Information processing apparatus, information processing method, and program

Abstract

To provide an information processing device, an information processing method, and a program capable of restoring missing weather data.SOLUTION: An information processing device according to an embodiment includes a processing unit that processes weather data. The processing unit creates partial data that is a part of the weather data other than the missing range, creates partially-missing data by missing a part of the partial data, acquires missing weather data based on the partial data and the partially-missing data, and outputs weather data that is not missing.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] An embodiment of the present invention relates to an information processing device, an information processing method, and a program for processing meteorological data. [Background technology]

[0002] Due to recent climate change, localized heavy rain, tornadoes, hail, etc. are becoming more frequent. The risk of water disasters caused by these localized weather phenomena has become a social issue. Weather radars have been put into practical use to observe these weather phenomena. Weather radars transmit radio waves to an observation area centered on an antenna, receive reflected waves from rain or snow particles present in the observation area, and generate weather data. Based on the weather data, current weather phenomena can be observed and future weather phenomena can be predicted.

[0003] If there is a structure such as a building or a mountain between the weather radar and the rain cloud or snow cloud, the weather data will be lost.

[0004] Various proposals have been made to restore missing data through calculations. In one proposal, data from points surrounding the missing area is weighted and calculated, and the weighted data is used as the data for the missing area. Therefore, it is difficult to accurately observe the shape of a minute and complex rain area, such as a localized heavy rain, and predict changes in the rain area using meteorological data obtained using this proposal. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2002-312760 [Patent Document 2] Japanese Patent Application Laid-Open No. 2011-252822 [Non-patent literature]

[0006] [Non-Patent Document 1] S. Tan, “A Conditional Generative Adversarial Network for Weather Radar Beam Blockage Correction”, IEEE Transactions on Geoscience and Remote Sensing, vol. 61, June 2023 Summary of the Invention [Problem to be solved by the invention]

[0007] An object of the present invention is to provide an information processing device, an information processing method, and a program for restoring missing meteorological data. [Means for solving the problem]

[0008] An information processing device according to an embodiment includes a processing unit that processes meteorological data. The processing unit creates partial data that is a portion of the meteorological data other than a missing range, creates partially missing data by missing a portion of the partial data, and creates a model that inputs the missing meteorological data and outputs the non-missing meteorological data based on the partial data and the partially missing data. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram for explaining an example of a radar system according to a first embodiment. [Figure 2] FIG. 3 is a diagram for explaining an example of quality information according to the first embodiment. [Figure 3] FIG. 2 is a diagram for explaining an example of functional blocks of the information processing device according to the first embodiment. [Figure 4] FIG. 3 is a diagram for explaining an example of weather data according to the first embodiment. [Figure 5] FIG. 3 is a diagram for explaining an example of quality information according to the first embodiment. [Figure 6] FIG. 4 is a diagram for explaining an example of processing by an extraction unit according to the first embodiment. [Figure 7]FIG. 4 is a diagram for explaining an example of processing of a missing portion according to the first embodiment. [Figure 8] FIG. 4 is a diagram for explaining an example of processing by a machine learning unit according to the first embodiment. [Figure 9] FIG. 4 is a diagram for explaining an example of processing by an extraction unit and a model application unit according to the first embodiment. [Figure 10] FIG. 4 is a diagram for explaining an example of processing by a replacement unit according to the first embodiment. [Figure 11] FIG. 2 is a diagram for explaining an example of restoration of lost data according to the first embodiment. [Figure 12] FIG. 10 is a diagram for explaining an example of functional blocks of an information processing device according to a second embodiment. [Figure 13] FIG. 10 is a diagram for explaining an example of processing by a machine learning unit according to the second embodiment. [Figure 14] FIG. 10 is a diagram for explaining an example of weather data according to the second embodiment. [Figure 15] FIG. 11 is a diagram for explaining an example of processing by a machine learning unit according to the third embodiment. [Figure 16] FIG. 11 is a diagram for explaining an example of weather data according to the third embodiment. [Figure 17] FIG. 10 is a diagram for explaining an example of functional blocks of an information processing device according to a fourth embodiment. [Figure 18] FIG. 13 is a diagram for explaining an example of interference according to the fourth embodiment. [Figure 19] FIG. 13 is a diagram for explaining an example of functional blocks of an information processing device according to a fifth embodiment. [Figure 20] FIG. 13 is a diagram for explaining an example of processing by a machine learning unit according to the fifth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] The following describes embodiments with reference to the drawings. The following description exemplifies devices and methods embodying the technical concepts of the embodiments. The technical concepts of the embodiments are not limited to the structures, shapes, arrangements, materials, etc. of the components described below. Modifications that can be easily conceived by those skilled in the art are naturally within the scope of the disclosure. For clarity of explanation, the drawings may schematically depict elements with different sizes, thicknesses, planar dimensions, shapes, etc. compared to the actual elements. Elements with different dimensional relationships or ratios may be included in multiple drawings. Corresponding elements may be designated by the same reference numerals in multiple drawings, and redundant description may be omitted. Some elements may be designated by multiple names, but these names are merely examples and do not exclude the use of other names for these elements. Elements without a plural name may also be designated by other names. "Connection" may include not only direct connection but also connection via other elements. Unless the number of elements is specified as being plural, the element may be a singular element or multiple elements.

[0011] [First embodiment] 1 is a diagram illustrating an example of a radar system 10 according to the first embodiment. The radar system 10 includes a radar device 12, a storage device 14, and an information processing device 16. The radar device 12, the storage device 14, and the information processing device 16 are connected to a network 18.

[0012] The radar device 12 includes an antenna 6 that can rotate around an axis extending in the vertical direction, and a signal processing unit 8. The radar device 12 transmits radio waves from the antenna 6 while rotating the antenna 6. An example of the radio waves is microwaves in the C band (4 to 8 GHz) or X band (8 to 12 GHz). The radio waves may be single polarized or dual polarized. An example of single polarized waves is horizontal polarization. Dual polarized waves include horizontal polarization and vertical polarization.

