People flow estimating device, people flow estimating method, and people flow estimating program

Abstract

Provided is a people flow estimating device capable of preventing an overestimation of a flow of people, and capable of estimating a number that does not exceed a cell capacity, using a substantially actual number. The people flow estimating device according to the present invention estimates a flow of people from communications between a plurality of base stations that communicate with a center, and user terminals, the people flow estimating device comprising: an acquiring unit that acquires information relating to the strength of sounding reference signals (SRS) received from the user terminals by the plurality of base stations at a predetermined timing; and a people flow estimating unit that estimates the flow of people by acquiring the information relating to the strength of the SRSs transmitted from the user terminals around the plurality of base stations and inputting the information relating to the strength of the SRSs into a learning model that is trained using the information relating to the strength of SRSs as training data, and using people flow data around the plurality of base stations, acquired in the past, as teacher data.

Description

People flow estimation device, people flow estimation method, and people flow estimation program

[0001] The present invention relates to a people flow estimation device, and more particularly to a people flow estimation device, a people flow estimation method, and a people flow estimation program that can realize people flow estimation based on signal strength analysis of an SRS (Sounding Reference Signal) and cell accommodation number counting.

[0002] Conventional people flow estimation involves incorporating a library into an application on a user's mobile device that acquires GPS (Global Positioning System) signals and periodically uploads them, and collecting data with the user's permission.

[0003] A technology has been proposed that provides a people flow estimation system that can grasp people flow regardless of whether or not the above-mentioned permission is obtained and that estimates people flow with high accuracy (see Patent Document 1, etc.).

[0004] In Patent Document 1, a people flow estimation system 1 is configured by a retention point extraction unit, a statistical correction unit, an estimation parameter calculation unit that uses spatial base station data related to communication terminal devices located within a specified space, acquired by a base station, and spatial trajectory data related to the trajectories of communication terminal devices moving within the space, to calculate estimation parameters related to the ratio between the number of communication terminal devices indicated by the spatial base station data and the number of communication terminal devices remaining within the space indicated by the spatial trajectory data, a visited location estimation unit that uses the spatial trajectory data to calculate the number of communication terminal devices remaining within a feature present in the space, and an estimation unit that uses the calculated number of communication terminal devices and the estimation parameters to estimate the number of communication terminal devices present within the feature.

[0005] Japanese Patent Application Laid-Open No. 2023-149528

[0006] However, the technology in Patent Document 1 uses linear interpolation based on the ratio of the region's actual population to the number of installed applications, resulting in significant errors in extreme areas where the estimated ratio of population to application users is 20:1. Furthermore, cells, which are areas centered on base stations, are classified as large cells and small cells, and the antenna coverage area varies depending on the cell. The cell capacity is unclear across administrative boundaries and changes if there are buildings or other structures in between. Furthermore, sending location information to a server increases the amount of data communication and battery usage on the user's mobile device.

[0007] The present invention has been made in consideration of the above points, and provides a people flow estimation device that can estimate people flow with nearly actual numbers, preventing the erroneous estimation that there are more people than there actually are, which is caused by using the above-mentioned linear interpolation estimation method (e.g., Fermi estimation) when estimating people flow.

[0008] That is, the people flow estimation device of the embodiment is a people flow estimation device that estimates people flow from communications between a plurality of base stations that communicate with a center and user terminals, and is characterized by comprising: an acquisition unit that acquires information about the strength of SRS (Sounding Reference Signal) received by the plurality of base stations from user terminals at a predetermined timing; and a people flow estimation unit that acquires information about the strength of SRS transmitted from user terminals around the plurality of base stations as learning data and inputs the information about the strength of SRS into a learning model that is trained using previously acquired people flow data around the plurality of base stations as teacher data, to estimate people flow.

[0009] Furthermore, the people flow estimation device may further include a learning unit that acquires location information of the user's terminal at the time information on SRS strength is acquired, creates training data based on the acquired location information, and performs learning.

