Learning system, data selection device, learning method, data selection method, and program

Abstract

The purpose of the present disclosure is to provide a learning system capable of increasing an estimation accuracy of attenuation map in an environment such as an actual factory. A learning system according to the present disclosure comprises a simulation unit (11) that generates a plurality of attenuation simulation maps using a plurality of sight determination precisions relating to sight determination results within a simulation environment, a learning unit (12) that generates, for each sight determination precision, a learning model that updates an occluded region within each attenuation simulation map, a generation unit (13) that uses a sight determination result within an observation region to generate an attenuation map, an update unit (14) that inputs the attenuation map to each learning model to update an occluded region within the attenuation map, and a selection unit (15) that uses a result variation of the occluded region within the observation region between before and after the update to select one attenuation map from among a plurality of attenuation maps in each of which the occluded region has been updated.

Description

[Technical Field]

[0001] The present disclosure relates to a learning system, a data selection device, a learning method, a data selection method, and a non-transitory computer-readable medium. [Background technology]

[0002] In recent years, there has been a demand for improving the efficiency of manufacturing operations by establishing wireless network environments and digitizing operations at factories and other manufacturing sites. However, factories are densely packed with industrial machinery and products, and it is known that the radio wave propagation environment within a factory is different from that outdoors or in general work spaces. Therefore, it is necessary to properly understand the radio wave propagation environment within the factory before designing and building a wireless network within the factory.

[0003] Wireless tomography is a method for estimating the spatial distribution of signal attenuation within a three-dimensional space such as a factory. Wireless tomography may also be called Radio Tomographic Imaging (RTI). Wireless tomography irradiates a wireless signal within a three-dimensional space and uses the observation results. In this case, the spatial distribution of signal attenuation is called an attenuation map.

[0004] Non-Patent Document 1 discloses estimating a signal attenuation map in a three-dimensional space by performing radio tomography using LOS (Line Of Sight) / NLOS (Non Line Of Sight) discrimination. LOS / NLOS discrimination is to discriminate whether there is line of sight between a transmitting device and a receiving device communicating in a three-dimensional space. A line-of-sight state is called LOS, and a non-line-of-sight state is called NLOS.

[0005] Here, when the position or structure of an object blocking radio wave propagation is estimated in an actual factory or the like using the method disclosed in Non-Patent Document 1, reflected waves generated in the factory may become noise, reducing the estimation accuracy. Therefore, it has been considered to reproduce radio wave propagation in an actual factory environment through simulation, and to generate an attenuation map in the factory by applying the method disclosed in Non-Patent Document 1 to the simulation results. More specifically, it has been considered to generate a learning model for generating an attenuation map from the simulation results. [Prior art documents] [Non-patent literature]

[0006] [Non-Patent Document 1] Takahiro Matsuda and four others, "Binary Radio Tomography Based on LOS / NLOS Discrimination," Institute of Electronics, Information and Communication Engineers, 2022-09 Summary of the Invention [Problem to be solved by the invention]

[0007] However, the accuracy of LOS / NLOS determination in a simulation environment often differs from that in an actual factory environment. Specifically, the location and structure of obstructions can be clearly determined in a simulation environment, resulting in extremely high LOS / NLOS determination accuracy. On the other hand, when estimating the location of obstructions or the structure of obstructions with high attenuation in an actual factory environment, it is not possible to determine the location and structure of the obstructions in advance. Therefore, the accuracy of LOS / NLOS determination in an actual factory environment is lower than that in a simulation environment. Therefore, for example, even if a learning model generated for estimating an attenuation map in a simulation environment is used to estimate an attenuation map in an actual factory environment, the desired estimation accuracy cannot be obtained.

[0008] In view of the above-mentioned problems, one of the objectives of the present disclosure is to provide a learning system, a data selection device, a learning method, a data selection method, and a non-transitory computer-readable medium that can improve the estimation accuracy of attenuation maps in actual factory environments, etc. [Means for solving the problem]

[0009] A learning system according to a first aspect of the present disclosure includes a simulation unit that generates multiple attenuation simulation maps showing the spatial distribution of signal attenuation using multiple visibility assessment accuracies related to visibility assessment results within a simulation environment, a learning unit that generates a learning model that updates occluded areas in the attenuation simulation map in association with each visibility assessment accuracy, a generation unit that generates an attenuation map showing the spatial distribution of signal attenuation using the visibility assessment results, an update unit that inputs the attenuation maps to each of the learning models to update the occluded areas in the attenuation maps, and a selection unit that selects one attenuation map from the multiple attenuation maps with updated occluded areas based on the results of changes in the occluded areas in the observation area before and after the update.

[0010] A data selection device according to a second aspect of the present disclosure includes a generation unit that generates an attenuation map showing the spatial distribution of signal attenuation using a visibility assessment result within an observation area; an update unit that generates multiple attenuation simulation maps showing the spatial distribution of signal attenuation using multiple visibility assessment accuracies related to the visibility assessment result within a simulation environment, and inputs the attenuation map into a learning model that updates occluded areas in the attenuation simulation map, generated in association with each visibility assessment accuracy; and a selection unit that selects one attenuation map from the multiple attenuation maps with updated occluded areas based on the fluctuation results of the occluded areas in the observation area before and after the update.

[0011] A learning method according to a third aspect of the present disclosure generates multiple attenuation simulation maps showing the spatial distribution of signal attenuation using multiple visibility assessment accuracies related to visibility assessment results within a simulation environment, generates learning models that update occluded areas within the attenuation simulation maps in association with each visibility assessment accuracy, generates attenuation maps showing the spatial distribution of signal attenuation using the visibility assessment results within an observation area, inputs the attenuation maps into each of the learning models to update the occluded areas within the attenuation maps, and selects one attenuation map from the multiple attenuation maps with updated occluded areas based on the fluctuation results of the occluded areas within the observation area before and after the update.

