JP2020091171A - Weather forecasting system, weather forecasting method, and weather forecasting program - Google Patents

Abstract

To forecast weather at any point with a high degree of accuracy.SOLUTION: A weather forecasting system generates forecasting data that indicate future weather at a target point by giving a neural network input data that indicate past weather at each of a plurality of observation points. The input data include a plurality of data sets corresponding to the plurality of observation points. A data set includes time-sequence weather data at one observation point, a geographical location of the observation point and a geographical location of the target point. The neural network includes a plurality of point blocks corresponding to the plurality of data sets and an assembly block for processing intermediate values to generate forecasting data. Each of the point blocks includes a weather block corresponding to the time-sequence weather data, a location block corresponding to the geographical location of both the observation points and the target point, and a coupling block for generating the intermediate values.SELECTED DRAWING: Figure 4

Description

One aspect of the present disclosure relates to a weather forecast system, a weather forecast method, and a weather forecast program.

Mechanisms for predicting weather have been known for some time. For example, Patent Documents 1 and 2 each describe a device that predicts the development and decline of a precipitation area quickly and easily.

Japanese Patent No. 3377073 Patent No. 3377075

Regarding the weather forecast of a point where the weather is not observed by the meteorological observation device, the geographical range to be forecast may be relatively wide or the accuracy may be insufficient. Therefore, it is desired to predict the weather at an arbitrary point with high accuracy.

A weather forecast system according to one aspect of the present disclosure includes at least one processor. At least one processor generates prediction data indicating future weather at the target point by supplying the neural network with input data indicating past weather at each of the plurality of observation points. The input data includes a plurality of data sets corresponding to a plurality of observation points. One data set corresponding to one observation point includes time series meteorological data at the one observation point, the geographical position of the one observation point, and the geographical position of the target point. The neural network includes a plurality of point blocks that correspond to a plurality of data sets and are independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate prediction data. Each of the plurality of point blocks includes a meteorological block corresponding to time-series meteorological data, a position block corresponding to both geographical positions of an observation point and a target point and independent of the meteorological block, the meteorological block and the meteorological block. A coupling block that processes the output values from both of the position blocks to produce an intermediate value.

A weather forecasting method according to one aspect of the present disclosure is executed by a weather forecasting system including at least one processor. The weather forecast method includes a step of generating forecast data indicating future weather at a target point by applying input data indicating past weather at each of a plurality of observation points to a neural network. The input data includes a plurality of data sets corresponding to a plurality of observation points. One data set corresponding to one observation point includes time series meteorological data at the one observation point, the geographical position of the one observation point, and the geographical position of the target point. The neural network includes a plurality of point blocks that correspond to a plurality of data sets and are independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate prediction data. Each of the plurality of point blocks includes a meteorological block corresponding to time-series meteorological data, a position block corresponding to both geographical positions of an observation point and a target point and independent of the meteorological block, the meteorological block and the meteorological block. A coupling block that processes the output values from both of the position blocks to produce an intermediate value.

A meteorological prediction program according to an aspect of the present disclosure includes a step of generating prediction data indicating future weather at a target point by supplying input data indicating past weather at each of a plurality of observation points to a neural network. Let the computer run. The input data includes a plurality of data sets corresponding to a plurality of observation points. One data set corresponding to one observation point includes time series meteorological data at the one observation point, the geographical position of the one observation point, and the geographical position of the target point. The neural network includes a plurality of point blocks that correspond to a plurality of data sets and are independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate prediction data. Each of the plurality of point blocks includes a meteorological block corresponding to time-series meteorological data, a position block corresponding to both geographical positions of an observation point and a target point and independent of the meteorological block, the meteorological block and the meteorological block. A coupling block that processes the output values from both of the position blocks to produce an intermediate value.

In such an aspect, input data including not only the past weather at a plurality of observation points but also the geographical position of each observation point and the geographical position of the target point is given to the neural network. Since the geographical position between the observation point and the target point is considered when predicting the weather at the target point, it becomes possible to predict the weather at any point with high accuracy.