[0010] Furthermore, in the people flow estimation device, the people flow estimation unit may estimate people flow by using beamforming based on the strength of the SRS and directing the antenna directivity of multiple base stations toward the terminal that transmitted the SRS, thereby measuring how many people moved in which direction.

[0011] Furthermore, the people flow estimation device may further include a data distribution unit that displays data indicating how many people moved in which direction, as estimated by the people flow estimation unit, for each regional mesh that divides the area into meshes of equal size based on latitude and longitude, and distributes the regional mesh data.

[0012] Furthermore, the people flow estimation device may further include a data prediction unit that learns and infers the periodicity of people flow using machine learning including decision tree learning and ensemble learning based on the estimated data obtained by beamforming and the learning model.

[0013] Furthermore, in the people flow estimation device, the data prediction unit may determine the reliability of the estimated data based on whether the data estimated by the people flow estimation unit exceeds the cell capacity, which is the upper limit number of user terminals that can communicate with each of the multiple base stations.

[0014] Furthermore, the people flow estimation device may further include an alarm unit that, if there is a large discrepancy between the predicted data predicted by the data prediction unit and the estimated data obtained by beamforming and machine learning, sounds an alarm, regarding the discrepancy as an abnormality.

[0015] The people flow estimation method of the embodiment is a people flow estimation method that estimates people flow using a people flow estimation device that estimates people flow from communications between a plurality of base stations that communicate with a center and user terminals, and includes an acquisition step in which the plurality of base stations acquire information about SRS (Sounding Reference Signal) strength from the user terminals at predetermined timing, and a people flow estimation step in which information about SRS strength transmitted from user terminals around the plurality of base stations is acquired as learning data and previously acquired people flow data around the plurality of base stations is used as training data to estimate people flow by inputting the information about SRS strength into the learning model.

[0016] The people flow estimation program of the embodiment is a people flow estimation program for estimating people flow in a people flow estimation device that estimates people flow from communications between a plurality of base stations communicating with a center and user terminals, and causes a computer to execute an acquisition step of acquiring information regarding the strength of SRS (Sounding Reference Signal) received by a plurality of base stations from user terminals at a predetermined timing, and a people flow estimation step of acquiring information regarding the strength of SRS transmitted from user terminals around a plurality of base stations as learning data and inputting the information regarding the SRS strength into a learning model that is trained using previously acquired people flow data around the plurality of base stations as teacher data, to estimate people flow.

[0017] The people flow estimation device of the present invention is a people flow estimation device that estimates people flow from communications between multiple base stations communicating with a center and user terminals. The people flow estimation device includes: an acquisition unit that acquires information about the strength of SRS (Sounding Reference Signals) received from user terminals at predetermined times by the multiple base stations; and a people flow estimation unit that acquires information about the strength of SRS transmitted from user terminals around the multiple base stations and inputs the information about SRS strength into a learning model trained using the information about SRS strength as learning data and previously acquired people flow data around the multiple base stations as training data, thereby estimating people flow. This prevents erroneous estimation of people flow that is greater than the actual number, which is caused by using linear interpolation, etc., using the ratio of the actual population of the area to the number of installed applications when estimating people flow, and enables estimation based on nearly actual numbers. Furthermore, errors in people flow prediction values ​​in rural areas can be significantly reduced.

[0018] Fig. 1 is a diagram showing an overall configuration in which people flow estimation is performed in an embodiment. Fig. 2 is a block diagram showing functional units of a people flow estimation device of an embodiment. Fig. 3 is a diagram showing that the people flow estimation device of an embodiment is realized using 5G. Fig. 4 is a diagram showing that machine learning is performed in the people flow estimation device of an embodiment. Fig. 5 is a flowchart showing people flow estimation processing in an embodiment. Fig. 6 is a flowchart showing learning processing in an embodiment.

[0019] The people flow estimation device of the embodiment is a people flow estimation device that can prevent overestimation of people flow and estimate a number that is almost actual and does not exceed the cell capacity. The people flow estimation device of the embodiment is a device that uses RIC (RAN (Radio access network) Intelligent Controller). RIC is a technology that makes radio access networks (RANs) more open and intelligent, and uses AI / ML (artificial intelligence / machine learning). This improves the efficiency and automation of network operations that use RIC.

