High-resolution weather mesh data calculation device and calculation system

Abstract

Obtaining high-resolution weather data in real time. [Solution] The high-resolution weather mesh data calculation device 1 according to the present invention comprises: a pre-calculation unit 111 that pre-calculates a diffusion coefficient 131 for diffusing each data point of the first mesh data 132 to each point of the second mesh data 133; a weather mesh data acquisition unit 112 that acquires weather mesh data as the first mesh data 132; and a generation unit 113 that generates second mesh data of a specified resolution based on the first mesh data 132 and the diffusion coefficient 131. The pre-calculation unit 111 pre-calculates a diffusion coefficient 131 related to the second mesh data 133, which has a different resolution from the first mesh data 132; and the generation unit 113 generates the second mesh data 133 by a process that includes a procedure for diffusing the weather data at each point of the first mesh data 132 to each point of the second mesh data 133 based on the diffusion coefficient 131.

Description

[Technical Field]

[0001] This invention relates to a calculation device and calculation system for high-resolution meteorological mesh data. [Background technology]

[0002] In acquiring meteorological data, the resolution of the data may differ from the desired resolution. Reasons for this include, for example, the limitations of the resolution of the meteorological data or forecast data itself acquired from external sources, the limitations of the observation resolution of meteorological observation or forecasting means, and the cost of installing meteorological observation equipment at observation points.

[0003] As mentioned above, it can be difficult to obtain weather data with the desired resolution, so there is a demand for higher resolution weather data. This process of calculating more detailed weather data based on coarse weather data is also called "downscaling." For weather data that requires rapid reporting, such as the 30-minute atmospheric analysis data, real-time downscaling may be required.

[0004] Regarding a technology for increasing the resolution of meteorological data, Patent Document 1 discloses an information processing device comprising: a learning means for generating or updating and learning a model that generates meteorological information of a target spatiotemporal resolution based on meteorological information with a lower spatiotemporal resolution; and an inference means for generating and inferring first information, which is meteorological information of a desired resolution, based on the model and second information, which is meteorological information of a lower resolution.

[0005] The technology described in Patent Document 1 can improve the convenience of downscaling weather information. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2021-192026 [Overview of the Initiative] [Problems that the invention aims to solve]

[0007] Incidentally, the technology described in Patent Document 1 achieves high resolution through inference using a machine learning model, i.e., a learned model. Therefore, due to the large amount of computation required, this technology may not be able to complete the inference for high resolution in real time. Thus, the technology described in Patent Document 1 has room for further improvement in terms of obtaining high-resolution weather data in real time.

[0008] The object of the present invention is to provide a technical means for obtaining high-resolution weather data in real time. [Means for solving the problem]

[0009] As a result of diligent research to solve the above problems, the inventors of the present invention have found that the above objectives can be achieved by a process that includes a procedure for spreading meteorological data from each observation point onto a meteorological mesh data grid based on a pre-calculated diffusion coefficient, and other technical features. The inventors have now completed the present invention.

[0010] One aspect of the present invention comprises: a pre-calculation unit that pre-calculates a diffusion coefficient for diffusing each data point of a first mesh data to each point of a second mesh data; a weather mesh data acquisition unit that acquires weather mesh data as the first mesh data; and a generation unit that generates the second mesh data based on the first mesh data and the diffusion coefficient, wherein the pre-calculation unit calculates the diffusion coefficient for the second mesh data which has a different resolution from the first mesh data. Using random numbers that follow a Laplace distribution The present invention provides a high-resolution weather mesh data calculation device that pre-calculates and generates second mesh data by a process that includes a procedure for diffusing weather data at each point of the first mesh data to each point of the second mesh data based on the diffusion coefficient.

[0011] In the above calculation process, each data of the first mesh data, which is meteorological mesh data, is diffused to each point based on a pre-calculated diffusion coefficient, thereby generating the second mesh data, which is meteorological mesh data with different resolutions. As a result, the above calculation process can achieve both the generation of high-resolution mesh data with spatial consistency and the realization of real-time performance through reduction of the calculation load in the real-time process of generating meteorological mesh data with different resolutions. Furthermore, by employing a diffusion coefficient based on the Laplace distribution, it is possible to prevent meteorological data related to local weather events from excessively affecting a wide range of meshes.

[0012] As described above, the above calculation process can provide a technical means for obtaining high-resolution meteorological data in real time.

[0013] In addition to the above aspect (hereinafter referred to as (1)), the present invention can take various aspects exemplified below: (2) An aspect of realizing real-time high-resolution conversion of more meteorological mesh data by distributed processing on a distributed mesh server; (3) Providing a cloud that performs management of aggregating and compressing meteorological data corresponding to lower-layer meshes in units of upper-layer mesh codes, and making the calculation device acquire this meteorological data via an API, thereby separating the external data processing function and the calculation device and improving the storage utilization efficiency in the system. These aspects contribute to providing technical means for high-resolution conversion of meteorological data in real time by the effects brought about by adding their respective specific configurations.

Effects of the Invention

[0014] As described above, the present invention can provide a technical means for obtaining high-resolution meteorological data in real time.

Brief Description of the Drawings

[0015] [Figure 1] FIG. 1 is a block diagram showing an example of the hardware configuration and software configuration of the system S in the present embodiment. [Figure 2]Figure 2 shows an example of a diffusion coefficient of 131. [Figure 3] Figure 3 shows an example of the first mesh data 132. [Figure 4] Figure 4 shows an example of the second mesh data 133. [Figure 5] Figure 5 is a schematic diagram of the calculation logic for generating the second mesh data 133. [Figure 6] Figure 6 is a schematic diagram of the system configuration using the API. [Figure 7] Figure 7 is a main flowchart showing an example of a preferred flow of calculation processing performed by the computing device 1 of this embodiment. [Figure 8] Figure 8 is a continuation of Figure 7. [Modes for carrying out the invention]

[0016] Firstly, although the following disclosures, figures, and / or claims are described either individually or in combination with one or more other aspects, the subject matter of the immediate disclosure is not intended to be limited in that way. That is, the immediate disclosures, figures, and claims are intended to encompass the various aspects described herein, either individually or in one or more combinations with each other. For example, even if the immediate disclosure describes and illustrates the first, second, and third embodiments in such a way that the first embodiment is described and illustrated particularly in relation to the second embodiment, or the second embodiment is described and illustrated only in relation to the third embodiment, the immediate disclosures and illustrations are not limited in that way and may include only the first embodiment, only the second embodiment, only the third embodiment, or one or more combinations of the first, second, and / or third embodiments, such as the first and second embodiments, the first and third embodiments, the second and third embodiments, or the first, second, and third embodiments.