[0013] The transmitted radio waves are reflected by rain or snow particles present in a circular observation area of ​​a predetermined radius (e.g., several hundred kilometers) centered on the rotation center of the antenna 6. The radar device 12 receives the reflected waves at the antenna 6 while rotating the antenna 6. The signal processing unit 8 processes the reflected waves and creates weather data representing the state of rain clouds. The weather data is data for each of multiple lattice points that make up the observation area. In order to observe three-dimensional weather phenomena, the radar device 12 can also change the elevation angle of the antenna 6 at different times and collect weather data three-dimensionally at multiple elevation angles (multiple altitude layers). In the first embodiment, a certain altitude is set as the altitude of the observation area. The radar device 12 sets the elevation angle of the antenna 6 according to the altitude of the observation area.

[0014] When the radar device 12 transmits single-polarized radio waves, an example of meteorological data is the radar reflectivity factor Z of rain or snow particles. Another example of meteorological data is the movement speed of rain or snow particles (i.e., wind strength). Yet another example of meteorological data is the velocity range of rain or snow particles. When observing discretely distributed targets such as rain or snow particles, the movement of individual rain or snow particles is random, resulting in variations in phase and amplitude for each observation and a broadened spectrum. This spectral width is called the velocity range. When the radar device 12 transmits dual-polarized radio waves, an example of meteorological data is the rate of change of phase difference KDP of reflected waves between polarized waves. Another example of meteorological data is the correlation coefficient ρhv of reflected waves between polarized waves. Yet another example of meteorological data is the power difference Zdr of reflected waves between polarized waves.

[0015] If there is a structure between the radar device 12 and the rain cloud or snow cloud, the radio waves will not reach the rain cloud or snow cloud, so the radar device 12 will not be able to receive reflected waves from the rain or snow particles, resulting in a loss of weather data within a certain range of the observation area.

[0016] When creating weather data, the radar device 12 also creates quality information consisting of flags related to the quality of the weather data at each location. The quality information is one of eight flags. Flags 1, 2, and 4 indicate missing factors. The radar device 12 determines the quality information flag based on topographical information around the installation location and diffraction processing within the radar device.

[0017] Flag 1 represents missing factor 1 (masking range). The masking range is the range that the radar does not irradiate. The radar device 12 does not irradiate radio waves into the masking range. For example, it does not irradiate radio waves into an area that includes a reflector such as a steel tower. This is because the reflectivity of a reflector is much higher than the reflectivity of rain or snow particles, and the reflector has a negative impact on the observation of weather information. The quality information of the masking range is flag 1.

[0018] Flag 2 indicates missing cause 2 (structure range). The structure range is an area that includes structures such as mountains and buildings. Since reflected waves from the structure range cannot be received, the weather data is missing.

[0019] Flag 1 and Flag 2 relate to regular defects due to topographical factors. The radar device 12 can recognize the ranges of Flag 1 and Flag 2 before observing meteorological data.

[0020] Flag 4 represents the missing factor 3 (radio wave dissipation range). Radio waves are dissipated in the heavy rain layer, making it impossible to collect meteorological data.

[0021] Other flags relate to the quality of the weather data. Flag 3 indicates "obstructed." "Obstructed" indicates that some reflected waves cannot be received due to an obstruction or the like. Therefore, the radar device 12 corrects the weather data when "obstructed." Flag 3 indicates that the weather data is corrected data, not actual observation data, and is of low quality.

[0022] Flag 5 indicates the rainfall estimation result by KDP. Rainfall estimation can be done by intensity or by KDP. KDP rainfall estimation has better accuracy than intensity rainfall estimation. Flag 5 indicates high quality data.

[0023] Flag 6 indicates a rain layer. A layer of rain and snow particles can be a rain layer, a melting layer (a mixture of rain and hail), or a snow layer depending on the altitude. The radar device 12 observes a rain layer, a melting layer (a mixture of rain and hail), or a snow layer depending on the altitude of the observation area. Meteorological data for a rain layer has the highest quality. Flag 6 indicates that the data quality is high.

[0024] Flags 7 and 8 are reserved flags (undefined). The observation area includes a range that does not fall under the definition of the above flags, i.e., a "no flag" range where no flag is set.

[0025] The radar device 12 substitutes the lower limit value of the data for the missing range of weather data and outputs it. That is, weather data whose quality information is flagged as 1, 2, or 3, or weather data for which the lower limit value of the data has been input, is considered to be missing.

[0026] FIG. 2 is a diagram illustrating an example of quality information according to the first embodiment. The horizontal axis of FIG. 2 represents the longitude of the observation area, and the vertical axis represents the latitude of the observation area. The quality information in FIG. 2 represents the quality of meteorological data at each point in a two-dimensional observation area at a certain altitude corresponding to the elevation angle within the three-dimensional area when the elevation angle of the antenna 6 is set to a certain angle. In the example of FIG. 2, the quality information includes four flags: flags 1, 2, 3, and 6. The approximately circular range centered on the rotation center 2A of the antenna 6 is the rain layer range, and the quality information for the rain layer is flag 6. Multiple radial ranges centered on the rotation center 2A of the antenna 6 are the masked range, the obstructed range, and the structure range, and the quality information is flags 1, 3, and 2, respectively. In the example of FIG. 2, the observation area also includes a range without a flag.

[0027] The storage device 14 stores the weather data and quality information output by the radar device 12. The radar device 12 transmits radio waves at regular intervals and outputs the weather data and quality information at regular intervals.

[0028] Missing data in meteorological data is classified into predictable missing data and unpredictable missing data. The missing factors for predictable missing data are Missing Factor 1 and Missing Factor 2. The range of predictable missing data is the range of Flag 1 and Flag 2.

[0029] Returning to the description of FIG. 1 , the information processing device 16 includes an input device 22, a processing unit 24, a memory 26, and an output device 28. The input device 22, the processing unit 24, the memory 26, and the output device 28 are connected to a bus line 30. The processing unit 24 is implemented by at least one processing unit and executes the processing of the information processing device 16. This processing unit may include, for example, a control unit or an arithmetic unit, and may be implemented by analog or digital circuits. The processing unit may be a central processing unit (CPU), a general-purpose processor, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof. As an example, at least a portion of the processing unit 24 may have a functional configuration realized by the CPU executing a loss restoration program.