[0020] Conventional people flow estimation systems incorporate libraries into mobile device applications that acquire and periodically upload GPS signals, collecting data with user permission. This method estimates people flow using linear interpolation, which uses the ratio of the actual population of a region to the number of installed applications, assuming that one person's movement corresponds to 20 other people's movements. This approach results in larger errors in rural areas than in urban areas. For example, if a mobile phone company has 47 million users and a tracking app installation base of 2 million, applying the above-mentioned method to estimate the movement of one user would result in the movement of (4700 / 200) = 23.5 people. While the movement of 23.5 people is not a large movement in urban areas, in depopulated rural areas, it corresponds to the movement of a large group of people.

[0021] Furthermore, cells, which are areas centered on base stations, are divided into large cells and small cells. The antenna width range varies depending on the cell, making the cell capacity unclear across administrative boundaries and changing if buildings or other structures are present. Furthermore, sending data to a server for pedestrian flow estimation increases the data traffic and battery consumption of users' mobile devices. Furthermore, macro stations, which are base stations that cover a wide area, also require support for Wi-Fi connection status, etc. Macro stations are common in urban areas, where small base stations are also located, playing complementary roles. Furthermore, urban areas have large populations and many mobile phone users, making it difficult to track individual users or provide services that utilize their individual profiles.

[0022] Therefore, the people flow estimation device (method) of the embodiment is positioned as a technology that uses SRS to estimate people flow, thereby preventing the use of Fermi estimation or the like to estimate a larger number of people than the actual number, and enabling estimation based on approximately actual numbers. It is also positioned as a technology that, when it is estimated that a number of people flowing around a base station exceeds the cell capacity, notifies the user that the estimation may be incorrect. SRS is a response signal indicating the strength of signals received by a user's terminal from multiple base stations. People flow estimation using SRS may involve having a computer located adjacent to a base station that has jurisdiction over a user's terminal estimate people flow within that base station based on the signal strength derived from SRS. Note that the people flow estimation may also be performed by a computer in the base station.

[0023] The configuration of a people flow estimation device 1 according to an embodiment is shown as a schematic diagram in Fig. 1. The people flow estimation device 1 communicates with a center 2. The center 2 communicates with multiple base stations 3. A user's terminal 4 communicates with one of the multiple base stations 3 that has jurisdiction over the user's terminal 4. The people flow estimation device 1 communicates with the multiple base stations 3 via a network 5 that communicates with the multiple base stations 3, and acquires information regarding the strength of SRS received by the multiple base stations 3 from the user's terminal 4 at a predetermined timing.

[0024] 2 is a block diagram showing the functional units of the people flow estimation device 1. The people flow estimation device 1 includes an acquisition unit 10, a people flow estimation unit 20, a learning unit 30, a data prediction unit 40, a data distribution unit 50, and a notification unit 60. The center 2 includes a data accumulation unit 70.

[0025] The acquisition unit 10 acquires information about the strength of SRS received at predetermined times by the multiple base stations 3 from user terminals 4. Acquisition of this information about SRS strength includes both cases where it is acquired through communication from the multiple base stations 3, and where the people flow estimation device 1 is connected to the antennas of the multiple base stations 3 and receives the information about SRS strength.

[0026] Referring to Figure 3, the people flow estimation unit 20 uses information about SRS intensity as learning data and previously acquired people flow data around multiple base stations 3 as teacher data 21 to machine-learn (22) a learning model 23, and acquires information about SRS intensity transmitted from user terminals 4 around multiple base stations 3 and inputs the information about SRS intensity into the learning model 23 to estimate people flow.

[0027] The people flow estimation unit 20 estimates people flow using beamforming based on the strength of the SRS. Beamforming is a technology for directing the antenna directivity of a base station 3 toward a terminal 4. By directing the antenna directivities of multiple base stations 3 toward a terminal 4 that has transmitted an SRS, people flow is estimated by measuring the number and direction of people moving. A distributed processing platform is used for this estimation, which distributes calculations and inferences to computers distributed to each of the multiple base stations 3. Note that while people flow estimation requires estimation using the learning model described above, it is also possible to integrate estimation using the learning model and estimation using beamforming.

