Information processing device, information processing method, and program

Abstract

To execute an estimation of data of another object using estimated data of an object with higher accuracy.SOLUTION: An information processing device includes a processing part. The processing part selects one or more first target points from among a plurality of target points from which estimation data of a plurality of first attributes is obtained, selects one or more first attributes from among the plurality of first attributes for each of the one or more first target points, and selects one of a plurality of second attributes for an observation point from which observation data of the plurality of second attributes is obtained. The processing part acquires observation data of the second attribute and acquires estimation data of the first attribute for each of the first target points. The processing part generates learning data including the acquired observation data as an objective variable and the estimated data acquired for each of the first target points as an explanatory variable. The processing part learns an estimation model using the learning data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

[0002] A technique has been proposed for estimating weather data at a certain location using meteorological data estimated for each mesh. For example, a technique has been proposed for calculating visibility due to suspended particles using weather forecast mesh data and snow depth forecast mesh data.

[0003] Wind speed, snowstorm strength, and the resulting deterioration of visibility vary greatly depending on the characteristics of each location, making them difficult to estimate. Furthermore, because data that would allow for the estimation of wind speeds for a period of several hours in advance is unavailable, it is not anticipated that visibility for that period will be estimated. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 5950422 [Patent Document 2] Patent No. 6184148 [Non-patent literature]

[0005] [Non-Patent Document 1] Masaru Matsuzawa, "Study on Estimation Methods for Snowfall Amounts During Snowstorms," ​​Journal of the Japanese Society of Snow and Ice, Vol. 78, No. 5 (September 2016), pp. 255-268 Summary of the Invention [Problem to be solved by the invention]

[0006] An object of the present invention is to provide an information processing device, an information processing method, and a program that can more accurately estimate data of a given object using estimated data of another object. [Means for solving the problem]

[0007] An information processing device according to an embodiment includes a processing unit. The processing unit selects one or more first target locations from among a plurality of target locations from which estimation data of a plurality of first attributes is obtained, selects one or more first attributes from among the plurality of first attributes for each of the one or more first target locations, and selects one of a plurality of second attributes from among the plurality of observation locations from which observation data of a plurality of second attributes is obtained. The processing unit acquires observation data of the second attribute and acquires estimation data of the first attribute for each of the first target locations. The processing unit creates training data that uses the acquired observation data as a response variable and includes the estimation data acquired for each of the first target locations as an explanatory variable. The processing unit trains an estimation model using the training data. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram of an information processing apparatus according to an embodiment. [Figure 2] FIG. 3 is a diagram showing an example of a data structure of forecast data. [Figure 3] FIG. 2 is a diagram showing an example of a data structure of observation data. [Figure 4] FIG. 10 is a diagram showing an example of the relationship between forecast points and observation points. [Figure 5] FIG. 4 is a diagram showing an example of parameters stored in a parameter storage unit. [Figure 6] 10 is a flowchart of processing by a selection unit and an acquisition unit. [Figure 7] FIG. 10 is a diagram for explaining an example of forecast points that are individually specified. [Figure 8] FIG. 10 is a diagram showing an example of specifying a range by the radius of a circle. [Figure 9] FIG. 10 is a diagram showing an example of selecting a forecast point. [Figure 10] 10 is a flowchart of a learning data creation process. [Figure 11] FIG. 10 is a diagram showing an example of aggregating observation data. [Figure 12] FIG. 10 is a diagram showing an example of multiple observation data. [Figure 13] FIG. 10 is a diagram showing an example of learning data. [Figure 14] FIG. 10 is a diagram for explaining an example of creating learning data. [Figure 15] FIG. 10 is a diagram showing an example in which multiple reference times are specified. [Figure 16] 10 is a flowchart of a learning process. [Figure 17] 10 is a flowchart of an estimation process. [Figure 18] FIG. 10 is a diagram showing an example of an output by an output control unit. [Figure 19] FIG. 1 is a hardware configuration diagram of an information processing device. DETAILED DESCRIPTION OF THE INVENTION

[0009] Preferred embodiments of an information processing device according to the present invention will be described in detail below with reference to the accompanying drawings. The following describes an example in which meteorological objects (such as maximum wind speed) are estimated using meteorological estimation data (such as weather forecast data). The data to which the following embodiments can be applied is not limited to meteorological data, and may be any other data.

[0010] As described above, for example, with regard to visibility, there are cases where it is difficult to estimate due to changes in the data used for estimation, and cases where it is not assumed to estimate visibility for a time period several hours in the future. In order to estimate visibility for that time period, for example, it is necessary to repeat estimation from the time when data for estimation is available until that time period.

[0011] In the following embodiment, an estimation model is trained (constructed) using estimated values ​​of meteorological data (hereinafter, forecast data) estimated for each mesh as explanatory variables and other meteorological data (e.g., maximum wind speed data) at points where meteorological observation data is obtained (hereinafter, observation points) as objective variables. Furthermore, using the trained estimation model, meteorological data (such as maximum wind speed data) for a target time of estimation after the time when the forecast data was obtained is estimated. Other meteorological data such as visibility distance may be further calculated from the estimated meteorological data using an estimation formula or the like.

[0012] The forecast data is, for example, LFM (Local Forecast Model) data published by the Japan Meteorological Agency, which forecasts weather conditions at points in a 2 km mesh. Forecast data (LFM data) for multiple mesh points may be used as explanatory variables. The forecast data is not limited to LFM data and may be any data. For example, the forecast data may be a meteorological mesoscale model (MSM) with a square of 5 km.

[0013] The observation data is data observed by an observation system such as the Japan Meteorological Agency's AMeDAS (Automated Meteorological Data Acquisition System). The observation data may be observed in any manner, but is typically observed at regular time intervals such as 1-minute intervals, 10-minute intervals, or 60-minute intervals.

[0014] In this embodiment, an estimation model is trained while estimating a subset of explanatory variables that are effective for estimation from comprehensive forecast data obtained from multiple locations. Furthermore, in this embodiment, weather data (objective variables) at a target time of estimation, which is specified at a time interval (e.g., 10-minute intervals) different from the time interval (e.g., 30-minute intervals) at which forecast data is obtained, are used as the estimation target. This enables data estimation to be performed with higher accuracy.

[0015] 1 is a block diagram showing an example of the configuration of an information processing device 100 according to an embodiment. As shown in FIG. 1, the information processing device 100 includes a forecast data storage unit 121, an observation data storage unit 122, a selection data storage unit 123, a parameter storage unit 124, a model storage unit 125, a display unit 131, a selection unit 101, an acquisition unit 102, a creation unit 103, a learning unit 104, an estimation unit 111, a calculation unit 112, and an output control unit 113.

