way, all of the above-mentioned conventional techniques are techniques for analyzing the flow of people observed and predicting the flow rate of people based on the observation data. Therefore, it is difficult to analyze the flow of people and predict the flow rate of people for each cluster classified according to the nature of the user (for example, preference for the place to visit, means of transportation, etc.).

The present invention has been made in view of the above circumstances, and estimates a different flow of people for each cluster classified according to the nature of the user by inputting a set data of human position information and movement history data related to a small number of users. , It is an object of the present invention to provide a human flow forecasting device, a method, and a program capable of accurately predicting a human flow rate for each cluster.

In order to achieve the above object, the human flow forecaster according to the first invention has observed for a plurality of users, for each movement between pairs of observation points, a departure observation point, a departure time, an arrival observation point, and Based on the feature amount obtained from the movement history data associated with the arrival time, the plurality of users are classified into a plurality of types of clusters, and each cluster is classified using the movement history data of each user classified into each cluster. A clustering unit that calculates the ratio of the outflow of people and the ratio of the inflow at each time of each observation point, aggregate data consisting of the outflow of people at each time of each observation point, and each of the observation points. Each of the movements from one to the other for each of the pairs of observation points obtained for each cluster from the aggregate data consisting of the inflow of people at each time in the above, the movement history data, and the classification results of the plurality of users. The outflow for each cluster is based on the observation number data representing the number of people moving at the time, and the ratio of the outflow of people and the ratio of the inflow at each time of each observation point calculated for each cluster by the clustering unit. Amount set data, inflow set data, transition probability of a person moving from one to the other for each of the observation point pairs, and movement from one to the other for each of the observation point pairs. The parameter of the transition probability, the parameter estimation unit for estimating the latent variable, and the target observation point to be predicted so as to optimize the objective function expressed by using the latent variable representing the number of people moving at each time. Other than the target observation points estimated by the parameter estimation unit for the target cluster and the search unit that accepts the input of the predicted time, and the target observation points, the target cluster, and the predicted time that received the input by the search unit. The target observation point from the observation points other than the target observation point for the parameter of the transition probability that moves from the observation point to the target observation point and the time before the predicted time estimated by the parameter estimation unit. Observation points other than the target observation point for the latent variable representing the number of people moving toward, the ratio of the outflow of people from the target observation point in the target cluster, and the time before the predicted time. It is provided with a predictive person flow rate calculation unit that calculates the predicted inflow amount of the target observation point at the predicted time of the target cluster based on the outflow amount of people from.

Further, in the human flow rate predictor according to the second invention, in the first invention, the parameter estimation unit fixes the parameters of the transition probability and calculates the latent variable, and fixes the latent variable. By repeating the step of calculating the parameters of the transition probability, the parameters of the transition probability and the latent variables that optimize the objective function are estimated for each cluster.

Further, in the first or second invention, the person flow rate prediction device according to the third invention is based on the predicted inflow amount of the target observation point at the predicted time by the predicted person flow rate calculation unit for the target cluster. Then, the predicted outflow amount of the target observation point at the predicted time is calculated, and the inflow amount of people at each time of the target observation point, the outflow amount of people at each time, the predicted inflow amount, and the predicted outflow amount. Based on, the transition of the population at the target observation point is calculated.

On the other hand, in order to achieve the above object, in the human flow prediction method according to the fourth invention, the departure observation point and the departure time are obtained for each movement between a pair of observation points observed by the clustering unit for a plurality of users. , The arrival observation point, and the movement history data of each user classified into each cluster by classifying the plurality of users into a plurality of types of clusters based on the feature amount obtained from the movement history data associated with the arrival time. For each cluster, the step of calculating the ratio of the outflow of people and the ratio of the inflow at each time of each observation point, and the parameter estimation unit, the outflow of people at each time of the observation point About each of the set data consisting of the observation points, the set data consisting of the inflow of people at each time at each of the observation points, the movement history data, and the pair of observation points obtained for each cluster from the classification results of the plurality of users. Observed number data representing the number of people moving from one to the other at each time, and the ratio of the outflow amount and the inflow amount of people at each time of each observation point calculated for each cluster by the clustering unit. Based on, for each cluster, the set data of the outflow amount, the set data of the inflow amount, the transition probability of a person moving from one to the other for each of the pair of observation points, and each of the pair of observation points. To optimize the objective function represented by using a latent variable representing the number of people moving from one to the other at each time, the parameters of the transition probability and the step of estimating the latent variable, and the search. The step of accepting the input of the target observation point, the target cluster, and the predicted time to be predicted by the unit, and the target observation point, the target cluster, and the predicted time for which the predictor flow calculation unit receives the input by the search unit. On the other hand, the parameter of the transition probability of moving from an observation point other than the target observation point to the target observation point estimated by the parameter estimation unit, and before the predicted time estimated by the parameter estimation unit. With respect to the time of, the latent variable representing the number of people moving from an observation point other than the target observation point toward the target observation point, the ratio of the outflow of people from the target observation point in the target cluster, and A step of calculating the predicted inflow of the target observation point at the predicted time for the target cluster based on the outflow of people from observation points other than the target observation point at a time before the predicted time. And, including.