[0028] The learning unit 30 acquires location information of the user's terminal 4 at the time when information related to SRS strength was acquired. A plurality of base stations 3 can obtain TMSI (Temporary Mobile Subscriber Identity) and IMSI (International Mobile Subscriber Identity), i.e., user IDs, and the learning unit 30 acquires location information of the user's terminal 4 from the information service provider that receives this information. Training data 21 is created based on the acquired location information, and training is performed. This training involves constructing a multi-layer neural network using past people flow data as training data and information on SRS strength, cell capacity, and beamforming angle as causal variables.

[0029] The data prediction unit 40 learns and infers the periodicity of people flow using machine learning (22), including decision tree learning and ensemble learning, based on the estimated data obtained by beamforming and the learning model 23. Decision tree learning is a machine learning technique that creates decision trees from data. Decision trees are graphs used for making decisions in decision theory fields such as risk management, and are used to plan and reach goals, and are created to aid decision-making. Decision trees are predictive models in the field of machine learning that derive conclusions about the target value of a certain item from observed results of that item. Ensemble learning is a machine learning technique that trains multiple models and generates predicted values ​​by majority vote (or average). The periodicity of people flow refers to the tendency of how many people move from one point to another on an hourly, daily, or weekly basis, for example.

[0030] The data prediction unit 40 determines the reliability of the estimated data based on whether the data estimated by the people flow estimation unit 20 exceeds the cell capacity, which is the upper limit of the number of user terminals 4 that can communicate with each of the multiple base stations 3. The cell capacity is the upper limit of the number of user terminals 4 with which a base station 3 can communicate, and if the value obtained for the number of people moving during people flow estimation exceeds this upper limit, the value is deemed unreliable. This number is determined by the people flow estimation unit 20 determining the number of user terminals 4 present within a specified coverage area after people have moved within that coverage area over a specified period of time.

[0031] The data distribution unit 50 displays data indicating how many people moved in which direction, estimated by the people flow estimation unit 20, for each regional mesh, which divides the region into meshes of equal size based on latitude and longitude, and distributes people flow distribution data for the regional mesh. Note that the size of the mesh is arbitrary and may differ depending on the region.

[0032] If there is a large discrepancy between the predicted data predicted by the data prediction unit 40 and the estimated data obtained by beamforming and machine learning, the alarm unit 60 determines that the discrepancy is abnormal and issues an alarm. A large discrepancy refers to a difference in magnification or order of magnitude. For example, if the predicted data is twice or three times the estimated data, the alarm unit 60 sets an abnormality flag to determine that the discrepancy is abnormal. Furthermore, if the predicted data shows one person and the estimated data shows 100 people, the alarm unit 60 sets an abnormality flag to determine that the discrepancy is abnormal. A threshold value may also be set and varied depending on the cell capacity. When the cell capacity is relatively high, the number of mobile phone users is large and even slight movement does not have a significant impact, so the threshold value is set high. On the other hand, when the cell capacity is relatively low, the number of mobile phone users is small and even slight movement has a significant impact, so the threshold value is set low.

[0033] The center 2 includes a data collection unit 70. The data collection unit 70 collects the information estimated by the people flow estimation unit 20 as people flow data. The data collection unit 70 also aggregates the information collected in the data collection unit 70 and sends it to the people flow estimation device 1.

[0034] FIG. 4 is a diagram showing that the people flow estimation device 1 of the embodiment is realized using 5G (5th Generation, fifth generation communication standard). RU / DU / CU is a classification of functions constituting a wireless base station of a 5G mobile communication system, and organizes the functions and roles constituting the radio access network (RAN) connecting the base station and terminals, and is connected in the order of terminal - RU - DU - CU - core network. The "RU" (Radio Unit) controls the antenna and communicates radio waves with the terminal, and also controls MIMO (Multi Input Multi Output) and beamforming. The "DU" (Distributed Unit) performs signal modulation and demodulation, MAC layer communication control, etc. The "CU" (Central Unit) is responsible for processing "PDCP" (Packet Data Convergence Protocol), which controls subordinate DUs and RUs, connects to the core network, and encrypts packets, as well as "RRC" (Radio Resource Control), which manages radio resources for terminals. Base stations are divided into "child stations" with antennas and RUs, and "master stations" with CUs, and the master station aggregates nearby child stations and controls the sending and receiving of data.