[0017] In this text, the phrase "or" means a non-exclusive arrangement unless explicitly specified otherwise. This non-exclusive arrangement is also stipulated in the Japanese Industrial Standard (JIS) "Format and Preparation Method of Standards Documents JIS Z 8301".

[0018] The following describes in detail an example of an embodiment of the present invention with reference to the drawings.

[0019] <System S> Figure 1 is a block diagram showing an example of the hardware and software configuration of System S in this embodiment. The following is a description of a preferred example of the hardware and software configuration of System S in this embodiment, using Figure 1.

[0020] The high-resolution weather mesh data calculation system (System S) according to this embodiment comprises a high-resolution weather mesh data calculation device 1. Preferably, System S further comprises a terminal T that communicates with the calculation device 1 via a network N. This configuration allows for the realization of a System S with high user convenience by configuring the calculation device 1 as a high-performance server (e.g., a distributed mesh data server) and allowing the terminal T to utilize the calculation device 1.

[0021] [Computing device 1] Computing device 1 comprises a control unit 11, a storage unit 13, a communication unit 14, and other hardware components.

[0022] [Configuration as a Mesh Server] In order to perform efficient calculations using map data in mesh data format, it is preferable that the computing device 1 be a mesh server that manages mesh data. The mesh data may be, for example, primary mesh sections (approximately 80 km per side), secondary mesh sections (approximately 10 km per side), tertiary mesh sections (standard regional mesh, approximately 1 km per side), quaternary mesh sections (approximately 500 m per side), quintuple mesh sections (approximately 250 m per side), quaternary mesh sections (approximately 125 m per side), extended 100 m mesh, extended 10 m mesh, extended 1 m mesh, and other various regional meshes or meshes usable on a global scale. In addition, the mesh data may be hierarchical mesh data that combines regional meshes or meshes usable on a global scale with different granularities.

[0023] (Regarding the configuration as a distributed mesh data server) In order to enable real-time processing of a large amount of weather mesh data, it is preferable that the computing device 1 be configured as a distributed mesh data server that includes multiple server devices configured to perform parallel processing on mesh data.

[0024] A distributed mesh data server preferably divides the calculations for a large number of meshes among its servers and processes them in parallel. Hash partitioning, data range partitioning, and other conventional computation partitioning methods can be applied to the calculations. For example, if each server corresponds to or is associated with a region, a partitioning method is possible where the calculations are divided by region and assigned to each server.

[0025] To prevent data leakage between users, it is preferable for distributed mesh data servers to manage data access through login processing and encryption. To prevent data leakage during inter-server communication, it is preferable for distributed mesh data servers to encrypt communications using symmetric-key cryptography, public-key cryptography, or other encryption methods.

[0026] [Control Unit 11] The control unit 11 includes a central processing unit (CPU), random access memory (RAM), read-only memory (ROM), and other hardware components.

[0027] The control unit 11 cooperates appropriately with at least one of the storage unit 13 and the communication unit 14 to enable the use of software components related to the program of this embodiment executed on the computing device 1. The computing device 1 includes a pre-calculation unit 111, a weather data acquisition unit 112, and a generation unit 113 as software components.

[0028] Details of the computational processing implemented by the software components described above will be explained later with reference to Figures 5 and 6.

[0029] [Storage Unit 13] The storage unit 13 is a device on which data and / or files are stored, and has a storage unit for non-temporarily storing data. The storage unit includes, for example, a hard disk, semiconductor memory, recording medium, memory card, and other storage materials. The storage unit 13 stores programs that are executed by the control unit 11.

[0030] The memory unit 13 stores the diffusion coefficient 131. The memory unit 13 may also store first mesh data 132 and / or second mesh data 133.

[0031] (Diffusion coefficient 131) In the computing device 1, the diffusion coefficient 131 is stored in the storage unit 13 in order to generate the second mesh data 133 by diffusing the data related to each point of the first mesh data 132. The diffusion coefficient 131 stores the diffusion coefficient that has been pre-calculated by the pre-calculation unit 111. As a result, the computing device 1 can perform real-time high-resolution processing of meteorological mesh data with fewer computing resources than when sequential calculations are performed without using the pre-calculated coefficient.

[0032] One possible format for the diffusion coefficient 131 is one that corresponds to each mesh (diffusion source) on the first mesh data 132. In this format, for each diffusion source mesh, multiple meshes (diffusion destinations on the second mesh data 133) that are the destinations of the diffusion from that source, and the corresponding diffusion coefficients are associated and recorded.

[0033] In this format, diffusion processing can be performed sequentially or in parallel from each source mesh. This allows the computing device 1 to achieve parallel processing by focusing on the first mesh data 132, which is the source mesh, and dividing the processing accordingly. In the processing of the generation unit 113 based on this format, even if the diffusion coefficient has a spatially sparse spread based on a Laplace distribution or other random distribution, the sum can be normalized in a later stage, enabling processing that maintains numerical consistency.

[0034] Furthermore, the diffusion coefficient 131 can also be expressed in a format that corresponds to each mesh (diffusion destination) on the second mesh data 133. In this format, for each diffusion destination mesh, the group of source meshes and the diffusion coefficient necessary to generate its output value are recorded as a list.

