Population state determination system and model generation system

Abstract

The present invention suitably determines the state of a population.　This population state determination system 10 comprises: an acquisition unit 11 that acquires population information, indicating a time series population of an area for which the population state is to be determined, and reference information, which corresponds to an area and a period pertaining to the population information and indicates a time series population serving as a reference for the population information; a difference calculation unit 12 that calculates a time series difference between the time series population indicated by the population information acquired by the acquisition unit 11 and the time series population indicated by the reference information; a model calculation unit 13 that performs a calculation by inputting the time series difference calculated by the difference calculation unit 12 to a pre-stored encoder / decoder model for compressing and restoring the input data, and that obtains an output from the encoder / decoder model; and a determination unit 14 that determines the state of the population of the area by comparing the time series difference and the output obtained by the model calculation unit 13.

Description

Population status determination system and model generation system

[0001] The present invention relates to a population status determination system for determining the population status of an area, and a model generation system for generating an encoder-decoder model.

[0002] 2. Description of the Related Art Conventionally, a technique has been proposed for estimating the population for each area and time period using data from mobile terminals such as mobile phones (see, for example, Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2020-123011

[0004] Using the estimated population information, it is possible to detect areas with abnormal populations, such as areas with higher than normal populations. This makes it possible to detect sudden events and discover places where many people gather during disasters.

[0005] One method for detecting population anomalies is a statistical method based on the average and variance of population in an area and time period. This method can detect anomalies in a certain area and time period. However, this method does not take population trends into account, and therefore cannot necessarily detect anomalies with high accuracy.

[0006] One embodiment of the present invention has been made in consideration of the above, and aims to provide a population status determination system that can appropriately determine the population status, and a model generation system related to population determination.

[0007] In order to achieve the above object, a population status determination system according to one embodiment of the present invention comprises: an acquisition unit that acquires population information indicating the time-series population of an area for which the population status is to be determined, and reference information indicating the time-series population that corresponds to the area and period related to the population information and serves as the reference for the population information; a difference calculation unit that calculates the time-series difference between the time-series population indicated by the population information acquired by the acquisition unit and the time-series population indicated by the reference information; a model calculation unit that inputs the time-series difference calculated by the difference calculation unit into a pre-stored encoder-decoder model that compresses and decompresses input data, performs calculations, and obtains output from the encoder-decoder model; and a determination unit that compares the time-series difference calculated by the difference calculation unit with the output obtained by the model calculation unit to determine the population status of the area.

[0008] The population status determination system according to one embodiment of the present invention can determine the population status taking into account the difference between the area's time-series population and a reference value. Furthermore, the determination is made by comparing the input and output of the encoder-decoder model. Therefore, the population status determination system according to one embodiment of the present invention can appropriately determine the population status.

[0009] A model generation system according to one embodiment of the present invention comprises a learning acquisition unit that acquires learning population information indicating a time-series population, and learning reference information indicating a time-series population that corresponds to the area and period related to the learning population information and serves as a reference for the learning population information, and is used to generate an encoder-decoder model that compresses and decompresses input data; a learning difference calculation unit that calculates the time-series difference between the time-series population indicated by the learning population information acquired by the learning acquisition unit and the time-series population indicated by the learning reference information; and a model generation unit that performs machine learning based on the time-series difference calculated by the learning difference calculation unit to generate an encoder-decoder model that inputs information indicating the time-series difference.

[0010] According to a model generation system according to an embodiment of the present invention, an encoder-decoder model to be used in a population status determination system can be generated.

[0011] According to one embodiment of the present invention, the state of the population can be appropriately determined.

[0012] FIG. 1 is a diagram showing the configuration of a computer that is a population status determination system and a model generation system according to an embodiment of the present invention. FIG. 2 is a diagram showing an example of information used by a computer. FIG. 3 is a graph showing an example of a time series difference (difference component) based on population information and reference information, and an example of an output value from an encoder-decoder model when the time series difference is used as an input value. FIG. 4 is a diagram showing a schematic example of an encoder-decoder model generated and used by a computer. FIG. 5 is a flowchart showing processing executed in a model generation system according to an embodiment of the present invention. FIG. 6 is a flowchart showing processing executed in a population status determination system according to an embodiment of the present invention. FIG. 7 is a diagram showing the hardware configuration of a computer that is a population status determination system and a model generation system according to an embodiment of the present invention.

[0013] Hereinafter, embodiments of a population status determination system and a model generation system according to the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are given the same reference numerals, and duplicated explanations will be omitted.

[0014] FIG. 1 shows a computer 1 that is a population status determination system 10 and a model generation system 20 according to this embodiment. The population status determination system 10 is a system (device) that determines (estimates) the population status of a geographical area. The area to be determined is, for example, a 500m square area divided into regions. A half regional mesh may be used as the area. Furthermore, administrative divisions such as cities, towns, villages, or prefectures, or pre-set land use divisions may also be used as the area. In the following explanation, the area will be described as a mesh. Note that the area to be determined does not have to be the above and can be any geographical area.

[0015] The determination by the population status determination system 10 is based on population information indicating the time-series population of the area to be determined. For example, as described below, population information indicating the population for each hour of a day is used for this determination. This determination is, for example, a determination of whether the population in the area to be determined is in an abnormal state that differs from normal. That is, this determination detects an abnormality in the population trend in the area to be determined. An abnormal state in which the population is different from normal is, for example, a state in which the population trend differs excessively from normal population trends. This determination can, for example, detect sudden events or discover locations where large numbers of people are congregating during a disaster. Note that the determination by the population status determination system 10 may determine the degree of abnormality rather than whether an abnormal state is present. Alternatively, the determination by the population status determination system 10 may be any other determination of the population status of an area.

[0016] As will be described later, the population status determination system 10 performs a determination by performing calculations on information based on population information using an encoder-decoder model, which is a trained model generated by machine learning. The encoder-decoder model compresses and decompresses input data. The model generation system 20 generates the encoder-decoder model used for the determination by the population status determination system 10.

[0017] A conventional computer can be used as the computer 1 that is the population status determination system 10 and the model generation system 20 according to this embodiment. The computer 1 may also be a computer system including multiple computers.

[0018] Next, the functions of the population status determination system 10 and the model generation system 20 according to this embodiment will be described. First, the functions of the model generation system 20 will be described, followed by the functions of the population status determination system 10. As shown in FIG. 1 , the model generation system 20 is configured to include a learning acquisition unit 21, a learning difference calculation unit 22, and a model generation unit 23.

[0019] The learning acquisition unit 21 is a functional unit that acquires learning population information indicating a time-series population, which is used to generate an encoder-decoder model, and learning reference information indicating a time-series population that corresponds to the area and period related to the learning population information and serves as a reference for the learning population information. The learning acquisition unit 21 acquires each piece of information as follows.

