JP7412381B2 - Feature data generation device, wireless usage prediction system, feature data generation method, and computer program - Google Patents

Description

The present invention relates to a feature data generation device, a wireless usage prediction system, a feature data generation method, and a computer program.

In recent years, consideration has been given to allowing a plurality of different wireless communication systems to share the same frequency band. For example, consideration is being given to secondary use of a frequency band (shared frequency band) allocated to an existing wireless communication system (primary carrier system) by another wireless communication system (secondary carrier system). ing. In this frequency sharing between different systems, there are restrictions on the times and places at which the secondary carrier system can use the shared frequency band so as not to interfere with the operation of the primary carrier system. For this reason, it is desirable to be able to predict the future usage status of the shared frequency band of the primary carrier system from the perspective of making investment decisions and usage planning for the secondary carrier system.

Patent Document 1 describes a technique for predicting future time-series data of busy states and idle states of a wireless channel using a stochastic neural network from time-series data of busy states and idle states of a wireless channel. ing.

Japanese Patent Application Publication No. 2019-101792

In order to improve the prediction accuracy of future channel states (busy state and idle state) with respect to the technology described in Patent Document 1 mentioned above, in addition to time-series data of past channel states, we also use time-series data of channel states. It is conceivable to predict future channel conditions by adding feature quantities that become fluctuation factors. However, if feature quantities unrelated to channel state fluctuations are used for prediction, there is a problem in that prediction accuracy deteriorates on the contrary. For this reason, it is desirable to accurately generate feature amount time-series data indicating temporal changes in feature amounts used for prediction processing of future usage conditions of wireless communication (for example, channel conditions and received power).

The present invention has been made in consideration of these circumstances, and its purpose is to provide time-series feature data indicating temporal changes in feature values used in prediction processing of future usage status of wireless communications. The aim is to improve accuracy.

(1) One aspect of the present invention is a feature data generation device that generates feature time series data used in prediction processing of future usage status of wireless communication, the device comprising: Based on usage status time-series data showing temporal changes in past usage status and all-area feature time-series data showing past temporal changes in each area of feature quantities that affect the propagation of radio waves, an importance calculation unit that calculates the importance of the feature of each area; an importance calculation unit that calculates an important area based on the importance; extracts the feature of the important area from the all-area feature time series data; The present invention is a feature amount data generation device including a feature amount time series data generation unit that generates feature amount time series data of an important area indicating past temporal changes in the feature amount of the important area.
(2) In one aspect of the present invention, the importance calculation unit calculates a predicted value of the future usage status of the wireless communication from the usage status time series data and the total area feature amount time series data as time series data. The feature amount data generation device according to (1) above performs learning of a machine learning model so as to output the data, and calculates the degree of importance from the degree of contribution of each area to the learning.
(3) One aspect of the present invention is the feature amount data generation device according to any one of (1) or (2) above, wherein the feature amount is the number of objects that affect the propagation of the radio waves.
(4) One aspect of the present invention is that the importance calculation unit calculates the plurality of types of the feature amount based on the usage state time series data and the whole area feature amount time series data regarding the plurality of types of the feature amount. The importance of the feature of each area is calculated for each feature, and the feature time series data generation unit calculates the importance of the plurality of types of features based on the importance of each of the plurality of types of feature. This is the feature amount data generation device according to (1) above, which generates feature amount time series data of important areas for each amount.
(5) In one aspect of the present invention, the importance calculation unit calculates the future usage state of the wireless communication from the usage state time series data and the whole area feature amount time series data regarding the plurality of types of the feature amounts. A machine learning model is trained to output the predicted value of as time series data, and the degree of importance is calculated from the degree of contribution of each area to the learning for each of the plurality of types of feature amounts. 4) is a feature data generation device.
(6) One aspect of the present invention is the feature amount data generation device according to either (4) or (5), wherein the plurality of types of feature amounts are micro information and macro information.

(7) One aspect of the present invention provides a feature data generation device according to any one of (1) to (6) above, and a wireless usage prediction device that generates wireless usage prediction data indicating a prediction of a future usage state of wireless communication. a data generation device, the wireless usage prediction data generation device includes usage status time series data indicating temporal changes in past usage status of radio waves used for wireless communication, and the feature amount data generation device. The present invention is a wireless usage prediction system that generates the wireless usage prediction data based on the generated feature amount time series data of important areas.
(8) In one aspect of the present invention, the usage state time-series data is data indicating past temporal changes in received power of the radio wave at the reception point of the radio wave, and the wireless usage prediction data is The radio usage prediction system according to (7) above is the received power predicted value time series data indicating the prediction of the temporal change in the received power of the radio wave at the receiving point.