In the weather prediction system according to another aspect, the target point may be different from any of the plurality of observation points. In this case, the meteorological observation device can be used to predict the meteorology of a point where the meteorological observation is not made.

In the weather prediction system according to another aspect, the target point may be located within a geographical range defined by connecting two or more outermost observation points among the plurality of observation points. By setting the target point so as to be surrounded by a plurality of observation points, the weather at the target point is predicted based on the surrounding time series meteorological data. Therefore, it becomes possible to predict the weather at the target point with higher accuracy.

According to one aspect of the present disclosure, it is possible to predict the weather at an arbitrary point with high accuracy.

It is a map showing an example of a positional relationship between an observation point and a target point. It is a figure which shows an example of the hardware constitutions of the computer used with the weather prediction system which concerns on embodiment. It is a figure showing an example of functional composition of a weather forecasting system concerning an embodiment. It is a figure which shows an example of the neural network used with the weather prediction system which concerns on embodiment. It is a flowchart which shows an example of the learning process of the weather prediction system which concerns on embodiment. It is a flowchart which shows an example of the prediction process of the weather prediction system which concerns on embodiment. It is a graph which shows an example of a prediction result.

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements will be denoted by the same reference symbols, without redundant description.

[System overview]
The weather prediction system 1 according to the embodiment is a computer system that predicts future weather in a designated place. Meteorology refers to the state of the atmosphere. The forecasted weather element is not limited. For example, the meteorological prediction system 1 is a state of the atmosphere viewed from the ground (sunny, cloudy, rain, etc.), atmospheric pressure (sea level pressure), temperature, humidity, wind direction, wind speed, and feeling of rain (an index indicating whether there is precipitation). Alternatively, the rainfall intensity (precipitation amount per unit time) may be predicted.

The weather forecasting system 1 uses machine learning. Machine learning is a method of autonomously finding rules or rules by iteratively learning based on given information. The weather forecast system 1 executes machine learning using a calculation model including a neural network. A neural network is a model of information processing that imitates the mechanism of human cranial nerve system. For example, the weather prediction system 1 predicts weather by executing deep learning. Deep learning refers to machine learning using a multilayered neural network (deep neural network (DNN)).

The weather prediction system 1 can generate a calculation model by repeating learning and can acquire this calculation model as a learned model. This corresponds to the learning phase. It should be noted that the learned model is a computational model that is estimated to be optimal for predicting future weather, and is not necessarily a “computation model that is optimal in reality”. The weather forecast system 1 can also output forecast data indicating future weather by processing the input data using the learned model. This corresponds to the prediction phase.

Trained models are portable between computer systems. Therefore, the trained model generated by one computer system can be used by another computer system. Of course, one computer system may perform both generation and utilization of the trained model. That is, the weather prediction system 1 may execute both the learning phase and the prediction phase, or may not execute either one of the learning phase and the prediction phase.

The weather forecasting system 1 generates forecast data showing future weather at one target point by supplying input data showing past weather at each of a plurality of observation points to a calculation model (neural network). An observation point is a point where weather is observed using an observation device. For example, the observation point may be a point where the meteorological observation device is installed. The number of observation points, the arrangement of individual observation points, and the distance between observation points are not limited at all. The target point is a point where future weather is predicted. Typically, the target point is a point different from any of the plurality of observation points, specifically, a point where the weather is not observed using the meteorological observation device, for example, the meteorological observation device is not installed. It is a point. However, the observation point may be selected as the target point. The target point may be any position within a geographical range defined by at least a part of the plurality of observation points. The term "within geographic area" may be a concept that includes the boundaries of a geographic area.

FIG. 1 is a map showing an example of the positional relationship between an observation point and a target point. In this example, a large number of observation points 91 are shown. FIG. 1 also shows an example of a geographical range that includes the point of interest. When predicting the weather of the target point based on the past weather at each of the plurality of observation points 91, the geographical range 92 is defined by connecting two or more observation points 91 located on the outermost side. .. Therefore, the geographical range 92 is set to include all of the plurality of observation points 91. In this case, an arbitrary location within the geographical range 92 (eg, the point 93, the point 94, or the point 95) can be the target point.