[0035] When 5G is implemented, received radio waves will be combined in one place, resulting in more deployed computing resources than are needed. This will allow mobile operators to use excess computing power to provide services that only mobile operators can offer. Furthermore, by using a GPU to decode propagation, when there are excess computing resources, such as at night or during off-peak periods, the excess computing power can be allocated to profit-making businesses. GPU stands for "Graphics Processing Unit" and is a semiconductor chip (processor) that performs the calculations required to render images such as 3D graphics.

[0036] Currently, people flow data is mostly obtained by acquiring GPS data from smartphone users and performing statistical analysis, but this method may be avoided by users who are sensitive to battery life. Furthermore, people flow estimation is not feasible unless there is a certain number of users. Furthermore, a library that transmits GPS data every hour is incorporated into a highly convenient application, and data is acquired with the user's permission. This means that people flow estimation is not subject to Fermi estimation, and nonparametric processing may be required.

[0037] However, the people flow estimation device 1 of the embodiment estimates people flow based on the number of terminal connections by base stations, rather than corrections based on regional share or demographics, making people flow data statistically significant and reliable.In addition, it can respond in real time to sudden people flows that are not dependent on residential demographics.

[0038] The people flow estimation method and people flow estimation program of the embodiment will now be described using the flowchart of Figure 5. The people flow estimation method of the embodiment is executed by the computer (people flow estimation unit 20) of the people flow estimation device 1 of the embodiment based on the people flow estimation program (see Figure 2). The people flow estimation program of the embodiment enables the computer of the people flow estimation device 1 to realize an acquisition function, a people flow estimation function, a data prediction function, a data distribution function, and a notification function. Each function overlaps with the description of the people flow estimation device 1 of the embodiment described above, so details will be omitted.

[0039] 5 shows the flow of an information processing method according to an embodiment, and includes various steps, such as an acquisition step (S1), a people flow estimation step (S2), a data prediction step (S3), a data distribution step (S4), and a notification step (S5). The people flow estimation method may also include various other steps not shown in the drawings as needed.

[0040] The acquisition function acquires information about the strength of SRS received by multiple base stations 3 from user terminals 4 at predetermined times (S1: acquisition step). The people flow estimation function estimates people flow using beamforming based on the SRS strength (S2: people flow estimation function).

[0041] The data prediction function learns and infers the periodicity of people flow using machine learning (22) including decision tree learning and ensemble learning based on the estimated data obtained by the beamforming and learning model 23 (S3: data prediction step). The data distribution function displays data indicating how many people moved in which direction, estimated by the people flow estimation unit 20, for each regional mesh obtained by dividing the region into meshes of equal size based on latitude and longitude, and distributes the regional mesh data (S4: data distribution step).

[0042] If there is a large discrepancy between the predicted data predicted by the data prediction unit 40 and the estimated data obtained by beamforming and machine learning, the alarm function issues an alarm, regarding the discrepancy as an abnormality (S5: alarm function).

[0043] 6 shows the flow of an information processing method according to an embodiment, and includes various steps, such as a location information acquisition step (S6), a teacher data creation step (S7), and a learning step (S8). The people flow estimation method also includes various other steps not shown in the drawings as needed.

[0044] The location information acquisition function acquires location information of the user's terminal 4 when information about the SRS strength is acquired (S6: location information acquisition function). The teacher data creation function creates teacher data 21 based on the acquired location information (S7: teacher data creation function). The learning function learns from the acquired location information and the created teacher data (S8: learning function).