[0020] Each piece of learning population information is in the same format as the population information used to determine the population status. For example, the population information is information indicating the population of an area for each hour of a day (midnight, 1:00, ..., 11:00 p.m.). When such population information is used, the population status determination system 10 determines the population status of the area being determined on that day. Note that the overall time period (one day in the above example), time interval (every hour in the above example), and format of the population information to be determined do not necessarily have to be as described above.

[0021] Each piece of learning reference information is information in the same format as the learning population information. For example, if the population information is information indicating the population of an area for each hour of a day as described above, the learning population information and the learning reference information are also information in the same format. The learning reference information corresponds to the learning population information and has a specific relationship with the learning population information in terms of area and period. One piece of learning reference information may correspond to one piece of learning population information, or may correspond to multiple pieces of learning population information. The relationship between the learning population information and the learning reference information is similar to the relationship between the population information and reference information used to determine the population status.

[0022] For example, the learning reference information is information indicating the population over time in the same area as the area related to the learning population information, for a period with the same days of the week and times of the day as the period related to the learning population information. Furthermore, for example, the learning reference information is information indicating the average population for the same days of the week and times of the day for a past period of a predetermined length relative to the period related to the learning population information. Specifically, the learning reference information is information indicating the population over time in the area for the same days of the week in the week preceding the period related to the learning population information. Alternatively, the learning reference information is information indicating the average population over time in the area for each hour of the day (each hour) for the same days of the week in the N weeks preceding the period related to the learning population information. Here, N is a predetermined number.

[0023] A large amount of training population information is used to generate an encoder-decoder model. The large amount of training population information typically includes training population information for multiple areas. The training acquisition unit 21 acquires, for example, the data shown in FIG. 2( a) as the training population information and training reference information. The data shown in FIG. 2( a) is information in which a mesh code (information in the "meshcode" column), information indicating a time (information in the "timestamp" column), and information indicating a population (information in the "population" column) are associated with each other. The mesh code is information such as a character string that identifies a mesh, which is an area, and is set in advance for each area. The information indicating a time is, for example, information indicating a year, month, date, and time of day. The information indicating a population indicates the population in the area and at the time indicated by the corresponding mesh code and information indicating a time.

[0024] The data acquired as the learning population information and the learning reference information is population information for the same mesh. The data shown in Figure 2(a) is information when the learning reference information is the population for the previous week of the learning population information. Of the data shown in Figure 2(a), the data with a "timestamp" date of "2021 / 10 / 26" is the learning population information, and the data with a "timestamp" date of "2021 / 10 / 19", which is the same day of the week of the previous week, is the learning reference information.

[0025] The population data shown in Figure 2(a) is generated as spatial statistical information from information indicating the location of mobile phones and registered information about mobile phone subscribers using an existing method, for example. The population data shown in Figure 2(a) may also be generated using any method other than the above. The learning acquisition unit 21 acquires the population data shown in Figure 2(a) that is pre-stored in a database of the computer 1 or another device.

[0026] The learning acquisition unit 21 formats the acquired data as learning population information and learning reference information into data for each mesh code and for each hour of the day (0:00, 1:00, ..., 23:00) as shown in FIG. 2(b), i.e., data on population trends for each day in each area. The data in the first row of FIG. 2(b) (data with "meshcode" of "12345") is an example of learning population information corresponding to the population data shown in FIG. 2(a). The learning reference information is also formatted in a similar data format. The learning acquisition unit 21 acquires a sufficient amount of population trend data to generate an encoder-decoder model using machine learning. The population trend data may or may not include data on the area whose population status is to be determined. Note that the learning acquisition unit 21 may also acquire information indicating time-series population other than the above as learning population information and learning reference information. The learning acquisition unit 21 outputs the acquired learning population information and learning reference information to the learning difference calculation unit 22.

[0027] The learning difference calculation unit 22 is a functional unit that calculates the time series difference between the time series population indicated by the learning population information acquired by the learning acquisition unit 21 and the time series population indicated by the learning reference information.

[0028] The learning difference calculation unit 22 inputs the learning population information and the learning reference information from the learning acquisition unit 21. For each piece of input learning population information, the learning difference calculation unit 22 calculates the difference (difference component) between the input learning population information and the corresponding learning reference information. Specifically, the learning difference calculation unit 22 calculates a time-series difference value by subtracting, for each time-series population value in the learning population information, the time-series population value in the learning reference information corresponding to the learning population information from the time-series population value. The values ​​from which the difference is taken are corresponding population values ​​in time series that are in the same order, for example, in chronological order. The calculated difference is information having the same number of elements as the learning population information and the learning reference information. For example, the calculated difference is data for each mesh code and for each hour on a daily basis.

[0029] Figure 2(c) shows the calculated time series difference. The data in each row of the table in Figure 2(c) represents one time series difference data. The data in the first row of Figure 2(c) (data with a "meshcode" of "12345") represents the time series difference data calculated from the learning population information and learning reference information shown in Figure 2(a). For example, the value at midnight of the data in the first row of Figure 2(c) is -10 (1000-1010), which is the value obtained by subtracting 1010, the value at midnight of the learning reference information (data dated "10 / 19 / 2021"), which represents the previous week's population, from 1000, the value at midnight of the learning population information (data dated "10 / 26 / 2021"). As described above, in this example, the value of the difference data is the population value for the period being determined minus the population value for the previous week. Figure 3 shows a portion of graph G1, an example of a time series difference. The learning difference calculation unit 22 outputs information indicating the calculated time series difference to the model generation unit 23 .

[0030] The above-described time series differences are used to generate the encoder-decoder model. Furthermore, learning type information indicating the type of area related to the learning population information may be used to generate the encoder-decoder model. In this case, the learning acquisition unit 21 may acquire learning type information indicating the type of area related to the learning population information. The learning acquisition unit 21 may acquire the learning type information by performing clustering using the learning population information. The learning acquisition unit 21 may acquire the learning type information by performing clustering using the time series differences calculated by the learning difference calculation unit 22.

[0031] The area type is a type that can affect the population transition in the area. For example, the area type is an urban type such as an "office district," a "residential district," and a "downtown district."

[0032] For example, the learning acquisition unit 21 acquires learning type information that is pre-stored in a database of the computer 1 or another device. For example, the learning type information is information that is associated with a mesh code and indicates the type of area indicated by the mesh code. The information that indicates the area type is, for example, a type code. The area type is set in advance for each area. The learning acquisition unit 21 acquires learning type information that corresponds to the mesh code of the area related to the learning population information.

[0033] The learning acquisition unit 21 may acquire the learning type information by performing clustering using the learning population information, rather than acquiring pre-stored learning type information. The learning acquisition unit 21 performs clustering using the data of daily population changes in area units. For example, the learning acquisition unit 21 performs area clustering using the data of daily population changes in area units, as follows. By performing such clustering, areas with similar population changes can be divided into clusters.