(9) One aspect of the present invention is a feature data generation method executed by a feature data generation device that generates feature time series data used for prediction processing of future usage status of wireless communication, The feature data generation device generates usage state time series data indicating temporal changes in the past usage state of the radio waves used for the wireless communication, and the past time of each area of the feature amount that affects the propagation of the radio waves. an importance calculation step of calculating the importance of the feature of each area based on time-series data of all area feature values showing changes in the area; , extracting the feature amount of the important area from the all-area feature time series data, and generating feature time series data of the important area indicating past temporal changes of the feature amount of the important area. This is a feature amount data generation method including a step of generating amount time series data.

(10) One aspect of the present invention is to provide a computer of a feature data generation device that generates feature time series data used for prediction processing of the future usage state of wireless communication with radio waves used for the wireless communication. Based on usage status time-series data showing temporal changes in past usage status and all-area feature time-series data showing past temporal changes in each area of feature quantities that affect the propagation of radio waves, an importance calculation step of calculating the importance of the feature of each area; determining an important area based on the importance; extracting the feature of the important area from the all-area feature time series data; The present invention is a computer program for executing a feature amount time series data generation step of generating feature amount time series data of an important area indicating past temporal changes in the feature amount of the important area.

According to the present invention, it is possible to improve the accuracy of feature amount time-series data indicating temporal changes in feature amounts used for prediction processing of future usage status of wireless communication.

FIG. 1 is a block diagram illustrating a configuration example of a feature data generation device and a wireless usage prediction system according to an embodiment. FIG. 2 is a schematic explanatory diagram of a feature data generation method according to an embodiment. FIG. 2 is a detailed explanatory diagram of a feature data generation method according to an embodiment. 7 is a chart illustrating an example of the configuration of importance data of mesh/feature amount types according to an embodiment. FIG. 2 is an explanatory diagram of a first specific example of a feature data generation method according to an embodiment. FIG. 7 is an explanatory diagram of a second specific example of the feature amount data generation method according to an embodiment. FIG. 2 is an explanatory diagram for explaining an importance calculation method according to an embodiment.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a block diagram illustrating a configuration example of a feature data generation device and a wireless usage prediction system according to an embodiment. In FIG. 1, a wireless usage prediction system 1 includes a feature data generation device 10 and a wireless usage prediction data generation device 30.

The wireless usage prediction system 1 is a system that predicts the future usage status of wireless communication. The usage status of wireless communication includes, for example, the channel status of radio waves used for wireless communication (busy state and idle state of the radio channel), the received power of radio waves used for wireless communication, and the like.

The feature data generation device 10 generates feature time series data 130, based on input data 110 and 120, to be used in a process of predicting the future usage state of wireless communication. The feature amount time-series data 130 is time-series data indicating past temporal changes in feature amounts that are a factor of variation in the usage state of wireless communication.

The wireless usage prediction data generation device 30 generates usage status time series data 140 indicating temporal changes in the past usage status of radio waves used for wireless communication, and feature amount time series data 130 generated by the feature amount data generation device 10. Based on this, wireless usage prediction data 150 indicating a prediction of the future usage status of wireless communication is generated. The wireless usage prediction data generation device 30 uses, for example, machine learning to predict the future usage state of wireless communication.

For example, the wireless usage prediction data generation device 30 generates channel state time-series data (usage state time-series data ) 140 and the feature amount time series data 130 generated by the feature amount data generation device 10, the channel state prediction time indicates a prediction of a temporal change in the future channel state (busy state and idle state of the wireless channel). Sequence data (wireless usage prediction data) 150 is generated.

For example, the wireless usage prediction data generation device 30 generates received power time-series data (usage state time-series data) 140 that indicates temporal changes in past received power of radio waves used for wireless communication at a receiving point of the radio waves, Based on the feature amount time series data 130 generated by the feature amount data generation device 10, received power predicted value time series data (wireless Usage prediction data) 150 is generated.

Each function of the feature data generation device 10 and the wireless usage prediction data generation device 30 is performed by the feature data generation device 10 and the wireless usage prediction data generation device 30 using a computer such as a CPU (Central Processing Unit) and memory. It is realized by having hardware and having a CPU execute a computer program stored in memory. Note that the feature amount data generation device 10 and the wireless usage prediction data generation device 30 may be configured using general-purpose computer devices, or may be configured as dedicated hardware devices. For example, the feature data generation device 10 and the wireless usage prediction data generation device 30 may be configured using a server computer connected to a communication network such as the Internet. Further, each function of the feature amount data generation device 10 and the wireless usage prediction data generation device 30 may be realized by cloud computing. Further, the feature amount data generation device 10 and the wireless usage prediction data generation device 30 may be realized by a single computer, or the functions of the feature amount data generation device 10 and the wireless usage prediction data generation device 30 may be realized by a single computer. It may also be realized by being distributed over multiple computers.

(Feature amount data generation device)
The feature data generation device 10 includes an importance calculation section 11 and a feature time series data generation section 12.

