High-resolution conversion device

Abstract

This high-resolution conversion device comprises: a high-resolution data acquisition unit that acquires high-resolution data, which is time-series data in which the resolution is high for both the time granularity and the space granularity; a low-resolution conversion unit that, on the basis of the high-resolution data, generates pseudo-low-resolution data, which is time-series data in which the resolution for both the time granularity and the spatial granularity is rougher than in the high-resolution data; a statistical information generation unit that generates statistical information relating to the high-resolution data on the basis of the high-resolution data; and a model learning unit that generates a resolution conversion model for converting low-resolution data into high-resolution data, by performing machine learning of learning data in which the high-resolution data is used as an objective variable and the pseudo low-resolution data and the statistical information are each used as an explanatory variable.

Description

High resolution device

[0001] The disclosed technology relates to a resolution increasing device.

[0002] Traditionally, time-series observation data on weather, population, traffic, electricity, etc. have been easily obtained with high-resolution data in both time and space, while future forecast data has often only been available with low-resolution data in both time and space. For example, high-resolution time-series weather data can be obtained from meteorological satellites, weather stations, and AMeDAS, but medium- to long-term weather forecasts often only have low-resolution time-series data. Furthermore, while population data can capture the ever-changing population of each regional grid using mobile spatial statistics and other human flow data, future population estimates are often only available on a quarterly basis and by municipality.

[0003] To improve the performance of mathematical models and machine learning models, it is desirable to use high-resolution time series data, and there are methods for converting low-resolution time series data into high-resolution time series data. Common methods include linear interpolation and polynomial interpolation, which interpolate time series variables. For example, Non-Patent Document 1 proposes a recursive interpolation method that expands data without damaging the characteristics of the original time series data, making it possible to accurately convert time series data into high-resolution data.

[0004] Amine Aboussalah, Minjae Kwon, Raj Patel, Cheng Chi, Chi-Guhn Lee, “Recursive Time Series Data Augmentation”, ICLR 2023, https: / / doi.org / 10.48550 / arXiv.2207.02891

[0005] However, many conventional technologies only increase the temporal granularity, and are unable to increase the resolution of variables with varying granularity in both time and space. For example, when considering data storing the weekly maximum temperature for each weather station, the weather stations are not uniformly located, resulting in a spatially varying granularity. Furthermore, the observation time of the weekly maximum temperature is inherently indefinite, and the amount of change within a day is also not uniform, resulting in a temporally varying granularity. Therefore, there is a need to increase the resolution of low-resolution time series data with varying granularity in both time and space.

[0006] The disclosed technology has been developed in consideration of the above points, and aims to provide a high-resolution device that can increase the resolution of low-resolution time series data in which both temporal and spatial granularity are non-uniform and the data is a mixture of high and low resolution.

[0007] One aspect of the present disclosure is a high-resolution device comprising: a high-resolution data acquisition unit that acquires high-resolution data, which is time series data with high resolution in both temporal granularity and spatial granularity; a low-resolution data acquisition unit that acquires low-resolution time series data with coarser temporal granularity and spatial granularity than the high-resolution data; a low-resolution data acquisition unit that generates pseudo-low-resolution data, based on the high-resolution data, which is low-resolution time series data with coarser temporal granularity and spatial granularity than the high-resolution data; a statistical information generation unit that generates statistical information regarding the high-resolution data, based on the high-resolution data; and a model learning unit that generates a resolution conversion model for converting low-resolution data into high-resolution data by machine learning learning data with the high-resolution data as a target variable and the low-resolution data or the pseudo-low-resolution data, and each of the statistical information as explanatory variables.

[0008] The disclosed technology has an advantage that it is possible to increase the resolution of low-resolution time series data in which both temporal granularity and spatial granularity are non-uniform and in which coarse and fine resolution data are mixed.