[0045] According to each aspect of the present disclosure described above, by using Fermi estimation or the like to estimate people flow, it is possible to prevent the estimation that there are more people than there actually are, and it is possible to estimate numbers that are close to actual. Therefore, accurate estimation of people flow in order to suppress people flow in order to prevent the spread of infection during a pandemic such as COVID-19 has an extremely important effect in terms of protecting the safety and health of the public, and can contribute to the achievement of Goal 3 of the Sustainable Development Goals (SDGs), which is to "Ensure healthy lives and promote well-being for all at all ages."

[0046] The people flow estimation program of the embodiment can be implemented using, for example, a scripting language such as ActionScript, JavaScript (registered trademark), Python, or Ruby, or a compiler language such as C, C++, C#, Objective-C, Swift, or Java (registered trademark).

[0047] [Regarding Functions and Circuits] Next, the functions and circuits of the above-described pedestrian flow estimation device 1 will be described. Each unit of the pedestrian flow estimation device 1 may be realized as a function of a computer's arithmetic processing unit or the like. That is, the acquisition unit 10, the pedestrian flow estimation unit 20, the learning unit 30, the data prediction unit 40, the data distribution unit 50, and the notification unit 60 of the pedestrian flow estimation device 1 may be realized as an acquisition function, a pedestrian flow estimation function, a learning function, a data prediction function, a data distribution function, and a notification function, respectively, by a computer's arithmetic processing unit or the like. A pedestrian flow estimation program can cause a computer to realize each of the above-described functions. The pedestrian flow estimation program may be recorded in a computer-readable non-transitory storage medium, such as a memory, a solid-state drive, a hard disk drive, or an optical disk. The storage medium may also be referred to as a non-transitory computer-readable medium that stores the pedestrian flow estimation program. The pedestrian flow estimation program may also be transmitted online. As described above, each unit of the pedestrian flow estimation device 1 may be realized by a computer's arithmetic processing unit or the like. The arithmetic processing unit or the like may be configured, for example, by an integrated circuit or the like. For this reason, each unit of the people flow estimation device 1 may be realized as a circuit that constitutes a processing device or the like. That is, the acquisition unit 10, the people flow estimation unit 20, the learning unit 30, the data prediction unit 40, the data distribution unit 50, and the notification unit 60 of the people flow estimation device 1 may be realized as an acquisition circuit, a people flow estimation circuit, a learning circuit, a data prediction circuit, a data distribution circuit, and a notification circuit that constitute a processing device or the like of a computer. Furthermore, the acquisition unit 10, the people flow estimation unit 20, the learning unit 30, the data prediction unit 40, the data distribution unit 50, and the notification unit 60 of the people flow estimation device 1 may be realized as, for example, an acquisition function, a people flow estimation function, a learning function, a data prediction function, a data distribution function, and a notification function that include the functions of a processing device or the like. Furthermore, the acquisition unit 10, people flow estimation unit 20, learning unit 30, data prediction unit 40, data distribution unit 50 and notification unit 60 of the people flow estimation device 1 may be realized as an acquisition circuit, people flow estimation circuit, learning circuit, data prediction circuit, data distribution circuit and notification circuit, for example, by being configured using integrated circuits or the like.Furthermore, the acquisition unit 10, people flow estimation unit 20, learning unit 30, data prediction unit 40, data distribution unit 50, and notification unit 60 of the people flow estimation device 1 may be configured, for example, as an acquisition device, people flow estimation device, learning device, data prediction device, data distribution device, and notification device by being composed of multiple devices.

[0048] The people flow estimation device 1 can combine one or any two or more of the above-mentioned multiple units. In this disclosure, the term "data" is used, but the term "data" can be replaced with "information," and the term "information" can be replaced with "data."

[0049] [Aspects and Effects of the Present Embodiment] Next, one aspect of the present embodiment and the effects of each aspect will be described. Note that each aspect described below is an example at the time of filing, and the present embodiment is not limited to the aspects described below. In other words, the present embodiment is not limited to each aspect described below, and may be realized by appropriately combining each of the above-mentioned parts. Furthermore, a lower aspect may in some cases cite any of the higher aspects. Furthermore, the effects of the present embodiment described below are only examples, and the effects of each aspect are not limited to those described below. Furthermore, each aspect may, for example, achieve at least one of the effects described below.