[0030] The input device 22 is connected to the storage device 14 via the network 18 and acquires meteorological data and quality information from the storage device 14. The input device 22 may be connected to the radar device 12 and acquires meteorological data and quality information from the radar device 12. The information processing device 16 processes the meteorological data. Examples of the processing include a process for restoring missing meteorological data and a process for creating a model used for the restoration.

[0031] The memory 26 stores a program executed by the processing unit 24. An example of the program is a loss restoration program. A portion of the memory 26 is connected to a backup power supply. The loss restoration program is stored in a portion of the memory 26 connected to the backup power supply. The processing unit 24 executes the loss restoration program stored in the memory 26, and restores the loss if a portion of the weather data is missing. In one example of restoration, the processing unit 24 first creates a model for restoring the loss, and then uses the model to restore the missing weather data.

[0032] The output device 28 supplies the weather data from which the missing data has been restored to an external device (not shown). An example of the external device is a weather data analysis device that analyzes weather phenomena based on the weather data and predicts the occurrence of dangerous water disasters.

[0033] 1 shows a radar system 10 including a radar device 12, a storage device 14, and an information processing device 16, but the radar system 10 does not have to include the radar device 12. A radar device not included in the radar system 10 may be connected to a network 18. In this case, the radar system 10 does not have to include the network 18 either. A radar device not included in the radar system 10 may be directly connected to the storage device 14 and the information processing device 16. The information processing device 16 may be directly connected to the storage device 14.

[0034] 1 shows a radar system 10 including a radar device 12, a storage device 14, and an information processing device 16, each of which is separate from the other devices. However, some of these devices may include other devices. For example, the radar device 12 may include the storage device 14 and / or the information processing device 16. The information processing device 16 may include the storage device 14.

[0035] 3 is a diagram illustrating an example of functional blocks of the information processing device 16 according to the first embodiment. The functional blocks in FIG. 3 are implemented by the processing unit 24 executing a loss restoration program. The processing unit 24 implements a model creation unit 40 and a restoration unit 42 by executing the loss restoration program. The model creation unit 40 includes a loss identification unit 50, an extraction unit 52, a loss reduction unit 54, and a machine learning unit 56. The restoration unit 42 includes an extraction unit 74, a model application unit 76, and a replacement unit 78.

[0036] Weather data is input to the defect identification unit 50, the extraction unit 52, the extraction unit 74, and the replacement unit 78. Quality information is input to the defect identification unit 50. The defect identification unit 50 identifies the range of defects in the weather data and sends range information representing the range of defects to the extraction unit 52, the defect creating unit 54, the extraction unit 74, and the replacement unit 78. The output of the extraction unit 52 is sent to the defect creating unit 54 and the machine learning unit 56. The output of the extraction unit 74 is sent to the model application unit 76. The machine learning unit 56 creates a model and sends the created model to the model application unit 76. The output of the model application unit 78 is sent to the replacement unit 78.

[0037] The restoration unit 42 operates using the model created by the model creation unit 40, and therefore operates after the model creation unit 40 completes creation of the model. The weather data input to the model creation unit 40 and the weather data input to the restoration unit 42 are data from different periods. During a certain period (model creation period), the weather data is not input to the restoration unit 42, but is input to the model creation unit 40. During periods other than the model creation period, the weather data is not input to the model creation unit 40, but is input to the restoration unit 42.

[0038] The information processing device 16 may be realized by an information processing device such as a computer, a computer system configured by multiple computers or servers communicating with each other via a network, or a PC cluster in which multiple computers cooperate to execute information processing.

[0039] The information processing device 16 may include a plurality of processing units each realizing at least a part of the functional blocks shown in FIG. 3, instead of including one processing unit 24 that realizes the functional blocks shown in FIG.

[0040] The information processing device 16 may include one or more hardware blocks corresponding to the respective functional blocks shown in FIG. 3, instead of including the processing unit 24 that realizes the functional blocks.

[0041] FIG. 4 is a diagram illustrating an example of weather data according to the first embodiment. FIG. 4 shows weather data for each of a plurality of lattice points (square areas in FIG. 4) arranged two-dimensionally in a two-dimensional observation area when the elevation angle of the antenna 6 is set to a certain elevation angle θ1. The horizontal axis of FIG. 4 represents the direction of the observation area, and the vertical axis represents the range of the observation area. The direction represents a range from 0 degrees to 360 degrees around the center of rotation of the antenna 6. The range represents the distance from the center of rotation of the antenna 6. Weather data may be missing across the entire range in a certain direction range.

[0042] FIG. 5 is a diagram illustrating an example of quality information according to the first embodiment. The quality information includes flags related to the quality of meteorological data at grid points. In this example, the quality information includes flag 1 (masking range), flag 3 (obstructed), and flag 6 (rain layer). The quality information for the missing range of meteorological data is flag 1 (masking range) or flag 2 (structure range).

[0043] Returning to the explanation of FIG. 3, the defect identification unit 50 of the model creation unit 40 identifies the defect range of the weather data. If there are multiple defect ranges in the weather data, the defect identification unit 50 may identify each of the multiple defect ranges, or may identify at least one of the defect ranges or only one of the defect ranges. If only one defect range is identified, the defect identification unit 50 may identify the widest defect range.

[0044] The defect identifying unit 50 may identify the missing range based on the quality information. In this case, the defect identifying unit 50 may identify the entire range in the azimuth direction for which the quality information is flag 1 or flag 2, indicating a steady defect, as the missing range. The defect identifying unit 50 may identify the missing range based on the number of grid points in the azimuth direction and the number of grid points in the range direction of the missing range. If flag 1 or flag 2, indicating a steady defect, is continuous in the range direction in a certain azimuth, the defect identifying unit 50 may consider all grid points in the range direction to be the missing range, and may create range information indicating only the number of grid points in the azimuth direction for flag 1 or flag 2. The defect identifying unit 50 may use new data as range information, in which identification numbers such as 1 are assigned to missing data in the weather data and 0 are assigned to other data.