[0016] The forecast data storage unit 121 stores forecast data. Any method may be used to acquire the forecast data, but as described above, a method of acquiring LFM data published by the Japan Meteorological Agency as forecast data is applicable. The forecast data includes forecast data (past forecast data) to be used for learning the estimation model by the learning unit 104, and forecast data (current estimation forecast data) to be used during estimation using the learned estimation model.

[0017] FIG. 2 is a diagram showing an example of the data structure of forecast data. FIG. 2 shows an example in which the forecast data includes atmospheric pressure and east-west wind speed as elements. Hereinafter, the elements included in the forecast data may be referred to as attribute AT1 (first attribute). Furthermore, forecast data corresponding to attribute AT1 may be referred to as forecast data (estimated data) of attribute AT1. Attribute AT1 is not limited to atmospheric pressure and east-west wind speed, but may be any other attribute. For example, attribute AT1 may further include some or all of north-south wind speed, temperature, humidity, precipitation, and sunshine hours.

[0018] The forecast data is generated by a forecasting process that is executed at predetermined time intervals (for example, every 30 minutes). Hereinafter, the time at which the forecasting process is executed is referred to as the estimated execution time. In addition, in the forecasting process, each attribute AT1 is estimated for one or more times after the estimated execution time. Hereinafter, the time that is the subject of forecasting in the forecasting process is referred to as the forecast time.

[0019] For example, forecast data may be acquired up to 10 hours in advance at 30-minute intervals. In such a case, the forecast time will be 20 times ranging from 30 minutes to 600 minutes after the estimated execution time. Note that Figure 2 shows an example of forecast data for each attribute estimated at 30-minute intervals from 60 minutes in advance, with the estimated execution time at 30-minute intervals.

[0020] Returning to the explanation of Fig. 1, the observation data storage unit 122 stores observation data. Any method may be used to acquire the observation data, but for example, a method of acquiring observation data that records past weather information released by the Japan Meteorological Agency may be applied.

[0021] Fig. 3 is a diagram showing an example of the data structure of observation data. Fig. 3 shows an example of observation data observed at 10-minute intervals. Fig. 3 also shows an example of observation data including, as elements, local atmospheric pressure, sea level atmospheric pressure, precipitation, temperature, humidity, average wind speed, average wind speed and direction, maximum instantaneous wind speed, wind direction at maximum instantaneous wind speed, and sunshine hours.

[0022] Hereinafter, an element included in the observation data may be referred to as attribute AT2 (second attribute). Also, observation data corresponding to attribute AT2 may be referred to as observation data of attribute AT2. Attribute AT2 is not limited to the example shown in FIG. 3, and may include any other attribute.

[0023] Any one of the attributes AT2 is selected as the dependent variable to be estimated by the estimation model. In the following, an example in which the maximum instantaneous wind speed is selected as the dependent variable will be mainly described.

[0024] Returning to the explanation of Fig. 1, the selection data storage unit 123 stores selection data used by the selection unit 101 to select forecast data (attribute AT1) to be used as explanatory variables and observation data (attribute AT2) to be used as objective variables. For example, the selection data includes the following data: Data used as the target variable (attribute AT2): Maximum instantaneous wind speed of the observed data Data used as explanatory variables (attribute AT1): Maximum instantaneous wind speed at forecast point PA, temperature at forecast point PA, maximum instantaneous wind speed at forecast point PB, maximum instantaneous wind speed at forecast point PC, average wind speed at forecast point P7

[0025] In this way, the selection data includes data specifying a forecast location and the attribute AT1 of the forecast data at the forecast location. The selection data may include an initial value when the selection unit 101 selects a forecast location. The selection data may be specified by a user, etc. In this embodiment, the selection data can be updated to indicate explanatory variables that are valid for estimation.

[0026] The forecast point refers to the target point that is the subject of the forecast. If the forecast data is meteorological data estimated for each mesh, the forecast point is determined for each of the multiple meshes from which the forecast data is obtained. For example, in the case of LFM data, the forecast point is determined for each 2-km mesh.

[0027] The point (hereinafter referred to as the estimation point) at which the objective variable (such as maximum instantaneous wind speed) is estimated can be a point different from the forecast point. For example, the estimation point can be an observation point where observation data can be obtained. In the following, an example in which the estimation point is an observation point will be described.

[0028] FIG. 4 is a diagram showing an example of the relationship between forecast points and observation points. In FIG. 4, the 12 circles placed at the intersections of the solid lines correspond to forecast points. The forecast points can be interpreted as corresponding to the representative points of each mesh separated by dashed lines. The representative points are, for example, the center points of the mesh. For example, when a 2 km square mesh is used, the representative points are also set every 2 km. Point 201 in FIG. 4 corresponds to an observation point. In this way, the observation point can be a point different from the multiple forecast points.

[0029] All of the forecast data obtained at the 12 forecast points can be used as explanatory variables for estimating the dependent variable at the observation point. The information processing device 100 of this embodiment selects, from the forecast data at these multiple forecast points, forecast data that is deemed effective for estimating the dependent variable as explanatory variables, and trains an estimation model using training data including the selected explanatory variables. The forecast point 202 in FIG. 4 represents an example of a forecast point corresponding to the selected explanatory variables. Multiple explanatory variables corresponding to multiple forecast points may be selected. In the example of FIG. 4, the forecast point 202 is the point closest to point 201, which corresponds to the observation point, but the forecast point corresponding to the forecast data selected as an explanatory variable does not have to be the point closest to the observation point.

[0030] Returning to the description of Fig. 1, the parameter storage unit 124 stores the values ​​of parameters used by each unit of the information processing device 100. Fig. 5 is a diagram showing an example of parameters stored in the parameter storage unit 124.