The importance calculation unit 11 calculates usage status time-series data indicating temporal changes in the past usage status of radio waves used for wireless communication, and past temporal changes in each area of feature quantities that affect the propagation of the radio waves. The degree of importance of the feature amount of each area is calculated based on the time series data of all area feature amounts showing changes. The input data 110 is usage status time series data. The input data 120 is whole area feature amount time series data. The importance calculation unit 11 calculates the importance of the feature amount of each area based on the input data 110 (usage state time series data) and the input data 120 (all area feature amount time series data).

The all-area feature time-series data indicates past temporal changes in feature values that affect the propagation of radio waves used for wireless communication for each area for all areas serviced by the wireless usage prediction system 1. This is time series data. The feature amount that affects the propagation of radio waves used in wireless communication is, for example, the number of objects that affect the propagation of the radio waves. Objects that affect the propagation of the radio waves are, for example, pedestrians and vehicles. Note that objects such as pedestrians and vehicles that affect the propagation of radio waves are sometimes referred to as blockers. The feature amount that affects the propagation of radio waves used in wireless communication is, for example, the urban structure. The urban structure is related to buildings, such as the height and width of buildings, and roads, such as the width of roads.

The feature time series data generation unit 12 determines important areas based on the importance calculated by the importance calculation unit 11, extracts the feature values of the important areas from the input data 120 (all area feature time series data), The feature amount time series data 130 of the important area is generated, which indicates the past temporal changes in the important area of the feature amount. The feature amount time series data 130 of this important area is supplied to the wireless usage prediction data generation device 30. The wireless usage prediction data generation device 30 predicts the future usage state of wireless communication based on the usage state time series data 140 and the feature amount time series data 130 of important areas generated by the feature amount data generation device 10. The wireless usage prediction data 150 shown in FIG.

FIG. 2 is a schematic explanatory diagram of the feature data generation method according to the present embodiment. Although FIG. 2 shows the case where received power is predicted as an example of the prediction of the usage state of wireless communication, the same applies when predicting the channel state (busy state and idle state of the wireless channel). .

In FIG. 2, radio waves transmitted by a transmitting station 31 (transmission point) reach a radio wave sensor 32 (reception point) through various radio wave propagation paths 33. The radio wave sensor 32 receives radio waves transmitted by the transmitting station 31 (transmission point) and detects the reception strength of the received radio waves. As the reception strength, for example, received signal strength indicator (RSSI) is detected. The radio waves transmitted by the transmitting station 31 may directly reach the radio wave sensor 32, or may reach the radio wave sensor 32 after being reflected by the building 41.

Pedestrians 43 and vehicles 42 are present in the radio wave arrival range until the radio waves transmitted from the transmitting station 31 (transmission point) reach the radio wave sensor 32 (reception point). Pedestrians 43 and vehicles 42 that are present in the radio wave arrival range are objects that affect the propagation of the radio waves transmitted by the transmitting station 31. Therefore, the number of pedestrians 43 and vehicles 42 existing within the radio wave arrival range is a feature amount that is a factor in the fluctuation of the usage state of the radio waves transmitted by the transmitting station 31. For example, the number of pedestrians 43 and vehicles 42 existing within the radio wave arrival range becomes a factor that changes the channel state of the radio waves transmitted by the transmitting station 31 (the busy state and idle state of the radio channel). For example, the number of pedestrians 43 and vehicles 42 existing in the radio wave arrival range becomes a factor in the variation in the reception power of the radio waves transmitted by the transmitting station 31 at the radio wave sensor 32.

For this reason, it is preferable to extract the number of pedestrians 43 and vehicles 42 that are present in the radio wave arrival range among all the areas serviced by the wireless usage prediction system 1 as a feature quantity. However, even within the range of arrival of radio waves transmitted by the transmitting station 31, the degree of importance with respect to fluctuations in the state of use of the radio waves is considered to vary.

Therefore, in the feature amount data generation method according to the present embodiment, the importance level indicates how important the feature amount of each area is with respect to the usage status of radio waves transmitted by the transmitting station 31 (for example, channel status and received power). is determined, an important area is determined based on the degree of importance, and the determined important area is used as an area from which feature quantities (for example, the number of pedestrians 43 and vehicles 42 (number of blockers)) are to be extracted. Note that the degree of importance may be determined as a degree of importance indicating how important the feature amount of each area is for predicting the usage state (for example, channel state and received power) of radio waves transmitted by the transmitting station 31. .

As a result, the feature values (for example, the number of pedestrians 43 and vehicles 42) of areas that are particularly important (important areas) with respect to the usage status of radio waves transmitted by the transmitting station 31 (for example, channel status and received power) are extracted. Therefore, it is possible to efficiently extract the feature amount that is a factor of variation in the usage status of radio waves transmitted by the transmitting station 31.