[0045] The defect identification unit 50 may identify the defect range based on information input by the user. In this case, the user manually sets the defect range in a parameter file, etc. The parameter file is sent to the defect identification unit 50.

[0046] FIG. 6 is a diagram illustrating an example of processing by the extraction unit 52 according to the first embodiment. The extraction unit 52 inputs weather data, extracts data from the weather data in a direction range other than the missing range based on range information, and outputs the extracted data as partial data. The size A2 in the direction of the partial data is equal to or greater than the size A1 in the direction of the missing range of the weather data. For example, the size A2 may be α times the size A1 (A2 = α × A1), where α is a positive number equal to or greater than 1. Alternatively, the size A2 may be β times longer than the size A1 (A2 = A1 + β), where β is a positive number. α and β may be predetermined values ​​or may be values ​​set by the user.

[0047] Therefore, the azimuth size of the partial data increases or decreases depending on the azimuth size of the missing range of the weather data. The azimuth size of the partial data may be changed depending on the azimuth size of the missing range. For example, the azimuth size of the partial data may be set to be larger as the azimuth size of the missing range increases. As will be described later, the model creation unit 40 generates a model by machine learning using data in which some of the partial data is missing, so machine learning can be performed with a small amount of information, and the amount of calculation for machine learning and the capacity of the memory 26 for storing data during calculation can be small.

[0048] The extraction unit 52 may extract a range corresponding to the missing range of the weather data, for example, a nearby range, as partial data. By extracting partial data from the vicinity of the missing range, machine learning is performed based on the partial data in which the topography of structures, etc. is close to the missing range. This improves the quality of the model and the accuracy of missing data restoration.

[0049] The extraction unit 52 transmits the partial data to the missing data generation unit 54 , the machine learning unit 56 , and the extraction unit 74 .

[0050] FIG. 7 is a diagram illustrating an example of processing by the missing section 54 according to the first embodiment. The missing section 54 missing a portion of the azimuth range of the partial data to create partial missing data. Missing partial data means setting the partial data to a lower limit value. The missing section 54 determines the missing range to be deleted based on range information. The size B1 of the missing range in the azimuth direction is equal to or greater than the size A1 of the missing range in the azimuth direction of the weather data. In FIG. 7, the center of the partial data in the azimuth direction is missing, but the position of the missing range in the azimuth direction may be arbitrary. The missing section 54 transmits the partial missing data to the machine learning section 56.

[0051] The missing data generating unit 54 is not limited to outputting one piece of partial missing data, but may output multiple pieces of partial missing data whose missing ranges are located at different azimuth positions. For example, the missing data generating unit 54 may output partial missing data whose missing range is located in the left half range or partial missing data whose missing range is located in the right half range. When the machine learning unit 56 performs learning based on multiple pieces of partial missing data whose missing ranges are located at different azimuth positions, the accuracy of machine learning is improved.

[0052] FIG. 8 is a diagram illustrating an example of processing by the machine learning unit 56 according to the first embodiment. The machine learning unit 56 creates a model using known loss restoration techniques, such as a convolutional neural network (CNN) or a generative adversarial network (GAN). For example, when partially missing data is input, the machine learning unit 56 creates a model that outputs loss-free data in which the loss in the input data has been restored. The machine learning unit 56 writes the created model to the memory 26. The memory 26 is also connected to the model application unit 76. The model application unit 76 reads the model from the memory 26 and restores the loss using the model.

[0053] The machine learning unit 56 performs supervised learning using a training dataset. The training dataset includes input data and training data. The input data is partially missing data. The training data is partial data.

[0054] When the missing unit 54 outputs multiple pieces of partially missing data with different missing ranges, the machine learning unit 56 may generate multiple models according to the positions of the missing ranges of the partially missing data. In this case, the machine learning unit 56 adds information indicating the positions of the missing ranges at the time of model creation to the model and writes the model to the memory 26.

[0055] The defect identification unit 50, extraction unit 52, and defect creation unit 54 perform the above processing on multiple pieces of weather data. As a result, the model creation unit 40 generates a model through supervised learning using multiple training datasets. The multiple pieces of weather data may be data collected on different days over different time periods, or may be data collected on the same day over different time periods.

[0056] The machine learning unit 56 may perform machine learning offline using past weather data. The past weather data is stored in the storage device 14. Alternatively, the machine learning unit 56 may perform machine learning in real time using weather data obtained by the radar device 12, and update the model as needed.

[0057] The extraction unit 74 of the restoration unit 42 inputs weather data and range information. If the weather data contains a missing part, the extraction unit 74 extracts data of the azimuth range including the missing part from the weather data based on the range information, and outputs the extracted data as input data. The relative position in the azimuth direction of the missing part in the input data is equal to the relative position in the azimuth direction of the missing part in the partially missing data. For example, if the missing part creating unit 54 creates a missing part in the azimuth direction of the partial data, the extraction unit 74 extracts the input data so that the missing part is located in the azimuth center. The azimuth size C1 of the input data is equal to the azimuth size A2 of the partial data.

[0058] 9 is a diagram illustrating an example of the processing of the extraction unit 74 and the model application unit 76 according to the first embodiment. The extraction unit 74 acquires real-time weather data from the storage device 14 or the radar device 12. The extraction unit 74 identifies a missing range of the weather data from the range information and extracts input data including the missing range. The extraction unit 74 transmits the extracted input data to the model application unit 76.

[0059] The model application unit 76 inputs input data into a model CNN or GAN. The model application unit 76 outputs missing data in which missing data in the input data has been restored. In other words, the model application unit 76 restores weather data for points where weather data could not be collected. The model application unit 76 transmits the missing data to the replacement unit 78.

[0060] When the training dataset includes multiple partially missing data with different positions of the missing ranges and the machine learning unit 56 creates multiple models according to the positions of the missing ranges, the model application unit 76 reads from the memory 26 a model corresponding to the position of the missing range in the input data output by the extraction unit 74, and uses the model to restore the missing portion.