[System configuration]
FIG. 2 shows a general hardware configuration of the computer 100 that constitutes the weather forecast system 1. For example, the computer 100 includes a processor 101, a main storage unit 102, an auxiliary storage unit 103, a communication control unit 104, an input device 105, and an output device 106. Processor 101 executes an operating system and application programs. The main storage unit 102 is composed of, for example, a ROM and a RAM. The auxiliary storage unit 103 is composed of, for example, a hard disk or a flash memory, and generally stores a larger amount of data than the main storage unit 102. The communication control unit 104 is composed of, for example, a network card or a wireless communication module. The input device 105 includes, for example, a keyboard, a mouse, a touch panel, and the like. The output device 106 is composed of, for example, a monitor and a speaker.

Each functional element of the weather forecast system 1 is realized by the weather forecast program 110 stored in advance in the auxiliary storage unit 103. Specifically, each functional element is realized by reading the weather forecast program 110 onto the processor 101 or the main storage unit 102 and executing the weather forecast program 110. The processor 101 operates the communication control unit 104, the input device 105, or the output device 106 according to the weather forecast program 110 to read and write data in the main storage unit 102 or the auxiliary storage unit 103. The data or database required for processing is stored in the main storage unit 102 or the auxiliary storage unit 103.

The weather forecast program 110 may be provided after being fixedly recorded in a tangible recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory. Alternatively, the weather forecast program 110 may be provided as a data signal superimposed on a carrier wave via a communication network.

The weather forecasting system 1 may be configured by one computer 100 or may be configured by a plurality of computers 100. When a plurality of computers 100 are used, these computers 100 are connected via a communication network such as the Internet or an intranet to logically construct one weather prediction system 1.

FIG. 3 is a diagram showing an example of the functional configuration of the weather prediction system 1. In the present embodiment, the weather prediction system 1 includes a first system 10 that generates a learned model for predicting weather, and a second system 10 that generates and outputs prediction data indicating future weather using the learned model. And a system 20. The first system 10 includes a learning unit 11 that generates and acquires a learned model by executing machine learning as a functional element. The second system 20 includes, as a functional element, a prediction unit 21 that obtains prediction data from input data using the learned model.

As mentioned above, the trained model is portable between computer systems. Therefore, as in this embodiment, the calculation model generated by the first system 10 can be used by the second system 20 which is another computer system. Of course, it is not essential to divide the weather forecast system into the first system 10 and the second system 20.

FIG. 4 is a diagram showing an example of the neural network 30 used in the weather prediction system 1. The neural network 30 is divided into a plurality of blocks. Specifically, the neural network 30 includes a plurality of point blocks 31 corresponding to a plurality of observation points and one assembly block 32.

One point block 31 corresponds to one observation point. The plurality of point blocks 31 are independent of each other, which means that each point block 31 does not exchange data with any of the remaining point blocks 31. FIG. 4 particularly shows a point block 31a corresponding to the observation point A, a point block 31b corresponding to the observation point B, and a point block 31c corresponding to the observation point C. The number of observation points may be any number as long as it is 2 or more. Therefore, the number of point blocks 31 is an arbitrary number of 2 or more. For example, when the weather prediction system 1 predicts the weather at the target point based on the past weather at 48 observation points, the neural network 30 includes 48 point blocks 31.

The input data input to the neural network 30 includes a plurality of data sets corresponding to a plurality of observation points. One data set corresponds to one observation point. Therefore, it can be said that the plurality of spot blocks 31 correspond to a plurality of data sets. FIG. 4 particularly shows a data set A corresponding to the observation point A, a data set B corresponding to the observation point B, and a data set C corresponding to the observation point C.