[0009] 1 is a block diagram showing an example of a hardware configuration of a high resolution enhancement device according to a first embodiment. FIG. 2 is a block diagram showing an example of a functional configuration of a high resolution enhancement device according to a first embodiment in a learning phase. FIG. 3 is a block diagram showing an example of a functional configuration of a high resolution enhancement device according to a first embodiment in an estimation phase. FIG. 4 is a flowchart showing an example of the flow of a model generation process by a high resolution enhancement processing program according to the first embodiment. FIG. 5 is a flowchart showing an example of the flow of a resolution estimation process by a high resolution enhancement processing program according to the first embodiment. FIG. 6 is a block diagram showing an example of a functional configuration of a high resolution enhancement device according to a second embodiment in an estimation phase. FIG. 7 is a flowchart showing an example of the flow of a resolution estimation process with confidence intervals by a high resolution enhancement processing program according to the second embodiment. FIG. 8 is a diagram showing examples of high resolution data and low resolution data. FIG. 9 is a block diagram showing an example of the functional configuration of a high resolution enhancement device according to a third embodiment in a learning phase. FIG. 10 is a block diagram showing an example of the functional configuration of a high resolution enhancement device according to a third embodiment in an estimation phase. FIG. 11 is a diagram showing an example of the functional configuration of a high resolution enhancement device according to the third embodiment in an estimation phase. FIG. 12 is a diagram showing an example of high resolution enhancement targeting temperature. FIG. 13 is a diagram showing an example of a three-month-ahead power demand forecast.

[0010] An example of an embodiment of the disclosed technology will be described below with reference to the drawings. Note that in each drawing, the same or equivalent components and parts are given the same reference numerals. Also, the dimensional proportions in the drawings are exaggerated for the convenience of explanation and may differ from the actual proportions.

[0011] The high resolution apparatus according to the present invention provides certain improvements over conventional approaches to high resolution of low resolution data, and represents an advancement in the field of high resolution of low resolution data.

[0012] First Embodiment FIG. 1 is a block diagram showing an example of the hardware configuration of a resolution enhancing device 10 according to a first embodiment.

[0013] 1, the resolution enhancing device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other via a bus 18 so as to be able to communicate with each other.

[0014] The CPU 11 is a central processing unit that executes various programs and controls each component. That is, the CPU 11 reads programs from the ROM 12 or storage 14 and executes the programs using the RAM 13 as a work area. The CPU 11 controls the above components and performs various arithmetic processing in accordance with the programs stored in the ROM 12 or storage 14. The ROM 12 or storage 14 stores a high-resolution processing program for executing the high-resolution processing according to this embodiment. Note that, instead of the CPU, for example, a GPU (Graphics Processing Unit) may be used.

[0015] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including the operating system and various data.

[0016] The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used to input various information to the device itself.

[0017] The display unit 16 is, for example, a liquid crystal display, and displays various information. The display unit 16 may be a touch panel type and function as the input unit 15.

[0018] The communication interface 17 is an interface for communicating with, for example, other external devices, and may use a wired communication standard such as Ethernet (registered trademark) or FDDI (Fiber Distributed Data Interface), or a wireless communication standard such as 4G, 5G, or Wi-Fi (registered trademark).

[0019] The resolution enhancing apparatus 10 according to this embodiment is implemented by a general-purpose computer such as a server computer or a personal computer (PC).

[0020] The resolution enhancement device 10 according to this embodiment acquires high-resolution data and low-resolution data, generates pseudo-low-resolution data from the acquired high-resolution data, generates statistical information related to the acquired high-resolution data, and generates a resolution conversion model by machine learning training data using the high-resolution data as a response variable and the low-resolution data or pseudo-low-resolution data and the statistical information as explanatory variables. The high-resolution data is high-resolution time series data with uniform temporal and spatial granularity. The low-resolution data or pseudo-low-resolution data is low-resolution time series data with coarser temporal and spatial granularity than the high-resolution data. The low-resolution data or pseudo-low-resolution data is coarser data with coarser temporal granularity and a mixture of coarse and dense spatial granularity compared to the high-resolution data. The resolution conversion model is a trained model for converting low-resolution data into high-resolution data. Using this resolution conversion model, high-resolution data can be estimated from input low-resolution data. Therefore, low-resolution data with coarse temporal and spatial granularity can be enhanced to a higher resolution.

