JP7541470B2 - Prediction device and prediction model learning device - Google Patents

Description

The present invention relates to a prediction device and a prediction model learning device.

Technology for predicting people flow is known in the past.

For example, a technology is known that predicts the flow of people at a target time using external information that is correlated with the flow of people (e.g., Patent Document 1). The external information in the technology disclosed in Patent Document 1 is information that is correlated with the flow of people, such as the expected number of customers in the area by time period (e.g., paragraph [0040] of Patent Document 1, etc.).

JP 2017-227994 A

The flow of people in a certain area may be affected by the surrounding environment, such as the scenery around the area. For this reason, when predicting the flow of people in a certain area, it is necessary to take the surrounding environment of the area into consideration. There are also cases where it is desirable to predict what attributes people will have in that flow.

However, the technology disclosed in the above Patent Document 1 predicts people flow using external information such as the expected number of visitors, and does not take into account the surrounding environment such as the scenery of the area. In addition, the technology disclosed in the above Patent Document 1 does not take into account the attributes of people included in the people flow.

Taking the above facts into consideration, the present invention aims to predict the flow of people with certain attributes by taking into account the surrounding environment.

A first aspect of the present invention is a prediction device that includes an acquisition unit that acquires attribute information representing attributes of a person, time information representing a time to be predicted, and environmental information representing an external environment around the person, and a prediction unit that predicts the location information of the person at a time corresponding to the time information by inputting the attribute information, the time information, and the environmental information acquired by the acquisition unit into a trained model that is trained in advance to predict the location information of the person at a time corresponding to the time information from the attribute information, the time information, and the environmental information.

The prediction device according to a second aspect of the present invention further includes a clustering unit, which predicts the location information of each of the plurality of people by inputting the attribute information, the time information, and the environmental information of each of the plurality of people into the trained model, and the clustering unit generates a behavioral pattern of each of the plurality of people from the location information of each of the plurality of people predicted by the prediction unit, and classifies the attribute information of the plurality of people by clustering the behavioral patterns, as described in claim 1.

The third aspect of the present invention is a prediction model learning device that includes a learning unit that generates a trained model for predicting the location information of the person at the time corresponding to the time information from the attribute information, the time information, and the environmental information by learning a model for predicting the location information of the person at the time corresponding to the time information from the attribute information, the time information, and the environmental information based on learning data in which learning attribute information representing the attributes of a person, learning time information representing the time of the prediction target, environmental information representing the external environment around the person, and learning location information of the person are associated with each other.

The present invention has the advantage that when predicting the flow of people with certain attributes, it is possible to predict the flow of people while taking into account the surrounding environment.

1 is a block diagram showing a people flow prediction system according to a first embodiment. FIG. FIG. 1 is a diagram for explaining an overview of an embodiment. FIG. 4 is a diagram for explaining learning data according to the embodiment; FIG. 2 is a diagram for explaining a model of the first embodiment. 10 is a flowchart showing the contents of a learning process routine of the people flow prediction system according to the embodiment. 10 is a flowchart showing the contents of a people flow prediction processing routine of the people flow prediction system according to the embodiment. FIG. 11 is a block diagram showing a people flow prediction system according to a second embodiment. FIG. 13 is a diagram for explaining a model of the third embodiment. FIG. 13 is a diagram for explaining a ranking result of feature amounts.

The following describes in detail an embodiment of the present invention with reference to the drawings.

[First embodiment]

<Configuration of the people flow prediction system of this embodiment>

Figure 1 is a block diagram showing an example of the configuration of a people flow prediction system 10 according to the first embodiment. Functionally, as shown in Figure 1, the people flow prediction system 10 can be represented as a configuration including an input unit 12, a people flow prediction device 14, and a display unit 16.

The input unit 12 accepts data input by a user. The input unit 12 is realized by, for example, a mouse and a keyboard.

The people flow prediction device 14 is realized by a computer including a CPU (Central Processing Unit), a ROM (Read Only Memory) that stores programs for implementing each processing routine, a RAM (Random Access Memory) that temporarily stores data, memory as a storage means, and a network interface.