[0012] A data selection method according to a fourth aspect of the present disclosure generates an attenuation map showing the spatial distribution of signal attenuation using a visibility assessment result within an observation area, generates multiple attenuation simulation maps showing the spatial distribution of signal attenuation using multiple visibility assessment accuracies related to the visibility assessment result within a simulation environment, inputs the attenuation maps into a learning model that updates occluded areas in the attenuation simulation maps, generated in association with each visibility assessment accuracy, and updates the occluded areas in the attenuation maps, and selects one attenuation map from the multiple attenuation maps with updated occluded areas based on the fluctuation results of the occluded areas in the observation area before and after the update.

[0013] A program according to a fifth aspect of the present disclosure causes a computer to perform the following steps: generate multiple attenuation simulation maps showing the spatial distribution of signal attenuation using multiple visibility assessment accuracies related to visibility assessment results within a simulation environment; generate a learning model that updates occluded areas within the attenuation simulation maps in association with each visibility assessment accuracy; generate attenuation maps showing the spatial distribution of signal attenuation using visibility assessment results within an observation area; input the attenuation maps into each of the learning models to update the occluded areas within the attenuation maps; and select one attenuation map from the multiple attenuation maps with updated occluded areas based on the fluctuation results of the occluded areas within the observation area before and after the update. [Effects of the Invention]

[0014] The present disclosure provides a learning system, a data selection device, a learning method, a data selection method, and a non-transitory computer-readable medium that can improve the estimation accuracy of attenuation maps in actual factory environments, etc. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 1 is a configuration diagram of a learning system according to a first embodiment. [Figure 2] FIG. 10 is a diagram showing a flow of a learning process according to the first embodiment. [Figure 3] FIG. 10 is a configuration diagram of a learning system according to a second embodiment. [Figure 4] FIG. 10 is a diagram showing a process flow of a learning phase according to the second embodiment. [Figure 5] FIG. 10 is a diagram showing an attenuation map according to the second embodiment. [Figure 6] FIG. 10 is a diagram showing an attenuation map according to the second embodiment. [Figure 7] FIG. 10 is a diagram showing an attenuation map according to the second embodiment. [Figure 8] FIG. 10 is a diagram showing a processing flow of an inference phase according to the second embodiment. [Figure 9]FIG. 10 is a diagram illustrating measurement of a radio propagation environment in an observation area according to a second embodiment. [Figure 10] 1A and 1B are configuration diagrams of a learning device or a data selection device according to respective embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0016] (Embodiment 1) Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The learning system 10 in FIG. 1 includes a simulation unit 11, a learning unit 12, a generation unit 13, an update unit 14, and a selection unit 15. Each functional block constituting the learning system 10 may be included in a single device or may be distributed across two or more devices. The device constituting the learning system 10 may be a computer device that operates by a processor executing a program stored in a memory. The computer device may be, for example, a communication device or a control device, or may also be a server device.

[0017] The simulation unit 11, the learning unit 12, the generation unit 13, the update unit 14, and the selection unit 15 may be software or modules whose processes are performed by a processor executing a program stored in a memory. Alternatively, the simulation unit 11, the learning unit 12, the generation unit 13, the update unit 14, and the selection unit 15 may be hardware such as a circuit or a chip. The processes performed in each functional block may be performed by a single processor executing a program stored in a memory, or may be distributed and executed among two or more processors.

[0018] The simulation unit 11 generates multiple attenuation simulation maps showing the spatial distribution of signal attenuation using multiple line-of-sight assessment accuracies related to line-of-sight assessment results within the simulation environment. The simulation environment may be, for example, an environment in which a space similar to an actual observation area, such as a factory, is reproduced on a computer device. The line-of-sight assessment results may be determined based on a simulation observation position within the simulation environment, or may be determined using received power, reception direction, and transmission source direction. The simulation observation position may be any position within the simulation environment.

[0019] The visibility determination result is, for example, a determination result of whether the measurement position in the simulation environment is visible or not, based on the simulation observation position. For example, the simulation observation position may be a position that simulates the position of a signal transmission device in the observation area, and the measurement position may be a position that simulates the position of a signal reception device in the observation area. Alternatively, the simulation observation position may be a position that simulates the position of a signal reception device in the observation area, and the measurement position may be a position that simulates the position of a signal transmission device in the observation area. The determination result of whether the measurement position is visible or not may be referred to as, for example, LOS / NLOS determination.

[0020] The LOS / NLOS determination based on the simulation observation position may be performed based on the position of an obstruction that blocks signal propagation and is placed in the simulation environment. For example, if an obstruction is placed on the line connecting the simulation observation position and the measurement position, it may be determined to be NLOS, and if no obstruction is placed on the line, it may be determined to be LOS. Furthermore, the LOS / NLOS determination may be performed based on whether an obstruction is placed in an area defined as a radio wave propagation path between the simulation observation position and the measurement position, not limited to the line connecting the simulation observation position and the measurement position.

[0021] Furthermore, the LOS / NLOS determination may take into account the material of obstacles placed in the simulation environment. For example, if an obstacle placed on a line connecting the simulation observation position and the measurement position is made of metal, it may be determined to be NLOS, and if the obstacle placed on the line is made of a material that allows radio waves to pass through it, it may be determined to be LOS.