[0021] Specifically, for example, temperature varies seasonally and over time, and may have irregular fluctuations due to abnormal weather. Temperature also has spatial correlation, and there are invariant characteristics, such as when one point becomes hot, other points also become hot. In other words, by inputting low-resolution data that expresses irregular fluctuations into a resolution conversion model that has learned periodic fluctuations and invariant characteristics, high-resolution data that expresses irregular fluctuations can be obtained.

[0022] Next, the functional configuration of the resolution enhancement device 10 will be described with reference to Figures 2 and 3. The resolution enhancement device 10 may be realized as a single device that executes the learning phase and the estimation phase related to the resolution enhancement process, or may be realized as separate devices that execute the learning phase and the estimation phase.

[0023] 2 is a block diagram showing an example of the functional configuration of the resolution enhancing device 10 according to the first embodiment in the learning phase. Note that while the embodiment shows a form in which pseudo low-resolution data is generated and used, it is also possible to include a low-resolution data acquisition unit 107 (described later) and use the acquired low-resolution data instead of the generated pseudo low-resolution data.

[0024] The resolution enhancing device 10 according to this embodiment includes, as functional components, a high-resolution data acquisition unit 101, a low-resolution unit 102, a statistical information generation unit 103, and a model learning unit 104. Each functional component is realized by the CPU 11 reading out a resolution enhancement processing program stored in the ROM 12 or the storage 14, expanding it in the RAM 13, and executing it.

[0025] The high-resolution data acquisition unit 101 acquires high-resolution data. As described above, the high-resolution data is high-resolution time-series data with uniform temporal and spatial granularity, and is a group of data prepared in advance. The high-resolution data includes, for example, at least one of instantaneous values, maximum values, minimum values, and average values ​​in a predetermined spatial unit (e.g., area unit, location unit, etc.), and spatial information related to the space. The high-resolution data may be actual measured values ​​or predicted values. However, it is desirable that the predicted values ​​of the high-resolution data have a small enough error from the actual values ​​so that the error can be ignored.

[0026] The resolution reducing unit 102 generates pseudo low-resolution data based on the high-resolution data acquired by the high-resolution data acquiring unit 101. As described above, the pseudo low-resolution data is low-resolution time series data with coarser temporal and spatial granularity than the high-resolution data. The pseudo low-resolution data has the same specifications as the low-resolution data used in the estimation phase described below. For example, if the granularity of the low-resolution data is meteorological station and weekly value, the granularity of the pseudo low-resolution data is also meteorological station and weekly value.

[0027] The statistical information generating unit 103 generates statistical information related to the high-resolution data based on the high-resolution data acquired by the high-resolution data acquiring unit 101. The statistical information includes at least one value, for example, a daily value, a weekly value, a monthly value, an annual value, or a normal year value, as a statistical value by region or location. The statistical information is generated based on actual values ​​or predicted values ​​of the high-resolution data. The statistical information generating unit 103 stores the generated statistical information in a statistical information database (Data Base: DB) 105. The statistical information DB 105 is stored in, for example, the storage 14.

[0028] The model learning unit 104 generates a resolution conversion model 106 for converting low-resolution data into high-resolution data by machine learning learning data in which the high-resolution data is the objective variable and the pseudo-low-resolution data and statistical information are each explanatory variables. Note that the pseudo-low-resolution data is a variable value, and the statistical information is a fixed value. The resolution conversion model 106 may use, for example, a machine learning algorithm such as LightGBM (Gradient Boosting Machine) or RandomForest, or a deep neural network (DNN) that applies deep learning. The resolution conversion model 106 is stored in, for example, the storage 14.

[0029] Furthermore, the model learning unit 104 associates the model parameters used in the machine learning of the resolution conversion model 106 with the resolution conversion model 106 and stores them in, for example, the storage 14 .

[0030] FIG. 3 is a block diagram showing an example of the functional configuration of the resolution enhancing device 10 according to the first embodiment in the estimation phase.

[0031] The high-resolution device 10 according to this embodiment includes, as functional components, a low-resolution data acquisition unit 107, a resolution estimation unit 108, and an output unit 109. Each functional component is realized by the CPU 11 reading out a high-resolution processing program stored in the ROM 12 or the storage 14, expanding it in the RAM 13, and executing it.