[0034] The learning acquisition unit 21 averages the population for each time period for each area, and generates one population transition data set for each area. For example, the learning acquisition unit 21 averages the population transition data for each day for a predetermined period for each area (for example, the period from one month before the present time to the present time), and generates one population transition data set for each area. The learning acquisition unit 21 clusters the population transition data to cluster the areas. The clustering itself may be performed using a conventional method (for example, the k-means method).

[0035] Alternatively, the learning acquisition unit 21 may cluster the population transition data that may include multiple population transition data for one area. For each area, the learning acquisition unit 21 determines the cluster that includes the most population transition data as the cluster for that area.

[0036] The learning acquisition unit 21 assigns a different type code (cluster number) to each cluster. The learning acquisition unit 21 uses the type code of the cluster to which an area belongs as learning type information for that area. The learning acquisition unit 21 stores the correspondence between the mesh code and type code for each area in the computer 1 so that it can also be used in the population status determination system 10.

[0037] Furthermore, the learning acquisition unit 21 may acquire learning type information by performing clustering using the time-series differences calculated by the learning difference calculation unit 22, rather than the learning population information. In this case, the learning acquisition unit 21 may acquire learning type information by replacing the learning population information in the case of clustering using the above-mentioned learning population information with the time-series differences. The learning acquisition unit 21 also outputs the acquired learning type information to the model generation unit 23.

[0038] The model generation unit 23 is a functional unit that performs machine learning based on the time-series differences calculated by the learning difference calculation unit 22, and generates an encoder-decoder model that receives information indicating the time-series differences. The model generation unit 23 may generate an encoder-decoder model based on the learning type information acquired by the learning acquisition unit 21. The model generation unit 23 may generate an encoder-decoder model that also receives type information indicating the area type. The model generation unit 23 may generate multiple encoder-decoder models according to the learning type information that indicates the area type.

[0039] The encoder-decoder model generated by the model generation unit 23 will now be described. An example of the encoder-decoder model is shown in Fig. 4. The encoder-decoder model is configured using a neural network, and is a trained model that is trained to input information indicating the time series difference based on the time series population of an area, perform dimensional compression, and then output information indicating the original difference. As the encoder-decoder model, an autoencoder (Geoffrey Hinton and Salakhutdinov Ruslan, "Reducing the dimensionality of data with neural networks." Science, pp. 504-507, 2006), a Transformer (Ashish Vaswani et al., "Attention Is All You Need." Advances in neural information processing systems 2017), etc. can be used. The input layer of the encoder-decoder model is provided with neurons equal to the number of elements of information indicating the difference (the number of dimensions of the information indicating the difference). If the information indicating the difference is information (numerical values) indicating the difference in population of an area for each hour of the day (midnight, 1 a.m., ..., 11 p.m.), the input layer of the encoder-decoder model is provided with 24 neurons (vectors) that input the numerical values ​​of the difference in population of the area for each hour. The output layer of the encoder-decoder model is provided with neurons (vectors) of the same number as the neurons in the input layer, which correspond to the neurons in the input layer.

[0040] The configuration of the encoder-decoder model itself may be the same as that of a conventional encoder-decoder model. As shown in Figure 4, a hidden layer provided with a plurality of neurons (vectors) is provided between the input layer and the output layer. Each neuron in the input layer and each neuron in the hidden layer are connected with a weight w used for calculation. Furthermore, each neuron in the hidden layer and each neuron in the output layer are connected with a weight w used for calculation. The number of neurons provided in the hidden layer is less than the number of neurons in the input layer and the output layer. This allows dimensionality reduction in the hidden layer.

[0041] The model generation unit 23 generates an encoder-decoder model as follows: First, an example of a mode in which learning type information is not used will be described, and then an example of a mode in which learning type information is used will be described.

[0042] The model generation unit 23 receives information indicating the time series differences from the learning difference calculation unit 22. As shown in Fig. 4, the model generation unit 23 performs machine learning using the information data indicating the time series differences as input values ​​to the encoder-decoder model and output values ​​(correct answers) of the encoder-decoder model to generate an encoder-decoder model. The above machine learning itself for generating the encoder-decoder model can be performed in the same way as conventional machine learning methods. The above is an example in which learning type information is not used.

[0043] Next, an example of a mode in which the learning type information is used will be described. The model generation unit 23 inputs a type code, which is the learning type information, from the learning acquisition unit 21. In this case, the model generation unit 23 generates an encoder-decoder model that also inputs the type code. In addition to the encoder-decoder model shown in FIG. 4, this encoder-decoder model has neurons corresponding to the type codes in the input layer and output layer. For example, neurons corresponding to the number of types of type codes, each corresponding to a type code, are provided. Of these neurons, a "1" is input to neurons (neurons in the input layer) corresponding to the input type code, and a "0" is input to other neurons. In other words, this is an encoder-decoder model in which the type code is input and output by one-hot encoding. Note that the input and output of the learning type information in the encoder-decoder model does not have to be as described above and may be in any format.

[0044] The model generation unit 23 associates information indicating time-series differences with a type code of the area related to the time-series differences. This association is performed using the mesh code as a key. The model generation unit 23 performs machine learning to generate an encoder-decoder model using the associated information indicating time-series differences and the type code as input values ​​to the encoder-decoder model and as output values ​​(correct answers) of the encoder-decoder model.

[0045] Furthermore, the model generation unit 23 may generate multiple encoder-decoder models according to the type code. For example, the model generation unit 23 may generate an encoder-decoder model for each type code. The model generation unit 23 uses information indicating time series differences related to areas with the same type code to generate one encoder-decoder model. That is, the model generation unit 23 filters the information indicating time series differences for each type code, and uses the filtered information data indicating the time series differences to generate the encoder-decoder model.

[0046] In this case, the model generation unit 23 may generate an encoder-decoder model (an encoder-decoder model that does not receive a type code as input) that receives only information indicating the time-series difference as input, as shown in Fig. 4. The model generation unit 23 performs machine learning to generate an encoder-decoder model, using the information indicating the time-series difference as an input value to the encoder-decoder model and as an output value (correct answer) of the encoder-decoder model.

[0047] Alternatively, the model generation unit 23 may generate an encoder-decoder model that inputs a type code in addition to information indicating a time series difference. In this case, the model generation unit 23 performs machine learning to generate an encoder-decoder model, using the information indicating the time series difference and the type code, which correspond to each other, as input values ​​to the encoder-decoder model and as output values ​​(correct answers) of the encoder-decoder model. The model generation unit 23 performs machine learning as described above for each type code to generate an encoder-decoder model for each type code.

[0048] The model generation unit 23 outputs the generated encoder-decoder model to the population status determination system 10. When the model generation unit 23 generates an encoder-decoder model for each type code, the model generation unit 23 also outputs the type code corresponding to each encoder-decoder model to the population status determination system 10.