Further, in order to achieve the above object, the program according to the fifth invention causes the computer to function as each part included in the human flow rate predictor according to any one of the first to third inventions.

As described above, according to the human flow forecasting device, method, and program according to the present invention, a cluster classified according to the nature of a user by inputting a set data of human position information and a movement history data relating to a small number of users. It is possible to estimate a different flow of people for each cluster and accurately predict the flow of people for each cluster.

It is a block diagram which shows an example of the functional structure of the person flow rate predicting apparatus which concerns on embodiment. It is a schematic diagram which shows an example of the information stored in the inflow amount storage part which concerns on embodiment. It is a schematic diagram which shows an example of the information stored in the outflow amount storage part which concerns on embodiment. It is a schematic diagram which shows an example of the information stored in the movement history storage part which concerns on embodiment. It is a flowchart which shows an example of the flow of the predicted person flow rate calculation process executed by the person flow rate predictor which concerns on embodiment. It is a schematic diagram which shows an example of the output content from the output part when the search part of the person flow rate prediction apparatus which concerns on embodiment performs a search.

Hereinafter, an example of a mode for carrying out the present invention will be described in detail with reference to the drawings.

The human flow rate prediction device according to the present embodiment represents the connection between observation points where people move in a graph, and represents the flow of people on the graph by a probability model. The human flow prediction device according to the present embodiment learns a probabilistic model capable of estimating a different human flow for each cluster classified according to the nature of the user, based on a set data of human position information and movement history data related to a small number of users. By using a trained probabilistic model, the human flow rate for each cluster at a future time is predicted.

In this embodiment, observation data is based on a probability model in which "transition probability between observation points for each cluster" and "latent variable representing the number of people moving between observation points at each time for each cluster" are unknown variables. Estimate the unknown variables of the stochastic model using the optimization algorithm from.

In this embodiment, a set data of various position information is targeted, and can be flexibly applied according to the set data obtained by observation.

The position information referred to here is position information acquired by indoor and outdoor sensors, Bluetooth (registered trademark), a camera, Wi-Fi (registered trademark), and the like. Further, the set data referred to here refers to data obtained by aggregating samples at a predetermined particle size with respect to time and space. Further, here, it is assumed that the target set data cannot be tracked as "where the individual went from where". In a practical situation, in addition to the above-mentioned aggregate data, it is possible that a small number of users can be traced in detail. For example, a small number of users can be tracked using a Wi-Fi (registered trademark) identifier, but the majority of users can only obtain aggregated data with a camera placed at the observation point. And so on.

Hereinafter, as an embodiment, the number of people moving between observation points is estimated for each cluster representing the nature of the user under the condition that a set data of general location information and movement history data related to a small number of users are given. At the same time, we will explain the case of learning a probabilistic model and predicting the flow of people at a future time.

FIG. 1 is a block diagram showing an example of a functional configuration of the human flow rate prediction device 90 according to the present embodiment.
As shown in FIG. 1, the human flow rate prediction device 90 according to the present embodiment includes an inflow amount storage unit 1, an outflow amount storage unit 2, a movement history storage unit 3, an operation unit 4, a search unit 5, and a user clustering unit 6. It includes a human flow rate model learning unit 9, a predicted human flow rate calculation unit 13, and an output unit 14.

Further, the user clustering unit 6 includes a clustering unit 7 and a ratio storage unit 8, and the human flow rate model learning unit 9 includes a parameter estimation unit 10, a transition probability parameter storage unit 11, and a moving number storage unit 12. It is composed of.

Of these, the operation unit 4 and the parameter estimation unit 10 are connected to each of the inflow amount storage unit 1 for people at the observation point, the outflow amount storage unit 2 for people at the observation point, and the movement history storage unit 3 for a small number of users. ing. Further, the clustering unit 7 is connected to the movement history storage unit 3 for a small number of users. Further, the predictor flow rate calculation unit 13 is connected to the outflow amount storage unit 2.

The inflow amount storage unit 1 according to the present embodiment is set data that can be analyzed by the human flow rate prediction device 90, and stores the set data composed of the inflow amount of people at each time at each of the plurality of observation points. .. The inflow amount storage unit 1 reads out the requested aggregate data according to the request from the human flow rate model learning unit 9, and transmits the read aggregate data to the human flow rate model learning unit 9.

For example, when the observation point is i and the observation time is t, the inflow of people at the observation point i is expressed as N _ti ^IN . Further, the data format stored in the inflow amount storage unit 1 is represented as (i, t, N _ti ^IN ). This data format means that the number of N _ti ^IN people flowed in from the observation point i at time t. The inflow amount storage unit 1 may be composed of a Web server, a database server including a database, and the like.