[0031] Each parameter shown in FIG. 5 will be explained below. Learning data collection start time: The start time of the period during which learning data is collected. · Learning data collection end time: End time of the period for collecting learning data

[0032] One or more reference times t during the period from the start time of training data collection to the end time of training data collection ref The training data is collected based on the reference time t ref The following mainly describes an example where the time is set at a fixed interval (for example, 60 minutes). When using a trained estimation model, the reference time t ref is the reference time for performing estimation using the estimation model (e.g., the current time t now ) is equivalent to

[0033] The parameters shown in FIG. 5 will be further explained. Estimated execution time: One or more estimated execution times for which the forecast data to be used is obtained are specified. The estimated execution times are calculated based on the reference time t refIt may also be specified by a time going back from the base point. Forecast time: One or more forecast times to be used are specified from the multiple forecast times for which forecast data is available. The forecast time must specify the time for which forecast data is available. For example, if forecast data is available up to 10 hours into the future, a value of 10 hours (600 minutes) or less should be specified for the forecast time. Estimation target time: The time to estimate the objective variable is specified. The estimation target time is, for example, the reference time t ref A future time based on the current time is specified. A plurality of estimation target times may be specified. In FIG. 5, an example is shown in which three estimation target times (80, 90, 100) are specified. Offset: Equivalent to the time difference between the estimated execution time and the estimated target time. As mentioned above, the time interval between the time when forecast data (explanatory variables) are obtained (estimated execution time) and the time when observed data (target variables) are obtained (estimated target time) may differ. Furthermore, the estimated execution time is specified by the time going back from the reference time. Therefore, the offset can be interpreted as the time used to absorb the difference in the time interval between the reference time (estimated execution time) and the estimated target time. · Training data usage rate: This indicates the percentage of collected training data other than data used for validation (validation data).

[0034] Returning to the explanation of Fig. 1, the model storage unit 125 stores data indicating the estimation model learned by the learning unit 104.

[0035] Each memory unit (selection data memory unit 123, forecast data memory unit 121, observation data memory unit 122, parameter memory unit 124, model memory unit 125) can be configured using any commonly used memory medium such as flash memory, memory card, RAM (Random Access Memory), HDD (Hard Disk Drive), and optical disk.

[0036] Each storage unit may be implemented as a physically different storage medium, or as different storage areas of the same physically identical storage medium.Furthermore, each storage unit may be implemented as a plurality of physically different storage media.

[0037] The display unit 131 is a device that displays various data such as data indicating the estimation result by the estimation unit 111. The display unit 131 can be configured with a liquid crystal display or the like.

[0038] The selection unit 101 uses the selection data stored in the selection data storage unit 123 to select explanatory variables (forecast data) and objective variables (observation data) to be used in creating learning data.

[0039] For example, the selection unit 101 selects one or more forecast points P1 (first forecast points) specified by the selection data from among a plurality of forecast points (forecast points corresponding to each of a plurality of meshes) from which forecast data is obtained. Furthermore, the selection unit 101 selects one or more attributes AT1 specified by the selection data from among a plurality of attributes AT1 for each of the one or more forecast points P1. The selected attribute AT1 corresponds to a subset of explanatory variables that are effective in estimation. Furthermore, the selection unit 101 selects one of a plurality of attributes AT2 for the observation point as a response variable.

[0040] The selection unit 101 also selects an estimated execution time ET1 (first estimated execution time) for estimating the forecast data (attribute AT1) to be selected as an explanatory variable. For example, the selection unit 101 selects one or more estimated execution times ET1 according to the parameters (estimated execution times) stored in the parameter storage unit 124.

[0041] Furthermore, the selection unit 101 selects a forecast time OT1 (first forecast time) of the forecast data to be selected as an explanatory variable. For example, the selection unit 101 selects one or more forecast times OT1 according to the parameters (forecast times) stored in the parameter storage unit 124.

[0042] The acquisition unit 102 acquires various types of information used in the information processing device 100. For example, the acquisition unit 102 acquires forecast data of attribute AT1 selected as an explanatory variable from the forecast data stored in the forecast data storage unit 121 in accordance with the selection result by the selection unit 101. The acquisition unit 102 also acquires observation data of attribute AT2 selected as a target variable from the observation data stored in the observation data storage unit 122.

[0043] The acquisition unit 102 also determines the estimated execution time and forecast time of the forecast data to be acquired according to the selection result by the selection unit 101. That is, the acquisition unit 102 acquires forecast data of the selected attribute AT1 obtained at the selected estimated execution time ET1 for each of one or more forecast locations P1. The acquisition unit 102 also acquires forecast data of the selected attribute AT1 obtained as a forecast result at the forecast time OT1 for each of one or more forecast locations P1.

[0044] The creation unit 103 creates learning data to be used for learning the estimation model. For example, the creation unit 103 creates learning data that uses the observation data acquired by the acquisition unit 102 as a dependent variable and includes one or more forecast data acquired for one or more forecast points P1 as explanatory variables.

[0045] The learning unit 104 learns an estimation model using the learning data created by the creation unit 103. The estimation model may be a model of any format as long as it receives one or more explanatory variables as input and outputs an estimated value of a dependent variable. The estimation model may be learned by any method. Examples of estimation models are described below. Inference models obtained by decision tree-based machine learning such as CART (Classification And Regression Tree), Random Forest, XGBoost, and LightGBM (Light Gradient Boosting Machine) Regression models with L1 regularization terms such as Lasso (Least Absolute Shrinkage and Selection Operator)

[0046] In this embodiment, the objective variable at the estimation target time specified at an interval (e.g., 10 minutes) smaller than the interval of the forecast time (e.g., 30 minutes) is the object of estimation. In other words, the estimation model is a model that outputs an estimate of the objective variable at the estimation target time specified at an interval smaller than the interval of the multiple forecast times that are the object of estimation in the forecast data.

[0047] When multiple values ​​are set as the estimation target time, which is one of the parameters stored in the parameter storage unit 124, the learning unit 104 learns multiple estimation models corresponding to the multiple estimation target times, respectively. Similarly, when multiple attributes AT2 are specified as objective variables as the selection data stored in the selection data storage unit 123, the learning unit 104 learns multiple estimation models corresponding to the multiple attributes AT2, respectively.

[0048] During learning of the estimation model, the learning unit 104 may output an optimal combination of hyperparameters and the importance of each of one or more explanatory variables. The importance is an index indicating the degree to which an explanatory variable influences the estimation result obtained by the estimation model. The importance can also be interpreted as the contribution (degree of contribution to the estimation result), the influence, the effectiveness, etc.

[0049] For example, in Lasso, explanatory variables whose coefficients in the L1 norm do not become 0 correspond to explanatory variables that have a large influence on the estimation results. For this reason, the magnitude of the coefficient, for example, can be used as the importance. In addition, in XGBoost, an index value called Feature Importance is calculated for each explanatory variable after training. This index value is a numerical value based on the frequency with which the explanatory variable is used when training the estimation model and the degree to which the error for the target variable is reduced by adopting the explanatory variable. Such index values ​​can be used as the importance.