[0032] The low-resolution data acquisition unit 107 acquires low-resolution data. The low-resolution data is low-resolution time-series data with both a temporal granularity and a spatial granularity that are coarser than the high-resolution data. As described above, the low-resolution data has the same specifications as the pseudo-low-resolution data. The low-resolution data includes, for example, at least one actual or predicted value of daily values, weekly values, monthly values, annual values, and normal values ​​in a predetermined spatial unit (e.g., area unit, location unit, etc.), and spatial information related to the space.

[0033] The resolution estimation unit 108 uses the resolution conversion model 106 to estimate high-resolution data from the low-resolution data acquired by the low-resolution data acquisition unit 107 and statistical information acquired from the statistical information DB 105 .

[0034] The output unit 109 outputs the high-resolution data estimated by the resolution estimation unit 108. The output destination may be, for example, the display unit 16 or the storage 14.

[0035] Next, the operation of the resolution enhancing device 10 according to this embodiment will be described with reference to FIGS.

[0036] 4 is a flowchart showing an example of the flow of a model generation process using a high resolution processing program according to the first embodiment. The model generation process using the high resolution processing program is realized by the CPU 11 of the high resolution device 10 writing the high resolution processing program stored in the ROM 12 or the storage 14 into the RAM 13 and executing it.

[0037] In step S101 of FIG. 4, the CPU 11 acquires high-resolution data, as shown in FIG. 2 above, for example.

[0038] In step S102, the CPU 11 generates pseudo low-resolution data from the high-resolution data acquired in step S101, as shown in FIG. 2 above, for example.

[0039] In step S103, the CPU 11 generates statistical information on the high-resolution data from the high-resolution data acquired in step S101, as shown in Fig. 2. At this time, the CPU 11 stores the generated statistical information in the statistical information DB 105.

[0040] In step S104, the CPU 11 generates a resolution conversion model 106 for converting low-resolution data into high-resolution data by machine learning learning data with high-resolution data as the objective variable and pseudo-low-resolution data and statistical information as the explanatory variables, as shown in Figure 2 above, as an example, and terminates the model generation process using this high-resolution processing program.

[0041] 5 is a flowchart showing an example of the flow of resolution estimation processing by the high resolution processing program according to the first embodiment. The resolution estimation processing by the high resolution processing program is realized by the CPU 11 of the high resolution device 10 writing the high resolution processing program stored in the ROM 12 or the storage 14 into the RAM 13 and executing it.

[0042] In step S111 of FIG. 5, the CPU 11 acquires low-resolution data, as shown in FIG. 3 above, for example.

[0043] In step S112, the CPU 11 estimates high-resolution data from the low-resolution data acquired in step S111 and statistical information obtained from the statistical information DB 105 using the resolution conversion model 106, as shown in FIG. 3 above, as an example.

[0044] In step S113, the CPU 11 outputs the high resolution data estimated in step S112, and ends the resolution estimation process according to the high resolution processing program.

[0045] Thus, according to this embodiment, by inputting low-resolution data that expresses irregular fluctuations into a resolution conversion model that has learned the periodic fluctuations and invariant characteristics contained in statistical information, high-resolution data that expresses irregular fluctuations can be obtained.

[0046] Second Embodiment In a second embodiment, a form will be described in which a confidence interval is assigned to estimated high-resolution data using statistically estimated confidence interval data.

[0047] FIG. 6 is a block diagram showing an example of the functional configuration of a resolution enhancing device 10A according to the second embodiment in the estimation phase.

[0048] The high resolution device 10A according to this embodiment includes, as functional components, a low resolution data acquisition unit 107, a resolution estimation unit 108, an output unit 109, a section estimation unit 111, and a section assignment unit 112. Each functional component is realized by the CPU 11 reading out a high resolution processing program stored in the ROM 12 or the storage 14, expanding the program in the RAM 13, and executing it. The storage 14 of the high resolution device 10A stores a high resolution DB 110. The high resolution DB 110 stores first high resolution data acquired by the high resolution data acquisition unit 101.