[0022] Furthermore, in determining whether an LOS or NLOS condition exists, if the difference between the straight-line distance between the simulation observation position and the measurement position and the path of the reflected wave is extremely small, the condition may be determined as LOS even if an obstruction is located on the line connecting the simulation observation position and the measurement position. If the difference between the straight-line distance between the simulation observation position and the measurement position and the path of the reflected wave is greater than a predetermined value and an obstruction is located on the line connecting the simulation observation position and the measurement position, the condition may be determined as NLOS. The difference between the straight-line distance between the simulation observation position and the measurement position and the path of the reflected wave required to determine whether an LOS or NLOS condition exists may be predetermined.

[0023] The line-of-sight determination accuracy may be a value indicating the accuracy of LOS / NLOS determination. The accuracy of LOS / NLOS determination may be, for example, a value indicating the proportion of cases where a situation should actually be determined as LOS but is erroneously determined as NLOS, or a value indicating the proportion of cases where a situation should actually be determined as NLOS but is erroneously determined as LOS. In a simulation environment, multiple line-of-sight determination accuracies can be set. For example, the determination accuracy when no LOS / NLOS determination error occurs may be set to 100%, and the determination accuracy may be decreased as the proportion of LOS / NLOS determination errors increases.

[0024] The attenuation simulation map may be image data that indicates the degree or amount of attenuation of a radio signal by different colors. The different colors may be indicated by values ​​set for each of multiple pixels that make up the image data. For example, 8-bit values ​​for R (Red), G (Green), and B (Blue) may be set for each pixel. Alternatively, when the grayscale color set for each pixel is indicated by an 8-bit value, the area with the highest attenuation may be set to 255, which indicates white, and the area with the lowest attenuation may be set to 0, which indicates black. The area with the highest attenuation may be referred to as an area that does not transmit radio waves. The area with the lowest attenuation may be referred to as an area that transmits radio waves without attenuation.

[0025] The degree of attenuation of a wireless signal may be, for example, a value indicating the ratio of the strength of a signal received at a measurement position to the strength of a signal transmitted from a simulation observation position. In other words, the greater the amount of attenuation, the greater the degree of attenuation. The attenuation simulation map may be generated using the LOS / NLOS determination result using, for example, the method disclosed in Non-Patent Document 1, or other methods.

[0026] The learning unit 12 generates a learning model that updates the occluded areas in the attenuation simulation map in association with each visibility assessment accuracy. In the attenuation simulation map, it is assumed that an obstruction exists in a location where the degree of attenuation or the amount of attenuation is greater than a predetermined value, and the area where the obstruction exists is considered to be an obstructed area. The learning model may be, for example, a learning model that uses the attenuation simulation map as input and updates the occluded areas in the attenuation simulation map based on the shape of the obstruction, etc. When multiple visibility assessment accuracies are used in the simulation unit 11, attenuation simulation maps are generated for the visibility assessment results to which each visibility assessment accuracy is applied. In other words, the visibility assessment accuracies and the attenuation simulation maps are associated. Furthermore, a learning model is generated for each attenuation simulation map. This associates the visibility assessment accuracies with the learning models.

[0027] For example, the learning model may update a location in the attenuation simulation map where no occlusion is assumed to exist to a location where an occlusion is assumed to exist, or may update a location where an occlusion is assumed to exist to a location where an occlusion is not present.

[0028] The generation unit 13 generates an attenuation map indicating the degree or amount of attenuation of wireless signals using the line-of-sight assessment result within the observation area. The observation area is an actual three-dimensional space, such as a factory, where the radio wave propagation environment or attenuation state is to be measured. The line-of-sight assessment result may be determined based on the observation position within the observation area, or may be determined using the received power, the receiving direction, and the transmitting source direction. The attenuation simulation map indicates the spatial distribution of signal attenuation within the simulation environment, while the attenuation map indicates the spatial distribution of signal attenuation within an actual three-dimensional space.

[0029] The observation position may be the position of a signal transmitting device in the observation area, and the measurement position may be the position of a signal receiving device in the observation area. Alternatively, the observation position may be the position of a signal receiving device in the observation area, and the measurement position may be the position of a signal transmitting device in the observation area. The observation position may be, for example, the position of a wireless device that transmits a wireless signal within the observation area. The wireless device may be, for example, a base station or an AP (Access Point) used as a master device in a wireless LAN (Local Area Network). The method for generating the attenuation map using the visibility determination result may be the same as the method for generating the attenuation simulation map in the simulation unit 11, or a method different from that used by the simulation unit 11 may be used.

[0030] The update unit 14 inputs the attenuation map into each learning model generated by the learning unit 12, and updates the occlusion regions in the attenuation map. The attenuation map with the updated occlusion regions is output for each learning model.

[0031] The selection unit 15 selects one attenuation map from the plurality of attenuation maps in which the occluded areas have been updated based on the result of the change in the occluded areas in the observation area before and after the update. The result of the change in the occluded areas may be represented, for example, by the area of ​​the area updated from an occluded area to a non-occluded area and the area updated from a non-occluded area to an occluded area before and after the update. Alternatively, the result of the change in the occluded area may be represented by the number of pixels at the location where the change in the occluded area occurred in the image data. The change in the occluded areas in the attenuation maps before and after the update may be, for example, an area determined to be an incorrect area in the inference of each learning model.

[0032] For example, the selection unit 15 may select an attenuation map in which the difference between the occluded areas in the observation area before and after the update is the smallest as a result of the variation, or may select an attenuation map in which the difference between the occluded areas in the observation area before and after the update is the largest as a result of the variation.

[0033] Next, the flow of the learning process executed in the learning system according to the first embodiment will be described with reference to FIG.

[0034] First, the simulation unit 11 generates multiple attenuation simulation maps showing the degree of attenuation of wireless signals using multiple visibility assessment accuracies related to visibility assessment results based on simulation observation positions within the simulation environment (S11). Next, the learning unit 12 generates a learning model that updates the obstructed areas in the attenuation simulation maps in association with each visibility assessment accuracy (S12). Next, the generation unit 13 generates an attenuation map showing the degree of attenuation of wireless signals using the visibility assessment results based on observation positions within the observation area (S13).