[0035] In the processing of the generation unit 113 based on this format, the necessary calculations are completed for each diffusion destination. Therefore, the computing device 1 can efficiently derive the output value with a single vector operation between the first mesh data 132 and the coefficients related to the diffusion destination mesh. Thus, the computing device 1 can perform parallel processing without causing write conflicts by spatially dividing each diffusion destination mesh as a processing unit.

[0036] The diffusion coefficient 131 may be configured in which multiple sets of diffusion coefficients are pre-calculated and stored. In this configuration, when the computing device 1 continuously increases the resolution of weather mesh data along a time series, for example, it can appropriately select from multiple sets of diffusion coefficients, thereby suppressing the continuous propagation of bias originating from a single random number sequence. Examples of methods by which the computing device 1 makes the above selection include: a method of sequentially switching and repeating the diffusion coefficient sets; and a method of randomly selecting from multiple sets of diffusion coefficients.

[0037] Figure 2 shows an example of the diffusion coefficient 131. In this example, each mesh (diffusion source) in the first mesh data 132 is associated with the diffusion destination mesh in the second mesh data 133 and the diffusion coefficient for each diffusion destination mesh, and these are stored as a list. As a result, the generation unit 113 can generate the values ​​for each mesh in the second mesh data 133 by multiplying each mesh in the first mesh data 132 by the diffusion coefficient and then normalizing the diffusion coefficients so that the sum of the diffusion coefficients is 1 for each diffusion destination.

[0038] In this example, multiple meshes (e.g., "(617,786)", "(618,786)", "(617,787)", "(618,787)") included in the first mesh data 132 are used as diffusion sources, and the diffusion coefficients assigned from each diffusion source to the surrounding diffusion destination meshes are shown.

[0039] Of these, for the source mesh "(617,787)", high coefficients of 47, 48, 37, and 37 are assigned to the destination meshes that are close on the map ("5339-44-18", "5339-44-19", "5339-44-28", and "5339-44-29"). On the other hand, moderate coefficients of 11, 11, and 12 are assigned to the destination meshes that are a little further away on the map (e.g., the eight meshes adjacent to the aforementioned close destination meshes, such as "5339-44-08", "5339-44-09", and "5339-44-17"). Furthermore, low coefficients of 2 are assigned to the destination meshes that are located diagonally from the aforementioned close destination meshes and are farther away (e.g., "5339-44-07" and "5339-45-00").

[0040] The coefficients shown in this list were pre-calculated by the pre-calculation unit 111 using random numbers following a Laplace distribution. Therefore, each coefficient is designed to more strongly reflect the influence of the source mesh closer to the destination mesh. As a result, the computing device 1 using the diffusion coefficient 131 in this example prevents meteorological data related to local meteorological events (e.g., torrential downpours, strong winds) from excessively affecting a wide range of meshes. Thus, such a computing device 1 can perform simple calculations that can be executed in real time and generate appropriate mesh data that suppresses the wide-area impact mentioned above. In other words, the computing device 1 can balance the computational requirements for real-time processing with the appropriate resolution achieved through such processing.

[0041] (First mesh data 132) In order to avoid reacquisition when repeatedly generating the second mesh data 133, the storage unit 13 may also store the first mesh data 132.

[0042] The first mesh data 132 stores the weather mesh data acquired by the weather data acquisition unit 112. This weather mesh data has a resolution different from (usually lower than) the required output resolution. The computing device 1 of this embodiment is usually used to generate weather mesh data with a higher resolution than the acquired weather mesh data. However, the computing device 1 of this embodiment can also be used to generate data with a lower resolution based on the acquired weather mesh data, or to generate data with a different mesh shape.

[0043] The format of this weather mesh data is not particularly limited. For example, the format is the International Meteorological Reporting Code FM92GRIB Binary Format Grid Point Data Meteorological Reporting Code (Version 2) (hereinafter referred to as "GRIB2"). In order to match the coordinate system in high-resolution data, it is preferable that the storage unit 13 stores logic for converting the coordinate system. This logic may be, for example, logic for directly converting from the coordinate system of the format to the coordinate system of the second mesh data 133, or logic for converting to a common coordinate system (e.g., latitude and longitude coordinate system) that mediates between the coordinate system of the format and the coordinate system of the second mesh data 133.

[0044] In order to output weather mesh data related to wind, which is important in fields such as transportation, logistics, and disaster prevention, it is preferable that the first mesh data 132 stores weather mesh data indicating wind speed and wind direction. Examples of such weather mesh data include mesh data that includes both east-west wind speed and north-south wind speed, and mesh data that includes both wind direction and wind speed. It is preferable that the weather mesh data includes mesh data that includes both east-west wind speed and north-south wind speed because it is suitable for generation using the diffusion coefficient.

[0045] In order to output weather mesh data related to temperature, which is important in fields such as agriculture, daily life, disaster prevention, and other areas, it is preferable that the first mesh data 132 stores weather mesh data indicating temperature.

[0046] In order to output meteorological mesh data related to precipitation, which is important in fields such as transportation, agriculture, daily life, disaster prevention, and other areas, it is preferable that the first mesh data 132 stores meteorological mesh data indicating precipitation. Similarly, it is preferable that the first mesh data 132 stores meteorological mesh data indicating the probability of precipitation.

[0047] In addition to the above, the first mesh data 132 may also store meteorological mesh data showing numerical meteorological data such as snowfall amount, probability of snowfall, sunshine duration, effective visibility, atmospheric pressure, cloud cover, ground surface temperature, ground surface humidity, dew point temperature, freezing depth, snow depth, snow cover amount, etc. Furthermore, the first mesh data 132 is not limited to containing only meteorological data as measured values, but may also contain measured values, analyzed values, predicted values, and other meteorological data.

[0048] Figure 3 shows an example of the first mesh data 132. This example shows weather mesh data for an area of ​​2 squares in the east-west direction and 3 squares in the north-south direction, with the mesh number in the upper left corner being (617,786) and the mesh number in the lower right corner being (618,788). In this example, the east-west wind speed U [m / s], north-south wind speed V [m / s], and temperature T [°C] corresponding to each mesh on the map are shown.