As a result, temporal changes in feature quantities (for example, the number of blockers such as pedestrians 43 and vehicles 42) used for prediction processing of the future usage status of wireless communication (for example, channel status and received power) can be calculated. The accuracy of the feature amount time series data 130 shown can be improved, and this can contribute to improving the prediction accuracy of future usage conditions of wireless communication (for example, channel conditions and received power).

FIG. 3 is a detailed explanatory diagram of the feature data generation method according to this embodiment. Although FIG. 3 shows a case where reception power is predicted as an example of prediction of wireless communication usage status, the same applies to prediction of channel status (busy state and idle state of a wireless channel). .

The feature data generation method according to this embodiment will be described in detail with reference to FIG. 3. Note that hereinafter, the area may be referred to as a mesh.

Input data 110 and 120 are input to the feature data generation device 10 . The input data 110 is received power time series data (usage state time series data). The received power time-series data is time-series data that indicates a temporal change in the past received power of a radio wave used for wireless communication at a reception point of the radio wave. The received power time series data is composed of, for example, a set of time and received power intensity (RSSI). The received power time series data is generated using, for example, the RSSI detected by the radio wave sensor 32 installed at the reception point. The receiving point from which received power time series data is acquired is a location where future received power of wireless communication is desired to be predicted.

The input data 120 (120-1, . . . , 120-M, M is an integer of 1 or more) is time-series data of all area features. All area feature amount time series data is provided for each type of feature amount. For example, the input data 120-1 is all-area feature time series data regarding the first feature, and is used for wireless communication for each area of all the areas serviced by the wireless usage prediction system 1. This is time-series data showing past temporal changes in a first feature that affects the propagation of radio waves. For example, the input data 120-M is all-area feature time-series data regarding the M-th feature, which is used for wireless communication for each area of all the areas serviced by the wireless usage prediction system 1. This is time-series data showing past temporal changes in the M-th feature amount that affects the propagation of radio waves.

Received power time-series data (usage state time-series data) 140 is input to the wireless usage prediction data generation device 30 . The received power time series data 140 is similar to the received power time series data of the input data 110 described above.

(Step S1) The importance calculation unit 11 of the feature data generation device 10 calculates, for each of the M types of all-area feature time series data (input data 120-1, . . . , 120-M), Aggregating feature amounts for each area (mesh). As a result of this aggregation, feature amount time series data 200-1, . . . , 200-M of all meshes are generated for each of M types of feature amounts. For example, the importance calculation unit 11 aggregates the first feature amount for each time stamp for each mesh with respect to all area feature amount time series data (input data 120-1) regarding the first feature amount, and calculates the first feature amount for each time stamp for all meshes. For each mesh, time series data (first feature time series data of all meshes) 200-1 having the aggregate value of the first feature of each time stamp is generated. For example, the importance calculation unit 11 aggregates the M-th feature amount for each time stamp for each mesh with respect to all area feature amount time series data (input data 120-M) regarding the M-th feature amount, and calculates the M-th feature amount for each time stamp for all meshes. For each mesh, time-series data (M-th feature time-series data of all meshes) 200-M having the aggregate value of the M-th feature of each time stamp is generated.

(Step S2) The importance calculation unit 11 calculates the received power time-series data (input data 110) and the feature amount time-series data 200-1, . . . , 200-M of all meshes for each of the M types of feature amounts. Based on this, the importance of each feature of each mesh is calculated.

The importance calculation method will be explained. The importance calculation unit 11 calculates the wireless communication from the received power time series data (input data 110) and the feature value time series data 200-1, ..., 200-M of all meshes for each of M types of feature values. A machine learning model (prediction model) is trained to output a predicted value of future received power as time series data. The machine learning model is, for example, a random forest method. In learning the prediction model, the importance calculation unit 11 calculates the importance for each of the M types of feature amounts based on the contribution of each mesh to the learning. For each of the M types of feature amounts, the degree of contribution of each mesh exists for each time stamp. The importance calculation unit 11 calculates the importance for each of the M types of feature amounts from the contribution for a certain period (a certain number of timestamps) for each mesh. For example, the importance may be the sum of contributions for a certain period (a certain number of timestamps). For example, the importance may be the average value of the contribution for a certain period (a certain number of timestamps). For example, the maximum value of the contribution degrees for a certain period (a certain number of timestamps) may be set as the importance degree.

The importance calculation unit 11 generates importance data 210 of the mesh/feature amount type from the calculated importance. The mesh/feature type importance data 210 is sent to the feature time series data generation unit 12.

FIG. 4 shows a configuration example of the mesh/feature type importance data 210 according to this embodiment. As illustrated in FIG. 4, the mesh/feature type importance data 210 is for each mesh for each of M types of features (first feature, second feature, . . . M-th feature). This data has a degree of importance of (first mesh, second mesh, . . . , Nth mesh, where N is the total number of meshes).