[0050] (Aspect 1) A people flow estimation device of one aspect estimates people flow based on communications between multiple base stations communicating with a center and user terminals. The people flow estimation device includes: an acquisition unit that acquires information about the strength of SRS (Sounding Reference Signals) received from user terminals at predetermined times by the multiple base stations; and a people flow estimation unit that acquires information about the strength of SRS transmitted from user terminals around the multiple base stations and inputs the information about the SRS strength into a learning model trained using the information about the SRS strength as training data and previously acquired people flow data around the multiple base stations as training data, thereby estimating people flow. This allows the people flow estimation device to use Fermi estimation or the like to estimate people flow, thereby preventing it from estimating a greater number of people than the actual number, and enabling it to estimate people based on nearly actual numbers. Furthermore, the people flow estimation device can significantly reduce errors in people flow prediction values ​​in rural areas.

[0051] (Aspect 2) A people flow estimation device according to one aspect further includes a learning unit that acquires location information of a user's device at the time when information about SRS strength is acquired, creates training data based on the acquired location information, and performs training. This allows the people flow estimation device to calculate commercially reasonable data by using commercial data as training data.

[0052] (Aspect 3) In one aspect of the people flow estimation device, the people flow estimation unit may estimate people flow by measuring the number of people moving in which direction by using beamforming based on the strength of the SRS and directing the antenna directivities of multiple base stations toward the terminal that transmitted the SRS. In this way, the people flow estimation device can finely divide an area by analyzing the beamforming information and estimate people flow for each of the finely divided areas.

[0053] (Aspect 4) The people flow estimation device of one aspect may further include a data distribution unit that displays data indicating the number of people moving in each direction estimated by the people flow estimation unit for each area mesh obtained by dividing an area into meshes of equal size based on latitude and longitude, and distributes the area mesh data. This allows the people flow estimation device to perform accurate people flow estimation for each area mesh.

[0054] (Aspect 5) The people flow estimation device of one aspect may further include a data prediction unit that learns and infers the periodicity of people flow using machine learning including decision tree learning and ensemble learning based on the estimation data obtained by beamforming and the learning model. This enables the people flow estimation to accurately infer the periodicity of people flow.

[0055] (Aspect 6) In the people flow estimation device of one aspect, the data prediction unit may determine the reliability of the estimated data based on whether the data estimated by the people flow estimation unit exceeds a cell capacity, which is an upper limit number of user terminals that can communicate with each of a plurality of base stations. In this way, when the information processing device estimates that a number of people exceeding the cell capacity are flowing around the base station, it can notify that the estimation may be incorrect.

[0056] (Aspect 7) The people flow estimation device of one aspect may further include a notification unit that, when there is a large deviation between the prediction data predicted by the data prediction unit and the estimation data obtained by beamforming and machine learning, determines that the deviation is an abnormality and sounds an alarm. In this way, when there is a large deviation between the prediction data predicted by the data prediction unit and the estimation data obtained by beamforming and machine learning, the people flow estimation device can notify that the estimation may be incorrect.

[0057] (Aspect 8) A people flow estimation method according to one aspect uses a people flow estimation device that estimates people flow from communications between a plurality of base stations communicating with a center and user terminals, and includes: an acquisition step in which the plurality of base stations acquire information about SRS (Sounding Reference Signal) strength from the user terminals at predetermined timing; and a people flow estimation step in which a learning model is trained using the information about SRS strength as learning data and previously acquired people flow data around the plurality of base stations as training data, and acquires information about SRS strength transmitted from user terminals around the plurality of base stations and inputs the information about SRS strength into the learning model to estimate people flow. This allows the people flow estimation method to achieve the same effects as the people flow estimation device according to the above-described aspect.