[0061] 10 is a diagram illustrating an example of processing by the replacement unit 78 according to the first embodiment. The replacement unit 78 receives the weather data, range information, and the loss-free data output by the model application unit 76, and replaces the input data in the weather data with the loss-free data. As a result, the replacement unit 78 outputs the loss-free data.

[0062] When meteorological data contains multiple missing data, the restoration unit 42 extracts input data, applies a model, and performs substitution for each missing data range. Since the smaller the missing data range, the higher the restoration accuracy, the extraction unit 74 extracts input data containing missing data in ascending order of missing data range, and the model application unit 76 and substitution unit 78 perform the above processing.

[0063] FIG. 11 is a diagram illustrating an example of restoration of missing data according to the first embodiment. FIG. 11(a) shows an example of weather data. The weather data includes a missing range (angle range in the azimuth direction: 14°) from an azimuth angle of 90° to 104°. FIG. 11(b) shows an example of weather data after restoration of missing data output by the replacement unit 78. FIG. 11(c) shows true value data that does not include missing data. The true value data (weather data in FIGS. 11(a) and 11(c)) is simulation data. The weather data after restoration of missing data (FIG. 11(b)) matches the trend of the true value data (FIG. 11(c)). Therefore, it can be confirmed that restoration according to the first embodiment enables accurate restoration of rain area data in a range that the radar device 12 was unable to collect.

[0064] When part of the weather data is missing due to geographical conditions, the information processing device according to the first embodiment can acquire the missing weather data, restore the missing data based on the partial data and the partially missing data, and output the weather data without the missing data, thereby improving the quality of the weather data. The information processing device may create a model for restoring the missing data. The information processing device may also use the model to restore the missing weather data. This makes it possible to accurately detect signs of dangerous weather phenomena using the weather data created by the radar device 12.

[0065] [Second embodiment] FIG. 12 is a diagram illustrating an example of functional blocks of an information processing device 16 according to the second embodiment. The information processing device 16 according to the second embodiment includes the same functional blocks as the information processing device 16 according to the first embodiment shown in FIG. 3. The second embodiment differs from the first embodiment in the manner of connections between the functional blocks. In the second embodiment, the partial data output from the extraction unit 52 and the partially missing data output from the missing data generating unit 54 are written to the memory 26. In a first embodiment, multiple pieces of weather data are collected over a certain period of time, and multiple learning datasets are created from the multiple pieces of weather data. In a second embodiment, multiple pieces of weather data are collected over at least two periods of time. If there are two data collection periods, the first period is referred to as period tp0 and the second period is referred to as period tp1. After the first period tp0 ends, the second period tp1 begins. The data collection period may be repeated. That is, after the second period tp1 ends, the first period tp0 may begin again.

[0066] The weather data collected during the two time periods tp0 and tp1 is supplied to the information processing device 16. The extraction unit 52 outputs a plurality of partial data from the weather data collected during each of the two time periods tp0 and tp1, and writes the plurality of partial data to the memory 26. The missing unit 54 outputs a plurality of partial missing data from the partial data during each of the two time periods tp0 and tp1, and writes the plurality of partial missing data to the memory 26.

[0067] 13 is a diagram illustrating an example of processing by the machine learning unit 56 according to the second embodiment. The machine learning unit 56 reads out a plurality of first partial missing data for the first period tp0, a plurality of second partial missing data for the second period tp1, and a plurality of second partial data for the second period tp1 from the memory 26. For ease of explanation, FIG. 13 illustrates a single first partial missing data for the first period tp0, a single second partial missing data for the second period tp1, and a single second partial data.

[0068] The machine learning unit 56 uses multiple first partial missing data from the first period tp0 and multiple second partial missing data from the second period tp1 as input data, and multiple second partial data from the second period tp1 as training data, and performs machine learning to create a model.

[0069] The plurality of partial data and the plurality of partial missing data for the second time period tp1 do not have to be written to memory 26. In this case, machine learning unit 56 uses, as input data, the plurality of first partial missing data for the first time period tp0 read from memory 26 and the plurality of second partial missing data for the second time period tp1 output from missing unit 54, and, as training data, the plurality of second partial data for the second time period tp1 output from extraction unit 52, to perform machine learning to create a model.

[0070] In the second embodiment, it is possible to learn the movement of rain clouds by using time-series data for learning, as compared to the first embodiment. When a model is used to restore defects, the learning accuracy of the model is improved.

[0071] FIG. 14 is a diagram illustrating an example of weather data according to the second embodiment. FIG. 14(a) shows an example of weather data collected at time t0. FIG. 14(b) shows an example of weather data collected 10 minutes before time t0. FIG. 14(c) shows an example of weather data collected 20 minutes before time t0. By comparing FIG. 14(c) with FIG. 14(b), it can be recognized that the rain area indicated by the black frame is moving in a direction close to an azimuth angle of 0°. The model creation unit 40 can learn the movement of the rain area by using weather data from different periods. This improves the restoration accuracy of the restoration unit 42.

[0072] [Third embodiment] In the first and second embodiments, the elevation angle of the antenna 6 is set to an angle corresponding to the altitude at which weather data is desired to be observed. Depending on the altitude of the observation range, weather data may be missing due to mountains or buildings. Increasing the altitude of the observation range increases the likelihood of missing data. The radar device 12 sets the elevation angle of the antenna 6 to at least two elevation angles (here, θ0 and θ1) and performs observations in the observation range and a higher observation range. The elevation angle θ1 corresponds to the altitude of the observation range for weather data. The elevation angle θ0 is an angle corresponding to an altitude that is not affected by mountains or buildings. The elevation angle θ0 is equal to or greater than the elevation angle θ1. Therefore, at the elevation angle θ0, the influence of structures is reduced, and high-quality weather data without any missing data is collected.