The explanation returns to FIG. 3.
(Step S3) The feature time series data generation unit 12 creates important areas (important meshes) for each of the M types of features based on the mesh/feature type importance data 210 generated by the importance calculation unit 11. demand. In the example of FIG. 4, meshes whose importance is equal to or higher than a predetermined threshold (importance indicated by underlining in FIG. 4) are determined as important meshes. At this time, for a feature quantity for which there is no mesh with an importance level greater than or equal to a predetermined threshold, no important mesh exists (an important mesh cannot be found).

The feature amount time series data generation unit 12 extracts the aggregated value of the feature amount of the important mesh from the feature amount time series data 200-1, ..., 200-M of all meshes for M types of feature amounts. The feature time series data 130 of the important mesh (the x-th feature time series data 130-x of the important mesh, . . . , the y-th feature time series data 130-y of the important mesh) are generate. At this time, for a feature quantity for which there is no mesh with an importance level equal to or higher than a predetermined threshold value, no important mesh exists, and therefore, feature quantity time-series data of the important mesh is not generated for the feature quantity.

For example, since there is an important mesh for the first feature among M types of features, the feature time-series data generation unit 12 generates the first feature time-series data 200-1 of all the meshes from the first feature The aggregate value of one feature quantity is extracted, and first feature quantity time-series data 130-1 of the important mesh indicating the past temporal change of the important mesh of the first feature quantity is generated. On the other hand, for example, since there is no important mesh for the second feature among the M types of features, the second feature time series data of the important mesh for the second feature is not generated.

The feature data generation device 10 generates the feature time series data 130 of the important mesh generated by the feature time series data generation unit 12 (the x-th feature time series data 130-x of the important mesh, . . . The y-th feature amount time series data 130-y) is transmitted to the wireless usage prediction data generation device 30.

Note that the feature amount time series data 130 of the important mesh may be updated in accordance with (synchronized with) the prediction timing of the wireless usage prediction data generation device 30, or the prediction timing of the wireless usage prediction data generation device 30 may be updated. It may be updated asynchronously, for example, at regular intervals. The latest data of the input data 110 and 120 is input to the feature data generation device 10 in accordance with the update timing of the feature time series data 130 of the important mesh.

(Step S4) The latest received power time-series data (usage status time-series data) 140 is input to the wireless usage prediction data generation device 30 in accordance with the prediction timing. The wireless usage prediction data generation device 30 generates received power time-series data (usage state time-series data) 140 and important mesh feature time-series data 130 (x-th feature of the important mesh) received from the feature data generation device 10. Based on the time series data 130-x, . Received power predicted value time-series data (wireless usage predicted data) 150 indicating a predicted change in power is generated.

The wireless usage prediction data generation device 30 uses, for example, machine learning to predict future temporal changes in received power. In the learning stage, received power time-series data (usage state time-series data) 140 and feature time-series data 130 of important meshes (x-th feature time-series data of important meshes 130-x, . . . , important The received power time series data (usage state time series data) 140 and the feature time series data 130 of important meshes (the yth feature time series data of important meshes) are used as learning data. The machine learning model is configured to output the predicted value of future received power as time series data from the x feature time series data 130-x, ..., y feature time series data 130-y of the important mesh. Learn. The machine learning model is, for example, a neural network. At the prediction stage, received power time series data (usage state time series data) 140 and feature time series data 130 of important meshes (xth feature time series data 130-x, important mesh The y-th feature time series data 130-y) of the mesh is input as input data to a trained machine learning model, and the time series of predicted values of future received power is output from the machine learning model as a result of the input. Get data. The wireless usage prediction data generation device 30 generates received power predicted value time series data (wireless usage prediction data) 150 using time series data of future predicted received power values output from the machine learning model.

Note that the machine learning algorithm used in step S2 and the machine learning algorithm used in step S4 described above may be the same or different.

(Specific example 1 of feature data generation method)
FIG. 5 is an explanatory diagram of a first specific example of the feature amount data generation method according to the present embodiment. Although FIG. 5 shows a case where reception power is predicted as an example of prediction of the usage state of wireless communication, the same is true when predicting the channel state (busy state and idle state of the wireless channel). . In the first specific example of the feature data generation method, the feature is the number of blockers such as pedestrians 43 and vehicles 42 (number of blockers).

Input data 110 and 120 are input to the feature data generation device 10 . The input data 110 is received power time series data (usage state time series data).