[0049] The low-resolution data acquisition unit 107 acquires low-resolution data, which is time-series data with low resolution, with both temporal granularity and spatial granularity coarser than the first high-resolution data.

[0050] The resolution estimation unit 108 uses the resolution conversion model 106 to estimate second high-resolution data from the low-resolution data acquired by the low-resolution data acquisition unit 107 and statistical information acquired from the statistical information DB 105 .

[0051] The interval estimation unit 111 combines the first high-resolution data and the second high-resolution data together in time and space, and statistically estimates confidence interval data that represents the magnitude of an error that occurs when the resolution estimation unit 108 increases the resolution of the low-resolution data and the magnitude of an error contained in the low-resolution data. The confidence interval data is data that represents a confidence interval, such as a 50% confidence interval or a 95% confidence interval. The period of the second high-resolution data includes the period of the first high-resolution data, and the confidence interval data is statistically estimated within the range of the period of the first high-resolution data.

[0052] The interval assigning unit 112 assigns a confidence interval (for example, a 50% confidence interval, a 95% confidence interval, etc.) to the second high-resolution data using the confidence interval data estimated by the interval estimating unit 111 .

[0053] The output unit 109 outputs the second high-resolution data to which the confidence interval has been assigned by the interval assigning unit 112 .

[0054] 7 is a flowchart showing an example of the flow of a resolution estimation process with a confidence interval by a high resolution processing program according to the second embodiment. The resolution estimation process with a confidence interval by a high resolution processing program is realized by the CPU 11 of the high resolution device 10A writing the high resolution processing program stored in the ROM 12 or the storage 14 into the RAM 13 and executing it.

[0055] In step S121 of FIG. 7, the CPU 11 acquires low-resolution data, as shown in FIG. 6 above, for example.

[0056] In step S122, the CPU 11 estimates second high-resolution data from the low-resolution data acquired in step S121 and statistical information obtained from the statistical information DB 105 using the resolution conversion model 106, as shown in Figure 6 above, as an example.

[0057] In step S123, the CPU 11 combines the first high-resolution data and the second high-resolution data in time and space, as shown in Figure 6 above, as an example, and statistically estimates confidence interval data representing the magnitude of error that occurs when converting the low-resolution data to high resolution and the magnitude of error contained in the low-resolution data.

[0058] In step S124, the CPU 11 assigns a confidence interval to the second high-resolution data using the confidence interval data estimated in step S123, as shown in FIG. 6 above, for example.

[0059] In step S125, the CPU 11 outputs the second high-resolution data to which the confidence interval was added in step S124, as shown in FIG. 6, for example, and ends the resolution estimation process with confidence interval by this high-resolution processing program.

[0060] Thus, according to this embodiment, statistical estimation is used to evaluate the magnitude of the error that occurs when converting low-resolution data to high resolution, and the confidence interval that represents the magnitude of the error contained in the low-resolution data, and the evaluated confidence interval can be assigned to the estimated high-resolution data.

[0061] Third Embodiment Next, with reference to FIGS. 8 to 12, a case where low-resolution long-range weather forecast data is converted into high-resolution long-range weather forecast data will be described as an example of the high-resolution processing.

[0062] FIG. 8 shows examples of high-resolution data and low-resolution data. For example, second-order mesh / 30-minute weather record data provided by a weather company is used as the high-resolution data. Hereinafter, the second-order mesh / 30-minute weather record data is referred to as high-resolution weather record data. The second-order mesh refers to an area divided into 10-km squares across the entire country, and is defined by JIS X0410. Hereinafter, the second-order mesh / 30-minute value is used to predict power demand. However, higher resolution data, such as fourth-order mesh (500-m mesh) / 10-minute value, may also be used. For example, for low-resolution data, long-term weather forecast data of meteorological stations / weekly values ​​provided by a weather company is used. Hereinafter, the long-term weather forecast data of meteorological stations / weekly values ​​is referred to as long-term weather forecast low-resolution data.