[0049] This example illustrates a situation where weather conditions differ between the north and south. In the northern meshes ("(617,786)" and "(618,786)"), the wind speed components are all negative (east-west wind speed U: -2 m / s in both cases, north-south wind speed V: -2 m / s, -3 m / s). Also, the temperature in the north is low (23.0°C, 21.5°C, respectively). In contrast, in the southern meshes ("(617,788)" and "(618,788)"), the wind speed components are positive (east-west wind speed U: +2 m / s, +3 m / s, north-south wind speed V: +1 m / s, +2 m / s). The temperature in the south is high (28.0°C, 30.0°C, respectively). The central mesh ("(617,787)", "(618,787)") shows wind speed and temperature that are intermediate between the north and south meshes.

[0050] In other words, this example shows a weather condition where the temperature rises towards the south, and the wind direction reverses between north and south in conjunction with the temperature distribution (southwesterly winds in the north and northeasterly winds in the south). Furthermore, this example also contains localized low-temperature regions ("(618,786)") and localized high-temperature regions ("(618,788)"). Therefore, the process of increasing the resolution of the first mesh data 132 in this example requires real-time interpolation in line with the overall weather conditions, while also ensuring that these localized regions do not have an excessive impact on the surrounding areas.

[0051] (Second Mesh Data 133) In order to avoid regeneration when the second mesh data 133 is used repeatedly, it is preferable that the storage unit 13 further stores the second mesh data 133.

[0052] The second mesh data 133 stores weather mesh data generated by the generation unit 113, which has a different resolution from the first mesh data 132. The resolution of the second mesh data 133 is the required output resolution. Normally, this output resolution is higher resolution than the first mesh data 132, but as described in the section on the first mesh data 132, the computing device 1 of this embodiment is not limited to this output resolution.

[0053] The format of the second mesh data 133 is not particularly limited. Preferably, the storage unit 13 stores logic for converting coordinate systems in order to match the coordinate system with the first mesh data 132. This logic may be, for example, logic for directly converting from the coordinate system of the first mesh data 132 to the coordinate system of the format, or logic for converting from the common coordinate system described in the "First Mesh Data 132" section to the coordinate system of the format.

[0054] The weather mesh data included in the second mesh data 133 may be the same as that of the first mesh data 132. In order to reduce computational load and storage usage by focusing on important data and increasing the resolution, the second mesh data 133 may be configured to include only a portion of the weather mesh data included in the first mesh data 132 (e.g., only east-west wind speed and north-south wind speed, or only east-west wind speed, north-south wind speed, and temperature).

[0055] Figure 4 shows an example of the second mesh data 133. This example shows weather mesh data for an area of ​​4 squares in the east-west direction and 6 squares in the north-south direction, with the mesh code in the upper left corner being "5339-34-98" and the mesh code in the lower right corner being "5339-45-41". In this example, the east-west wind speed U [m / s], north-south wind speed V [m / s], and temperature T [°C] corresponding to each mesh on the map are shown. Note that the units of each data in each mesh in the figure have been omitted due to the narrow display area.

[0056] In this example, in the first row counting from the northern end (from "5339-34-98" to "5339-35-91"), a southwesterly wind is blowing, with both wind speed components U and V being negative. This is thought to be because the generation using the diffusion coefficient by the calculation device 1 of this embodiment was strongly influenced by the diffusion sources "(617,786)" and "(618,786)".

[0057] In the second row (from "5339-44-98" to "5339-45-01"), the wind is generally from the southwest, but only "5339-45-01" shows a southerly wind. This is likely due to the influence of the east-southeasterly wind data at the diffusion source "(618,787)" in addition to the diffusion source mentioned above.

[0058] In the third row (from "5339-44-18" to "5339-45-11"), the western half has southwesterly winds, and the eastern half has southeasterly winds. This is thought to be the result of mutual influence between diffusion sources "(617,787)" and "(618,787)", whose positions almost overlap, and diffusion sources "(617,786)" and "(618,786)", which are adjacent to them to the north.

[0059] As a result, in this example, despite the computational complexity being low enough to enable real-time processing—specifically, normalizing the first mesh data 132 (Figure 3), which is the source of diffusion, by multiplying it by the diffusion coefficient 131 (Figure 2)—high-resolution rendering that naturally reproduces wind vortices is achieved.

[0060] Furthermore, regarding temperature, the high temperatures of 28.0°C and 30.0°C at the diffusion sources "(617,788)" and "(618,788)" have a localized effect, while achieving high resolution without adversely impacting distant data. In this example, the high temperatures in these meshes strongly affect the 5th row ("5339-44-38" to "5339-45-21") and the 6th row ("5339-44-48" to "5339-45-41"). On the other hand, the effect of these high temperatures on the 4th row ("5339-44-28" to "5339-45-21") is considered to be limited.

[0061] [Communication Unit 14] The communication unit 14 is not particularly limited as long as it is configured to enable communication by connecting the computing device 1 to the network N. Examples of the communication unit 14 include a network card compatible with the Ethernet standard, and communication equipment compatible with wireless LAN and other communication equipment.

[0062] [Calculation Logic of This Embodiment] Figure 5 is a schematic diagram of the calculation logic for generating the second mesh data 133. First, the coordinate system and resolution of the first mesh data 132 are referenced by the pre-calculation unit 111, and the diffusion coefficient 131 for increasing the resolution of the second mesh data 133 is pre-calculated.

[0063] After the preliminary calculations are completed, the first mesh data 132 (e.g., 30-minute atmospheric analysis data for GRIB2) is processed by the meteorological data acquisition unit 112 to obtain data components that are subject to high-resolution enhancement (e.g., north-south wind component VGRD, east-west wind component UGRD, temperature TMP).