FIG. 2 is a schematic diagram showing an example of information stored in the inflow amount storage unit 1 according to the present embodiment.
As an example, as shown in FIG. 2, the inflow amount storage unit 1 stores the observation point ID for identifying the observation point, the inflow amount (number of people), and the observation time in association with each time step. There is.

The outflow amount storage unit 2 according to the present embodiment is set data that can be analyzed by the human flow rate prediction device 90, and stores the set data consisting of the outflow amount of people at each time at each of the plurality of observation points. .. The outflow amount storage unit 2 reads out the requested aggregate data according to the request from the human flow rate model learning unit 9, and transmits the read aggregate data to the human flow rate model learning unit 9.

For example, when the observation point is i and the observation time is t, the outflow amount of a person at the observation point i is expressed as N _ti ^OUT . Further, the data format stored in the outflow amount storage unit 2 is represented as (i, t, N _ti ^OUT ). This data format means that the number of N _ti ^OUT people leaked from the observation point i at time t. The outflow amount storage unit 2 may be composed of a Web server, a database server including a database, and the like.

FIG. 3 is a schematic diagram showing an example of information stored in the outflow amount storage unit 2 according to the present embodiment.
As an example, as shown in FIG. 3, the outflow amount storage unit 2 stores the observation point ID for identifying the observation point, the outflow amount (number of people), and the observation time in association with each time step. There is.

The movement history storage unit 3 according to the present embodiment stores a set of movement history data relating to a small number of users that can be analyzed by the human flow rate prediction device 90. The movement history storage unit 3 reads the requested aggregate data according to the requests from each of the user clustering unit 6 and the human flow model learning unit 9, and reads the read aggregate data into the user clustering unit 6 and the human flow model learning unit 9. Send to each of.

The data format stored in the movement history storage unit ^{3 (u, i, t out} , j, t in) represents the. This data format means that the user u is starting the observation point i at time t ^out, to arrive at the observation point j in time t ^in. The movement history storage unit 3 may be composed of a Web server, a database server including a database, and the like.

FIG. 4 is a schematic diagram showing an example of information stored in the movement history storage unit 3 according to the present embodiment.
As an example, as shown in FIG. 4, the movement history storage unit 3 is used to identify the user ID, the departure observation point ID for identifying the departure observation point, the departure time, and the arrival observation point for each movement of the user. The set data of the movement history including the arrival observation point ID and the arrival time (hereinafter, also referred to as the movement history data) is stored. That is, this movement history data is data in which a departure observation point, a departure time, an arrival observation point, and an arrival time are associated with each movement between a pair of observation points observed for a plurality of users.

The operation unit 4 according to the present embodiment receives various operations from the user for the set data stored in each of the inflow amount storage unit 1, the outflow amount storage unit 2, and the movement history storage unit 3. The various operations referred to here are an operation of registering a stored set data, an operation of modifying, an operation of deleting, and the like. The operation unit 4 is provided with a keyboard, a mouse, a touch panel, etc. as input means, and is realized by a device driver of these input means, a menu screen control software, or the like.

The search unit 5 according to the present embodiment accepts inputs of the target observation point i, the target cluster k, and the predicted time t ′ for which the human flow rate is predicted. For the target observation point i, the target cluster k, and the predicted time t'that have been input by the search unit 5, the parameters of the transition probability between the observation points and the latent number of people moving between the observation points will be described later. Variables, the ratio of the outflow from the target observation point i in the target cluster k, and the outflow from the observation points other than the target observation point i at the time (t'-1) one time before the predicted time t'. The predicted human flow rate at the target observation point i is output based on. The search unit 5 is provided with a keyboard, a mouse, a touch panel, and the like as input means, and is realized by a device driver of these input means, menu screen control software, and the like.

The human flow forecasting device 90 according to the present embodiment is configured as a computer including a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disk Drive), and the like. The human flow rate prediction program according to this embodiment is stored in the ROM. The human flow rate prediction program may be stored in the HDD.

The above-mentioned human flow rate prediction program may be pre-installed in, for example, the human flow rate prediction device 90. This human flow rate prediction program may be realized by storing it in a non-volatile storage medium or distributing it via a network and appropriately installing it in the human flow rate prediction device 90. Examples of the non-volatile storage medium include a CD-ROM (Compact Disc Read Only Memory), a magneto-optical disk, a DVD-ROM (Digital Versatile Disc Read Only Memory), a flash memory, and a memory card.

The CPU functions as the clustering unit 7, the parameter estimation unit 10, the search unit 5, the predicted person flow rate calculation unit 13, and the output unit 14 by reading and executing the person flow rate prediction program stored in the ROM. ..