[0063] 9 is a block diagram showing an example of the functional configuration of a resolution increasing device 10B according to the third embodiment in the learning phase. Note that while the embodiment shows an example in which pseudo long-term weather forecast low-resolution data is generated and used, a low-resolution data acquiring unit 107 (described later) may be provided and the acquired long-term weather forecast low-resolution data may be used instead of the generated pseudo long-term weather forecast low-resolution data.

[0064] The resolution enhancing device 10B according to this embodiment includes, as functional components, a high-resolution data acquisition unit 101, a low-resolution unit 102, a statistical information generation unit 103, and a model learning unit 104. Each functional component is realized by the CPU 11 reading out a high-resolution processing program stored in the ROM 12 or the storage 14, expanding it in the RAM 13, and executing it.

[0065] The high-resolution data acquisition unit 101 acquires the above-mentioned actual weather data. The resolution reduction unit 102 generates pseudo long-term weather forecast low-resolution data based on the actual weather data acquired by the high-resolution data acquisition unit 101. The statistical information generation unit 103 generates statistical information related to the actual weather data based on the actual weather data acquired by the high-resolution data acquisition unit 101. The statistical information generation unit 103 stores the generated statistical information in a statistical information DB 105. The statistical information DB 105 is stored in, for example, the storage 14. The model learning unit 104 generates a resolution conversion model 106 for converting the long-term weather forecast low-resolution data into long-term weather forecast high-resolution data by machine learning the learning data using the actual weather data as a target variable and the pseudo long-term weather forecast low-resolution data and the statistical information as explanatory variables. The resolution conversion model 106 is stored in, for example, the storage 14.

[0066] The statistical information includes at least one of daily values, weekly values, monthly values, annual values, and average values ​​created from the actual weather data. At least one of these daily values, weekly values, monthly values, annual values, and average values ​​is a value related to a weather element, such as maximum temperature, minimum temperature, or average temperature. The resolution enhancement device 10B may create a resolution conversion model 106 for each element of the high resolution enhancement, such as a resolution conversion model for temperature, a resolution conversion model for humidity, or a resolution conversion model for solar radiation. Furthermore, for weather elements that are not present in the long-term weather forecast but are present in the actual weather data, if the relationship between the weather elements can be incorporated as an explanatory variable, the weather element that is not present in the long-term weather forecast may be added as an element of the high resolution enhancement.

[0067] FIG. 10 is a block diagram showing an example of the functional configuration of a resolution enhancing device 10B according to the third embodiment in the estimation phase.

[0068] The high-resolution device 10B according to this embodiment includes, as functional components, a low-resolution data acquisition unit 107, a resolution estimation unit 108, an output unit 109, a section estimation unit 111, and a section assignment unit 112. Each functional component is realized by the CPU 11 reading out a high-resolution processing program stored in the ROM 12 or the storage 14, expanding it in the RAM 13, and executing it. The storage 14 of the high-resolution device 10B stores a weather record DB 113. The weather record DB 113 stores high-resolution weather record data acquired by the high-resolution data acquisition unit 101.

[0069] The low-resolution data acquisition unit 107 acquires long-term weather forecast low-resolution data. The resolution estimation unit 108 uses the resolution conversion model 106 to estimate long-term weather forecast high-resolution data from the long-term weather forecast low-resolution data acquired by the low-resolution data acquisition unit 107 and statistical information acquired from the statistical information DB 105. The interval estimation unit 111 combines the long-term weather forecast high-resolution data and the actual weather data in time and space, and statistically estimates confidence interval data representing the magnitude of error that occurs when the resolution estimation unit 108 converts the long-term weather forecast low-resolution data into high-resolution data and the magnitude of error contained in the long-term weather forecast low-resolution data. The long-term weather forecast high-resolution data includes the period of the actual weather high-resolution data, and the confidence interval data is statistically estimated within the period of the actual weather high-resolution data. The interval assignment unit 112 assigns a confidence interval (e.g., a 50% confidence interval, a 95% confidence interval, etc.) to the long-term weather forecast high-resolution data using the confidence interval data estimated by the interval estimation unit 111. The output unit 109 outputs the long-range weather forecast high-resolution data to which the confidence intervals have been assigned by the interval assigning unit 112 .