As shown in FIG. 1, the people flow prediction device 14 functionally comprises a learning data storage unit 20, a learning unit 22, a trained model storage unit 24, an acquisition unit 26, and a prediction unit 28.

2 and 3 are diagrams for explaining an overview of this embodiment. When predicting a pedestrian flow representing the flow of people, the pedestrian flow prediction device 14 of the embodiment predicts the pedestrian flow based on attribute information representing the attributes of the person and environmental information representing the external environment around the person. The attribute information includes, for example, information such as the person's gender, age, occupation, and place of residence. The environmental information includes, for example, information such as the weather conditions, greenery view rate, scenery, buildings, stores, the attractiveness of the stores (for example, scores on gourmet sites, etc.), and transportation convenience in the area. These pieces of information are converted into information that can be input into the trained model described later. The external environment around a person refers to the environment within a predetermined range from the position of the person.

When predicting the flow of people within a certain area, it is necessary to take into account the environment within that area. For example, if there is an attractive store within the area, it is expected that people will gather around it. Furthermore, if that attractive store is aimed at people in their 20s, it is expected that people with the attributes of people in their 20s will gather around the store.

As shown in FIG. 2, the people flow prediction device 14 of this embodiment predicts people flow by taking into account attribute information At representing the attributes of a person and environmental information En representing the external environment around the person. For example, the people flow prediction device 14 predicts that a person will be present at the position of the black circle shown in FIG. 2 at the time t to be predicted. In this case, the people flow prediction device 14 predicts that the person to be predicted will be present at position P of the black circle shown in FIG. 2 at the time t to be predicted, based on the attribute information At of the person to be predicted and environmental information En within the area. The person to be predicted who is present at position P has attribute information At. In this way, according to this embodiment, it is possible to predict people flow by taking into account the attributes and environment of the person.

The details are explained below.

The learning data storage unit 20 stores multiple pieces of learning data. The learning data in this embodiment is data in which learning attribute information representing the attributes of a person, learning time information representing the time of the prediction target, learning environment information representing the external environment around the person, and learning location information of the person at that time are associated with each other. The learning data is, for example, data based on past performance.

Figure 3 is a diagram for explaining the learning data of this embodiment. As shown in Figure 3, the learning data is stored for each person in such a manner that learning attribute information, learning time information, learning environment information, and learning location information are associated with each other.

The learning data for person ID "00001" shown in Figure 3 indicates that when a person's learning attribute information is "A1, ...", learning time information indicating the time to be predicted is "D1, ...", and learning environmental information is "F1, ...", the person's learning location information is "Z1, ...". Note that the attribute information and environmental information are replaced with numerical data. For example, the gender of the attribute information is converted into a format in which males are -1 and females are 1, and the format is made one that can be input into the trained model described below. The same applies to the environmental information.

The learning unit 22 learns a model for predicting the location information of a person from the person's attribute information, time information, and environmental information representing the external environment around the person, based on multiple learning data stored in the learning data storage unit 20. Then, the learning unit 22 generates a trained model for predicting the location information of the person from the person's attribute information, the time information of the prediction target, and the environmental information.

Figure 4 shows a diagram for explaining the model of the first embodiment. As shown in Figure 4, the model of the first embodiment is a neural network. For this reason, a trained neural network M1 is generated by the learning unit 22. When a person's attribute information, time information of the prediction target, and environmental information are input to the trained neural network M1, latitude and longitude are output as the position information of the person. Note that the position information output from the neural network M1 at this time is the position information of the person at the time corresponding to the time information of the prediction target. Note that data such as "time" and "day of the week" are included as input data to the trained neural network M1 in Figure 4, and data corresponding to the time information is input into the places corresponding to this "time" and "day of the week".

For this reason, the learning unit 22 trains a neural network by known supervised machine learning based on multiple pieces of learning data stored in the learning data storage unit 20, and generates a trained neural network. Note that any method of supervised machine learning may be used.