[0050] The learning unit 104 may store information (such as a table) that associates explanatory variables with importance in the storage unit. The learning unit 104 may update the selection data stored in the selection data storage unit 123 according to the importance. For example, the learning unit 104 updates the selection data so as to select attribute AT1 corresponding to an explanatory variable whose importance is greater than that of other explanatory variables. An explanatory variable whose importance is greater than that of other explanatory variables is identified, for example, as an explanatory variable whose importance in estimation results for all estimation target times is greater than a threshold.

[0051] After this, the selection unit 101 can select the attribute AT1 corresponding to the explanatory variable whose importance is greater than that of the other explanatory variables according to the updated selection data. This makes it possible to create training data including a subset of explanatory variables that are effective in estimation and to train an estimation model.

[0052] When the learning data usage rate is specified as a parameter, the learning unit 104 may select learning data from the created (collected) learning data in accordance with the learning data usage rate, and learn an estimation model using the selected learning data. The learning unit 104 uses the remaining learning data as validation data to validate the learned estimation model.

[0053] The estimation unit 111 performs estimation using the trained estimation model. For example, the estimation unit 111 inputs input data, which is forecast data for the selected attribute AT1 acquired for each of one or more forecast points P1, into the trained estimation model to obtain an estimate of the observation data for the selected attribute AT2. The input data corresponds to forecast data at the time of estimation used during estimation using the trained estimation model. The input data (forecast data at the time of estimation) is created, for example, by extracting explanatory variables that are the same as the explanatory variables selected during training of the estimation model from forecast data (such as LFM data) acquired for estimation.

[0054] The calculation unit 112 is used when calculating other information (such as other weather data) using estimated values ​​obtained by the estimation model. For example, when the estimation model is a model that estimates maximum wind speed data, the calculation unit 112 calculates visibility information from the maximum wind speed data estimated by the estimation model. The visibility information is information that indicates the presence or absence of visibility impairment or the visibility distance. Note that the estimation model may be trained to estimate the visibility distance. In such a case, the calculation unit 112 may not be provided.

[0055] The output control unit 113 controls the output of various information used in the information processing device 100. For example, the output control unit 113 outputs (displays) output information such as an estimated value obtained by the estimation unit 111 or visibility information calculated by the calculation unit 112 to the display unit 131. The output method by the output control unit 113 is not limited to a method of displaying on the display unit 131, and any other method may be used. For example, a method of transmitting the output information to an external device connected via a network may be used.

[0056] At least a part of each of the above units (selection unit 101, acquisition unit 102, creation unit 103, learning unit 104, estimation unit 111, calculation unit 112, and output control unit 113) may be realized by one or more processing units. Each of the above units is realized, for example, by one or more processors. For example, each of the above units may be realized by having a processor such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) execute a program, that is, by software. Each of the above units may be realized by a processor such as a dedicated IC (Integrated Circuit), that is, by hardware. Each of the above units may be realized by a combination of software and hardware. When multiple processors are used, each processor may realize one of the units, or may realize two or more of the units.

[0057] Furthermore, the information processing device 100 may be physically configured as one device or may be physically configured as multiple devices. For example, the information processing device 100 may be constructed in a cloud environment. Furthermore, each unit within the information processing device 100 may be distributed across multiple devices.

[0058] Next, a detailed description will be given of the processing performed by the selection unit 101 and the acquisition unit 102. Fig. 6 is a flowchart showing an example of the processing performed by the selection unit 101 and the acquisition unit 102.

[0059] The selection unit 101 selects, in accordance with the selection data, a forecast location from which forecast data will be adopted, an attribute AT1 to be used as an explanatory variable, and an attribute AT2 to be used as a target variable from the observation data observed at the observation location (step S101). The acquisition unit 102 acquires forecast data of the selected attribute AT1 from the forecast data of the selected forecast location from the forecast data storage unit 121 (step S102). The acquisition unit 102 also acquires observation data of the selected attribute AT2 from the observation data storage unit 122 (step S103).

[0060] As described above, the selection data can be expressed in a format that individually specifies the forecast locations and attributes to be selected. FIG. 7 is a diagram for explaining an example of individually specified forecast locations. FIG. 7 shows an example in which three forecast locations 202 are specified as forecast locations to be selected. Although not shown in FIG. 7, the selection data specifies the attribute AT1 to be adopted for each of the three forecast locations 202.

[0061] The selection data may be in a format indicating a predetermined rule (hereinafter, "selection rule") for selecting a forecast point or an attribute. That is, the selection unit 101 may select one or more forecast points P1 from the plurality of forecast points based on the selection rule.

[0062] In the initial state, selection is made according to selection data indicating the selection rules, and the selection data may then be updated to specify that forecast locations and attributes effective for estimation are selected according to the importance of each explanatory variable obtained during learning.

[0063] Below are some example selection rules: (R1) A rule that selects all forecast locations and attributes. (R2) A rule for selecting forecast points within a specified range. Any method for specifying the range may be used, but for example, a method for specifying a rectangular range or a method for specifying a range by the radius of a circle centered on the observation point may be applied. (R3) A rule for clustering (classifying) multiple forecast locations into multiple clusters and selecting a forecast location from each of the multiple clusters.

[0064] Figure 8 is a diagram showing an example of specifying a range by the radius of a circle according to (R2). In the example of Figure 8, forecast points within a circular range 801, which is centered at point 201 corresponding to the observation point and has a specified radius of 5 km, are selected.

[0065] Selection rule (R3) will now be described. When selection rule (R3) is used, the selection unit 101 first selects multiple forecast locations within a predetermined range in the east-west and north-south directions, starting from the observation location or the forecast location closest to the observation location. The range is specified as one of the selection rules. The selection unit 101 classifies the multiple forecast locations into multiple clusters based on the distances between multiple forecast data obtained at the selected multiple forecast locations.

[0066] Any clustering (classification) method may be used, but for example, the following method can be applied: In the following method, candidates for the number of clusters (e.g., 3, 4, 5, 6, 7) and the number of forecast points to be adopted from each cluster are further specified as selection rules.

[0067] First, the selection unit 101 extracts forecast data of multiple forecast times and multiple attribute values ​​obtained for multiple selected forecast locations. The selection unit 101, for example, randomly selects forecast times and attribute values. Using the selected forecast data, the selection unit 101 performs clustering using a clustering method such as the k-means method, based on the distance between the forecast data of the multiple forecast locations.