[0064] The data component D corresponding to the second mesh j in the second mesh data 133 j is generated by the generation unit 113 using the formula D j = Σ i d i n ji / Σ i n ji Here, d i is the data component corresponding to the first mesh i in the first mesh data 132, and n ji is the diffusion coefficient 131 from the first mesh i to the second mesh j. The numerator of this formula indicates that, within the appropriate range of the first mesh i, the sum of the values obtained by multiplying the physical quantity d i at each point by the coefficient n ji is calculated. The denominator of this formula indicates that the sum of the diffusion coefficients n ji within the above-mentioned range is calculated and used as the divisor for normalization.

[0065] That is, the generation unit 113 combines the physical quantity and the diffusion coefficient to generate a normalized diffusion result, which is used as the second mesh data 133. In this series of calculation logics, the generation unit 113 responsible for real-time high-resolution generation calculates the sum of the values obtained by multiplying the pre-calculated diffusion coefficient 131 by the first mesh data 132 and normalizes it, generating the second mesh data 133 in a procedure without complex processing. As a result, the computing device 1 can achieve both the generation of high-resolution mesh data with spatial consistency and the realization of real-time performance through the reduction of computational load in the real-time process of generating meteorological mesh data with different resolutions.

[0066] Incidentally, the Laplace distribution, also known as the double exponential distribution, is a probability density function with a location parameter μ and a scale parameter κ, and has the characteristic of high locality. The Laplace distribution centered on the first mesh i is used to calculate n ji by extracting from an independent and identical distribution in the latitude and longitude directions for a predetermined number N. Due to its high locality, the diffusion using the Laplace distribution prevents meteorological data related to local meteorological events (e.g., so-called guerrilla heavy rain, gusts) from overly affecting a wide range of meshes.

[0067] Therefore, especially when the diffusion coefficient 131 is a coefficient pre-calculated using random numbers following a Laplace distribution, the computing device 1 prevents the appropriateness of the forecast data from being compromised by the wide-ranging effects described above, even though it performs simple calculations that can be executed in real time. In other words, the computing device 1 using this coefficient can achieve both the computational requirements for appropriately increasing the resolution of meteorological data in real time and the appropriateness of the forecast data.

[0068] [Network N] The type of network N is not particularly limited, as long as it enables mutual communication between information processing devices included in system S. Examples of network N include the internet, a mobile phone network, and a wireless LAN.

[0069] [Terminal T] The type of terminal T is not particularly limited and includes, for example, a desktop personal computer, a laptop computer, a smartphone, a tablet device, and other terminal devices.

[0070] Terminal T is configured to perform, for example, the process of instructing the computing device 1 to generate high-resolution data, and the process of acquiring and displaying the high-resolution data from the computing device 1. The manner in which this process is implemented is not particularly limited, and the following examples can be given: a mode implemented by a program installed on terminal T; a mode implemented by a browser that processes static and / or dynamic data provided by the computing device 1.

[0071] [Example of a system configuration using API] Figure 6 is a schematic diagram of a system configuration using API. In this configuration, a "cloud 2" is further provided that acquires weather data from data source D and provides the pre-processed data to computing device 1 via API.

[0072] [Cloud 2 Configuration] Cloud 2 comprises a data acquisition unit (not shown), a data extraction unit (sign omitted), an API (sign omitted), and a data management unit (not shown) as software components. The hardware configuration of Cloud 2 is not particularly limited, as long as it is configured to communicate with data source D and computing device 1 via network N.

[0073] (Data Acquisition Unit) The data acquisition unit performs the process of acquiring weather data from data source D. This acquisition process is implemented, for example, as a batch process that is executed at regular intervals. As a result, Cloud 2 can provide processing target mesh data based on substantially the latest weather data in response to commands from computing device 1.

[0074] (Data Extraction Unit) The data extraction unit extracts mesh data (preprocessed meteorological data) that is subject to high-resolution processing from the above-mentioned meteorological data (first mesh data 132). The preprocessed meteorological data includes data components that are subject to high-resolution processing (e.g., north-south wind component VGRD, east-west wind component UGRD, temperature TMP). This allows Cloud 2 to provide data components subject to high-resolution processing in a format suitable for processing by the computing device 1, regardless of the format of the meteorological data (e.g., GRIB2 format).

[0075] (API) The API performs processing as an interface that provides pre-processed weather data in response to access from computing device 1. The API may also be configured to allow the user to specify the processing period or data type depending on the request. This enables system S to achieve flexible data linkage in accordance with the application processing on the computing device 1 side.

[0076] In this configuration, the computing device 1 can acquire necessary weather data at the required time using a standardized communication method with the cloud 2. As a result, the computing device 1 can delegate resource-intensive data extraction processing to the cloud 2, allowing it to concentrate on subsequent processing such as high-resolution image enhancement while maintaining both real-time capabilities and flexibility.

[0077] (Data Management Unit) When Cloud 2 performs the extraction process described above, it is preferable that the pre-processed weather data, which is the weather mesh data after preprocessing, is managed by the Data Management Unit. The Data Management Unit manages the pre-processed weather data, for example, based on the hierarchical structure of the mesh codes. More specifically, the Data Management Unit aggregates and compresses each data (pre-processed weather data) corresponding to the lower-level meshes in units of the higher-level mesh codes (e.g., the first 6 digits of the regional mesh code) for storage.

[0078] With this configuration, when Cloud 2 receives a request via API for a portion of the first mesh data 132, it can decompress that portion and provide it to computing device 1. The decompressed data can be discarded after being provided. Therefore, this configuration contributes to improving the storage utilization efficiency of system S.

[0079] In this embodiment, in order to increase the resolution of weather mesh data, the amount of data to be processed may become enormous (e.g., hundreds of thousands of files). This configuration suppresses the increase in the number of data files and contributes to improving the storage utilization efficiency of system S. Furthermore, because this configuration separates the cloud 2, which is responsible for external data processing functions, from the computing device 1, it contributes to real-time high-resolution processing of a large amount of weather mesh data.

[0080] Furthermore, from a security standpoint, it is preferable to configure a firewall between computing device 1 and cloud 2 in this configuration, allowing only the communication necessary for API use to pass through.