[0073] The functional block configuration of the processing unit 24 in the third embodiment is the same as the functional block configuration of the second embodiment shown in FIG. 12. In the first embodiment, multiple pieces of weather data are collected during a certain period, and a learning data set is created from the multiple pieces of weather data. In the second embodiment, multiple pieces of weather data are collected during at least two periods. The elevation angle of the antenna 6 is set to the same angle during at least two periods. In the third embodiment, multiple pieces of weather data are also collected during at least two periods. The elevation angle of the antenna 6 is set to different angles during at least two periods. For example, the elevation angle is set to θ0 during the first period tp0, and the elevation angle is set to θ1 during the second period tp1. The data collection period may be repeated. That is, the first period tp0 may start again after the second period tp1 ends.

[0074] The weather data collected during the two time periods tp0 and tp1 is supplied to the information processing device 16. The extraction unit 52 outputs a plurality of partial data from the weather data collected during each of the two time periods tp0 and tp1, and writes the plurality of partial data to the memory 26. The missing unit 54 outputs a plurality of partial missing data from the partial data during each of the two time periods tp0 and tp1, and writes the plurality of partial missing data to the memory 26.

[0075] 15 is a diagram illustrating an example of processing by the machine learning unit 56 according to the third embodiment. The radar device 12 sets the elevation angle of the antenna 6 to θ0 during a first time period tp0 and collects weather data, and sets the elevation angle of the antenna 6 to θ1 during a second time period tp1 and collects weather data. The extraction unit 52 outputs a plurality of first partial data for the first time period tp0 and writes the plurality of first partial data to the memory 26, and outputs a plurality of second partial data for the second time period tp1 and writes the plurality of second partial data to the memory 26. The missing data generation unit 54 outputs a plurality of second partial missing data for the second time period tp1 and writes the plurality of second partial missing data to the memory 26.

[0076] The machine learning unit 56 reads out a plurality of first partial data for the first period tp0, a plurality of second partial missing data for the second period tp1, and a plurality of second partial data for the second period tp1 from the memory 26. For ease of explanation, FIG. 15 shows a single first partial data for the first period tp0, a single second partial missing data for the second period tp1, and a single second partial data for the second period tp1.

[0077] The machine learning unit 56 uses multiple first partial data for the first period tp0 and elevation angle θ0 and multiple second partial missing data for the second period tp1 and elevation angle θ1 as input data, and uses multiple second partial data for the second period tp1 and elevation angle θ1 as training data, and performs machine learning to create a model.

[0078] The plurality of partial data for the second period tp1 and the plurality of partial missing data do not have to be written to memory 26. In this case, machine learning unit 56 uses as input data the plurality of first partial data for the first period tp0 and elevation angle θ0 read from memory 26 and the plurality of second partial missing data for the second period tp1 and elevation angle θ1 output from missing data generating unit 54, and as training data the plurality of second partial data for the second period tp1 and elevation angle θ1 output from extraction unit 52, and performs machine learning to create a model.

[0079] In the third embodiment, although there is a gap in the weather data for the second period, there is no gap in the weather data for the preceding first period, which improves the accuracy of machine learning and also improves the quality of the restored weather data.

[0080] FIG. 16 is a diagram illustrating an example of weather data according to the third embodiment. The weather data shown in FIG. 16 assumes that heavy rain exceeding 50 mm / h occurred in a small area of ​​10 km square. FIG. 16(a) shows partial data extracted from weather data collected during the second period tp1 (elevation angle θ1). FIG. 16(b) shows partially missing data created from weather data collected during the second period tp1 (elevation angle θ1). FIG. 16(c) shows partial data extracted from weather data collected during the first period tp0 (elevation angle θ0). In the partial data for the second period tp1 (elevation angle θ1) shown in FIG. 16(a), there is a heavy rain area surrounded by a black frame. However, in the partially missing data for the second period tp1 (elevation angle θ1) shown in FIG. 16(b), the heavy rain area has disappeared. In this case, it is difficult to restore the missing data simply by using the data for the second period tp1 (elevation angle θ1). On the other hand, when model learning is performed using the partially missing data for the first period tp0 (elevation angle θ0) shown in FIG. 16(c), the machine learning unit 56 can recognize the presence of a heavy rain area and can restore the missing data with high accuracy.

[0081] [Fourth embodiment] 17 is a diagram showing an example of functional blocks of an information processing device 16 according to the fourth embodiment. The functional blocks of the fourth embodiment are configured by adding an interference detection unit 72 to the functional blocks of the first embodiment shown in FIG. 3. The interference detection unit 72 receives weather data 72 and detects the occurrence of interference. The detection result of the interference detection unit 72 is supplied to a detection unit 74.

[0082] The radar device 12 may not be able to observe weather data due to interference from radio waves from other weather radars or wireless systems. Interference occurs continuously for a short period of time when the main lobe of the interfered radar device 12 and the main lobe of the interfering system face each other.

[0083] FIG. 18 is a diagram illustrating an example of interference according to the fourth embodiment. Interference occurs in a radial range centered on the rotation axis of the antenna 6 of the radar device 12. The azimuth range of the interference range is relatively narrow. It is difficult to collect weather data while interference is occurring. If weather data in the interference range cannot be collected, minute weather phenomena will be overlooked. The quality information does not include a flag indicating the interference range. Therefore, the information processing device 16 cannot detect interference based on the quality information. The fourth embodiment detects interference from weather data and restores missing data that suddenly occurs due to interference.

[0084] The model creation unit 40 creates a model (referred to as a first model) in the same manner as in the first to third embodiments. Furthermore, the model creation unit 40 also creates a second model. When creating the second model, the defect identification unit 50 sets the maximum azimuth range of interference that may occur based on past observation results and identifies that range as the missing range. The maximum azimuth range of interference that may occur may be set from the interference range that occurs in observations on clear days. The extraction unit 52 extracts partial data whose azimuth size is equal to or greater than the maximum interference range. The azimuth position of the partial data in the weather data is arbitrary. The missing unit 54 deletes a range of azimuth directions in the partial data whose azimuth size is equal to or greater than the maximum interference range, thereby creating partially missing data. The machine learning unit 56 creates a second model that restores the missing range of the partially missing data (i.e., the maximum azimuth range of interference that may occur). The machine learning unit 56 writes the first and second models to the memory 26.