[0068] The selection unit 101 performs clustering for each candidate number of clusters specified by the selection rule, and determines the number of clusters to adopt, for example, by the elbow method. The elbow method is a method of adopting the first number of clusters that minimizes the change in the sum of squared errors within a cluster as the optimal number of clusters.

[0069] The selection unit 101 selects, for example, randomly, one or more forecast locations P1 specified as a selection rule from each of a plurality of clusters. The selection unit 101 may select, from each cluster, a forecast location that represents the cluster (for example, a forecast location closest to the center of the cluster).

[0070] Fig. 9 is a diagram showing an example of selecting forecast points according to (R3). Range 901 corresponds to a range that is predetermined in the east-west and north-south directions starting from point 201, which is an observation point. In the example of Fig. 9, the multiple forecast points included in range 901 are classified into six clusters A to F. Forecast points 911 to 916 correspond to forecast points selected from each of the six clusters.

[0071] The data (attribute AT1) to be used as an explanatory variable may be selected by the following method. First, the selection unit 101 adds a new forecast time point, and constructs an estimation model by partially selecting attributes (explanatory variables) included in the forecast time point. The selection unit 101 calculates an evaluation value of the estimation model, such as AIC (Akaike Information Criterion), and if the evaluation value of the estimation model is improved, adds the selected attribute as an attribute to be used as an explanatory variable. The selection unit 101 may add explanatory variables by repeatedly executing such processing.

[0072] The evaluation of the estimation model is not limited to the above method. For example, the selection unit 101 may use the verification data to calculate a likelihood ratio between an estimation model constructed including a newly selected attribute and an existing estimation model before construction, and determine that the evaluation value of the estimation model has been improved if the calculated likelihood ratio exceeds a predetermined threshold (e.g., 10).

[0073] Next, a detailed description will be given of the process of creating learning data by the creating unit 103. The creating unit 103 creates learning data based on a reference time t ref As a reference, learning data is created in which the explanatory variables (forecast data of attribute AT1) acquired by the acquisition unit 102 correspond to the objective variables (observation data of attribute AT2).

[0074] Typically, an estimation model is trained using multiple training data. For example, the multiple training data are obtained by multiple specified reference times t within a specified period (e.g., a period from the start time of training data collection to the end time of training data collection). ref Therefore, the target variable is collected at the specified reference time t ref It is represented by a vector containing T elements corresponding to the number T of elements. Hereinafter, the vector representing this objective variable will be referred to as the objective variable vector.

[0075] Similarly, the explanatory variables corresponding to one attribute AT1, one estimated execution time, one forecast time, and one forecast location can be represented by a vector containing T elements. Hereinafter, a vector representing the explanatory variables for each attribute AT1, estimated execution time, forecast time, and forecast location will be referred to as an explanatory variable vector. The creation unit 103 creates a matrix by concatenating the explanatory variable vectors for each attribute AT1, estimated execution time, forecast time, and forecast location. Hereinafter, this matrix will be referred to as an explanatory variable matrix.

[0076] The creation unit 103 creates and outputs learning data in a matrix format including an explanatory variable matrix and a target variable vector. Note that the format of the learning data is not limited to a matrix format and may be any other format. A specific example of a method for creating learning data will be described below.

[0077] First, definitions of terms, variables, etc. used in the explanation will be given. The number J of attributes AT1 used as explanatory variables: The number J can also be interpreted as representing the number of dimensions of the explanatory variables. Number of reference times T: Learning data is collected (created) for each reference time. Therefore, the number T can also be interpreted as the number of instances in the learning data. · Reference time t∈[t1, t2, . . . , tT]: When distinguishing T reference times t, the reference time t is expressed as one of t1, t2, . . . , tT. Number of forecast times set L: For example, if the forecast times are set in 30-minute intervals from 30 to 600, L = 20. · Forecast time l∈[l1,l2,···,l L ]: Indicates the forecast time based on the estimated execution time. For example, the forecast time is set to a value in 30-minute increments (30, 60, ..., 600, etc.). · Number of forecast points U: The number U corresponds to the number of forecast points selected according to the selected data. Forecast point u∈[1,2,···,U]: When distinguishing U forecast points u, the forecast points u are u1,u2,···,u U It is expressed as one of the following. Objective variable vector y: Objective variable vector y is expressed by the following equation (1): t represents the value (scalar value) of the objective variable corresponding to the reference time t. Note that the symbol "T" in the upper right corner of the parentheses represents transposition.

number

number

number

[0078] The flow of the learning data creation process by the creation unit 103 will be described with reference to Fig. 10. Fig. 10 is a flowchart showing an example of the learning data creation process.

[0079] The creation unit 103 acquires the value of the observation data of the attribute AT2 selected as the objective variable at the time obtained by adding the estimation target time to the reference time (step S201). For example, when the estimation target time of 100 minutes (estimation of the objective variable after 100 minutes) is specified as a parameter, the creation unit 103 acquires the value of the observation data of the attribute AT2 selected as the objective variable at the time obtained by adding the estimation target time to the reference time t ref The time t (=t ref+100 ) the target variable y t Get the value of .

[0080] If the time intervals at which forecast data is obtained are different from the time intervals at which observation data is obtained, the creation unit 103 may acquire values ​​by aggregating the observation data so that the time intervals match the time intervals of the forecast data. Fig. 11 is a diagram showing an example of aggregating the observation data.

[0081] FIG. 11 shows an example of aggregating observation data observed at 10-minute intervals into data at 60-minute intervals. The aggregation method is determined for each of multiple attributes AT2 included in the observation data. The aggregation method may include calculating the average value, the sum value, the mode value, or the maximum value. For example, as shown in the center of FIG. 11, an aggregation method that calculates the average value is specified for atmospheric pressure, and an aggregation method that calculates the maximum value is specified for maximum instantaneous wind speed. The bottom of FIG. 11 shows the result of aggregating the observation data for the six times shown in the top into one observation data.

[0082] The process of aggregating the observed data is repeated every 60 minutes as shown in Fig. 11. Fig. 12 is a diagram showing an example of a plurality of observed data obtained by repeating the aggregating process.

[0083] Returning to the explanation of Fig. 10, the creation unit 103 converts the time of the acquired objective variable into a value before adding the estimation target time, that is, the reference time t ref (Step S202). ref It is possible to associate the obtained explanatory variables with the

[0084] The creation unit 103 acquires forecast data corresponding to the forecast time specified by the parameters (step S203). The forecast time may be, for example, the forecast time closest to the time obtained by adding the target time to the reference time, or a forecast time before or after the forecast time closest to this time. In other words, one or more times included in a certain time range including the time obtained by adding the target time to the reference time may be specified as the forecast time. This can improve the accuracy of estimation by the estimation model.