[0049] When a second encoder-decoder model is used in addition to the above encoder-decoder model for the determination by population status determination system 10, model generation unit 23 may generate the second encoder-decoder model by performing machine learning based on the learning population information acquired by learning acquisition unit 21. The second encoder-decoder model is an encoder-decoder model that receives as input information indicating a time-series population, and is different from the above encoder-decoder model that receives as input information indicating a time-series difference.

[0050] The second encoder-decoder model may also have the same model configuration as the example of the encoder-decoder model shown in FIG. 4 , except for the type of input information. The second encoder-decoder model is configured using a neural network and is a trained model that inputs population information indicating the time-series population of an area, performs dimensionality reduction, and then outputs the original population information. As with the encoder-decoder model described above, an autoencoder or the like can be used as the second encoder-decoder model. The input layer of the second encoder-decoder model has neurons equal to the number of elements in the population information (the number of dimensions of the population information). If the population information is information (numerical values) indicating the population of an area for each hour of the day (0:00, 1:00, ..., 23:00), the input layer of the encoder-decoder model has 24 neurons (vectors) that input the population values ​​of the area for each hour. The output layer of the encoder-decoder model has neurons (vectors) corresponding to the neurons in the input layer, and the same number of neurons as the neurons in the input layer.

[0051] The configuration of the second encoder-decoder model itself may be the same as that of a conventional encoder-decoder model. As shown in Figure 4, a hidden layer provided with a plurality of neurons (vectors) is provided between the input layer and the output layer. Each neuron in the input layer and each neuron in the hidden layer are connected with a weight w used for calculation. Furthermore, each neuron in the hidden layer and each neuron in the output layer are connected with a weight w used for calculation. The number of neurons provided in the hidden layer is less than the number of neurons in the input layer and the output layer. This allows dimensionality reduction in the hidden layer.

[0052] The model generation unit 23 generates the second encoder-decoder model as follows. In this case, the learning acquisition unit 21 outputs the learning population information acquired for use in generating the second encoder-decoder model to the model generation unit 23. The learning population information here may be used by the learning difference calculation unit 22 to calculate the time series difference. That is, the learning population information may be used to generate the encoder-decoder model. Alternatively, the learning population information may not be used by the learning difference calculation unit 22 to calculate the time series difference. Moreover, the learning population information may include both of these.

[0053] The model generation unit 23 receives population transition data, which is the learning population information, from the learning acquisition unit 21. The model generation unit 23 performs machine learning using the population transition data as input values ​​to the encoder-decoder model and output values ​​(correct answers) of the encoder-decoder model to generate an encoder-decoder model. The above machine learning itself for generating the encoder-decoder model can be performed in the same way as conventional machine learning methods.

[0054] The model generation unit 23 may also generate the second encoder-decoder model using the learning type information in the same manner as the above-mentioned encoder-decoder model. For example, the model generation unit 23 may generate a second encoder-decoder model to which a type code, which is the learning type information, is also input as described above. The model generation unit 23 may also generate a plurality of second encoder-decoder models according to the type code.

[0055] The model generation unit 23 outputs the generated second encoder-decoder model to the population status determination system 10. When the model generation unit 23 generates a second encoder-decoder model for each type code, it also outputs the type code corresponding to each second encoder-decoder model to the population status determination system 10. These are the functions of the model generation system 20 according to this embodiment.

[0056] Next, we will explain the functions of the population status determination system 10. As shown in Figure 1, the population status determination system 10 is configured to include an acquisition unit 11, a difference calculation unit 12, a model calculation unit 13, and a determination unit 14.

[0057] The acquisition unit 11 is a functional unit that acquires population information indicating the time-series population of an area whose population status is to be determined, and reference information indicating the time-series population that corresponds to the area and period related to the population information and serves as a reference for the population information. The acquisition unit 11 may acquire, as the reference information, information indicating the time-series population for a period having the same days of the week and times of the day as the period related to the population information. The acquisition unit 11 may acquire, as the reference information, information indicating the average population for the same days of the week and times of the day for a predetermined length of the past period related to the population information. The acquisition unit 11 acquires the above-mentioned population information for the area and time period whose status is to be determined. The acquisition unit 11 may acquire type information indicating the type of the area whose population status is to be determined.

[0058] For example, the acquisition unit 11 accepts a specification from a user of the population status determination system 10 of the area and time period (date) to be determined, and acquires population information relating to the specified area and time period in the same manner as the acquisition of learning population information by the learning acquisition unit 21 described above.

[0059] The reference information is information in the same format as the population information. For example, if the population information is information indicating the population of an area for each hour of a day, as described above, the reference information is also information in the same format. The reference information corresponds to the population information and has a specific relationship with the population information in terms of area and time period. For example, the reference information is information indicating the population in a time series for the same area as the area related to the population information (the area to be determined), for a period (the time period to be determined) that has the same days of the week and times of the day as the period related to the population information. Furthermore, for example, the reference information is information indicating the average population for the same days of the week and times of the day for a predetermined length of the past period related to the population information. Specifically, the reference information is information indicating the population in the area for the same days of the week in the week preceding the period related to the population information. Alternatively, the reference information is information indicating the average population for each hour of the day in the area for the same days of the week in the N weeks preceding the period related to the population information. Here, N is a predetermined value. The acquisition unit 11 acquires reference information corresponding to the population information in addition to acquiring the population information, in the same manner as the acquisition of the learning reference information by the learning acquisition unit 21 described above.

[0060] The acquisition unit 11 may acquire type information indicating the type of an area for which the population state is to be determined. The acquisition unit 11 acquires type information for the area related to the acquired population information that is the same as the learning type information acquired by the learning acquisition unit 21. For example, when the learning acquisition unit 21 acquires learning type information that is previously stored in association with the mesh code of the area, the acquisition unit 11 acquires, from the same information, a type code that corresponds to the mesh code of the area related to the population information as type information.

[0061] In addition, when clustering is performed by the learning acquisition unit 21, the acquisition unit 11 acquires, as type information, a type code corresponding to the mesh code of the area related to the population information from the information corresponding to the mesh code and type code stored in the computer 1 as a result of the clustering.

[0062] The acquisition unit 11 outputs the acquired population information and reference information to the difference calculation unit 12. When acquiring type information, the acquisition unit 11 also outputs the acquired type information to the model calculation unit 13.

[0063] The difference calculation unit 12 is a functional unit that calculates the time-series difference between the population time-series indicated by the population information acquired by the acquisition unit 11 and the population time-series indicated by the reference information.

[0064] The difference calculation unit 12 inputs the population information and the reference information from the acquisition unit 11. The difference calculation unit 12 calculates the difference between the input population information and the reference information. Specifically, the difference calculation unit 12 calculates the time-series difference value by subtracting the time-series population value of the reference information from each time-series population value of the population information. The values ​​from which the difference is taken are corresponding population values ​​in the same time-series order, such as in the same chronological order. The calculated difference is information having the same number of elements as the population information and the reference information. For example, the calculated difference is data for each mesh code and for each hour on a daily basis.