Then, the learning unit 22 stores the generated trained model in the trained model storage unit 24.

The trained model memory unit 24 stores the trained model generated by the training unit 22.

The acquisition unit 26 acquires data of the prediction target. Specifically, the acquisition unit 26 acquires attribute information of the person whose position is to be predicted, time information of the prediction target, and environmental information that represents the external environment around the person.

The prediction unit 28 predicts the location information of the person to be predicted by inputting the attribute information of the person to be predicted, the time information of the person to be predicted, and the environmental information acquired by the acquisition unit 26 into the learned model stored in the learned model storage unit 24.

In this way, the people flow prediction device 14 of the first embodiment uses people attribute information and environmental information that represents the external environment around the people when predicting people flow. This makes it possible to predict people flow taking into account the environment when predicting the flow of people with certain attributes.

The display unit 16 displays the information output from the people flow prediction device 14. The display unit 16 is realized, for example, by a display. The display unit 16 displays the prediction results obtained by the prediction unit 28. The user checks the prediction results displayed on the display unit 16 and uses them, for example, in building planning.

<Function of the people flow prediction system>

Next, the operation of the people flow prediction system 10 of the first embodiment will be described with reference to the figures. When the people flow prediction device 14 of the people flow prediction system 10 receives learning data via the input unit 12, it stores the data in the learning data storage unit 20. Then, when the people flow prediction device 14 of the people flow prediction system 10 receives an instruction signal to start the learning process, it executes the learning process routine shown in FIG. 5.

＜Learning processing routine＞

In step S100, the learning unit 22 acquires multiple pieces of learning data stored in the learning data storage unit 20.

In step S102, the learning unit 22 generates a trained neural network for predicting a person's position information by training a neural network, which is an example of a model, based on the multiple pieces of learning data acquired in step S100.

In step S104, the learning unit 22 stores the trained model generated in step S102 in the trained model storage unit 24 and ends the learning processing routine.

Next, when the people flow prediction device 14 of the people flow prediction system 10 receives the data to be predicted input from the input unit 12, it executes the people flow prediction processing routine shown in FIG. 6.

<People flow prediction processing routine>

In step S200, the acquisition unit 26 acquires attribute information of the person whose location is to be predicted, which is input from the input unit 12.

In step S202, the acquisition unit 26 acquires the time information of the target location prediction input from the input unit 12.

In step S204, the acquisition unit 26 acquires environmental information of the person whose position is to be predicted, which is input from the input unit 12.

In step S206, the prediction unit 28 reads out the trained neural network, which is the trained model stored in the trained model storage unit 24.

In step S208, the prediction unit 28 predicts the location information of the person whose location is to be predicted by inputting the person's attribute information acquired in step S200, the time information acquired in step S202, and the environmental information acquired in step S204 to the trained neural network read out in step S206.

In step S210, the prediction unit 28 outputs the location information obtained in step S208 as a result, and ends the people flow prediction processing routine.

The display unit 16 displays the information output from the people flow prediction device 14. The user checks the information displayed on the display unit 16 and uses it, for example, for building planning.

As described above in detail, the people flow prediction device of the first embodiment acquires attribute information that represents the attributes of a person, time information of the prediction target, and environmental information that represents the external environment around the person. The people flow prediction device then predicts the location information of a person at a time corresponding to the time information of the prediction target by inputting the acquired attribute information, time information, and environmental information into a pre-trained model for predicting the location information of a person at the prediction target time. In this way, when predicting the flow of people with certain attributes, it is possible to predict the people flow taking into account the surrounding environment. Specifically, it is predicted what attributes people will have and what positions they will be in at a certain time of the prediction target.

More specifically, the people flow prediction device of the first embodiment can predict people flow taking into account the environment around people by considering the environment as an external factor as a parameter.

In addition, the people flow prediction device of the first embodiment can reproduce or predict people flow in a certain environment and with certain attribute information. For example, it can predict the places (location information) that a car-loving man in his 30s is likely to visit on a sunny day.