[0081] [Main Flowchart of Calculation Process] Figure 7 is a main flowchart showing an example of a preferred flow of the calculation process performed by the computing device 1 of this embodiment. Figure 8 is a continuation of Figure 7. The following is an example of a preferred flow of the calculation process performed by the computing device 1 of this embodiment, using Figures 7 and 8.

[0082] First, the computing device 1 performs a series of processes to pre-calculate the diffusion coefficient 131. As described above, the diffusion coefficient 131 is a coefficient used in the process of diffusing each data point of the first mesh data 132 to each point of the second mesh data 133. Steps S1 to S5 are an example of this process.

[0083] [Step S1: Determine whether to pre-calculate] The control unit 11 works in cooperation with the storage unit 13 and the communication unit 14 to enable the pre-calculation unit 111. The control unit 11 then performs a process to determine whether or not to pre-calculate the diffusion coefficient 131 using the pre-calculation unit 111 (pre-calculation determination step). If the control unit 11 determines that pre-calculation is to be performed, it moves the process to step S2; otherwise, it moves the process to step S6.

[0084] In this step, the pre-calculation unit 111 performs the above-mentioned determination by, for example, a procedure for determining whether to perform a pre-calculation if the pre-calculated diffusion coefficient 131 is not stored in the storage unit 13, a procedure for determining whether to perform a pre-calculation if a command for pre-calculation is received, and other procedures.

[0085] [Step S2: Loop through each pre-calculation target] The control unit 11 executes a process that loops through the processes from step S3 to step S4 for each pre-calculation target using the pre-calculation unit 111 (pre-calculation loop start step). The control unit 11 then moves the process to step S3.

[0086] (Subjects of Pre-calculation) In this step, "each subject of pre-calculation" refers to, for example, each point corresponding to each mesh in the first mesh data 132, which is the source of diffusion. Alternatively, "each subject of pre-calculation" may refer to each point corresponding to each mesh in the second mesh data 133, which is the destination of diffusion, in accordance with the procedure for calculating diffusion described later.

[0087] (Parallel processing of pre-computation) Furthermore, if the processing in this step is executed in parallel, each unit processing constituting the loop may be distributed among multiple processing units and executed in parallel. In this case, it is preferable that each unit processing is configured to avoid write contention in parallel processing.

[0088] One example of this configuration is one in which the entire mesh described above is divided into submeshes of roughly similar shape (for example, rectangular meshes of roughly the same size) and distributed among them. In this configuration, it is expected that the relative positions of the meshes to be processed on the submesh will be roughly the same at the same time. Therefore, the distance between processing targets tends to increase with the size of the submesh, and the probability of processing the same diffusion destination at the same time becomes very small. Thus, this configuration can basically avoid write contention in parallel processing.

[0089] [Step S3: Calculate diffusion to the second mesh data] The control unit 11 uses the pre-calculation unit 111 to perform a process to calculate the diffusion to the second mesh data 133 for the data to be processed in the loop (diffusion calculation step). The control unit 11 then moves the process to step S4.

[0090] (Use of Random Numbers) In this step, the pre-calculation unit 111 performs a calculation to spread each point (mesh; source mesh) of the first mesh data to each point (mesh; destination mesh) of the second mesh data. In order to prevent bias in the data referenced when increasing resolution, it is preferable that this calculation is performed using random numbers (physical random numbers or pseudo-random numbers).

[0091] (Distribution based on distance index) To prevent meteorological data relating to local meteorological events from affecting a wide range of meshes, it is preferable that the calculation uses random numbers that follow a distribution in which the diffusion probability decreases as the distance index between the source mesh and the destination mesh increases. The distance index in the calculation is based, for example, on the distance between representative points (e.g., center points) of the meshes. Examples of such distributions include the Laplace distribution and the Gaussian distribution.

[0092] (Use of Laplace distribution) In particular, for the technical significance explained in "Calculation logic of this embodiment," it is preferable that the calculation be performed using random numbers that follow a Laplace distribution.

[0093] The following are examples of diffusion calculation procedures (diffusion procedure using two-axis independent random numbers) and alternative examples (inverse diffusion procedure using two-axis independent random numbers). These procedures can achieve diffusion based on distance, even when the mesh shape and / or size are not uniform. The diffusion procedure using two-axis independent random numbers does not necessarily result in a constant sum of diffusion weights, but it can faithfully reproduce diffusion according to a defined probability distribution. The inverse diffusion procedure using two-axis independent random numbers does not strictly reproduce the probability distribution, but it avoids write contention during parallel processing and contributes to simplifying the normalization process of diffusion weights.

[0094] (Diffusion procedure using two-axis independent random numbers) In this procedure, the "processing target in the loop" becomes the source mesh for diffusion. (Step 1) Representative point (u i ,v i (Step 2) Identify the coordinates (u i +r1,v i (Step 3) Identify the destination mesh j corresponding to +r2). (Step 4) Increase the diffusion weight from the source mesh i by 1 in the destination mesh j. (Step 5) Repeat steps 2 and 3 a predetermined number of times.

[0095] (Descending procedure using two-axis independent random numbers) In this procedure, the "target of processing in the loop" becomes the destination mesh. (Step 1) Representative point (u' of the destination mesh j) j ,v' j (Step 2) Determine the coordinate (u' j +r1,v' j (Step 3) Identify the source mesh i corresponding to +r2). (Step 4) Increase the diffusion weight from source mesh i to destination mesh j by 1. (Step 5) Repeat steps 2 and 3 a predetermined number of times.

[0096] [Step S4: Determining whether the diffusion calculation is complete] The control unit 11 performs a process to determine whether the diffusion calculation is complete using the pre-calculation unit 111 (pre-calculation completion determination step). If the control unit 11 determines that it is complete, it moves the process to step S5; otherwise, it returns the process to step S2 and continues the loop.

[0097] In this step, the pre-calculation unit 111 achieves the above determination by, for example, a procedure for determining completion when a completion flag is set for all pre-calculation targets, a procedure for determining completion when calculation completion is notified from all processing units of parallel processing, and other procedures.