[0065] The calculated difference has the same format as the difference between the learning population information and the learning reference information shown in Fig. 2(c). The difference calculation unit 12 outputs information indicating the calculated time-series difference to the model calculation unit 13 and the determination unit 14.

[0066] The model calculation unit 13 is a functional unit that inputs the time-series differences calculated by the difference calculation unit 12 into a pre-stored encoder-decoder model, performs calculations, and obtains output from the encoder-decoder model. The model calculation unit 13 may perform calculations using the encoder-decoder model based on the type information acquired by the acquisition unit 11. The model calculation unit 13 may also input the type information into the encoder-decoder model and obtain output from the encoder-decoder model. The model calculation unit 13 may select an encoder-decoder model to be used for calculations from a plurality of pre-stored encoder-decoder models based on the type information, and perform calculations using the selected encoder-decoder model.

[0067] The model calculation unit 13 receives and stores the encoder / decoder model generated by the model generation system 20. The model calculation unit 13 receives information indicating the time-series difference from the difference calculation unit 12.

[0068] The model calculation unit 13 performs calculations using the weights w of the encoder-decoder model, with information indicating the time series difference as an input value to the encoder-decoder model, to obtain an output value from the encoder-decoder model. That is, the model calculation unit 13 performs inference using the encoder-decoder model. The output value from the encoder-decoder model is restored data of the time series difference data, and is information in the same format as the information indicating the time series difference. Graph G2 shows an example of output values ​​when the time series difference shown by graph G1 in Figure 3 is used as an input value.

[0069] In an aspect in which type information is used, the model calculation unit 13 inputs the type information from the acquisition unit 11 and performs the following processing. In this case, for example, an encoder-decoder model that also inputs type information is generated by the model generation system 20 as described above. The model calculation unit 13 performs a calculation using the weight w of the encoder-decoder model, with the type information and information indicating the time-series difference as input values ​​to the encoder-decoder model, to obtain an output value from the encoder-decoder model. When the type information is used as an input value to the encoder-decoder model, it is input in the same manner as the type information input to the encoder-decoder model during machine learning by the model generation unit 23. For example, a type code, which is type information, is input to the encoder-decoder model by one-hot encoding.

[0070] In this case, for example, as described above, the model generation system 20 generates a plurality of encoder-decoder models corresponding to the type code. The model calculation unit 13 selects, from the plurality of encoder-decoder models, an encoder-decoder model corresponding to the type code, which is the input type information. Using the selected encoder-decoder model, the model calculation unit 13 obtains an output value from the encoder-decoder model in the same manner as described above.

[0071] The model calculation unit 13 outputs the output value from the obtained encoder-decoder model to the determination unit 14. Note that the output value output to the determination unit 14 may be only the portion corresponding to the information indicating the time-series difference.

[0072] In addition to the calculation using the above-mentioned encoder-decoder model, the model calculation unit 13 may input the population information acquired by the acquisition unit 11 into a second encoder-decoder model different from the above-mentioned encoder-decoder model, perform calculations, and obtain output from the second encoder-decoder model.

[0073] In this case, the acquisition unit 11 outputs the acquired population information to the model calculation unit 13. The model calculation unit 13 receives and stores the second encoder-decoder model generated by the model generation system 20. The model calculation unit 13 receives the population information from the acquisition unit 11.

[0074] The model calculation unit 13 uses the population information as an input value to the second encoder-decoder model and performs calculations using the weights w of the second encoder-decoder model to obtain output values ​​from the second encoder-decoder model. The output values ​​from the second encoder-decoder model are restored data on population transitions, which is population information, and have the same format as the population information.

[0075] In a configuration in which type information is also used for the second encoder / decoder model, the model calculation unit 13 can obtain an output value from the second encoder / decoder model using type information, as in the case of the above-mentioned encoder / decoder model in which information indicating a time series difference is used as an input value.

[0076] The model calculation unit 13 outputs the output value obtained from the second encoder-decoder model to the determination unit 14. Note that the output value output to the determination unit 14 may be only the portion corresponding to the population information.

[0077] The determination unit 14 is a functional unit that compares the time series difference calculated by the difference calculation unit 12 with the output obtained by the model calculation unit 13 to determine the state of the population of the area. The determination by the determination unit 14 is, for example, as described above, a determination of whether the population in the area to be determined is in an abnormal state different from normal. However, any determination other than the above may be made as long as it can be made by comparing information indicating the time series difference input to the encoder-decoder model with the output from the encoder-decoder model. The determination unit 14 determines the state of the population of the area as follows.

[0078] The determination unit 14 receives input of information indicating the time series difference from the difference calculation unit 12. The determination unit 14 receives input of an output value corresponding to the information indicating the time series difference from the model calculation unit 13. The determination unit 14 compares the input to the encoder-decoder model (time series difference data, for example, graph G1 in FIG. 3 ) with the output from the encoder-decoder model (reconstructed data of the time series difference data, for example, graph G2 in FIG. 3 ), and calculates an error as the degree of anomaly. For example, the determination unit 14 calculates the absolute value of the difference between the input and output for each time period every hour, and determines the sum of all time periods as the error.

[0079] The determination unit 14 compares the calculated error with a preset threshold. If the error is equal to or greater than the threshold, the determination unit 14 determines that the population in the area to be determined is in an abnormal state. For example, the determination unit 14 sets the determination result to 1 in this case. In this case, it is estimated that an event or other phenomenon that differs from normal times is occurring in the area to be determined. If the error is not equal to or greater than the threshold, the determination unit 14 determines that the population in the area to be determined is not in an abnormal state. For example, the determination unit 14 sets the determination result to 0 in this case.

[0080] The above judgment utilizes the fact that if an encoder-decoder model is generated by machine learning using only normal data, the model cannot be successfully restored when abnormal data is input to the encoder-decoder model. Therefore, under normal circumstances, the judgment is based on the time series differences used when generating the encoder-decoder model in the model generation system 20, i.e., the training population information and the training reference information.

[0081] Therefore, in this embodiment, the learning population information and learning reference information are used that are (considered to be) indicative of population trends under normal conditions. Furthermore, if the population trend indicated by the reference information used for the judgment is abnormal, the time series difference used for the judgment may be judged to be an abnormal state even if the population indicated by the population information is not in an abnormal state. Therefore, the reference information used for the judgment is also used that is (considered to be) indicative of population trends under normal conditions.

[0082] The determination unit 14 may compare the population information acquired by the acquisition unit 11 with the output from the second encoder-decoder model obtained by the model calculation unit 13 to determine the population status of the area.

[0083] In this case, the acquisition unit 11 outputs the acquired population information to the determination unit 14. The determination unit 14 receives the population information from the acquisition unit 11. The determination unit 14 receives the output value corresponding to the population information obtained from the second encoder-decoder model from the model calculation unit 13. The determination unit 14 compares the input to the second encoder-decoder model (data on population transitions) with the output from the second encoder-decoder model (reconstructed data on the population transitions), and calculates an error as the degree of anomaly. For example, the determination unit 14 calculates the absolute value of the difference between the input and output for each time period every hour, and determines the sum for all time periods as the error.