[0058] (Aspect 9) A people flow estimation program of one aspect is a people flow estimation program for estimating people flow in a people flow estimation device that estimates people flow from communications between a plurality of base stations that communicate with a center and user terminals, the program causing a computer to execute an acquisition step of acquiring information about the strength of SRS (Sounding Reference Signals) received by the plurality of base stations from user terminals at predetermined timings, and a people flow estimation step of acquiring information about the strength of SRS transmitted from user terminals around the plurality of base stations as learning data and inputting the information about the SRS strength into a learning model that is trained using previously acquired people flow data around the plurality of base stations as training data, thereby estimating people flow. As a result, the people flow estimation program can achieve the same effect as the people flow estimation device of the above-mentioned aspect.

[0059] REFERENCE SIGNS LIST 1 people flow estimation device 2 center 3 multiple base stations 4 terminal 5 network 10 acquisition unit 20 people flow estimation unit 21 training data 23 learning model 30 learning unit 40 data prediction unit 50 data distribution unit 60 notification unit 70 data accumulation unit

Claims

1. A people flow estimation device that estimates people flow from communications between a plurality of base stations that communicate with a center and user terminals, comprising: an acquisition unit that acquires information about the strength of an SRS (Sounding Reference Signal) received by the plurality of base stations from the user terminal at a predetermined timing; and a people flow estimation unit that acquires information about the strength of the SRS transmitted from the user terminal around the plurality of base stations, using the information about the strength of the SRS as learning data and previously acquired people flow data around the plurality of base stations as training data, and inputs the information about the strength of the SRS into the learning model to estimate people flow.

2. The people flow estimation device according to claim 1, further comprising a learning unit that acquires location information of the user's terminal at the time when information about the strength of the SRS is acquired, creates training data based on the acquired location information, and performs learning.

3. The people flow estimation device according to claim 1, wherein the people flow estimation unit estimates people flow by measuring how many people are moving in which direction by using beamforming based on the strength of the SRS and directing the antenna directivity of the plurality of base stations toward the terminal that transmitted the SRS.

4. The people flow estimation device of claim 1, further comprising a data distribution unit that displays data indicating how many people moved in which direction, estimated by the people flow estimation unit, for each area mesh that divides an area into meshes of equal size based on latitude and longitude, and distributes the data for all area meshes.

5. The people flow estimation device according to claim 1, further comprising a data prediction unit that learns and infers the periodicity of people flow using machine learning including decision tree learning and ensemble learning based on the estimation data obtained by beamforming and the learning model.

6. The people flow estimation device of claim 5, wherein the data prediction unit determines the reliability of the estimated data based on whether the data estimated by the people flow estimation unit exceeds a cell capacity, which is the upper limit number of terminals of the users that can communicate with each of the plurality of base stations.

7. The people flow estimation device according to claim 6, further comprising an alarm unit that, if there is a large deviation between the predicted data predicted by the data prediction unit and the estimated data obtained by the beamforming and the machine learning, sounds an alarm, regarding the deviation as an abnormality.

8. A people flow estimation method for estimating people flow using a people flow estimation device that estimates people flow from communications between a plurality of base stations communicating with a center and user terminals, comprising: an acquisition step in which the plurality of base stations acquire information regarding the strength of an SRS (Sounding Reference Signal) from the user terminal at a predetermined timing; and a people flow estimation step in which a learning model is trained using the information regarding the strength of the SRS as learning data and previously acquired people flow data around the plurality of base stations as teacher data, and acquires information regarding the strength of the SRS transmitted from the user terminal around the plurality of base stations, and inputs the information regarding the strength of the SRS into the learning model to estimate people flow.

9. A people flow estimation program for estimating people flow in a people flow estimation device that estimates people flow from communications between multiple base stations communicating with a center and user terminals, comprising: a people flow estimation program that causes a computer to execute the following steps: an acquisition step of acquiring information regarding the strength of SRS (Sounding Reference Signal) received by the multiple base stations from the user's terminal at a predetermined timing; and a people flow estimation step of acquiring information regarding the strength of the SRS transmitted from the user's terminal around the multiple base stations as learning data and inputting the information regarding the strength of the SRS into a learning model that is trained using previously acquired people flow data around the multiple base stations as teacher data, to estimate people flow.

Images