[0098] The pre-calculation unit 111 performs a process to normalize the diffusion coefficient 131 as needed (e.g., in the case of a diffusion procedure where the sum of diffusion weights is not constant). Step S5 is an example of this process.

[0099] [Step S5: Normalize the diffusion coefficient] The control unit 11 uses the pre-calculation unit 111 to perform a process to normalize the diffusion coefficient 131 (normalization step). The control unit 11 then moves the process to step S6.

[0100] In this step, the pre-calculation unit 111 calculates, for example, the diffusion weight n in the diffusion destination mesh j. ji The sum Σ i n ji The diffusion weight n from the source mesh i to the destination mesh j is calculated, and the diffusion weight n ji The sum Σ i n ji By dividing by this, a normalized diffusion coefficient of 131 is obtained.

[0101] After a series of pre-calculation processes, the computing device 1 executes a series of processes to increase the resolution of the weather mesh data. Steps S6 to S11 are an example of such processes. This process can also be used to generate lower-resolution data or data with a different mesh shape based on the acquired weather mesh data.

[0102] [Step S6: Determine whether to generate the second mesh data] The control unit 11 works in cooperation with the storage unit 13 and the communication unit 14 to determine whether or not to generate the second mesh data 133 (generation determination step). If the control unit 11 determines that it will generate the data, it moves the process to step S7; otherwise, it returns the process to step S1 and repeats the processes from step S1 to step S10.

[0103] In this step, the control unit 11 achieves the above-mentioned determination by, for example, a procedure for determining whether to generate when a generation command is received, or by other procedures.

[0104] [Step S7: Acquire weather data] The control unit 11 works in cooperation with the storage unit 13 and the communication unit 14 to make the weather data acquisition unit 112 available. The control unit 11 then uses the weather data acquisition unit 112 to acquire weather data (weather mesh data) as the first mesh data 132 (weather data acquisition step). The control unit 11 then moves the process to step S8.

[0105] The weather data acquisition unit 112 acquires weather mesh data from, for example, an external data source D. If the system S includes a cloud 2 that stores the weather mesh data acquired from data source D, the weather data acquisition unit 112 may acquire the weather mesh data from the cloud 2 via an API.

[0106] To achieve high-resolution processing independent of the weather mesh data format, it is preferable that the weather data acquisition unit 112 further performs processing to extract weather data to be processed for high resolution from the weather mesh data. When Cloud 2 performs this extraction processing, it is preferable that the weather data acquisition unit 112 acquires the extracted weather data from Cloud 2 via API. This allows the computing device 1 to delegate the resource-intensive data extraction processing to Cloud 2, and concentrate on subsequent processing such as high-resolution processing while maintaining both real-time capabilities and flexibility.

[0107] The stored weather data, format, and other characteristics of the weather mesh data shall be in accordance with the description in section "First Mesh Data 132".

[0108] [Step S8: Loop through each mesh] The control unit 11 works in cooperation with the storage unit 13 and the communication unit 14 to enable the generation unit 113. Then, the control unit 11 uses the generation unit 113 to execute a process that loops through the processes from step S9 to step S10 for each mesh (generation loop start step). The control unit 11 then moves the process to step S9.

[0109] (Each mesh to be looped) In this step, "each mesh" refers to, for example, each point corresponding to each mesh in the first mesh data 132 that serves as the source of diffusion. This allows the computing device 1 to retain only a portion of the first mesh data 132. Therefore, the method of looping through the first mesh data 132 enables high-resolution processing with excellent memory efficiency. If processing speed in parallel processing is more important than memory efficiency, "each mesh" may be each point corresponding to each mesh in the second mesh data 133 that serves as the destination of diffusion, in accordance with the procedure for calculating diffusion described later.

[0110] (Parallel Processing) Furthermore, if the processing in this step is executed in parallel, each unit processing constituting the loop may be distributed among multiple processing units and executed in parallel. In this case, it is preferable that each unit processing is configured to avoid write contention in parallel processing.

[0111] One example of this configuration is one in which the entire mesh described above is divided into submeshes of roughly similar shape (for example, rectangular meshes of roughly the same size) and distributed among them. In this configuration, it is expected that the relative positions of the meshes to be processed on the submesh will be roughly the same at the same time. Therefore, the distance between processing targets tends to increase with the size of the submesh, and the probability of processing the same diffusion destination at the same time becomes very small. Thus, this configuration can basically avoid write contention in parallel processing.

[0112] [Step S9: Generate second mesh data] The control unit 11 uses the generation unit 113 to generate second mesh data 133 for the processing target in the loop based on the first mesh data 132 and the diffusion coefficient 131 (generation step). The control unit 11 then moves the process to step S10.

[0113] In this step, the generation unit 113 generates the second mesh data 133 by a process that includes a procedure for diffusing the meteorological data at each point of the first mesh data 132 to each point of the second mesh data 133 based on the diffusion coefficient 131.

[0114] The following are examples of the generation procedure in this step (procedure for processing the source mesh) and alternative examples (procedure for processing the destination mesh). The procedure for processing the source mesh contributes to improving memory efficiency for the first mesh data 132 and improving cache utilization efficiency for the first mesh data 132. The procedure for processing the destination mesh avoids write contention during parallel processing and also contributes to simplifying the normalization process of diffusion weights.

[0115] (Procedure for processing the source mesh of diffusion) In this procedure, the "target of processing in the loop" is the source mesh i of diffusion. The generation unit 113 is a coefficient related to the source mesh i of diffusion and is a coefficient n included in the diffusion coefficient 131. ij For each of these, weather data d in the diffusion source mesh i. i Diffusion coefficient n ij The value obtained by multiplying by is the weather data D in the diffusion destination mesh j. j Add it to the total.

[0116] (Procedure for processing the destination mesh) In this procedure, the "target of processing in the loop" is the destination mesh j. The generation unit 113 is given by equation D j =Σ i d i n ji / Σ i n jiUsing this, the weather data D of the second mesh data 133 related to the diffusion destination mesh j. j Generates.