[0085] In addition, if the forecast time is not specified as a parameter, the creation unit 103 may be configured to identify one or more forecast times that fall within a certain time range that includes the time obtained by adding the estimated target time to the reference time, for example.

[0086] An example of the process in step S203 will be described. For example, at a certain forecast point u, the reference time t ref The explanatory variable indicating the forecast data with a forecast time of 30 minutes later obtained by the forecast with an estimated execution time of 30 minutes before is X -30,30,u It is expressed as:

[0087] Since the expression including the selection of the subset by the selection unit 101 becomes complicated, in the following, j and u represent the subsets selected in advance. Also, t is originally a reference time t ref However, to avoid the complexity of the notation, we will use the reference time t ref Only the difference value from is expressed as the value of t.

[0088] As above, the reference time t ref Consider the explanatory variable matrix X to find the estimated value of the objective variable 100 minutes after the reference time t ref Of the forecast times of the forecast data obtained by a forecast executed with the estimated execution time as the estimated execution time, the forecast time closest to 100 minutes that is more than 100 minutes ahead is 120 minutes from now. Note that the forecast time is assumed to be in 60-minute increments between 0 and 600 minutes. When configuring explanatory variables using only forecast data 120 minutes from now, the explanatory variable matrix X is expressed by the following equation (4).

number

[0089] Furthermore, if forecast data for 90 minutes later is also added to the explanatory variables, an explanatory variable matrix X expressed by the following equation (5) is further added.

number

[0090] Furthermore, the reference time t ref The forecast data obtained by the forecast executed at the estimated execution time 30 minutes before the reference time t ref If the forecast data for the next 120 minutes is also added as an explanatory variable, the explanatory variable matrix X expressed by the following equation (6) is further added.

number

[0091] As shown in equations (4) to (6), the explanatory variable matrix X has a format in which explanatory variable vectors x are concatenated in the column direction in the number of columns equal to the number (J) of attribute AT1 and the number (U) of forecast locations. The creation unit 103 further creates an explanatory variable matrix X by concatenating the explanatory variable matrices X obtained for each estimated execution time and each forecast time.

[0092] For example, the reference time t ref The estimated execution times are 0 minutes and 60 minutes before the reference time t ref When the forecast data for the forecast times 120 minutes and 90 minutes after the current time are used as explanatory variables, the four explanatory variable matrices X 0,120 , X 0,90 , X -60,180 , X -60,150 The explanatory variable matrix X is created by concatenating the above in the column direction, as shown in the following equation (7).

number

[0093] The creating unit 103 outputs the learning data D={y, X} that is obtained by concatenating the explanatory variable matrix X created in this way and the objective variable vector y.

[0094] Fig. 13 is a diagram showing an example of learning data. Fig. 13 shows an example of linking explanatory variable matrices for three forecast points PA and PB. Note that Fig. 13 shows an example in which atmospheric pressure and east-west wind speed are selected as the attribute AT1. If another attribute AT1 is further selected, columns for that attribute AT1 are further linked for each forecast point.

[0095] It is desirable that the estimation of the objective variable is performed using as new a forecast data as possible (forecast data predicted at the most recent estimation execution time) as the explanatory variables. Therefore, it is desirable that the estimation model be trained using as new a forecast data as possible as the explanatory variables during training.

[0096] However, due to delays in transmitting forecast data, when an estimation is performed, only forecast data for the past estimation execution time may be available. Also, the estimation target time may be specified at intervals (e.g., 10 minutes) different from the interval at which the reference time is set (e.g., 60 minutes). An offset is used to adjust for such differences in time intervals.

[0097] As an example, consider the case where the estimated value of the objective variable at 4:20 is estimated, and only the forecast data from the forecast executed at the estimated execution time before 3:00 has arrived. In this case, 3:00, counted back from 4:20, is used as the reference time t ref and consider the following: "Estimate the objective variable 80 minutes from now using the forecast data up to 3:00."

[0098] FIG. 14 is a diagram for explaining an example of creating learning data in such a case. As shown in FIG. 14, ref is set to time 1401 (3:00), which is the latest estimated execution time for which forecast data is available. In this example, the estimated target time is 4:20. Therefore, the reference time t ref is 80 minutes before the estimated target time.

[0099] In the example of Figure 14, the reference time t ref The forecast is executed with the estimated execution time (0 minutes before) as the reference time tref The forecast data obtained from the forecast executed with the estimated execution time as 60 minutes before the reference time t ref Explanatory variables are created using forecast data with forecast times of 90 minutes and 60 minutes after the event.

[0100] In the example of FIG. 14, as described above, the reference time t ref To reflect that the estimated execution time is 80 minutes before the estimated target time, an offset of −80 minutes is added to each specified estimated execution time. That is, −80 is added to the reference time t. The offset is stored, for example, as one of the parameters in the parameter storage unit 124. The explanatory variable matrix X using the reference time with the offset is expressed, for example, by the following equation (8).

number

[0101] In addition, since the objective variable after 80 minutes is estimated, the objective variable vector shown in the following equation (9) is created. T The time of the objective variable (t1+80) corresponds to the value before the process of reassigning it to the value before adding the estimated target time (80 minutes) (for example, step S202 above). After this, the time of the objective variable is reassigned to the value before adding the estimated target time (80 minutes). For example, t1+80 is reassigned to t1.

number

[0102] The response variable vector y and explanatory variable matrix X are both arranged in the row direction as [t1, t2, , t T The creation unit 103 can link the response variable vector y and the explanatory variable matrix X using this timestamp.

[0103] Figure 14 shows the time t ref As mentioned above, when collecting training data, multiple reference times tref FIG. 15 shows the relationship between multiple reference times t ref 15 is a diagram showing an example in which eight reference times t ref An example where T=8 is specified is shown. ref For each reference time t, the estimated execution time going back 60 minutes is specified, and the forecast times 120 minutes and 150 minutes in the future are specified. ref The training data corresponding to the data set d t (corresponding to equation (3)).

[0104] Next, the flow of the learning process by the learning unit 104 will be described with reference to Fig. 16. Fig. 16 is a flowchart showing an example of the learning process.

[0105] The learning unit 104 learns an estimation model using the learning data created by the creation unit 103 (step S301). The learning unit 104 stores data indicating the estimation model obtained by learning in the model storage unit 125 (step S302). The learning unit 104 updates the selection data stored in the selection data storage unit 123 using the importance of each explanatory variable obtained during learning (step S303), and ends the learning process.