The input data 120 is blocker information (entire area feature amount time series data). Blocker information is provided to pedestrians 43 and vehicles 42, which are objects that affect the propagation of radio waves used for wireless communication, for each area (mesh) for all areas (mesh) serviced by the wireless usage prediction system 1. This is time series data showing past temporal changes in the number of blockers (number of blockers). The blocker information is composed of, for example, a set of time and blocker position (latitude, longitude). The range (latitude, longitude) of each mesh is set in advance. The blocker information is generated using, for example, position information obtained by a mobile terminal such as a smartphone using a GPS (Global Positioning System) or the like. Further, the blocker information may be generated using connected terminal number data indicating the number of mobile terminals such as smartphones connected to the base station (number of connected terminals).

(Step S1) The importance calculation unit 11 of the feature amount data generation device 10 aggregates the number of blockers for each mesh with respect to the blocker information (input data) 120 for each time stamp, and calculates each time stamp for each mesh for all meshes. time series data (blocker number time series data for all meshes) 200 having a total value of the number of blockers is generated.

(Step S2) The importance calculation unit 11 calculates the importance of each mesh based on the received power time series data (input data 110) and the blocker number time series data 200 of all meshes. The importance calculation method is the same as the method described above. However, in this specific example 1, there is only one type of feature amount (only the number of blockers).

The importance calculation unit 11 generates mesh importance data 210 from the calculated importance. However, in this specific example 1, only mesh importance data 210 regarding the number of blockers is generated. The mesh importance data 210 is sent to the feature amount time series data generation unit 12.

(Step S3) The feature amount time series data generation unit 12 determines important areas (important meshes) based on the mesh importance data 210 generated by the importance calculation unit 11. The feature amount time series data generation unit 12 extracts the aggregated value of the number of blockers in important meshes from the time series data 200 on the number of blockers in all meshes, and generates time series data 130 on the number of blockers in important meshes. The blocker number time-series data 130 of important meshes is time-series data showing past temporal changes in the number of blockers for each important mesh.

The feature data generation device 10 transmits the important mesh blocker number time series data 130 generated by the feature time series data generation unit 12 to the wireless usage prediction data generation device 30.

(Step S4) The latest received power time-series data (usage status time-series data) 140 is input to the wireless usage prediction data generation device 30 in accordance with the prediction timing. The wireless usage prediction data generation device 30 makes predictions for wireless communication based on the received power time series data (usage state time series data) 140 and the important mesh blocker number time series data 130 received from the feature data generation device 10. Reception power predicted value time series data (wireless usage prediction data) 150 indicating a prediction of a temporal change in future reception power of the radio wave at the reception point of the radio wave to be used is generated. The method of generating received power predicted value time series data is the same as the method described above. However, in this specific example 1, there is only one type of feature amount time series data (only the number of blockers).

(Specific example 2 of feature data generation method)
FIG. 6 is an explanatory diagram of a second specific example of the feature data generation method according to the present embodiment. Although FIG. 6 shows a case where reception power is predicted as an example of prediction of the usage state of wireless communication, the same applies when predicting the channel state (busy state and idle state of the wireless channel). . In the second specific example of the feature amount data generation method, the feature amounts are micro information and macro information. The micro information is the number of blockers such as pedestrians 43 and vehicles 42 (number of blockers). Macro information is the urban structure.

Input data 110 and 120 are input to the feature data generation device 10 . The input data 110 is received power time series data (usage state time series data).

Blocker information 120-1 is input to the feature data generation device 10 as micro information. Additionally, satellite image time series data 120-2 is input to the feature amount data generation device 10 as macro information. The satellite image time series data 120-2 is time series data of satellite images (images taken from a satellite) including all the areas serviced by the radio usage prediction system 1. The satellite image has a resolution that allows analysis of urban structure. In this specific example 2, the satellite image is an image having a resolution that allows calculation of at least the area ratio of buildings (building area ratio) within the area.

(Step S1) The importance calculation unit 11 of the feature amount data generation device 10 aggregates the number of blockers for each mesh for each time stamp with respect to the blocker information (input data) 120-1, and calculates the number of blockers for each mesh for all meshes. Time series data (blocker number time series data for all meshes) 200-1 having a total value of the number of blockers in the time stamp is generated.

In addition, the importance calculation unit 11 calculates the building area ratio for each time stamp for each mesh with respect to the satellite image time series data (input data) 120-2, and calculates the building area ratio for each time stamp for each mesh for all meshes. time series data (building area ratio time series data of all meshes) 200-2 is generated.

(Step S2) The importance calculation unit 11 calculates the received power time series data (input data 110) and the feature quantity time series data of all meshes for each of two types of feature quantities (number of blockers, building area ratio) (of all meshes). The importance of each feature (number of blockers, building area ratio) of each mesh is calculated based on blocker number time series data 200-1 and building area ratio time series data 200-2 of all meshes. The importance calculation method is the same as the method described above. However, in this specific example 2, there are two types of feature amounts (number of blockers and building area ratio).