The clustering unit 7 according to the present embodiment acquires a set data of the movement history of a small number of users stored in the movement history storage unit 3. The clustering unit 7 classifies a small number of users into a plurality of types of clusters based on the feature amount obtained from the movement history data relating to the small number of users. The clustering unit 7 uses the movement history data of each user classified into each cluster, and for each cluster, the ratio of the outflow amount of people and the inflow amount ratio at each time of each observation point (hereinafter, simply, “outflow amount”). The ratio of the outflow amount and the inflow amount ratio are calculated respectively, and the calculated outflow amount ratio and inflow amount ratio for each cluster are stored in the ratio storage unit 8.

In the present embodiment, the parameter estimation unit 10 stores the inflow amount set data stored in the inflow amount storage unit 1, the outflow amount set data stored in the outflow amount storage unit 2, and the movement history storage unit 3. The aggregated data of the movement history, the ratio of the inflow amount and the ratio of the outflow amount for each cluster stored in the ratio storage unit 8 are acquired. The parameter estimation unit 10 moves from one to the other for each pair of observation points for each cluster based on the set data of the movement history acquired from the movement history storage unit 3 and the classification result of a small number of users. Obtain the observation number data representing the number of people moving at each time. The parameter estimation unit 10 sets the set data of the inflow amount and the outflow amount for each cluster based on the set data of the inflow amount, the set data of the outflow amount, the number of observers data, the ratio of the inflow amount, and the ratio of the outflow amount. Estimate the transition probability parameters and latent variables so as to optimize the objective function represented using the set data, transition probabilities, and latent variables of. The transition probability represents the probability that a person moves from one to the other for each pair of observation points, and the latent variable means each time that a person moves from one to the other for each pair of observation points. Represents the number of people moving.

The parameter estimation unit 10 stores the transition probability parameter estimated above in the transition probability parameter storage unit 11. Further, the parameter estimation unit 10 stores the number of moving people estimated as a latent variable above in the moving number storage unit 12.

The predictor flow rate calculation unit 13 according to the present embodiment has a transition probability parameter storage unit 11 for an observation point other than the target observation point with respect to the target observation point, the target cluster, and the predicted time for which input has been received by the search unit 5. The parameter of the transition probability of moving from to the target observation point is acquired, and the number of people moving from the observation point other than the target observation point to the target observation point for the time before the predicted time is calculated from the moving number storage unit 12. The latent variable to be represented is acquired, the ratio of the outflow amount of a person from the target observation point in the target cluster is acquired from the ratio storage unit 8, and the target observation point for the time before the predicted time is obtained from the outflow amount storage unit 2. Obtain the outflow of people from observation points other than. The predictor flow rate calculation unit 13 calculates the predicted inflow of the target observation point at the predicted time for the target cluster based on the transition probability parameter, the latent variable, the ratio of the outflow amount, and the outflow amount acquired above.

The output unit 14 according to the present embodiment outputs the predicted inflow amount calculated by the predictor flow rate calculation unit 13.

In the present embodiment, in addition to incorporating the above-mentioned clustering unit 7, the set data of human position information and the movement history data related to a small number of users are input, and the connection between observation points is expressed as a graph for each cluster. , The flow of people is expressed by a stochastic model on the graph.

Hereinafter, the probabilistic model expressing the flow of people will be specifically described.

<Observation data>
Here, we consider a situation in which a person moves on a complete graph G = (V, E), which is a graph having branches between arbitrary two vertices. Note that V represents a node set and E represents an edge set.

Here, the node i corresponds to the observation point, and the edge E corresponds to the passage connecting the observation points. Let N _ti ^{OUT be} the number of people flowing out of node i ∈ V at time t, and let N _{t + 1, i} ^IN be the number of people flowing into node i ∈ V at time (t + 1). When the number of observation time steps is T, the outflow of people at the observation point is

The inflow of people at the observation point is expressed as

It is expressed as.

<Observed number data>
Also, movement history storage unit 3 stored in and aggregate data ^{(u, i, t out,} j, t in) can be calculated using the number of people moving from observation point i to measurement point j at time t L _Let it be _tij, and collect the number of people moving from observation point i to other observation points j.

It is expressed as. here,

Means that node i is excluded from the node set V. Also,

And.

<Model parameters>
The parameter of the transition probability that people belonging to cluster k move from node i to another node j,

And.

Meet. Collecting the parameters of transition probability between all nodes,

It is expressed as. Here, K represents the number of clusters.

<Latent variable>
The number of people belonging to the cluster k moving from node i to node j at time t is defined as _Mktij, and the number of people moving from node i to another node j is collectively _defined .

It is expressed as. These values shall be unobservable

Is a latent variable.

<Clustering>
First, the clustering section 7, movement history data stored in the movement history storage unit 3 (u, i, t out , j, t in) was used to analyze the movement tendency of the user, performs clustering of users ..

First, let U be a small number of user sets, and set a feature vector representing the movement tendency of the users u included in the user set U.