[0117] [Step S10: Determining if generation is complete] The control unit 11 performs a process to determine whether generation is complete by the generation unit 113 (generation completion determination step). If the control unit 11 determines that it is complete, it terminates the calculation process; otherwise, it returns to step S8 and continues the loop.

[0118] In this step, the generation unit 113 achieves the above determination by, for example, a procedure for determining completion when a completion flag is set for all processing targets, a procedure for determining completion when all processing units of parallel processing notify that the calculation is complete, and other procedures.

[0119] [Effects of the calculation process] The calculation process described above in this embodiment contributes to solving the problem of obtaining high-resolution weather data in real time by achieving the following technical significance.

[0120] In the calculation process described above, each data point of the first mesh data 132, which is weather mesh data, is dispersed to each point based on the diffusion coefficient 131 pre-calculated in steps S1 to S5, thereby generating the second mesh data 133, which is weather mesh data with a different resolution (steps S6 to S10). As a result, the calculation process described above can achieve both the generation of spatially consistent high-resolution mesh data and real-time performance through a reduction in computational load in a real-time process that generates weather mesh data with different resolutions.

[0121] Therefore, the above-described calculation process can provide a technical means for obtaining high-resolution weather data in real time.

[0122] Furthermore, the above-described calculation process may adopt a diffusion coefficient based on the Laplace distribution (step S3). This configuration prevents meteorological data relating to local meteorological events from excessively affecting a wide-area mesh.

[0123] Furthermore, the above-described calculation process may be carried out in a distributed manner on a distributed mesh server (step S8). This manner contributes to realizing real-time high-resolution processing for a larger amount of weather mesh data.

[0124] In addition, the calculation process described above can take the form of sequentially acquiring the necessary weather mesh data via API from Cloud 2, which manages the aggregation and compression of weather data corresponding to lower-level meshes on a higher-level mesh code basis (Step S7).

[0125] In this embodiment, in order to increase the resolution of weather mesh data, the amount of data to be processed may become enormous (e.g., hundreds of thousands of files). This embodiment suppresses the increase in the number of data files and contributes to improving the storage utilization efficiency of system S. Furthermore, this embodiment contributes to real-time high-resolution processing for a larger amount of weather mesh data by making it possible to separate the external data processing function from the computing device 1.

[0126] These embodiments, through the effects brought about by the addition of their respective unique configurations, contribute to providing technical means for real-time, high-resolution enhancement of meteorological data.

[0127] <Example of Use> The following is an example of using System S of this embodiment.

[0128] [Preparing Cloud 2] The user configures the data acquisition unit of Cloud 2 so that Cloud 2 acquires weather data from data source D at the desired time intervals. The data acquisition unit acquires weather data from data source D according to the configuration.

[0129] The acquired weather data is preprocessed by the data extraction unit. The preprocessed weather data is then aggregated and compressed by the data management unit, with the weather mesh data corresponding to the lower-level mesh being stored in units of the higher-level mesh code.

[0130] [Pre-calculation of diffusion coefficient 131] The user instructs the computing device 1 to pre-calculate the diffusion coefficient 131 via terminal T. The computing device 1 pre-calculates the diffusion coefficient 131 based on the specifications of the source mesh, the destination mesh, and the distribution used for diffusion.

[0131] [Real-time high-resolution processing] The user, via terminal T, specifies the range of the second mesh data 133 to be generated and commands the computing device 1 to perform real-time high-resolution processing. The computing device 1 acquires the weather data necessary for generating the specified second mesh data 133 as the first mesh data 132 from cloud 2.

[0132] Then, the computing device 1 generates the second mesh data 133 by dispersing the meteorological data to each point of the second mesh data 133 based on a pre-calculated diffusion coefficient 131. The generated second mesh data 133 is provided externally in real time and used as high-resolution meteorological data.

[0133] Within the scope of the concept of the present invention, those skilled in the art can conceive of various modifications and alterations. Therefore, such modifications and alterations are understood to fall within the scope of the present invention. For example, any addition, deletion, or design change of components, or addition, omission, or modification of processes, made by a person skilled in the art to the above-described embodiments, is also included within the scope of the present invention, as long as it retains the gist of the present invention. [Explanation of Symbols]

[0134] S System 1 computing device 11 Control Unit 111 Pre-calculation section 112 Weather Data Acquisition Unit 113 Generation part 13 Storage section 131 Diffusion coefficient 132 First Mesh Data 133 Second Mesh Data 14 Communications Department 2 Cloud D Data Source N Network T terminal

Claims

1. A pre-calculation unit that pre-calculates the diffusion coefficient for diffusing each data point in the first mesh data to each point in the second mesh data, A weather mesh data acquisition unit acquires weather mesh data as the first mesh data, A generation unit that generates the second mesh data based on the first mesh data and the diffusion coefficient, Equipped with, The pre-calculation unit pre-calculates the diffusion coefficient for the second mesh data, which has a different resolution from the first mesh data, using random numbers following a Laplace distribution. The generation unit generates the second mesh data by a process that includes a procedure for diffusing the meteorological data at each point of the first mesh data to each point of the second mesh data based on the diffusion coefficient. A computing device for high-resolution weather mesh data.

2. The computing device according to claim 1, wherein the generation unit generates the second mesh data by distributed processing on a distributed mesh server.

3. A computing device according to claim 1 or 2, A cloud that manages the aforementioned weather mesh data, It consists of, The aforementioned cloud includes a data management unit that manages the weather mesh data based on the hierarchical structure of the mesh code, The aforementioned data management unit aggregates, compresses, and stores each data corresponding to a lower-level mesh in units of a higher-level mesh code. The aforementioned hierarchical structure is such that multiple meshes with different resolutions are associated with each other in a one-to-many relationship, with higher-level layers having lower resolutions than lower-level layers, and each mesh of the lower-level layer corresponding to one of the higher-level layers is contained within the mesh of the first higher-level layer. A calculation system for high-resolution weather mesh data.

Images