[0070] That is, the high-resolution device 10B estimates a high-resolution second-mesh / 30-minute long-term weather forecast by inputting the long-term weather forecast of the meteorological station / weekly value and the statistical information obtained from the statistical information DB 105 as low-resolution data into the resolution conversion model 106. The high-resolution device 10B then assigns a confidence interval to the high-resolution long-term weather forecast data using the confidence interval data estimated from the high-resolution long-term weather forecast data and the high-resolution weather performance data, and outputs high-resolution long-term weather forecast data with a confidence interval.

[0071] Here, the confidence interval data is estimated by, for example, combining the actual weather data of a second-order mesh / 30-minute value with the high-resolution second-order mesh / 30-minute value long-term weather forecast in space and time, expressing the magnitude of the error in a confidence interval such as a 50% confidence interval or a 95% confidence interval, and outputting it as confidence interval data. Note that it is desirable to estimate the confidence interval data for each element of high resolution, and it may be calculated, for example, as confidence interval data for temperature, confidence interval data for humidity, confidence interval data for solar radiation, etc.

[0072] FIG. 11 shows an example of high-resolution temperature forecasting. In this embodiment, high-resolution temperature forecasting involves upgrading a long-term weather forecast based on meteorological station / weekly values ​​to a second-order mesh / 30-minute value. It can be seen that the temperature distribution in the high-resolution long-term weather forecast is relatively close to the true value. A long-term weather forecast based on meteorological station / weekly values ​​cannot precisely consider the temperature for each second-order mesh, and only provides temperatures such as the maximum temperature, minimum temperature, and average temperature for the week, without providing hourly temperatures. In contrast, a high-resolution long-term weather forecast can consider the temperature for each second-order mesh and can also provide hourly temperatures. It has been confirmed that high-resolution long-term weather forecasts reduce errors in meteorological elements such as temperature, humidity, and solar radiation compared to when using normal values.

[0073] Fig. 12 is a diagram showing an example of a three-month-ahead power demand forecast. In Fig. 12, the vertical axis represents power demand, and the horizontal axis represents date and time (in 30-minute increments). Solid lines represent actual values, and dotted lines represent predicted values. Dotted hatching indicates a 95% confidence interval, and solid hatching indicates a 50% confidence interval.

[0074] As shown in Figure 12, by performing an electricity demand forecast using the above-mentioned high-resolution long-term weather forecast data with confidence intervals, it is possible to assign a confidence interval to the forecast result. Although it is difficult to accurately forecast the weather several months in advance, for example, the confidence interval can be used to understand how much risk there is of the predicted temperature value being over or underestimated, and by applying the result to an electricity demand forecast, it is possible to express as a forecast confidence interval how much risk there is of the demand amount being over or underestimated.

[0075] As described above, according to this embodiment, high-resolution long-term weather forecast data can be obtained by inputting low-resolution long-term weather forecast data into a resolution conversion model that has learned periodic fluctuations and invariant characteristics.

[0076] In addition, statistical estimation can be used to evaluate the magnitude of the error that occurs when converting low-resolution long-term weather forecast data into high-resolution data, and the confidence interval that represents the magnitude of the error contained in the low-resolution long-term weather forecast data, and the evaluated confidence interval can be assigned to the estimated high-resolution long-term weather forecast data.

[0077] The high-resolution processing executed by the CPU after reading the program in the above embodiment may be executed by various processors other than the CPU. Examples of such processors include a programmable logic device (PLD) (such as a field-programmable gate array (FPGA)) whose circuit configuration can be changed after manufacture, and a dedicated electrical circuit, such as an application-specific integrated circuit (ASIC), which is a processor having a circuit configuration designed specifically for executing specific processing. The high-resolution processing may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, or a combination of a CPU and an FPGA). The hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor elements.

[0078] In the above embodiment, the high-resolution processing program is described as being pre-stored (also referred to as "installed") in a ROM or storage, but the present invention is not limited to this. The high-resolution processing program may be provided in a form stored in a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The high-resolution processing program may also be downloaded from an external device via a network.