And. Here, D represents the number of dimensions of the feature vector. Any feature quantity x _ud representing the feature vector may be used. For example, the time the user u stayed at the observation point d, the number of times the user u stayed, and the like are applied to the feature amount x _ud . The user cluster Uk is _obtained by applying the clustering method using this feature vector. The user cluster U _k represents a set of users U assigned to the cluster k. Here, the clustering method is not particularly limited, but for example, the k-means method or the like is applied.

Then, the clustering unit 7, to the user set _{U k} included in each cluster, using the movement history data of each user ^{(u, i, t out,} j, t in) the starting number _R ^{KTI out} and Calculate the number of arrivals R _{k, t + 1, i} ⁱⁿ . Here, the number of departures R _kti ^out represents the number of people who departed the node i at time t by the user belonging to the cluster k, and the number of arrivals R _{k, t + 1, i} ⁱⁿ indicates the time of the user belonging to the cluster k. (T + 1) represents the number of people arriving at node i. Using these departure number R _kti ^out and arrival number R _{k, t + 1, i} ⁱⁿ , the ratio of outflow from node i in cluster k

And the ratio of the inflow to node i in cluster k

Is calculated by the following equations (1) and (2).

(1)

(2)

Put these together

Is stored in the ratio storage unit 8.

<Model>
First, the parameter estimation unit 10 divides the observed inflow amount N ⁱⁿ and outflow amount N ^out for each cluster by using the inflow amount ratio and the outflow amount ratio for each cluster stored in the ratio storage unit 8. To do. Specifically, the ratio of the outflow amount calculated from the above movement history data.

And the ratio of inflow

Is used to perform division according to the following equations (3) and (4).

(3)

(4)

A set of inflows of cluster k,

A set of outflows of cluster k,

And.

Then, the parameter estimation unit 10, the movement history data stored in the movement history storage unit ^{3 (u, i, t out} , j, t in) and a starting number _R ^{KTI out} obtained by the clustering section 7 _{Is used} to calculate the number of people _Lktij who have moved from place i to place j in time step t for a small number of users belonging to cluster k. Collecting the number of people moving in cluster k,

And.

In the following, independently for each cluster k,

The method of estimating is described. _{Given the} transition probability parameter θ _ki , the set of outflows N _{kt i} ^out, and the number of departures R _{kt i} ^out , M _{kt i} + L _{kt i} is a multinomial distribution.

(5)

It is assumed that it is generated from. At this time, it is _assumed that the latent variable _Mktij representing the number of people to be moved satisfies the following two relational expressions.

(6)

(7)

Here, the equation (6) means that the sum of the number of people moving from the observation point i, which belongs to the cluster k and departs from the observation point i, is equal to the actually observed outflow amount. Further, the equation (7) means that the sum of the number of people who belong to the cluster k and have moved to the observation point i at the time (t + 1) is equal to the actually observed inflow amount.

The parameter estimation unit 10 estimates the parameter Θ of the transition probability between the observation points and the latent variable M representing the number of people moving between the observation points, assuming that the observation data is generated according to the above probability model. These unknown variables can be estimated by various optimization methods (for example, gradient-based optimization method, Markov chain Monte Carlo sampling method, etc.), but here, as an example, Non-Patent Document 3 shown below is used. With reference to this, a method for estimating parameters based on the L-BFGS-B method, which is one of the gradient-based optimization methods, will be described.

Non-patent document 3. R. H. Byrd, P. Lu, J. Nocedal, and C. Zhu, “A Limited Memory Algorithm for Bound Constrained Optimization”, SIAM Journal on Scientific Computing, vol. 16, pp. 1190-1208, 1995.

<Parameter estimation method based on L-BFGS-B method>
From the above equation (5), the log-likelihood function is

(8)

It is expressed as. Here, Stalin approximation

Was used.

Latent variable _Mk and transition probability parameters that maximize the above equation (8) while taking into account the constraints of the above equations (6) and (7).

Consider estimating. In reality, it is difficult to strictly satisfy the above equations (6) and (7) due to observation noise. Therefore, consider maximizing the objective function with soft constraints. The objective function,

(9)

And. Here, σ is a hyperparameter that determines the degree to which constraints are added, and can be determined by using a cross-validation method or the like. A locally optimal solution can be obtained by alternately maximizing the above equation (9) with respect to the latent variable M _k and the parameter θ _k of the transition probability. That is, the parameter estimation unit 10 may be repeated calculating the latent variable M _k by fixing the parameter theta _k of the transition probabilities, and calculating a parameter theta _k of the transition probabilities to secure the latent variable M _k Therefore, the parameters of the transition probability and the latent variables that optimize the objective function represented by the above equation (9) are estimated for each cluster. First, in the optimization of the latent variable M _k , the following optimization problem is solved after fixing the parameter θ _{k of the} transition probability.

(10)

This optimization problem can be solved by the L-BFGS-B method (Non-Patent Document 3), which can consider the domain of the latent variable _Mktij as a constraint.

Next, when the latent variable M _k is fixed, the maximum likelihood estimation value of the transition probability parameter θ _k can be obtained in a closed form.