[0085] In the restoration unit 42, an interference detection unit 72 determines whether or not sudden interference has occurred and identifies the interference range (missing range). The interference detection unit 72 may determine that a point where the distance deviation of the correlation coefficient ρhv of reflected waves between polarized waves or the distance deviation of the phase difference of reflected waves between polarized waves exceeds a threshold is a point where interference has occurred. The interference detection unit 72 may be the interference determination unit described in paragraphs 0009-0012 of Patent Document 2.

[0086] The extraction unit 74 extracts input data including the interference range detected by the interference detection unit 72 from the weather data. The processing by the model application unit 76 and the replacement unit 78 is similar to the processing in the first to third embodiments. However, if the interference detection unit 72 detects interference, the model application unit 76 reads the second model from the memory 26. If the interference detection unit 72 does not detect interference, the model application unit 76 reads the first model from the memory 26.

[0087] The above processing makes it possible to restore data lost due to sudden interference, enabling accurate detection of signs of dangerous weather phenomena.

[0088] [Fifth embodiment] FIG. 19 is a diagram illustrating an example of functional blocks of an information processing device 16 according to a fifth embodiment. The fifth embodiment has the functions of the second, third, and fourth embodiments. The information processing device 16 according to the fifth embodiment has the same functional blocks as the information processing device 16 according to the fourth embodiment shown in FIG. 17. The fifth embodiment differs from the fourth embodiment in the manner in which the functional blocks are connected to each other. In the fifth embodiment, partial data output from the extraction unit 52 and partial missing data output from the missing data generating unit 54 are written to the memory 26.

[0089] In the fifth embodiment, multiple weather data are collected during at least three periods tp0, tp1, and tp2. After the first period tp0 ends, the second period tp1 begins. After the second period tp1 ends, the third period tp2 begins. The data collection periods may be repeated. That is, after the third period tp2 ends, the first period tp0 may begin again.

[0090] 20 is a diagram illustrating an example of processing by the machine learning unit 56 according to the fifth embodiment. The radar device 12 sets the elevation angle of the antenna 6 to θ1 during a first period tp0 and collects weather data, sets the elevation angle of the antenna 6 to θ0 during a second period tp1 and collects weather data, and sets the elevation angle of the antenna 6 to θ1 during a third period tp2 and collects weather data. Alternatively, the radar device 12 may set the elevation angle of the antenna 6 to θ0 during a first period tp0 and collects weather data, sets the elevation angle of the antenna 6 to θ1 during a second period tp1 and collects weather data, and sets the elevation angle of the antenna 6 to θ1 during a third period tp2 and collects weather data.

[0091] The weather data collected during three periods tp0, tp1, and tp2 is supplied to the information processing device 16. The extraction unit 52 outputs a plurality of partial data from the weather data collected during each of the three periods tp0, tp1, and tp2, and writes the plurality of partial data to the memory 26. The missing unit 54 outputs a plurality of partial missing data for each of the partial data during the two periods tp0 and tp2, and writes the plurality of partial missing data to the memory 26.

[0092] The machine learning unit 56 reads from the memory 26 a plurality of first partial missing data for the first period tp0 and elevation angle θ1, a plurality of second partial data for the second period tp1 and elevation angle θ0, a plurality of third partial missing data for the third period tp2 and elevation angle θ1, and a plurality of third partial data for the third period tp2 and elevation angle θ1.

[0093] 20 shows a single first partial missing data set for a first period tp0 and an elevation angle θ1, a single second partial data set for a second period tp1 and an elevation angle θ0, a single third partial data set for a third period tp2 and an elevation angle θ1, and a single third partial data set for a third period tp2 and an elevation angle θ1. The azimuth sizes of the missing ranges of the first partial missing data set and the third partial missing data set are equal to or greater than the azimuth size of the maximum range of interference that can occur.

[0094] The machine learning unit 56 performs machine learning using multiple first partial missing data for the first period tp0 and elevation angle θ1, multiple second partial data for the second period tp1 and elevation angle θ0, and multiple third partial missing data for the third period tp2 and elevation angle θ1 as input data, and multiple third partial data for the third period tp2 and elevation angle θ1 as training data.

[0095] According to the fifth embodiment, since the functions of the second, third and fourth embodiments are provided, missing meteorological data can be restored with high accuracy.

[0096] The present invention is not limited to the above-described embodiments, and the components can be modified and embodied in practice without departing from the spirit of the invention. Furthermore, various inventions can be created by appropriately combining multiple components disclosed in the above-described embodiments. For example, some components may be omitted from all the components shown in the embodiments. Furthermore, components from different embodiments may be appropriately combined. [Explanation of symbols]

[0097] 12... radar device, 14... storage device, 16... information processing device, 40... model creation unit, 42... restoration unit, 50... defect identification unit, 52... extraction unit, 54... defect creation unit, 56... machine learning unit, 74... extraction unit, 76... model application unit, 78... replacement unit

Claims

1. Create partial data that is a portion of the meteorological data other than the missing range, creating partially missing data by missing a portion of the partial data; a processing unit for acquiring missing meteorological data and outputting non-missing meteorological data based on the partial data and the partially missing data; Information processing device.

2. The processing unit Creating a plurality of partial data that are portions of each of the plurality of meteorological data other than the missing range, creating a plurality of partially missing data by missing a portion of each of the plurality of partial data; The information processing apparatus according to claim 1 , wherein the non-missing meteorological data is output based on the plurality of partially missing data and the plurality of partial data.

3. The processing unit obtaining quality information regarding the quality of the meteorological data; The information processing apparatus according to claim 1 , wherein a missing range of the meteorological data is identified based on the quality information.

4. the meteorological data includes data for each point in an observation area defined by an azimuth direction and a range direction; The information processing apparatus according to claim 1 , wherein the size of the partial data in the azimuth direction is equal to or larger than the size of the missing range of the meteorological data in the azimuth direction.