[0035] Next, the update unit 14 inputs the attenuation maps into each learning model to update the occluded areas in the attenuation maps (S14). Next, the selection unit 15 selects one attenuation map from the plurality of attenuation maps whose occluded areas have been updated, based on the result of the change in the occluded areas in the observation area before and after the update (S15).

[0036] As described above, the learning system 10 according to the first embodiment uses multiple visibility assessment accuracies to generate a learning model associated with each visibility assessment accuracy. Furthermore, the learning system 10 selects one attenuation map from the attenuation maps updated using each learning model as the attenuation map representing the position of the obstruction in the actual observation area based on the result of the change in the obstructed area. This allows the learning system 10 to compensate for the difference between the visibility assessment accuracies in the simulation environment and those in the actual observation area and select the optimal attenuation map.

[0037] (Embodiment 2) Next, an example of the configuration of a learning system will be described using Fig. 3. The learning system in Fig. 3 includes a learning device 20 and a data selection device 30. The learning device 20 and the data selection device 30 may be computer devices that operate by a processor executing a program stored in a memory.

[0038] 3 shows that the simulation unit 11 and learning unit 12 described in FIG. 1 are arranged in a learning device 20, and the generation unit 13, update unit 14, and selection unit 15 are arranged in a data selection device 30. FIG. 3 shows an example of a distributed arrangement of the simulation unit 11, learning unit 12, generation unit 13, update unit 14, and selection unit 15, and the respective functions may be distributed in an arrangement different from that shown in FIG.

[0039] In FIG. 3, the learning device 20 executes a learning phase in which a learning model is generated, and the data selection device 30 executes an inference phase in which data acquired from an actual observation area is applied to the learning model.

[0040] Here, the flow of processing in the learning phase executed by the learning device 20 will be described with reference to FIG. 4. First, the simulation unit 11 simulates a wireless propagation environment in a simulation environment (S21). The simulation environment may be, for example, an environment that simulates a closed space inside a building such as a factory. In addition, the placement of shields that block the propagation of wireless signals in the simulation environment is also simulated. For example, the simulation unit 11 generates simulation environments with various shielding patterns by changing the position, material, structure, size, etc. of the shields in the simulation environment. In addition, the simulation unit 11 may generate simulation environments with a pattern in which the shields move, a pattern in which the shields are fixed, or a pattern in which moving and fixed shields are mixed.

[0041] Furthermore, the simulation unit 11 determines the position of a transmitting device that transmits a wireless signal set at an arbitrary position within the simulation environment as a simulation observation position. The simulation unit 11 determines the value of received power when a receiving device receives a wireless signal transmitted from the transmitting device at an arbitrary position within the simulation environment, taking into account the position of an obstruction. In other words, the simulation unit 11 may generate a received power distribution within the simulation environment as a wireless propagation environment.

[0042] Next, the simulation unit 11 performs LOS / NLOS determination based on the simulation observation position in the simulation environment (S22). Here, the simulation unit 11 has a plurality of LOS / NLOS determination accuracies, and performs LOS / NLOS determination in the simulation environment for each LOS / NLOS determination accuracy.

[0043] The LOS / NLOS determination may be made based on whether an obstruction is located on the line between the transmitting device and the receiving device, and further based on the material or structure of the obstruction. That is, the LOS / NLOS determination may be made to determine whether the receiving device can receive the direct wave transmitted from the transmitting device. For example, if the receiving device can receive the direct wave from the transmitting device, the determination is LOS, and if the receiving device cannot receive the direct wave, the determination is NLOS. Furthermore, the LOS / NLOS determination may be made taking into account the influence of reflected waves received by the receiving device. For example, a location where the received power value is greater than a predetermined value may be determined as LOS, and a location where the received power value is less than the predetermined value may be determined as NLOS.

[0044] LOS / NLOS determination is a determination of whether the receiving device is visible from the simulated observation position. Therefore, in a simulation environment where the location, material, and structure of obstructions, the location of the transmitting device, and the location of the receiving device are known, the accuracy of LOS / NLOS determination can be 100%. However, when estimating the location and structure of obstructions using the received power distribution in an actual observation area, the value of the received power is used. Here, in an actual observation area, reflected waves may arrive at locations that are not reached by direct waves from the observation position. Therefore, locations that are not reached by direct waves from the observation position cannot be seen from the observation position, and therefore, when a wireless signal with a higher received power than the specified level is received, it may be determined to be LOS even though it should be determined to be NLOS. Thus, when LOS / NLOS determination is performed using the received power distribution of wireless signals in an actual observation area, a location that should be determined to be NLOS may be determined to be LOS due to the influence of reflected waves. Therefore, the accuracy of LOS / NLOS determination in an actual observation area is generally not 100%.

[0045] Therefore, the simulation unit 11 may perform LOS / NLOS determination with multiple determination accuracies by taking into account the influence of reflected waves. The multiple determination accuracies may be set to values ​​less than 100%, such as 99%, 98%, and 90%, for example.

[0046] The simulation unit 11 performs LOS / NLOS determination using a plurality of LOS / NLOS determination accuracies, thereby obtaining LOS / NLOS determination results for each determination accuracy.

[0047] Next, the simulation unit 11 performs radio tomography to generate an attenuation simulation map (S23). The attenuation simulation map and the attenuation map may be referred to as an SLF (Spatial Loss Field). Performing radio tomography may be referred to as performing RTI. Generating the attenuation simulation map may be referred to as estimating the attenuation simulation map.