(11)

Is described.

By alternately repeating the above equations (10) and (11) until the values of the above equation (9), which is the objective function, converge.

To get. Furthermore, by performing the above processing independently for each cluster k,

Can be obtained. And latent variables

Is stored in the moving number storage unit 12, and is a parameter of the transition probability.

Are stored in the transition probability parameter storage unit 11.

The transition probability parameter storage unit 11 is not particularly limited as long as it stores the parameters of the transition probability and can be restored. For example, the transition probability parameter storage unit 11 itself may be a database, and the transition probability parameter storage unit 11 may be a specific area of a general-purpose storage device (memory, hard disk device, etc.) provided in advance in the human flow rate prediction device 90. There may be.

Similarly, the moving number storage unit 12 is not particularly limited as long as it stores the latent variable and can be restored. For example, the moving number storage unit 12 itself may be a database, and the moving number storage unit 12 is a specific area of a general-purpose storage device (memory, hard disk device, etc.) provided in advance in the person flow rate prediction device 90. May be good.

When the predictor flow rate calculation unit 13 receives the input of the target observation point i, the target cluster k, and the predicted time t'via the search unit 5, it is stored in the transition probability parameter storage unit 11 as described above. Transition probability parameters

And the latent variable stored in the moving number storage unit 12

And the ratio of the outflow amount stored in the ratio storage unit 8.

And the outflow amount N _{t'-1, j} ^out from the observation points other than the observation point i at the time (t'-1) one time step before the predicted time t'stored in the outflow amount storage unit 2. Based on the above, the predicted human flow rate (inflow amount to the observation point i) of the observation point i at the prediction time t'for the target cluster k is calculated.

Next, the operation of the human flow rate predictor 90 according to the present embodiment will be described with reference to FIG. Note that FIG. 5 is a flowchart showing an example of the flow of the predicted human flow rate calculation process executed by the human flow rate predictor 90 according to the present embodiment.

The program of the predicted person flow rate calculation process according to the present embodiment is started at the timing when a predetermined operation is performed by using the operation unit 4.

In step 100 of FIG. 5, the predictor flow rate calculation unit 13 receives the input of the target observation point i, the target cluster k, and the predicted time t'via the search unit 5.

In step 102, the predictor flow rate calculation unit 13 initializes the predicted inflow amount _Ckt'ji to 0.

In step 104, the predictor flow rate calculation unit 13 initializes the temporary variable j to 0.

In step 106, the predictor flow rate calculation unit 13 determines whether or not the observation point j belongs to the node set V _\ i. When it is determined that the observation point j belongs to the node set V _\ i (in the case of affirmative determination), the _{process proceeds} to step 108. _Further , when it is determined that the observation point j does not belong to the node set V _\ i (in the case of a negative determination), a series of predicted person flow rate calculation processes by this program are terminated.

In step 108, the predictor flow rate calculation unit 13 sets the time one hour _before the predicted time t'from the observation point j _∈ V _\ i other than the observation point i with respect to the cluster k specified by the search unit 5. number _{N k} flowing out in _(t'-1), a ^{_t'-1, j out,} be calculated using equation (12) shown below.

(12)

In addition, it should be noted

Indicates the ratio of the outflow amount stored in the ratio storage unit 8, and N _{t'-1, j} ^out is an observation point j _∈ V _\ i other than the observation point i observed at the time (t'-1). Shows the outflow of people from.

In step 110, the predictor flow rate calculation unit 13 sets the time one hour _before the predicted time t'from the observation point j _∈ V _\ i other than the observation point i with respect to the cluster k specified by the search unit 5. Number of people moving to (t'-1) and moving to observation point i

Is calculated using the formula (13) shown below.

(13)

In addition, it should be noted

Shows the parameters of the stored transition probabilities in the transition probability parameter storage unit _{11, N k, t'-1} , j out indicates the outflow number for the cluster k calculated in step 108.

In step 112, the predictor flow rate calculation unit 13 flows out from the observation points j _∈ V _\ i other than the observation point i at a time (t'-1) before the predicted time t', and the observation point is at the predicted time t'. The total C _kt'ji of the inflow amount flowing into i is calculated using the formula (14) shown below.

(14)

In addition, it should be noted

Indicates the number of moving people calculated in step 110, and _{Ckt'ji on} the right side indicates the inflow amount calculated in the previous step 112.

In step 114, the predictor flow rate calculation unit 13 adds 1 to the temporary variable j, shifts to step 106, and repeats the process.