[0106] Next, a specific example of the calculation process by the calculation unit 112 will be described. The calculation unit 112 calculates the visibility distance using, for example, maximum wind speed data (maximum instantaneous wind speed) output by the estimation model. The visibility distance can be calculated by, for example, the method described in Non-Patent Document 1. The method for calculating the visibility distance is not limited to this. Furthermore, the meteorological data (attributes) calculated by the calculation unit 112 is not limited to the visibility distance, and may be any meteorological data.

[0107] If an estimated value of an attribute required for calculation cannot be obtained from the estimation model, the calculation unit 112 may obtain an estimated value of the attribute from forecast data with a closer forecast time and use the obtained value for calculation. For example, assume that precipitation and maximum instantaneous wind speed are used to calculate visibility, but only the maximum instantaneous wind speed is obtained from the estimation model. In such a case, the calculation unit 112 obtains an estimated value of precipitation from the forecast data and uses the obtained value for calculating visibility.

[0108] Next, the flow of the estimation process by the estimation unit 111 will be described with reference to Fig. 17. Fig. 17 is a flowchart showing an example of the estimation process.

[0109] The estimation unit 111 acquires forecast data for a specified forecast time (estimation time forecast data) (step S401). The estimation unit 111 acquires the estimation time forecast data (explanatory variables) according to, for example, parameters stored in the parameter storage unit 124.

[0110] The estimation unit 111 acquires data indicating an estimation model corresponding to an unprocessed estimation target time among the specified estimation target times from the model storage unit 125 (step S402). The estimation unit 111 converts the acquired forecast data into the format of input data for the acquired estimation model (step S403). For example, the estimation unit 111 converts the forecast data into the format of an explanatory variable matrix X in which explanatory variable vectors x are concatenated, as in the case of creating training data. The estimation unit 111 inputs the input data to the estimation model and acquires an estimated value output by the estimation model (step S404).

[0111] When estimating meteorological data different from the estimated value by the estimation model, for example, when estimating the visibility distance, the calculation unit 112 calculates the visibility distance using the estimated value by the estimation model (step S405).

[0112] The estimation unit 111 determines whether all the specified estimation target times have been processed (step S406). If all the estimation target times have not been processed (step S406: No), the estimation unit 111 returns to step S402 and repeats the process for the next unprocessed estimation target time.

[0113] If all estimation target times have been processed (step S406: Yes), the output control unit 113 outputs (displays) the calculated visibility distance to, for example, the display unit 131 (step S407), and ends the estimation process.

[0114] Fig. 18 is a diagram showing an example of output by the output control unit 113. Fig. 18 shows an example of a display screen that outputs the calculated visibility distance for each of the three execution times of the estimation processes and for each of the four target estimation times. Every time an estimation process is executed (for example, every hour), the result of the estimation process may be added to a new row on the display screen.

[0115] In this manner, in the embodiment, the estimation model is trained using a subset (attributes) of explanatory variables that are effective for estimation among forecast data that are forecast at multiple forecast locations and include multiple attributes, thereby enabling more accurate data estimation.

[0116] Next, the hardware configuration of the information processing apparatus according to the embodiment will be described with reference to Fig. 19. Fig. 19 is an explanatory diagram illustrating an example of the hardware configuration of the information processing apparatus according to the embodiment.

[0117] The information processing device of the embodiment includes a control device such as a CPU (Central Processing Unit) 51, a storage device such as a ROM (Read Only Memory) 52 and a RAM (Random Access Memory) 53, a communication I / F 54 that connects to a network and communicates, and a bus 61 that connects each part.

[0118] The programs executed by the information processing apparatus according to the embodiment are provided in advance in the ROM 52 or the like.

[0119] The program executed by the information processing device of the embodiment may be configured to be provided as a computer program product by being recorded in an installable or executable format on a computer-readable recording medium such as a CD-ROM (Compact Disk Read Only Memory), a flexible disk (FD), a CD-R (Compact Disk Recordable), or a DVD (Digital Versatile Disk).

[0120] Furthermore, the program executed by the information processing apparatus of the embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the program executed by the information processing apparatus of the embodiment may be provided or distributed via a network such as the Internet.

[0121] The programs executed by the information processing device of the embodiment can cause a computer to function as each of the above-mentioned parts of the information processing device. In this computer, the CPU 51 can read the programs from a computer-readable storage medium onto a main storage device and execute them.