In addition, since the prediction of people flow takes into account the environment, which represents the external factors of the town, and actual human behavior, it is possible to predict the future bustle of the town. Furthermore, by inputting changes in external factors due to new redevelopment into the people flow prediction device, it becomes possible to predict people flow after redevelopment. Furthermore, by predicting people flow taking into account attributes and the environment, it is also possible to predict the effect of attracting customers, which can also be used for event content and inventory management of commercial facilities.

[Second embodiment]

Next, a second embodiment will be described. The second embodiment differs from the first embodiment in that a behavioral pattern of each of a plurality of people is generated from the prediction results of the location information of each of the plurality of people, and the attribute information of the plurality of people is classified by clustering the behavioral patterns.

Figure 7 is a block diagram showing an example of the configuration of a people flow prediction system 210 according to the second embodiment. Functionally, the people flow prediction system 210 can be represented as a configuration including an input unit 12, a people flow prediction device 214, and a display unit 16, as shown in Figure 7.

As shown in FIG. 7, the people flow prediction device 214 functionally comprises a learning data storage unit 20, a learning unit 22, a trained model storage unit 24, an acquisition unit 26, a prediction unit 228, and a clustering unit 30.

The prediction unit 228 predicts the location information of each of the multiple people at the target time by inputting the attribute information of the person, the time information of the prediction target, and the environmental information into the trained model.

The clustering unit 30 generates behavioral patterns for each of the multiple people from the location information for each of the multiple people at the time of the prediction target, predicted by the prediction unit 228. The clustering unit 30 then classifies the attribute information of the multiple people by clustering the multiple behavioral patterns.

For example, consider a case where the prediction unit 228 predicts that a woman in her twenties is in region R at the time t to be predicted. In this case, the clustering unit 30 generates a behavior pattern (time t: region R) for the woman in her twenties. Here, for example, if a man in his sixties also has a similar behavior pattern (time t: region R), the clustering unit 30 clusters the woman in her twenties and the man in his sixties as having the same behavior pattern.

In this way, by clustering attribute information based on behavioral patterns, it is possible to identify the attributes of people with similar tendencies. Such information can be used, for example, in urban or building planning.

Note that other configurations and operations of the people flow prediction system according to the second embodiment are similar to those of the first embodiment, and therefore will not be described.

As described above in detail, the people flow prediction device of the people flow prediction system of the second embodiment generates behavioral patterns of each of the multiple people from the predicted location information of each of the multiple people at the time of the prediction target, and classifies the attribute information of the multiple people by clustering the behavioral patterns. This makes it possible to identify the attributes of people with similar behavioral tendencies.

[Third embodiment]

Next, the third embodiment will be described. The third embodiment differs from the first and second embodiments in that the trained model is a decision tree model. Note that the configuration of the people flow prediction system according to the third embodiment is similar to that of the second embodiment, so the same reference numerals are used and the description will be omitted.

Figure 8 shows an example of a model of the third embodiment. As shown in Figure 8, the model of the third embodiment is a decision tree model. The decision tree model M2 of this embodiment is generated for each person's attribute information.

For example, when time information and environmental information for a person to be predicted are input as input data into decision tree model M2 for a certain attribute, a judgment is made at the node "Weekday?", and if it is a weekday, the process moves to the node on the left. Note that the time information at this time also includes information equivalent to the date. On the other hand, if it is a holiday, the process moves to the node on the right. Finally, a predicted value for location information, such as "latitude xx, longitude xx", is output from decision tree model M2.

The learning unit 22 of the people flow prediction device 214 of the second embodiment, like the first and second embodiments, trains a decision tree model using a supervised machine learning algorithm for a known decision tree model based on multiple learning data, and generates a trained decision tree model for each person's attribute information. Then, the learning unit 22 stores the trained decision tree model in the trained model storage unit 24.

The prediction unit 228 of the people flow prediction device 14 of the second embodiment predicts the location information of a person whose location is to be predicted by inputting time information and environmental information of the person whose location is to be predicted into a trained decision tree model for each attribute. In addition, a predicted value of the location information is output for each attribute information.