[0048] Here, an example of an attenuation simulation map generated by the simulation unit 11 will be described with reference to FIG. 5. In the attenuation simulation map of FIG. 5, the shaded areas indicate areas where wireless signals do not propagate due to the influence of obstructions. Areas other than the shaded areas indicate areas where wireless signals propagate. In other words, in FIG. 5, it is estimated that obstructions exist in the shaded areas. In the attenuation simulation map of FIG. 5, for example, the simulation observation position may be set around the center of FIG. 5. For ease of explanation, the attenuation simulation map of FIG. 5 is shown using shaded and unshaded areas. An actual attenuation simulation map may be, for example, image data composed of multiple pixels, and each pixel may be represented by a grayscale color using an 8-bit value. Furthermore, in FIG. 5, the boundary between the shaded area and the other area is clearly indicated by a straight line or a curve, but in an actual attenuation simulation map, the boundary may be represented as an area with a blurred color.

[0049] The simulation unit 11 generates an attenuation simulation map for each LOS / NLOS determination result with different determination accuracy.

[0050] Returning to Fig. 4, the learning unit 12 performs learning to update the occluded area for each attenuation simulation map (S24). For example, the learning unit 12 may generate a learning model that updates the occluded area in the attenuation simulation map using the shape of the occluding object as training data or correct answer data. For example, when the attenuation simulation map is input to the learning model that updates the occluded area in the attenuation simulation map, the occluded area may be updated to approximate the shape of the occluding object, as shown in Fig. 6.

[0051] Furthermore, the lower the LOS / NLOS determination accuracy, the more likely it is that erroneous LOS / NLOS determination results will occur. For example, as the number of erroneous LOS / NLOS determination results increases, an area other than the occluded area may exist within the occluded area, as shown in area A in Figure 7, resulting in a so-called occluded area being missing. In this way, when updating an attenuation simulation map generated based on a determination result with low LOS / NLOS determination accuracy, a learning model may be constructed to repair the missing area, assuming that there is a missing area in the occluded area.

[0052] In other words, a learning model for updating an attenuation simulation map generated based on higher LOS / NLOS determination accuracy will tend to make the boundaries of occluded areas clearer. Making the boundaries of occluded areas clearer means removing noise that exists near the boundaries. Also, a learning model for updating an attenuation simulation map generated based on lower LOS / NLOS determination accuracy will tend to make the boundaries of occluded areas clearer and to repair missing areas of occluded areas. As the LOS / NLOS determination accuracy decreases, the learning model will tend to repair missing areas of occluded areas.

[0053] Next, the flow of processing in the inference phase in the data selection device 30 will be described with reference to Fig. 8. First, the generation unit 13 acquires the radio propagation environment in the observation area (S31). Here, measurement of the radio propagation environment in the observation area will be described with reference to Fig. 9.

[0054] FIG. 9 shows that a moving wireless receiving device 50 receives a wireless signal transmitted from a wireless transmitting device 40 within an observation area. The observation area may be, for example, a three-dimensional space having a finite area, such as a factory or a room, with a wall separating the observation area from the external environment. The wireless transmitting device 40 may be, for example, a base station or an access point. The wireless receiving device 50 may be, for example, a smartphone terminal mounted on a vehicle, an IoT (Internet of Things) terminal, or other communication device. Alternatively, a person may walk while holding the wireless receiving device 50 and receive a wireless signal from the wireless transmitting device 40 while moving.

[0055] The shields 61 and 62 may be, for example, objects such as metal that do not transmit radio waves. Also, Fig. 9 shows a state in which the wireless transmitting device 40 emits radio waves 70. The radio waves 70 represent direct waves received by the wireless receiving device 50. The wireless receiving device 50 may receive reflected waves along with the radio waves 70, which are direct waves. Alternatively, the wireless receiving device 50 may receive only reflected waves without receiving the radio waves 70, which are direct waves.

[0056] The wireless receiving device 50 receives a wireless signal and measures its received power. Since the wireless receiving device 50 receives wireless signals while moving, it measures the received power at any location within its observation area. The generating unit 13 acquires a measurement result that associates the received power measured by the wireless receiving device 50 with the location at which the received power was measured. The measurement result that associates the received power measured by the wireless receiving device 50 with the location at which the received power was measured may indicate a wireless propagation environment. The wireless propagation environment may be referred to as a received power distribution. The wireless receiving device 50 may identify its current location from, for example, a moving speed, a moving time, a moving direction, or the like, or may identify its current location using a global positioning system (GPS) or a sensor. The generating unit 13 may acquire the measurement result from the wireless receiving device 50 via a network. Alternatively, the generating unit 13 may acquire information measured by the wireless receiving device 50 offline. It is also assumed that the generating unit 13 stores the location of the wireless transmitting device 40 in advance.

[0057] Returning to FIG. 8 , next, the generation unit 13 performs LOS / NLOS determination using the received power distribution indicating the received power measured by the wireless receiving device 50 and the position where the received power was measured (S32). The generation unit 13 does not have information indicating the position, material, structure, etc. of an obstruction present in the observation area. Therefore, the generation unit 13 performs LOS / NLOS determination using the received power distribution of the wireless signal. For example, a position where the received power is higher than a predetermined value may be determined as LOS, and a position where the received power is lower than the predetermined value may be determined as NLOS. Alternatively, if the received power is higher than the surrounding received power, it may be determined as LOS, and if it is lower than the surrounding received power, it may be determined as NLOS. The LOS / NLOS determination method using the received power distribution is not limited to this, and various methods may be used.

[0058] Next, the generating unit 13 performs radio tomography on the LOS / NLOS determination result to generate an attenuation map (Spatial Loss Field: SLF) (S33).