[0084] The determination unit 14 compares the calculated error with a predetermined threshold. Note that this threshold may be different from the threshold used to compare the error obtained from the time-series difference. For example, the determination unit 14 sets the determination result to 1 if the error is equal to or greater than the threshold, and sets the determination result to 0 if the error is less than the threshold. The determination unit 14 determines that the population in the area to be determined is abnormal if both the error obtained from the time-series difference using the encoder-decoder model and the error obtained from the population information using the second encoder-decoder model are equal to or greater than the threshold (e.g., if both determination results are 1). Alternatively, the determination unit 14 may determine that the population in the area to be determined is abnormal if at least one of the two errors is equal to or greater than the threshold (e.g., if at least one of the determination results is 1). If the determination unit 14 does not determine that the population in the area to be determined is abnormal, it determines that the population in the area to be determined is not abnormal.

[0085] The above determination utilizes the fact that if a second encoder-decoder model is generated by machine learning using only normal data, the encoder-decoder model cannot be successfully restored when abnormal data is input to the encoder-decoder model. Therefore, under normal circumstances, the determination is based on the training population information used when the model generation system 20 generated the second encoder-decoder model.

[0086] The determination by the determining unit 14 may be performed by a method other than the above, as long as it is performed by comparing the input and output of the encoder-decoder model and the second encoder-decoder model.

[0087] The determination unit 14 outputs information indicating the determination result. For example, the determination unit 14 may display the determination result on a display device provided in the computer 1 so that the user can refer to the determination result. Alternatively, the determination unit 14 may transmit the information indicating the determination result to another device. Furthermore, the determination unit 14 may output the information indicating the determination result to an output destination other than the above by a method other than the above. These are the functions of the population status determination system 10 according to this embodiment.

[0088] Next, the processing executed by the computer 1 according to this embodiment (the operating method performed by the computer 1) will be described using the flowcharts of Figures 5 and 6. First, the processing executed by the model generation system 20 will be described using the flowchart of Figure 5.

[0089] In this process, first, the learning acquisition unit 21 acquires learning population information and learning reference information (S01). Next, the learning difference calculation unit 22 calculates the time-series difference between the time-series population indicated by the learning population information and the time-series population indicated by the learning reference information (S02). Next, the learning acquisition unit 21 acquires learning type information (S03). Note that, in a mode in which learning type information is not used, the acquisition of learning type information (S03) may not be performed. Furthermore, the acquisition of learning type information (S03) may be performed before the acquisition of learning population information and learning reference information (S01) or the calculation of the time-series difference (S02), as long as such acquisition processing is possible.

[0090] Next, the model generation unit 23 performs machine learning based on the time series differences to generate an encoder-decoder model (S04). In addition, in an aspect where learning type information is used, the encoder-decoder model is generated based on the learning type information. In addition, in an aspect where a second encoder-decoder model is used, machine learning is performed based on learning population information to generate the second encoder-decoder model. The generated encoder-decoder model is output to the population status determination system 10 and stored by the model calculation unit 13. The above is the processing executed by the model generation system 20 according to this embodiment.

[0091] Next, the processing executed by the population status determination system 10 will be described using the flowchart of FIG. 6 . In this processing, first, the acquisition unit 11 acquires population information, reference information, and type information (S11). Note that in a mode where type information is not used, acquisition of type information is not necessary. Next, the difference calculation unit 12 calculates the time-series difference between the time-series population indicated by the population information and the time-series population indicated by the reference information (S12). Next, the model calculation unit 13 inputs information indicating the time-series difference into an encoder-decoder model, performs calculations, and obtains output from the encoder-decoder model (S13). Also, in a mode where type information is used, calculations are performed using the encoder-decoder model based on the type information. Also, in a mode where a second encoder-decoder model is used, population information is input into a second encoder-decoder model, performs calculations, and obtains output from the second encoder-decoder model.

[0092] Next, the determination unit 14 compares the input to the encoder-decoder model with the output from the encoder-decoder model (S14). Furthermore, in a mode in which a second encoder-decoder model is used, the input to the second encoder-decoder model is compared with the output from the second encoder-decoder model. The determination unit 14 determines the population status of the area based on the above comparison (S15). Next, the determination unit 14 outputs information indicative of the determination result (S16). The above is the processing executed by the population status determination system 10 according to this embodiment.

[0093] The population status determination system 10 according to this embodiment uses the time-series difference based on the time-series population information and the reference information for determination, making it possible to determine the population status taking into account the difference between the time-series population of the area and the reference value. Furthermore, the determination is made by comparing the input and output of the encoder-decoder model. Therefore, the population status determination system 10 according to this embodiment can appropriately determine the population status with high accuracy.

[0094] In stations or event venues where population fluctuates greatly over time, it is thought to be difficult to reconstruct population trends using a general encoder-decoder model. Therefore, in such cases, it is thought to be difficult to detect whether there are any abnormalities in population trends simply using population trends. From another perspective, anomaly detection using a general encoder-decoder model can be thought of as predicting population from population trends over a certain period of time. In congestion prediction, population trends are cyclical, and it is easier to predict the difference (residual) from the cyclical component than to directly predict the congestion value (population), and there is research that shows that predicting the difference improves accuracy. Considering this, it is thought that the difference component is also effective in anomaly detection.

[0095] Furthermore, as in the above-described embodiment, the reference information may be information indicating a time-series population for a period that is the same as the period related to the population information, on the same day of the week, and at the same time of day. Furthermore, the reference information may be information indicating an average population for the same time on the same day of the week and for a period of a predetermined length in the past related to the population information. This configuration allows the time-series difference used for the determination to be appropriate, enabling a more appropriate determination of the population status. However, the reference information may be information other than the above, as long as it corresponds to the area and period related to the population information used for the determination and indicates a time-series population that serves as a reference for the population information.

[0096] Furthermore, in addition to the encoder-decoder model that inputs the time-series difference as in the above-described embodiment, a second encoder-decoder model that inputs population information may be used to determine the population status. With this configuration, the population status can be determined taking into account not only the difference between the area's time-series population and a reference value, but also the time-series population, allowing for more accurate and appropriate determination of the population status. However, the second encoder-decoder model does not necessarily have to be used.

[0097] Furthermore, type information may be used as in the above-described embodiment. By using type information, it is possible to appropriately determine the population status according to the characteristics of the area. For example, it is possible to make a determination that takes into account the functional characteristics of a city, such as an office district, a residential district, and a downtown area. This allows for more accurate and appropriate determination compared to a determination based on average population fluctuations that does not take type information into account.