[0079] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[0080] The following additional notes are provided regarding the above-described embodiments.

[0081] (Supplementary Item 1) A high-resolution device comprising: a memory; and at least one processor connected to the memory, wherein the processor acquires high-resolution data, which is time series data with high resolution in both temporal granularity and spatial granularity; acquires low-resolution data, which is time series data with low resolution in both temporal granularity and spatial granularity coarser than the high-resolution data; generates pseudo-low-resolution data, which is time series data with low resolution in both temporal granularity and spatial granularity coarser than the high-resolution data, based on the high-resolution data; generates statistical information related to the high-resolution data, based on the high-resolution data; and generates a resolution conversion model for converting low-resolution data into high-resolution data by machine learning learning data that uses the high-resolution data as a target variable and each of the low-resolution data or the pseudo-low-resolution data and the statistical information as explanatory variables.

[0082] (Supplementary Item 2) A non-transitory storage medium storing a program executable by a computer to execute a high-resolution processing, wherein the high-resolution processing: acquires high-resolution data, which is time series data with high resolution in both temporal granularity and spatial granularity; acquires low-resolution data, which is time series data with low resolution in both temporal granularity and spatial granularity coarser than the high-resolution data; generates pseudo-low-resolution data, which is time series data with low resolution in both temporal granularity and spatial granularity coarser than the high-resolution data, based on the high-resolution data; generates statistical information related to the high-resolution data, based on the high-resolution data; and generates a resolution conversion model for converting low-resolution data into high-resolution data by machine learning learning data with the high-resolution data as a target variable and the low-resolution data or the pseudo-low-resolution data and each of the statistical information as explanatory variables.

[0083] 10, 10A, 10B High-resolution device 11 CPU 12 ROM 13 RAM 14 Storage 15 Input unit 16 Display unit 17 Communication I / F 18 Bus 101 High-resolution data acquisition unit 102 Low-resolution unit 103 Statistical information generation unit 104 Model learning unit 105 Statistical information DB 106 Resolution conversion model 107 Low-resolution data acquisition unit 108 Resolution estimation unit 109 Output unit 110 High-resolution DB 111 Section estimation unit 112 Section assignment unit 113 Weather record DB

Claims

1. A high-resolution data acquisition unit that acquires high-resolution data, which is time series data with high resolution in both temporal granularity and spatial granularity; a low-resolution data acquisition unit that acquires low-resolution data, which is time series data with low resolution in both temporal granularity and spatial granularity that are coarser than the high-resolution data; a low-resolution data acquisition unit that generates pseudo-low-resolution data, which is time series data with low resolution in both temporal granularity and spatial granularity that are coarser than the high-resolution data, based on the high-resolution data; a statistical information generation unit that generates statistical information regarding the high-resolution data based on the high-resolution data; and a model learning unit that generates a resolution conversion model for converting low-resolution data into high-resolution data by machine learning learning data that uses the high-resolution data as a target variable and each of the low-resolution data or the pseudo-low-resolution data and the statistical information as explanatory variables.

2. The high resolution device according to claim 1, further comprising: a resolution estimation unit that uses the resolution conversion model to estimate high resolution data from the low resolution data and the statistical information; and an output unit that outputs the high resolution data estimated by the resolution estimation unit.

3. The high-resolution data generation device according to claim 1, further comprising: a high-resolution database that stores the first high-resolution data acquired by the high-resolution data acquisition unit; a resolution estimation unit that uses the resolution conversion model to estimate second high-resolution data from the low-resolution data and the statistical information; and an interval estimation unit that combines the first high-resolution data and the second high-resolution data in time and space, and statistically estimates confidence interval data that represent the magnitude of the error that occurs when the low-resolution data is converted to high resolution by the resolution estimation unit, and the magnitude of the error contained in the low-resolution data.

4. The high-resolution device described in claim 3, further comprising: an interval assignment unit that assigns a confidence interval to the second high-resolution data using the confidence interval data estimated by the interval estimation unit; and an output unit that outputs the second high-resolution data to which the confidence interval has been assigned by the interval assignment unit.

Images