[0122] A configuration example of the embodiment will be described below. (Configuration example 1) selecting one or more first target points from among a plurality of target points at which estimated data of a plurality of first attributes can be obtained; selecting one or more first attributes from the plurality of first attributes for each of the one or more first points of interest; selecting one of the second attributes for an observation point from which observation data of a plurality of second attributes is obtained; acquiring the observed data for the selected second attribute and acquiring the estimated data for the selected first attribute for each of one or more of the first points of interest; creating training data including the acquired observation data as a response variable and one or more of the estimation data acquired for each of the one or more first target locations as an explanatory variable; Using the training data, an estimation model is trained to input one or more of the explanatory variables and output an estimated value of the objective variable. Processing section An information processing device comprising: (Configuration example 2) The processing unit selecting one or more first target points from among the plurality of target points based on a predetermined rule; The information processing device according to configuration example 1. (Configuration example 3) The rule is: a rule for classifying the plurality of target points into a plurality of clusters based on distances between the plurality of estimated data obtained at the plurality of target points, and selecting one or more first target points from each of the plurality of clusters; The processing unit selecting one or more of the first points of interest from each of the plurality of clusters based on the rule; The information processing device according to configuration example 2. (Configuration example 4) The processing unit outputting the importance of each of the one or more explanatory variables during training of the estimation model; selecting the first attribute corresponding to the explanatory variable whose importance is greater than that of other explanatory variables; The information processing device according to any one of configuration examples 1 to 3. (Configuration Example 5) the estimation model outputs the estimated value at a target time of estimation, the target time being specified at intervals smaller than the intervals of the plurality of forecast times targeted for estimation in the estimation data; The information processing device according to any one of configuration examples 1 to 4. (Configuration Example 6) The processing unit inputting input data, which is the estimated data of the selected first attribute acquired for each of the one or more first target locations, into the trained estimation model to obtain the estimated value of the observation data of the selected second attribute; 6. The information processing device according to any one of configuration examples 1 to 5. (Configuration Example 7) The processing unit Select the second attribute used to calculate visibility information, which is the presence or absence of visibility impairment or the visibility distance; Calculating the visibility information using the estimated value. The information processing device according to configuration example 6. (Configuration Example 8) The processing unit selecting the second attribute, which is the presence or absence of visibility impairment or the visibility distance; The information processing device according to any one of configuration examples 1 to 6. (Configuration Example 9) The processing unit selecting one or more first estimation execution times representing times over which estimation of the estimation data is to be performed; obtaining, for each of the one or more first points of interest, the estimated data for the selected first attribute obtained at the first estimated execution time; The information processing device according to any one of configuration examples 1 to 8. (Configuration Example 10) The processing unit selecting one or more first forecast times from among a plurality of forecast times to be estimated in the estimation data; obtaining, for each of the one or more first target locations, the estimated data for the selected first attribute resulting from the estimation of the first forecast time; The information processing device according to any one of configuration examples 1 to 9. (Configuration Example 11) An information processing method executed by an information processing device, selecting one or more first target points from among a plurality of target points from which estimation data of a plurality of first attributes can be obtained; selecting, for each of the one or more first points of interest, one or more first attributes from a plurality of the first attributes; selecting one of the second attributes for an observation point from which observation data of the second attributes is obtained; obtaining the observed data for the selected second attribute and obtaining the estimated data for the selected first attribute for each of one or more of the first points of interest; creating training data including the acquired observation data as a response variable and one or more of the estimation data acquired for each of the one or more first target locations as an explanatory variable; using the training data to train an estimation model that inputs one or more of the explanatory variables and outputs an estimate of the objective variable; An information processing method including: (Configuration Example 12) On the computer, selecting one or more first target points from among a plurality of target points from which estimation data of a plurality of first attributes can be obtained; selecting, for each of the one or more first points of interest, one or more first attributes from a plurality of the first attributes; selecting one of the second attributes for an observation point from which observation data of the second attributes is obtained; obtaining the observed data for the selected second attribute and obtaining the estimated data for the selected first attribute for each of one or more of the first points of interest; creating training data including the acquired observation data as a response variable and one or more of the estimation data acquired for each of the one or more first target locations as an explanatory variable; using the training data to train an estimation model that inputs one or more of the explanatory variables and outputs an estimate of the objective variable; A program to execute.

[0123] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0124] 100 Information processing device 101 Selection section 102 Acquisition Department 103 Creation Department 104 Learning Department 111 Estimation Department 112 Calculation Unit 113 Output control section 121 Forecast data storage unit 122 Observation data storage unit 123 Selection data storage unit 124 Parameter storage section 125 Model Memory Unit 131 Display section

Claims

1. selecting one or more first target points from among a plurality of target points at which estimated data of a plurality of first attributes can be obtained; selecting one or more first attributes from the plurality of first attributes for each of the one or more first points of interest; selecting one of the second attributes for an observation point from which observation data of a plurality of second attributes is obtained; acquiring the observed data for the selected second attribute, and acquiring the estimated data for the selected first attribute for each of the one or more first points of interest; creating training data including the acquired observation data as a response variable and one or more of the estimation data acquired for each of the one or more first target points as an explanatory variable; using the training data to train an estimation model that inputs one or more of the explanatory variables and outputs an estimated value of the objective variable; Processing section An information processing device comprising:

2. The processing unit selecting one or more first target points from among the plurality of target points based on a predetermined rule; The information processing device according to claim 1 .

3. The rule is: a rule for classifying the plurality of target points into a plurality of clusters based on distances between a plurality of pieces of estimated data obtained at the plurality of target points, and selecting one or more first target points from each of the plurality of clusters; The processing unit selecting one or more of the first points of interest from each of the plurality of clusters based on the rule; The information processing device according to claim 2 .

4. The processing unit outputting the importance of each of the one or more explanatory variables during training of the estimation model; selecting the first attribute corresponding to the explanatory variable whose importance is greater than that of other explanatory variables; The information processing device according to claim 1 .

5. the estimation model outputs the estimated value at a target time of estimation, the target time being specified at intervals smaller than the intervals of the plurality of forecast times targeted for estimation in the estimation data; The information processing device according to claim 1 .

6. The processing unit inputting input data, which is the estimated data of the selected first attribute acquired for each of the one or more first target locations, into the trained estimation model to obtain the estimated value of the observation data of the selected second attribute; The information processing device according to claim 1 .

7. The processing unit Selecting the second attribute to be used for calculating visibility information, which is the presence or absence of visibility impairment or the visibility distance; Calculating the visibility information using the estimated value. The information processing device according to claim 6 .

8. The processing unit selecting the second attribute, which is the presence or absence of a visibility impairment or the visibility distance; The information processing device according to claim 1 .

9. The processing unit selecting one or more first estimation execution times representing times over which estimation of the estimation data is to be performed; obtaining, for each of the one or more first points of interest, the estimated data for the selected first attribute obtained at the first estimated execution time; The information processing device according to claim 1 .

10. The processing unit selecting one or more first forecast times from among a plurality of forecast times to be estimated in the estimation data; obtaining, for each of the one or more first target locations, the estimated data for the selected first attribute resulting from the estimation of the first forecast time; The information processing device according to claim 1 .

11. An information processing method executed by an information processing device, selecting one or more first target points from among a plurality of target points from which estimation data of a plurality of first attributes can be obtained; selecting one or more first attributes from a plurality of first attributes for each of the one or more first points of interest; selecting one of the second attributes for an observation point from which observation data of the second attributes is obtained; acquiring the observed data for the selected second attribute and acquiring the estimated data for the selected first attribute for each of one or more of the first points of interest; creating training data including the acquired observation data as a response variable and one or more of the estimation data acquired for each of the one or more first target locations as explanatory variables; using the training data to train an estimation model that inputs one or more of the explanatory variables and outputs an estimate of the dependent variable; An information processing method including:

12. On the computer, selecting one or more first target points from among a plurality of target points from which estimation data of a plurality of first attributes can be obtained; selecting one or more first attributes from a plurality of first attributes for each of the one or more first points of interest; selecting one of the second attributes for an observation point from which observation data of the second attributes is obtained; acquiring the observed data for the selected second attribute and acquiring the estimated data for the selected first attribute for each of one or more of the first points of interest; creating training data including the acquired observation data as a response variable and one or more of the estimation data acquired for each of the one or more first target locations as explanatory variables; using the training data to train an estimation model that inputs one or more of the explanatory variables and outputs an estimate of the dependent variable; A program to execute.

Images