In addition, from the trained decision tree model for each attribute, it is possible to obtain the importance that indicates how much the division of the feature located at the node contributes to the prediction. Note that the feature corresponds to "temperature" in the division of "Is the temperature below 10 degrees Celsius?", for example. The importance is calculated, for example, from known Gini impurity, etc.

For this reason, the clustering unit 30 calculates the importance of the feature for each piece of attribute information. Then, the clustering unit 30 ranks the importance of the feature for each piece of attribute information. Figure 9 shows a diagram for explaining the ranking of the importance of the feature. In the example of Figure 9, feature A has the highest importance, and feature B has the second highest importance. The clustering unit 30 ranks the feature in this way for each piece of attribute information.

Then, the clustering unit 30 classifies the attribute information of multiple people based on the result of ranking the features based on the importance. For example, consider the case where the ranking result of the features of a woman in her twenties is that feature A is ranked first, feature B is ranked second, and feature C is ranked third.

For example, suppose that the ranking results for the features of a man in his 60s are also such that feature A is ranked first, feature B is ranked second, and feature C is ranked third. In this case, the clustering unit 30 clusters the woman in her 20s and the man in his 60s into the same cluster.

In this way, by clustering attribute information based on the features used to predict location information, it is possible to identify attributes of people with similar tendencies. Such information can be used, for example, in urban or building planning.

Note that other configurations and operations of the people flow prediction system according to the third embodiment are similar to those of the first or second embodiment, and therefore will not be described.

As described above in detail, the people flow prediction device of the people flow prediction system of the third embodiment predicts people's location information using a decision tree model as a trained model. This makes it possible to cluster attribute information based on the features used when predicting location information, making it possible to identify the attributes of people with similar tendencies. It is also possible to understand which features are important for predicting people flow.

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

For example, in each of the above embodiments, a neural network or a decision tree model is used as an example of a trained model, but the present invention is not limited to this. Any model obtained by machine learning may be used.

In addition, while the above describes an embodiment in which the program according to the present invention is pre-stored (installed) in a storage unit (not shown), the program according to the present invention can also be provided in a form in which it is recorded on a recording medium such as a CD-ROM, DVD-ROM, or microSD card.

10, 210 People flow prediction system 12 Input unit 14, 214 People flow prediction device 16 Display unit 20 Learning data storage unit 22 Learning unit 24 Learned model storage unit 26 Acquisition unit 28, 228 Prediction unit 30 Clustering unit

Claims (3)

An acquisition unit that acquires attribute information representing attributes of a person, time information representing a time of a prediction target, and environmental information representing an external environment around the person;
a prediction unit that predicts the position information of the person at a time corresponding to the time information by inputting the attribute information, the time information, and the environmental information acquired by the acquisition unit into a trained model that has been trained in advance for predicting the position information of the person at a time corresponding to the time information from the attribute information, the time information, and the environmental information;
Including,
The prediction unit predicts the location information of each of a plurality of people by inputting the attribute information, the time information, and the environmental information of each of a plurality of people into the trained model;
a clustering unit that generates a behavioral pattern of each of the plurality of people from the position information of each of the plurality of people predicted by the prediction unit, and classifies the attribute information of the plurality of people by clustering the behavioral patterns;
Prediction device. The prediction device is a prediction device used for city planning or building planning,
The prediction result of the person's location information is used for urban planning or building planning.
The prediction device according to claim 1 . the trained model is a plurality of decision tree models generated in advance for each piece of attribute information, and the decision tree model is a model that outputs location information of a person having the attribute information when the time information and the environmental information are input;
From the decision tree model for each attribute information, it is possible to obtain an importance indicating a contribution of a feature value located at a node to a prediction,
the clustering unit calculates the importance of a feature for each of the attribute information items based on a decision tree model for each of the attribute information items, ranks the importance of the feature for each of the attribute information items, and classifies the attribute information of a plurality of people based on a result of ranking the feature items based on the importance.
The prediction device according to claim 1 or 2.