[0059] Next, the update unit 14 applies each learning model generated in the learning phase to the generated attenuation map to generate an attenuation map with updated occlusion regions (S34). In other words, the update unit 14 inputs the generated attenuation map to each learning model and obtains the attenuation map with updated occlusion regions output from each learning model.

[0060] Next, the selection unit 15 selects one attenuation map that most accurately indicates the position of the obstructing object in the observation area from the plurality of attenuation maps whose obstructing areas have been updated (S35). Here, the selection process of the attenuation map will be described.

[0061] It is desirable that the selector 15 selects, as the updated attenuation map, an updated attenuation map using a learning model associated with a determination accuracy equivalent to the LOS / NLOS determination accuracy within the observation area. However, the selector 15 cannot grasp the LOS / NLOS determination accuracy within the observation area. Therefore, the selector 15 needs to estimate and select an updated attenuation map using a learning model associated with a determination accuracy equivalent to the LOS / NLOS determination accuracy within the observation area.

[0062] For example, a case will be described in which it is assumed that the LOS / NLOS determination accuracy within the observation area is 90%, and the update unit 14 updates the attenuation map using a learning model associated with a determination accuracy of 100%.

[0063] Because the wireless transmission device 40 does not move throughout the entire observation area, the received power distribution generated based on the measurement results of the wireless transmission device 40 also includes unmeasured areas. In such cases, the unmeasured areas are treated as areas determined to be NLOS, and the attenuation map shows the unmeasured areas as areas where obstructions exist. Here, a learning model associated with a determination accuracy of 100% has a stronger tendency to reshape areas of obstructions containing noise to match the shape of the obstructions. Therefore, among areas where obstructions exist, areas that do not match the shape of the obstructions are likely to be updated to areas where no obstructions exist. Therefore, if an attenuation map generated based on a determination result with an LOS / NLOS determination accuracy of 90% is updated using a learning model associated with a determination accuracy of 100%, it is estimated that the areas with obstructions will be significantly reduced.

[0064] Here, we will explain the case where an attenuation map generated based on a determination result with a LOS / NLOS determination accuracy of 90% is updated using a learning model associated with a determination accuracy lower than 100%.

[0065] As the accuracy associated with the learning model decreases, the tendency to update missing parts of the occluded region to occluded regions increases. Therefore, it is estimated that the learning model associated with a lower accuracy value will have more occluded regions in the updated attenuation map.

[0066] However, learning models associated with low judgment accuracy values ​​assume low reliability for LOS / NLOS judgment results. Therefore, based on the LOS / NLOS judgment results, the values ​​of each pixel that make up the updated attenuation map, and the range of values ​​that each pixel can take (for example, 0 to 255 in an 8-bit system), will be closer to the median. This means that when the reliability of the LOS / NLOS judgment results is high, for example, each pixel can be set to a value far from the median, making it possible to clearly indicate whether or not it is an occlusion. However, when the reliability of the LOS / NLOS judgment results is low, it becomes difficult to clearly indicate whether or not it is an occlusion, and each pixel is set to a value close to the median, which can make the determination of whether or not it is an occlusion ambiguous. Therefore, it is expected that the lower the judgment accuracy of a learning model associated with a value, the smaller the occlusion area in the updated attenuation map will be.

[0067] Therefore, taking into consideration the increase in the area of ​​the obstruction and the decrease in the area of ​​the obstruction, the selector 15 selects an attenuation map based on the change in the area of ​​the obstruction before and after the update. Specifically, the selector 15 may select an attenuation map in which the ratio of the area of ​​the obstruction after the update to the area of ​​the obstruction before the update is the highest, as the attenuation map updated by the learning model associated with a determination accuracy similar to the determination accuracy of LOS / NLOS in the observation area.

[0068] Alternatively, the selector 15 may select an attenuation map using a value obtained by subtracting the number of pixels representing the area of ​​the obstruction after the update from the number of pixels representing the area of ​​the obstruction before the update. In this case, the selector 15 may select the attenuation map with the largest value after the subtraction as the attenuation map updated by the learning model associated with the same level of determination accuracy as the LOS / NLOS determination accuracy in the observation area.

[0069] As described above, the learning system according to the second embodiment selects an attenuation map based on the fluctuation of the occluded area in the attenuation map updated by the learning model when the accuracy of LOS / NLOS determination obtained from the received power distribution in the observation area is unknown. This allows the data selection device 30 to select an appropriate attenuation map that indicates the status of the observation area. As a result, a user of the learning system can appropriately position wireless devices in the observation area using an attenuation map that appropriately indicates the status of the observation area.

[0070] Fig. 10 is a block diagram showing an example configuration of the communication device 10. Referring to Fig. 10, the communication device 10 includes a network interface 1201, a processor 1202, and a memory 1203. The network interface 1201 is used to communicate with a network node (e.g., eNB, MME, P-GW). The network interface 1201 may include, for example, a network interface card (NIC) conforming to the IEEE 802.3 series. Here, eNB stands for evolved Node B, MME stands for Mobility Management Entity, and P-GW stands for Packet Data Network Gateway. IEEE stands for Institute of Electrical and Electronics Engineers.

[0071] The processor 1202 reads and executes software (computer programs) from the memory 1203 to perform the processing of the communication device 10 described using the flowcharts in the above-described embodiments. The processor 1202 may be, for example, a microprocessor, an MPU, or a CPU. The processor 1202 may include multiple processors.

[0072] The memory 1203 is configured by a combination of volatile memory and non-volatile memory. The memory 1203 may include storage located remotely from the processor 1202. In this case, the processor 1202 may access the memory 1203 via an I / O (Input / Output) interface (not shown).