The importance calculation unit 11 generates importance data 210 of the mesh/feature amount type from the calculated importance. The mesh/feature amount type importance data 210 is data having the importance of each mesh for each of two types of feature amounts (number of blockers, building area ratio). The mesh/feature type importance data 210 is sent to the feature time series data generation unit 12.

(Step S3) The feature time series data generation unit 12 generates data for each of the two types of feature quantities (number of blockers, building area ratio) based on the importance data 210 of the mesh/feature type generated by the importance calculation unit 11. Find important areas (important mesh).

The feature value time series data generation unit 12 extracts the aggregated value of the number of blockers of the important mesh from the time series data 200-1 of the number of blockers of all meshes for the important mesh of the number of blockers, and generates the time series data 130 of the number of blockers of the important mesh. -1 is generated. The important mesh blocker number time series data 130-1 is time series data showing past temporal changes in the number of blockers for each important mesh of the number of blockers.

In addition, the feature time series data generation unit 12 extracts the building area ratio of the important mesh from the building area ratio time series data 200-2 of all meshes, and extracts the building area ratio of the important mesh from the building area ratio time series data 200-2 of the important mesh. Data 130-2 is generated. The building area ratio time-series data 130-2 of important meshes is time-series data showing past temporal changes in building area ratios for each important mesh of building area ratios.

Note that for a feature quantity for which a mesh of importance does not exist, since no important mesh exists, feature quantity time-series data of an important mesh is not generated for the feature quantity. For example, for the number of blockers in which an important mesh exists, time-series data 130-1 on the number of blockers in an important mesh is generated, but for the building area ratio in which an important mesh does not exist, time-series data 130-2 on the building area ratio in an important mesh. is not generated, etc.

The feature data generation device 10 generates feature time series data 130 of important meshes (blocker number time series data 130-1 of important meshes, building area ratio time series data of important meshes) generated by the feature time series data generation unit 12. 130-2) to the wireless usage prediction data generation device 30.

(Step S4) The latest received power time-series data (usage status time-series data) 140 is input to the wireless usage prediction data generation device 30 in accordance with the prediction timing. The wireless usage prediction data generation device 30 generates received power time series data (usage status time series data) 140 and feature amount time series data 130 of important meshes received from the feature amount data generation device 10 (blocker number time series of important meshes). Data 130-1, important mesh building area ratio time series data 130-2), reception indicating a prediction of the temporal change in future received power of radio waves used for wireless communication at the reception point of the radio waves. Power predicted value time series data (wireless usage predicted data) 150 is generated. The method of generating received power predicted value time series data is the same as the method described above. However, in this specific example 2, there are two types of feature quantity time series data (number of blockers and building area ratio).

(Importance calculation method)
The importance calculation method according to this embodiment will be explained. In learning a prediction model (machine learning model), the importance calculation unit 11 calculates the importance for each of the M types of feature amounts based on the contribution of each mesh to the learning. For example, a random forest method or a gradient boosting method can be used as a machine learning algorithm for a predictive model. The Random Forest method and the gradient boosting method are machine learning algorithms called "ensemble learning" that learn multiple decision trees (CART: Classification and Regression Trees) to construct an overall model. These machine learning algorithms can quantify and output the degree of contribution indicating how much the input feature amount has contributed to learning. As the degree of contribution, in CART, as shown in FIG. 7, the degree of error improvement ΔG before and after branching based on a certain feature amount is calculated by the following equation (1) using the Gini coefficient.

According to this embodiment, it is possible to accurately extract from a data science perspective the feature values that are a factor of fluctuations in the usage status of wireless communication (for example, channel status and received power), so that the future usage status of wireless communication can be accurately extracted. The effect of improving the accuracy of feature amount time series data indicating temporal changes in feature amounts used in prediction processing can be obtained. This makes it possible to contribute to improving the accuracy of predicting the future usage status of wireless communications.

Furthermore, according to the present embodiment, the feature amount time series data is composed of input data suitable for the prediction process of the future usage state of wireless communication, which contributes to reducing the number of dimensions of the input data to the prediction process. can do. Thereby, the amount of processing of the prediction process is reduced, and the effect of shortening the processing time can be obtained.

Furthermore, according to the present embodiment, the process can be carried out as long as there is usage state time series data (for example, received power time series data) and actual measurement data of feature quantities, so there is no need to prepare other data unrelated to prediction. .

Furthermore, as this can improve the overall quality of service in frequency sharing between different systems, for example, goal 9 of the Sustainable Development Goals (SDGs) led by the United Nations, ``Develop resilient infrastructure.'' It will be possible to contribute to "promoting sustainable industrialization and expanding innovation."

Although the embodiments of the present invention have been described above in detail with reference to the drawings, the specific configuration is not limited to these embodiments, and may include design changes without departing from the gist of the present invention.