The output unit 14 outputs the predicted inflow amount to the observation point i designated by the search unit 5 based on the total inflow amount _Ckt'ji calculated by the predictor flow rate calculation unit 13 as described above. .. At the same time, the output unit 14 outputs the predicted value of the population of the observation point i at the future time t'by using the predicted inflow amount to the observation point i. At this time, in order to calculate the population, the predicted outflow amount is required in addition to the predicted inflow amount at the future time t'. As a method of calculating the predicted outflow, the ratio of the outflow to the population at the observation point i is obtained based on the outflow from the observation point i at the past time, and the predicted outflow at the future time t'according to this ratio. Use the method of calculating the quantity. In this way, based on the predicted inflow of the observation point i at the future time t', the predicted outflow of the observation point i at the future time t'is calculated, and the inflow of people at each time of the observation point i, The transition of the population at the observation point i is calculated based on the outflow of people at each time, the predicted inflow at the future time t', and the predicted outflow at the future time t'. Further, the output unit 14 is a transition probability parameter stored in the transition probability parameter storage unit 11.

Outputs the result of visualizing a human flow graph showing the ease of flow of people between observation points using. Here, the output is a concept including display on a display, printing on a printer, sound output, transmission to an external device, and the like. Further, the output unit 14 may or may not include an output device such as a display and a speaker. The output unit 14 is realized by the driver software of the output device, the driver software of the output device, the output device, or the like.

Hereinafter, a specific example of the output screen 30 output by the output unit 14 will be described with reference to FIG.

FIG. 6 is a schematic diagram showing an example of the output contents from the output unit 14 when the search unit 5 of the human flow rate prediction device 90 according to the present embodiment searches.

As an example, as shown in FIG. 6, the output screen 30 has a search content display unit 31 for displaying the search content. In the search content display unit 31, for example, data indicating the cluster number, data indicating the observation point ID, and data indicating the predicted time are displayed. Thus, the user, the people belonging to the cluster k, can be known at a future time t ', the predictive value of the population of the observation points i ₁ shown above observation point ID.

Further, the output screen 30 includes a transition display unit 32 for displaying the transition of the population in the observation point i _1. The transition display unit 32 is, for example, the search unit 5 by the accepted cluster number, observation point ID, and in accordance with the predicted time, changes in the population at observation points i ₁ is displayed. As a result, the user is, for the people belonging to the cluster k, can be analyzed for whether the population in the specified observation point i ₁ is what kind of transition.

Further, the output screen 30 has a human flow graph display unit 33 for displaying a human flow graph between observation points. The pedestrian flow graph display unit 33, for example, two-dimensional map in which the flow is shown in humans to observation point i ₁ from each observation point is displayed. Thus, the user, the people belonging to the cluster k, it is possible to know the flow rate from each observation point at specified prediction time t 'to the observation point i _1.

Further, the output screen 30 has an inflow amount display unit 34 for displaying the inflow amount from each observation point. The inflow amount display unit 34 is, for example, table inflow is indicated by the number of the observation points i ₁ from each observation point is displayed. As a result, the user is, for the people belonging to the cluster k, can be analyzed for the number of people flowing from each observation point in the specified prediction time t 'to the observation point i _1.

In the present embodiment, the case where the search unit 5 accepts the input of the future time t'has been described, but the present invention is not limited to this. For example, the search unit 5 may receive the past time to analyze the past human flow rate and output the analysis result of the past human flow rate.

As described above, according to the present embodiment, by using the estimated "transition probability between observation points for each cluster", the flow of people between each observation point can be analyzed for each cluster. For example, cluster 1 (people who like games) tends to flow in the order of exhibition booth A, exhibition booth B, and exhibition booth C, and cluster 2 (people who like sports) has exhibition booth D, exhibition booth B, and exhibition booth. It is possible to extract knowledge that people tend to flow in the order of E.

In addition, by using the estimated "number of people moving between observation points at each time in each cluster", "when", "where to where", and "how many people" moved are analyzed for each cluster. be able to. For example, when congestion occurs at the observation point A at time t, it is possible to extract knowledge for each cluster that the number of inflows from the observation point B is large as the cause.

Also, consider predicting the number of inflows of people to a certain observation point A at a future time t'. By predicting the number of people moving in each cluster, it is possible to know not only the amount of inflow to a certain observation point A, but also what kind of cluster the people are included in. ..

As described above, the human flow rate predictor has been illustrated and described as an embodiment. The embodiment may be in the form of a program for causing the computer to function as each part included in the human flow rate predictor. The embodiment may be in the form of a storage medium that can be read by a computer that stores this program.

In addition, the configuration of the human flow rate predictor described in the above embodiment is an example, and may be changed depending on the situation within a range that does not deviate from the gist.

Further, the processing flow of the program described in the above embodiment is also an example, and even if unnecessary steps are deleted, new steps are added, or the processing order is changed within a range that does not deviate from the purpose. Good.

Further, in the above embodiment, the case where the processing according to the embodiment is realized by the software configuration by using the computer by executing the program has been described, but the present invention is not limited to this. The embodiment may be realized by, for example, a hardware configuration or a combination of a hardware configuration and a software configuration.