[0073] 10, the memory 1203 is used to store software modules. The processor 1202 reads and executes these software modules from the memory 1203, thereby performing the processing of the communication device 10 described in the above embodiment.

[0074] As described with reference to FIG. 10, each of the processors included in the communication device 10 in the above-described embodiment executes one or more programs including a set of instructions for causing a computer to perform the algorithm described with reference to the drawings.

[0075] In the above example, the program can be stored and supplied to a computer using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic storage media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical storage media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / Ws, and semiconductor memories (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, and RAMs (Random Access Memory)). The program may also be supplied to a computer by various types of transitory computer-readable media. Examples of transitory computer-readable media include electrical signals, optical signals, and electromagnetic waves. The transitory computer-readable media can supply the program to a computer via a wired communication path such as an electric wire or optical fiber, or via a wireless communication path.

[0076] The present disclosure is not limited to the above-described embodiment, and can be modified as appropriate within the scope of the present disclosure.

[0077] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes. (Appendix 1) a simulation unit that generates a plurality of attenuation simulation maps showing a spatial distribution of signal attenuation using a plurality of visibility assessment accuracies related to visibility assessment results in a simulation environment; a learning unit that generates a learning model for updating the occluded area in the attenuation simulation map in association with each visibility determination accuracy; a generating unit that generates an attenuation map indicating a spatial distribution of signal attenuation using a visibility determination result within an observation area; an update unit that inputs the attenuation map into each of the learning models to update occluded regions in the attenuation map; A learning system comprising: a selection unit that selects one attenuation map from among the plurality of attenuation maps in which the occluded area has been updated based on the result of the change in the occluded area within the observation area before and after the update. (Appendix 2) The generation unit generating the attenuation map as image data; The selection unit A learning system as described in Appendix 1, which selects one attenuation map from among the multiple attenuation maps in which the occlusion area has been updated based on the result of a change in the number of pixels that constitute the occlusion area in the image data. (Appendix 3) The selection unit A learning system as described in Appendix 2, which selects one attenuation map from among the multiple attenuation maps in which the occluded areas have been updated based on the ratio of the number of pixels that make up the occluded area in the image data after the update to the number of pixels that make up the occluded area in the image data before the update. (Appendix 4) The selection unit A learning system as described in Appendix 2 or 3, which selects the attenuation map that has the highest ratio of the number of pixels constituting the occluded area in the image data after updating to the number of pixels constituting the occluded area in the image data before updating from among multiple attenuation maps in which the occluded areas have been updated. (Appendix 5) The selection unit A learning system as described in Appendix 2, which selects one attenuation map from among the multiple attenuation maps in which the occluded areas have been updated based on the difference between the number of pixels that make up the occluded area in the image data before the update and the number of pixels that make up the occluded area in the image data after the update. (Appendix 6) The selection unit A learning system as described in Appendix 5, which selects the attenuation map from among multiple attenuation maps in which the occluded areas have been updated, the attenuation map that has the smallest difference between the number of pixels that make up the occluded area in the image data before the update and the number of pixels that make up the occluded area in the image data after the update. (Appendix 7) a generating unit that generates an attenuation map indicating a spatial distribution of signal attenuation using a visibility determination result within an observation area; an update unit that generates a plurality of attenuation simulation maps showing the spatial distribution of signal attenuation using a plurality of visibility assessment accuracies related to visibility assessment results in a simulation environment, and inputs the attenuation maps into a learning model that updates occluded areas in the attenuation simulation maps, the learning model being generated in association with each visibility assessment accuracy, and updates occluded areas in the attenuation maps; A data selection device comprising: a selection unit that selects one attenuation map from the plurality of attenuation maps in which the occluded area has been updated based on the result of the change in the occluded area within the observation area before and after the update. (Appendix 8) The generation unit generating the attenuation map as image data; The selection unit A data selection device as described in Appendix 7, which selects one attenuation map from among the multiple attenuation maps in which the occlusion area has been updated based on the result of a change in the number of pixels constituting the occlusion area in the image data. (Appendix 9) generating a plurality of simulated attenuation maps illustrating a spatial distribution of signal attenuation using a plurality of line-of-sight assessment accuracies related to the line-of-sight assessment results within the simulated environment; generating a learning model for updating the occluded area in the attenuation simulation map in association with each visibility determination accuracy; Using the visibility assessment results within the observation area, an attenuation map is generated that shows the spatial distribution of signal attenuation; inputting the attenuation map into each of the learning models to update occluded regions in the attenuation map; A learning method for selecting one attenuation map from among the plurality of attenuation maps in which the occluded areas have been updated, based on the result of the change in the occluded areas within the observation area before and after the update. (Appendix 10) generating an attenuation map showing the spatial distribution of signal attenuation using the visibility determination results within the observation area; A plurality of attenuation simulation maps showing the spatial distribution of signal attenuation are generated using a plurality of visibility determination accuracies related to the visibility determination results in the simulation environment, and the attenuation maps are input into a learning model that updates occluded areas in the attenuation simulation maps and is generated in association with the respective visibility determination accuracies; and the occluded areas in the attenuation maps are updated; A data selection method for selecting one attenuation map from among a plurality of attenuation maps in which the occluded area has been updated, based on the result of the change in the occluded area within the observation area before and after the update. (Appendix 11) generating a plurality of simulated attenuation maps illustrating a spatial distribution of signal attenuation using a plurality of line-of-sight assessment accuracies related to the line-of-sight assessment results within the simulated environment; generating a learning model for updating the occluded area in the attenuation simulation map in association with each visibility determination accuracy; Using the visibility assessment results within the observation area, an attenuation map is generated that shows the spatial distribution of signal attenuation; inputting the attenuation map into each of the learning models to update occluded regions in the attenuation map; A non-transitory computer-readable medium storing a program that causes a computer to select one attenuation map from among a plurality of attenuation maps in which the occluded area has been updated based on the result of the change in the occluded area within the observation area before and after the update. (Appendix 12) generating an attenuation map showing the spatial distribution of signal attenuation using the visibility determination results within the observation area; A plurality of attenuation simulation maps showing the spatial distribution of signal attenuation are generated using a plurality of visibility determination accuracies related to the visibility determination results in the simulation environment, and the attenuation maps are input into a learning model that updates occluded areas in the attenuation simulation maps and is generated in association with the respective visibility determination accuracies; and the occluded areas in the attenuation maps are updated; A non-transitory computer-readable medium storing a program that causes a computer to select one attenuation map from among a plurality of attenuation maps in which the occluded area has been updated based on the result of the change in the occluded area within the observation area before and after the update. [Explanation of symbols]