Further, a computer program for realizing the functions of each device described above may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed. Note that the "computer system" here may include hardware such as an OS and peripheral devices.
Furthermore, "computer-readable recording media" refers to flexible disks, magneto-optical disks, ROMs, writable non-volatile memories such as flash memory, portable media such as DVDs (Digital Versatile Discs), and media built into computer systems. A storage device such as a hard disk.

Furthermore, "computer-readable recording medium" refers to volatile memory (for example, DRAM (Dynamic It also includes those that retain programs for a certain period of time, such as Random Access Memory).
Further, the program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in a transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line.
Moreover, the above-mentioned program may be for realizing a part of the above-mentioned functions. Furthermore, it may be a so-called difference file (difference program) that can realize the above-described functions in combination with a program already recorded in the computer system.

DESCRIPTION OF SYMBOLS 1...Radio usage prediction system, 10...Feature amount data generation device, 11...Importance calculation unit, 12...Feature amount time series data generation unit, 30...Wireless usage prediction data generation device

Claims (10)

A feature data generation device that generates feature time series data used in prediction processing of future usage status of wireless communication,
Usage status time series data showing temporal changes in the past usage status of radio waves used for the wireless communication, and whole-area features showing past temporal changes in each area of feature quantities that affect the propagation of the radio waves. an importance calculation unit that calculates the importance of the feature amount of each area based on the amount time series data;
An important area is determined based on the degree of importance, the feature amount of the important area is extracted from the all-area feature time series data, and the feature of the important area indicates a past temporal change of the feature amount of the important area. a feature amount time series data generation unit that generates amount time series data;
A feature data generation device comprising: The importance calculation unit includes:
training a machine learning model to output a predicted value of the future usage status of the wireless communication as time series data from the usage status time series data and the total area feature value time series data;
calculating the degree of importance from the degree of contribution of each area to the learning;
The feature data generation device according to claim 1. The feature quantity is the number of objects that affect the propagation of the radio waves,
The feature data generation device according to claim 1 or 2. The importance calculation unit calculates the characteristics of each area for each of the plurality of types of feature amounts based on the usage state time series data and the all area feature time series data regarding the plurality of types of the feature amounts. Calculate the importance of quantity,
The feature amount time series data generation unit generates feature amount time series data of an important area for each of the plurality of types of the feature amounts based on the importance of each of the plurality of types of the feature amounts.
The feature data generation device according to claim 1. The importance calculation unit includes:
Learning a machine learning model to output a predicted value of the future usage status of the wireless communication as time series data from the usage status time series data and the whole area feature value time series data regarding the plurality of types of the feature quantities. and
Calculating the degree of importance for each of the plurality of types of feature amounts from the degree of contribution each area has made to the learning;
The feature data generation device according to claim 4. The plurality of types of feature amounts are micro information and macro information,
The feature data generation device according to claim 4 or 5. A feature data generation device according to any one of claims 1 to 6;
A wireless usage prediction data generation device that generates wireless usage prediction data indicating a prediction of future usage status of wireless communication,
The wireless usage prediction data generation device generates usage status time-series data indicating temporal changes in past usage status of radio waves used for wireless communication, and feature amount time of important areas generated by the feature amount data generation device. generating the wireless usage prediction data based on the series data;
Wireless usage prediction system. The usage state time series data is data indicating past temporal changes in received power of the radio wave at the reception point of the radio wave,
The wireless usage prediction data is received power predicted value time series data indicating a prediction of a temporal change in the received power of the radio wave at the reception point of the radio wave.
The wireless usage prediction system according to claim 7. A feature data generation method executed by a feature data generation device that generates feature time series data used for prediction processing of future usage status of wireless communication, the method comprising:
The feature amount data generation device generates usage status time series data indicating temporal changes in the past usage status of radio waves used for the wireless communication, and past usage status data of each area of feature quantities that affect the propagation of the radio waves. an importance calculation step of calculating the importance of the feature amount of each area based on the time series data of all area feature amounts showing temporal changes;
The feature amount data generation device obtains an important area based on the importance level, extracts the feature amount of the important area from the all-area feature time series data, and calculates the past time of the important area of the feature amount. a feature time series data generation step of generating feature time series data of important areas showing changes in characteristics;
Feature data generation method including. A computer of a feature data generation device that generates feature time series data used for prediction processing of future usage status of wireless communication,
Usage status time series data showing temporal changes in the past usage status of radio waves used for the wireless communication, and whole-area features showing past temporal changes in each area of feature quantities that affect the propagation of the radio waves. an importance calculation step of calculating the importance of the feature amount of each area based on the amount time series data;
An important area is determined based on the degree of importance, the feature amount of the important area is extracted from the all-area feature time series data, and the feature of the important area indicates a past temporal change of the feature amount of the important area. a feature quantity time series data generation step for generating quantity time series data;
A computer program for running.

Images