1 Inflow amount storage unit 2 Outflow amount storage unit 3 Movement history storage unit 4 Operation unit 5 Search unit 6 User clustering unit 7 Clustering unit 8 Percentage storage unit 9 Person flow rate model learning unit 10 Parameter estimation unit 11 Transition probability parameter storage unit 12 Movement Number of people storage unit 13 Predicted person flow rate calculation unit 14 Output unit 30 Output screen 31 Search content display unit 32 Transition display unit 33 Person flow graph display unit 34 Inflow amount display unit 90 Person flow rate prediction device

Claims (5)

Based on the feature quantity obtained from the movement history data associated with the departure observation point, the departure time, the arrival observation point, and the arrival time for each movement between the observation point pairs observed for a plurality of users. Multiple users are classified into multiple types of clusters, and the movement history data of each user classified into each cluster is used to determine the ratio of human outflow and inflow at each time of each observation point for each cluster. The clustering part to calculate each and
Cluster data from the set data consisting of the outflow of people at each time at each of the observation points, the set data consisting of the inflow of people at each time at each of the observation points, the movement history data, and the classification results of the plurality of users. Observed number data representing the number of people moving from one to the other at each time for each of the pair of observation points obtained for each, and each of the observation points calculated for each cluster by the clustering unit. Based on the ratio of the outflow of people and the ratio of the inflow at the time, for each cluster, one person to the other for each of the set data of the outflow amount, the set data of the inflow amount, and the pair of the observation points. The transition probability so as to optimize the objective function represented by the moving transition probability and the latent variable representing the number of people moving from one to the other for each of the pairs of observation points. And the parameter estimation unit that estimates the latent variable
A search unit that accepts input of the target observation point, target cluster, and predicted time to be predicted,
The transition probability of moving from an observation point other than the target observation point to the target observation point estimated by the parameter estimation unit with respect to the target observation point, the target cluster, and the predicted time for which input is received by the search unit. And the latent variable representing the number of people who have moved from an observation point other than the target observation point to the target observation point at a time before the predicted time estimated by the parameter estimation unit. The target cluster is based on the ratio of the outflow of people from the target observation point in the target cluster and the outflow of people from observation points other than the target observation point at a time before the predicted time. And the predictor flow rate calculation unit that calculates the predicted inflow of the target observation point at the predicted time of
A human flow rate predictor equipped with. The parameter estimation unit performs the objective function by repeating a step of fixing the parameter of the transition probability and calculating the latent variable and a step of fixing the latent variable and calculating the parameter of the transition probability. The person flow rate predictor according to claim 1, wherein the parameters of the transition probability and the latent variables to be optimized are estimated for each cluster. The predictor flow rate calculation unit calculates the predicted outflow amount of the target observation point at the predicted time based on the predicted inflow amount of the target observation point at the predicted time for the target cluster, and calculates the predicted outflow amount of the target observation point at the predicted time. The invention according to claim 1 or 2, wherein the transition of the population at the target observation point is calculated based on the inflow amount of people at each time, the outflow amount of people at each time, the predicted inflow amount, and the predicted outflow amount. Human flow forecaster. For each movement between pairs of observation points observed by the clustering unit for multiple users, the feature quantity obtained from the movement history data associated with the departure observation point, departure time, arrival observation point, and arrival time Based on this, the plurality of users are classified into a plurality of types of clusters, and using the movement history data of each user classified into each cluster, the ratio of the outflow amount and the inflow of people at each observation point at each time for each cluster. Steps to calculate each quantity ratio and
The parameter estimation unit uses a set data consisting of the outflow amount of people at each time at each of the observation points, a set data consisting of an inflow amount of people at each time at each of the observation points, the movement history data, and the plurality of users. The number of observed people data representing the number of people moving from one to the other at each time for each of the pair of observation points obtained for each cluster from the classification result of the above, and calculated for each cluster by the clustering unit. One of the set data of the outflow amount, the set data of the inflow amount, and each of the pair of the observation points for each cluster based on the ratio of the outflow amount of the person and the ratio of the inflow amount at each time of each observation point. To optimize the objective function expressed using the transition probability of a person moving from one to the other and the latent variable representing the number of people moving from one to the other at each time for each of the pairs of observation points. In addition, the step of estimating the transition probability parameter and the latent variable,
The step in which the search unit accepts the input of the target observation point, the target cluster, and the predicted time to be predicted,
The target observation point, the target cluster, and the predicted time received by the predictor flow calculation unit from the target observation point other than the target observation point estimated by the parameter estimation unit. The number of people who have moved from an observation point other than the target observation point to the target observation point for the parameter of the transition probability to move to the point and the time before the predicted time estimated by the parameter estimation unit. The latent variable representing the above, the ratio of the outflow of people from the target observation point in the target cluster, and the outflow of people from observation points other than the target observation point at a time before the predicted time. Based on the step of calculating the predicted inflow amount of the target observation point at the predicted time for the target cluster, and
Person flow rate prediction method including. A program for causing a computer to function as each part included in the human flow rate predictor according to any one of claims 1 to 3.