Each data set includes time series meteorological data at the corresponding observation point, the geographical position of the observation point, and the geographical position of the target point. The time-series meteorological data is time-series data indicating the change in the weather at the observation point in a certain time width, and more specifically, data indicating the weather at the observation point at each of a plurality of consecutive observation times. .. The time series meteorological data includes a plurality of sets of one observation time and one or more kinds of meteorological values indicating the meteorology at that time. The geographical position means a position on the ground surface in the real world, and is represented by latitude and longitude, for example.

FIG. 4 illustrates a data set A at the observation point A. In this example, the time-series meteorological data indicates meteorological values obtained at 10-minute intervals from the reference time t ₀ to one hour before the reference time t ₀ . In other words, the time series meteorological data includes seven sets of time and meteorological value. In this example, the meteorological value is composed of temperature, humidity, sea level pressure, wind (u, v), and rain. The expression "wind (u, v)" means that the wind is expressed by a vector (wind direction and wind speed) in the east-west direction and a vector (wind direction and wind speed) in the north-south direction. In the example of FIG. 4, the geographical positions of the observation point and the target point are both indicated by latitude and longitude. The geographic location of the point of interest is the same for all datasets.

Each point block 31 includes a weather block 33 corresponding to time-series meteorological data, a position block 34 corresponding to geographical positions of both an observation point and a target point, and both the weather block 33 and the position block 34. The weather block 33 and the position block 34 are independent of each other, including a coupling block 35 for processing the output values of the, which means that no data is exchanged between the weather block 33 and the position block 34. The weather block 33 receives and processes the time-series weather data to generate an output value, and passes the output value to the coupling block 35 as a processing result. The position block 34 receives and processes the geographical positions of both the observation point and the target point to generate an output value, and passes the output value to the coupling block 35 as a processing result. It can be said that the output value passed from the weather block 33 and the position block 34 to the coupling block 35 is an intermediate value obtained during the processing in the neural network 30. The coupling block 35 processes the intermediate value to generate an output value, and passes the output value to the assembly block 32 as a processing result. It can be said that the output value passed from the point block 31 (more specifically, the coupling block 35) to the assembly block 32 is an intermediate value obtained during the processing in the neural network 30.

The assembly block 32 is a block that processes a plurality of output values obtained from the plurality of spot blocks 31 (more specifically, a plurality of coupling blocks 35) to generate and output prediction data (prediction result). For example, when the weather prediction system 1 predicts the weather at the target point based on the past weather at 48 observation points, the assembly block 32 receives the output values from the 48 point blocks 31, and Process the output values to generate and output predicted data.

The configuration of each block in the neural network 30 is not limited. For example, the number of layers in each block and the number of nodes in each fully-connected layer may be arbitrarily determined. FIG. 4 also shows an example of the configuration of individual blocks. In this example, the weather block 33 is composed of two fully connected layers “FC120” having 120 nodes and one fully connected layer “FC20” having 20 nodes. The position block 34 is composed of three fully connected layers "FC10" having 10 nodes. The coupling block 35 is also composed of three fully coupled layers “FC10” having 10 nodes. The assembly block 32 is composed of two fully connected layers “FC10” having 10 nodes and an output layer “FC1”.

In the example of FIG. 4, the neural network 30 outputs prediction data indicating the temperature 60 minutes after the reference time t ₀ . This means that the neural network 30 is used only to predict the temperature 60 minutes after any given point. When predicting the temperature (for example, temperature after 30 minutes) at another target point or another meteorological element (for example, rain feeling) after 60 minutes at any target point Therefore, the weather forecasting system 1 needs to prepare another neural network. Also, when changing the number of observation points, the weather forecasting system 1 needs to prepare another neural network.

[System operation]
The operation of the weather prediction system 1 will be described and the weather prediction method according to the present embodiment will be described with reference to FIGS. 5 and 6. FIG. 5 is a flowchart showing an example of the learning process of the weather prediction system 1 as a process flow S1. FIG. 6 is a flowchart showing an example of a prediction process of the weather prediction system 1 as a process flow S2. 5 and 6 correspond to the learning phase and the prediction phase, respectively. Both of the processing flows S1 and S2 can correspond to the weather forecasting method according to the present disclosure.