[0098] The type information may be used as input to the encoder-decoder model as described above. Alternatively, the encoder-decoder model to be used for calculation may be selected based on the type information. With this configuration, the type information can be used reliably and appropriately, and determination can be made reliably and appropriately. However, the type information may be used in a manner other than the above. Also, the type information does not necessarily have to be used.

[0099] The model generation system 20 according to this embodiment can generate an encoder-decoder model for use in the population status determination system 10. Furthermore, when generating the encoder-decoder model, training type information corresponding to the type information may also be used. Furthermore, the training type information may be obtained by performing clustering using the training population information, as described above. Alternatively, the training type information may be obtained by performing clustering using time-series differences, as described above. With this configuration, even if type information is not previously associated with an area, it is possible to generate an encoder-decoder model based on the area type and perform determination using the encoder-decoder model.

[0100] In this embodiment, the computer 1 includes the population status determination system 10 and the model generation system 20, but the population status determination system 10 and the model generation system 20 may be implemented independently of each other.

[0101] The block diagrams used to explain the above embodiments show functional blocks. These functional blocks (components) are realized by any combination of hardware and / or software. Furthermore, the method for realizing each functional block is not particularly limited. That is, each functional block may be realized using a single device that is physically or logically coupled, or may be realized using two or more physically or logically separated devices that are directly or indirectly connected (e.g., wired, wireless, etc.) and these multiple devices. The functional block may also be realized by combining software with the single device or multiple devices.

[0102] Functions include, but are not limited to, judgment, determination, judgment, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, consideration, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating, mapping, and assignment. For example, a functional block (component) that performs transmission is called a transmitting unit or transmitter. As mentioned above, there are no particular limitations on how these functions are implemented.

[0103] For example, the computer 1 according to an embodiment of the present disclosure may function as a computer that performs information processing according to the present disclosure. Fig. 7 is a diagram illustrating an example of a hardware configuration of the computer 1 according to an embodiment of the present disclosure. The computer 1 described above may be physically configured as a computer device including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like.

[0104] In the following description, the term "apparatus" can be interpreted as a circuit, a device, a unit, etc. The hardware configuration of the computer 1 may be configured to include one or more of the apparatuses shown in the drawings, or may be configured to exclude some of the apparatuses.

[0105] Each function in computer 1 is realized by loading specified software (programs) onto hardware such as processor 1001 and memory 1002, causing processor 1001 to perform calculations, control communication via communication device 1004, and control at least one of reading and writing data in memory 1002 and storage 1003.

[0106] The processor 1001 controls the entire computer by running, for example, an operating system. The processor 1001 may be configured as a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, a register, etc. For example, each function of the computer 1 described above may be realized by the processor 1001.

[0107] The processor 1001 also reads programs (program codes), software modules, data, etc. from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes in accordance with these. The programs used are those that cause the computer to execute at least some of the operations described in the above-described embodiments. For example, each function of the computer 1 may be implemented by a control program stored in the memory 1002 and running on the processor 1001. While the above-described various processes have been described as being executed by one processor 1001, they may also be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The programs may also be transmitted from a network via a telecommunications line.

[0108] The memory 1002 is a computer-readable recording medium and may be configured, for example, by at least one of a read-only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a random access memory (RAM), etc. The memory 1002 may also be called a register, a cache, a main memory (primary storage device), etc. The memory 1002 can store executable programs (program codes), software modules, etc. for performing information processing according to an embodiment of the present disclosure.

[0109] Storage 1003 is a computer-readable recording medium, and may be composed of at least one of, for example, an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy (registered trademark) disk, a magnetic strip, etc. Storage 1003 may also be called an auxiliary storage device. The storage medium provided in computer 1 may be, for example, a database, a server, or other appropriate medium including at least one of memory 1002 and storage 1003.

[0110] The communication device 1004 is hardware (transmission / reception device) for communicating between computers via at least one of a wired network and a wireless network, and is also called, for example, a network device, a network controller, a network card, or a communication module.

[0111] The input device 1005 is an input device (e.g., a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that receives input from the outside. The output device 1006 is an output device (e.g., a display, a speaker, an LED lamp, etc.) that outputs to the outside. The input device 1005 and the output device 1006 may be integrated into one device (e.g., a touch panel).

[0112] Furthermore, each device, such as the processor 1001 and the memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or may be configured using different buses between each device.

[0113] The computer 1 may also be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, the processor 1001 may be implemented using at least one of these pieces of hardware.

[0114] The order of the procedures, sequences, flowcharts, etc. of each aspect / embodiment described in this disclosure may be changed unless it is consistent. For example, the methods described in this disclosure present elements of various steps using an example order, and are not limited to the particular order presented.

[0115] Input and output information may be stored in a specific location (for example, memory) or may be managed using a management table. Input and output information may be overwritten, updated, or added to. Output information may be deleted. Input information may be transmitted to another device.

[0116] The determination may be made based on a value represented by one bit (0 or 1), a Boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0117] The aspects / embodiments described in this disclosure may be used alone, in combination, or switched depending on the implementation. Notification of predetermined information (e.g., notification that "X is true") is not limited to explicit notification, but may be implicit (e.g., not notifying the predetermined information).

[0118] Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the spirit and scope of the present disclosure as defined by the claims. Therefore, the description of the present disclosure is intended to be illustrative and does not have any limiting meaning on the present disclosure.

[0119] Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0120] Software, instructions, information, etc. may also be transmitted or received over a transmission medium. For example, if software is transmitted from a website, server, or other remote source using wired technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and / or wireless technologies (such as infrared, microwave), then these wired and / or wireless technologies are included within the definition of transmission media.

[0121] As used in this disclosure, the terms "system" and "network" are used interchangeably.

[0122] Furthermore, the information, parameters, etc. described in this disclosure may be expressed using absolute values, may be expressed using relative values ​​from a predetermined value, or may be expressed using other corresponding information.

[0123] As used in this disclosure, the terms "determining" and "determining" may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, inquiring (e.g., searching in a table, database, or other data structure), ascertaining, and the like. "Determining" and "determining" may also include receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in memory), and the like. Furthermore, "judgment" and "decision" can include regarding resolving, selecting, choosing, establishing, comparing, etc. as having been "judged" or "decided." In other words, "judgment" and "decision" can include regarding some action as having been "judged" or "decided." Furthermore, "judgment (decision)" can be interpreted as "assuming," "expecting," "considering," etc.

[0124] The terms "connected," "coupled," or any variation thereof, refer to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, "connected" may be read as "access." As used in this disclosure, two elements may be considered to be "connected" or "coupled" to each other using one or more wires, cables, and / or printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and optical (both visible and invisible) range, as some non-limiting and non-exhaustive examples.

[0125] As used in this disclosure, the phrase "based on" does not mean "based only on," unless expressly stated otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0126] As used in this disclosure, any reference to an element using a designation such as "first," "second," etc. does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient method of distinguishing between two or more elements. Thus, a reference to a first and a second element does not imply that only two elements may be employed or that the first element must in some way precede the second element.