5. The information processing device according to claim 4 , wherein the processing unit omits a central portion of the partial data in the azimuth direction.

6. The information processing apparatus according to claim 5 , wherein a size in the azimuth direction of the missing range of the partially missing data is equal to or larger than a size in the azimuth direction of the missing range of the meteorological data.

7. The processing unit creating first partial data that is a portion of the weather data for the first period other than the missing range; creating first partial missing data by missing a portion of the first partial data; creating second partial data that is a portion of the meteorological data for a second period after the first period other than the missing range; creating second partial missing data by missing a portion of the second partial data; The information processing apparatus according to claim 1 , wherein the non-missing weather data is output based on the first partial data, the first partially missing data, the second partial data, and the second partially missing data.

8. The processing unit creating a plurality of first partial data, which are portions of each of the plurality of weather data for the first period other than the missing range; creating a plurality of first partial missing data by respectively missing a portion of the plurality of first partial data; creating a plurality of second partial data that are portions other than the missing range of each of the plurality of weather data for a second period after the first period; creating a plurality of second partial missing data by respectively missing a portion of the plurality of second partial data; The information processing apparatus according to claim 1 , wherein the non-missing weather data is output based on the plurality of first partial data, the plurality of first partial missing data, the plurality of second partial data, and the plurality of second partial missing data.

9. The processing unit creating first partial data that is a portion of the meteorological data for the observation area at the first altitude other than the missing area; creating second partial data that is a portion of the meteorological data of the observation area at a second altitude different from the first altitude other than the missing range; creating second partial missing data by missing a portion of the second partial data; The information processing apparatus according to claim 1 , further comprising: an information processing device configured to output the non-missing weather data based on the first partial data, the second partial missing data, and the second partial data.

10. The processing unit creating a plurality of first partial data, which are portions other than the missing range of each of the plurality of meteorological data of the observation area at the first altitude; creating a plurality of second partial data, which are portions other than the missing range of each of the plurality of meteorological data of the observation area at a second altitude different from the first altitude; creating a plurality of second partial missing data by respectively missing a portion of the plurality of second partial data; The information processing apparatus according to claim 1 , further comprising: an information processing device configured to output the non-missing meteorological data based on the plurality of first partial data, the plurality of second partial missing data, and the plurality of second partial data.

11. The processing unit creating first partial data that is a portion of the meteorological data of the observation area at the second altitude or a first altitude different from the second altitude other than the missing range; creating first partial missing data by missing a portion of the first partial data; creating second partial data that is a portion of the meteorological data of the observation area at the first altitude or the second altitude other than the missing range; creating second partial missing data by missing a portion of the second partial data; creating third partial data that is a portion of the meteorological data for the observation area at the second altitude other than the missing range; creating third partial missing data by missing a portion of the third partial data; The information processing apparatus according to claim 1 , wherein the non-missing weather data is output based on the first partial missing data, the second partial data, the third partial missing data, and the third partial data.

12. The processing unit creating a plurality of first partial data, which are portions other than the missing range of each of the plurality of meteorological data of the observation area at the second altitude or a first altitude different from the second altitude; creating a plurality of first partial missing data by respectively missing a portion of the plurality of first partial data; creating a plurality of second partial data that are portions other than the missing range of each of the plurality of meteorological data of the observation area at the first altitude or the second altitude; creating a plurality of second partial missing data by respectively missing a portion of the plurality of second partial data; creating a plurality of third partial data, which are portions other than the missing range of each of the plurality of meteorological data of the observation area at the second altitude; creating a plurality of third partial missing data by respectively missing a portion of the plurality of third partial data; The information processing apparatus according to claim 1 , further comprising: an information processing device configured to output the non-missing weather data based on the plurality of first partial missing data, the plurality of second partial data, the plurality of third partial missing data, and the plurality of third partial data.

13. The information processing apparatus according to claim 1 , wherein the processing unit creates the partial data that is a portion of the weather data at a position corresponding to a position of a missing range of the weather data.

14. The information processing device according to claim 1 , wherein the processing unit creates a model based on the partial data and the partial missing data, in which the missing weather data is used as an input and the non-missing weather data is used as an output, and restores the missing weather data using the model.

15. The processing unit creating a plurality of partially missing data by respectively missing a plurality of portions at different positions of the partial data; creating a plurality of the models based on the plurality of partially missing data; The information processing apparatus according to claim 14 , wherein after the plurality of models are created, a model according to the position of a missing range of the meteorological data is used.

16. The processing unit The information processing apparatus according to claim 15 , wherein after the plurality of models are created, a model corresponding to the position of the missing range in ascending order of the missing ranges of the meteorological data is used.

17. The processing unit creating a second model that reconstructs the maximum extent of interference in the weather data; The information processing apparatus according to claim 14 , wherein, when interference is detected from the meteorological data, a loss in the meteorological data due to the interference is restored using the second model.

18. The information processing device according to claim 7, wherein the processing unit creates a model based on the second partial data, the first partial missing data, and the second partial missing data, in which the missing weather data is input and the non-missing weather data is output, and restores the missing weather data using the model.

19. 10. The information processing device according to claim 9, wherein the processing unit creates a model based on the first partial data, the second partial missing data, and the second partial data, in which the missing weather data is input and the non-missing weather data is output, and restores the missing weather data using the model.

20. The information processing device according to claim 11, wherein the processing unit creates a model based on the second partial data, the third partial data, the first partial missing data, and the third partial missing data, in which the missing weather data is used as input and the non-missing weather data is used as output, and restores the missing weather data using the model.

21. Create partial data that is a portion of the meteorological data other than the missing range, creating partially missing data by missing a portion of the partial data; A method for obtaining missing weather data and outputting non-missing weather data based on the partial data and the partially missing data.

22. The computer creates partial data that is a portion of the meteorological data other than the missing area, causing the computer to delete a portion of the partial data to create partially deleted data; A program that causes the computer to acquire missing meteorological data and output non-missing meteorological data based on the partial data and the partially missing data.

Images