[0078] 10 Learning System 11 Simulation Section 12 Learning Department 13 Generation part 14 Update section 15 Selection section 20 Learning Device 30 Data selection device 40 Radio transmitting device 50 Radio receiving device 61 Shield 62 Shield 70 Radio Waves

Claims

1. a simulation means for generating a plurality of attenuation simulation maps showing a spatial distribution of signal attenuation using a plurality of visibility assessment accuracies related to visibility assessment results in a simulation environment; a learning means for generating a learning model for updating the occluded area in the attenuation simulation map in association with each visibility determination accuracy; a generating means for generating an attenuation map showing a spatial distribution of signal attenuation using a visibility determination result within an observation area; updating means for inputting the attenuation map into each of the learning models to update occluded regions in the attenuation map; A learning system comprising: a selection means for selecting one attenuation map from the plurality of attenuation maps in which the occluded area has been updated based on the result of the change in the occluded area within the observation area before and after the update.

2. The generating means generating the attenuation map as image data; The selection means The learning system of claim 1 , wherein one attenuation map is selected from the plurality of attenuation maps in which the occlusion areas have been updated based on the result of a change in the number of pixels constituting the occlusion areas in the image data.

3. The selection means The learning system of claim 2, wherein one attenuation map is selected from among a plurality of attenuation maps in which occlusion areas have been updated based on the ratio of the number of pixels constituting the occlusion area in the image data after the update to the number of pixels constituting the occlusion area in the image data before the update.

4. The selection means The learning system according to claim 2 or 3, wherein the attenuation map that has the highest ratio of the number of pixels constituting the occluded area in the image data after the update to the number of pixels constituting the occluded area in the image data before the update is selected from among the plurality of attenuation maps whose occluded areas have been updated.

5. The selection means The learning system of claim 2, wherein one attenuation map is selected from the plurality of attenuation maps in which the occlusion areas have been updated based on the difference between the number of pixels constituting the occlusion area in the image data before the update and the number of pixels constituting the occlusion area in the image data after the update.

6. The selection means The learning system of claim 5, wherein the attenuation map that has the smallest difference between the number of pixels constituting the occluded area in the image data before the update and the number of pixels constituting the occluded area in the image data after the update is selected from among the plurality of attenuation maps in which the occluded areas have been updated.

7. a generating means for generating an attenuation map showing a spatial distribution of signal attenuation using a visibility determination result within an observation area; a plurality of attenuation simulation maps showing the spatial distribution of signal attenuation are generated using a plurality of visibility assessment accuracies related to the visibility assessment results in the simulation environment, and the attenuation maps are input into a learning model that updates the occluded areas in the attenuation simulation maps and is generated in association with the respective visibility assessment accuracies; and an updating means that updates the occluded areas in the attenuation maps. and a selection means for selecting one attenuation map from the plurality of attenuation maps in which the occluded area has been updated based on the result of the change in the occluded area within the observation area before and after the update.

8. Using a plurality of visibility assessment accuracies related to visibility assessment results within a simulation environment, a plurality of attenuation simulation maps showing a spatial distribution of signal attenuation are generated; generating a learning model for updating the occluded area in the attenuation simulation map in association with each visibility determination accuracy; Using the visibility assessment results within the observation area, an attenuation map is generated that shows the spatial distribution of signal attenuation; inputting the attenuation map into each of the learning models to update occluded regions in the attenuation map; A learning method for selecting one attenuation map from among the plurality of attenuation maps in which the occluded areas have been updated, based on the result of the change in the occluded areas within the observation area before and after the update.

9. Using the visibility assessment results within the observation area, an attenuation map is generated that shows the spatial distribution of signal attenuation; A plurality of attenuation simulation maps showing the spatial distribution of signal attenuation are generated using a plurality of visibility determination accuracies related to the visibility determination results in the simulation environment, and the attenuation maps are input into a learning model that updates occluded areas in the attenuation simulation maps and is generated in association with the respective visibility determination accuracies; and the occluded areas in the attenuation maps are updated; A data selection method for selecting one attenuation map from among a plurality of attenuation maps in which the occluded area has been updated, based on the result of the change in the occluded area within the observation area before and after the update.

10. Using a plurality of visibility assessment accuracies related to the visibility assessment results in the simulation environment, a plurality of attenuation simulation maps showing the spatial distribution of signal attenuation are generated; generating a learning model for updating the occluded area in the attenuation simulation map in association with each visibility determination accuracy; Using the visibility assessment results within the observation area, an attenuation map is generated that shows the spatial distribution of signal attenuation; inputting the attenuation map into each of the learning models to update occluded regions in the attenuation map; A program that causes a computer to select one attenuation map from among a plurality of attenuation maps in which the occluded area has been updated, based on the result of the change in the occluded area within the observation area before and after the update.

Images