[0127] When the terms "include," "including," and variations thereof are used in this disclosure, these terms are intended to be inclusive, similar to the term "comprising." Furthermore, when the term "or" is used in this disclosure, it is not intended to be an exclusive or.

[0128] In this disclosure, where articles are added by translation, such as a, an, and the in English, the disclosure may include that the nouns following these articles are in the plural form.

[0129] In the present disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "coupled" may also be interpreted in the same way as "different."

[0130] The population status determination system and model generation system disclosed herein have the following configuration: [1] A population status determination system comprising: an acquisition unit that acquires population information indicating time-series population of an area whose population status is to be determined, and reference information indicating time-series population that corresponds to the area and period related to the population information and serves as a reference for the population information; a difference calculation unit that calculates a time-series difference between the time-series population indicated by the population information acquired by the acquisition unit and the time-series population indicated by the reference information; a model calculation unit that inputs the time-series difference calculated by the difference calculation unit to a pre-stored encoder-decoder model that compresses and decompresses input data, performs calculations, and obtains output from the encoder-decoder model; and a determination unit that compares the time-series difference calculated by the difference calculation unit with the output obtained by the model calculation unit to determine the population status of the area. [2] The population status determination system described in [1], wherein the acquisition unit acquires, as the reference information, information indicating time-series population for a period that has the same days of the week and times of day as a period related to the population information. [3] The population status determination system according to [2], wherein the acquisition unit acquires, as the reference information, information indicating an average population for the same time on the same day of the week and in one day for a past period of a preset length related to the population information. [4] The population status determination system according to any of [1] to [3], wherein the model calculation unit inputs the population information acquired by the acquisition unit into a pre-stored second encoder-decoder model different from the encoder-decoder model that compresses and decompresses input data, performs calculations, and obtains output from the second encoder-decoder model, and the determination unit compares the population information acquired by the acquisition unit with the output from the second encoder-decoder model obtained by the model calculation unit to determine the population status of the area. [5] The population status determination system according to any of [1] to [4], wherein the acquisition unit acquires type information indicating the type of an area whose population status is to be determined, and the model calculation unit performs calculations using the encoder-decoder model based on the type information acquired by the acquisition unit.[6] The population status determination system according to [5], wherein the model calculation unit selects an encoder-decoder model to be used for calculation from a plurality of pre-stored encoder-decoder models based on the type information, and performs calculation using the selected encoder-decoder model. [7] A model generation system comprising: a learning acquisition unit that acquires learning population information indicating time-series population and learning reference information that corresponds to an area and period related to the learning population information and indicates time-series population that serves as a reference for the learning population information, used to generate an encoder-decoder model that compresses and decompresses input data; a learning difference calculation unit that calculates a time-series difference between the time-series population indicated by the learning population information acquired by the learning acquisition unit and the time-series population indicated by the learning reference information; and a model generation unit that performs machine learning based on the time-series difference calculated by the learning difference calculation unit to generate an encoder-decoder model that inputs information indicating the time-series difference. [8] The model generation system according to [7], wherein the learning acquisition unit acquires learning type information indicating the type of area related to the learning population information, and the model generation unit generates the encoder-decoder model based on the learning type information acquired by the learning acquisition unit. [9] The model generation system according to [8], wherein the learning acquisition unit acquires the learning type information by performing clustering using the learning population information.

[10] The model generation system according to [8] or [9], wherein the learning acquisition unit acquires the learning type information by performing clustering using the time series difference calculated by the learning difference calculation unit.

[0131] 1...computer, 10...population status determination system, 11...acquisition unit, 12...difference calculation unit, 13...model calculation unit, 14...judgment unit, 20...model generation system, 21...learning acquisition unit, 22...learning difference calculation unit, 23...model generation unit, 1001...processor, 1002...memory, 1003...storage, 1004...communication device, 1005...input device, 1006...output device, 1007...bus.

Claims

1. A population status determination system comprising: an acquisition unit that acquires population information indicating a time-series population of an area for which a population status is to be determined, and reference information indicating a time-series population that corresponds to the area and period related to the population information and serves as a reference for the population information; a difference calculation unit that calculates a time-series difference between the time-series population indicated by the population information acquired by the acquisition unit and the time-series population indicated by the reference information; a model calculation unit that inputs the time-series difference calculated by the difference calculation unit into a pre-stored encoder-decoder model that compresses and decompresses input data, performs calculations, and obtains an output from the encoder-decoder model; and a determination unit that compares the time-series difference calculated by the difference calculation unit with the output obtained by the model calculation unit to determine the population status of the area.

2. A population status determination system according to claim 1, wherein the acquisition unit acquires, as the reference information, information indicating a time series of population for a period that is the same as the period, day of the week, and time of day related to the population information.

3. A population status determination system as described in claim 2, wherein the acquisition unit acquires, as the reference information, information indicating the average population for the same time on the same day of the week and in the same day for a past period of a preset length related to the population information.

4. The population status determination system of claim 1, wherein the model calculation unit inputs the population information acquired by the acquisition unit into a second encoder-decoder model different from the encoder-decoder model that has been stored in advance and compresses and restores input data, performs calculations, and obtains output from the second encoder-decoder model, and the determination unit compares the population information acquired by the acquisition unit with the output from the second encoder-decoder model obtained by the model calculation unit to determine the population status of the area.

5. The population status determination system according to claim 1, wherein the acquisition unit acquires type information indicating the type of area for which the population status is to be determined, and the model calculation unit performs calculations using the encoder-decoder model based on the type information acquired by the acquisition unit.

6. A population status determination system as described in claim 5, wherein the model calculation unit selects an encoder-decoder model to be used for calculation from a plurality of pre-stored encoder-decoder models based on the type information, and performs calculation using the selected encoder-decoder model.

7. A model generation system comprising: a learning acquisition unit that acquires learning population information indicating a time-series population, and learning reference information indicating a time-series population that corresponds to an area and period related to the learning population information and serves as a reference for the learning population information, used to generate an encoder-decoder model that compresses and restores input data; a learning difference calculation unit that calculates a time-series difference between the time-series population indicated by the learning population information acquired by the learning acquisition unit and the time-series population indicated by the learning reference information; and a model generation unit that performs machine learning based on the time-series difference calculated by the learning difference calculation unit, and generates an encoder-decoder model that inputs information indicating the time-series difference.

8. The model generation system described in claim 7, wherein the learning acquisition unit acquires learning type information indicating the type of area related to the learning population information, and the model generation unit generates the encoder-decoder model based on the learning type information acquired by the learning acquisition unit.

9. The model generating system according to claim 8, wherein the learning acquisition unit acquires learning type information by performing clustering using the learning population information.

10. A model generation system according to claim 8, wherein the learning acquisition unit acquires learning type information by performing clustering using the time series differences calculated by the learning difference calculation unit.

Images