The processing flow S1 executed by the learning unit 11 will be described with reference to FIG. It can be said that this processing flow S1 is an example of a method of generating a learned model.

In step S11, the learning unit 11 selects one of the plurality of observation points as a target point. The plurality of observation points used for learning may be selected based on arbitrary criteria.

In step S12, the learning unit 11 executes machine learning using input data including a data set of all observation points other than the target point (that is, all remaining observation points). It can be said that the input data used in the learning phase is training data for machine learning. The method of acquiring input data is not limited. For example, the learning unit 11 may access a given database to read the input data, or may receive the input data sent from another computer. In this machine learning, the actual measurement value at the observation point selected as the target point is used as the correct answer. The learning unit 11 generates prediction data by giving the input data to the neural network 30. Then, the learning unit 11 updates the weight in the neural network 30 using a method such as back propagation (error back propagation method) based on the difference (that is, error) between the predicted data (prediction result) and the correct answer. To do.

In step S13, the learning unit 11 determines whether or not the end condition is satisfied. This end condition is a condition for ending machine learning and acquiring a learned model. The termination condition may be set arbitrarily. For example, the learning unit 11 may end the machine learning when the performance of the neural network 30 satisfies a predetermined standard. Alternatively, the learning unit 11 may end the machine learning when the learning is performed a predetermined number of times. For example, the learning unit 11 may end the machine learning when all of the selected plurality of observation points are processed as the target points.

If the ending condition is not satisfied, the process proceeds to step S14. In step S14, the learning unit 11 selects one observation point that has not been selected as the target point yet as the next target point. Then, the learning unit 11 executes the process of step S12 again.

If the end condition is satisfied, the process proceeds to step S15. In step S15, the learning unit 11 acquires the machine-learned calculation model (that is, the neural network 30) as a learned model. This processing means that a trained model has been generated. The learning unit 11 outputs the learned model. The output method of the learned model is not limited. For example, the learning unit 11 may store the learned model in a storage device such as a memory or a database, or may transmit the model to another computer system.

For example, when using the observation results at 48 observation points, the learning unit 11 selects one observation point as the target point (step S11), and inputs data including the data set of the remaining 47 observation points. Is used to execute machine learning (step S12). If the end condition is to process all the observation points as target points, the learning unit 11 repeats machine learning while replacing the target points. Therefore, in this case, the learning unit 11 acquires the learned model after executing the process of step S12 48 times.

The processing flow S2 executed by the prediction unit 21 will be described with reference to FIG. In step S21, the prediction unit 21 receives input data. The method of acquiring input data is not limited. For example, the prediction unit 21 may access a given database to read the input data, or may receive the input data sent from another computer.

In step S22, the prediction unit 21 inputs the input data to the learned model (neural network) to generate prediction data.

In step S23, the prediction unit 21 outputs the prediction data. As a result, the user of the weather forecast system 1 can obtain the weather forecast of the target point. The output method of the prediction data is not limited. For example, the prediction unit 21 may display the prediction data on a monitor, store it in a predetermined database, or send it to another computer system. The method of expressing the prediction data is also arbitrary, and for example, the prediction data can be expressed by any method such as text, table, graph, or illustration.

[effect]
As described above, the weather prediction system according to one aspect of the present disclosure includes at least one processor. At least one processor generates prediction data indicating future weather at the target point by supplying the neural network with input data indicating past weather at each of the plurality of observation points. The input data includes a plurality of data sets corresponding to a plurality of observation points. One data set corresponding to one observation point includes time series meteorological data at the one observation point, the geographical position of the one observation point, and the geographical position of the target point. The neural network includes a plurality of point blocks that correspond to a plurality of data sets and are independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate prediction data. Each of the plurality of point blocks includes a meteorological block corresponding to time-series meteorological data, a position block corresponding to both geographical positions of an observation point and a target point and independent of the meteorological block, the meteorological block and the meteorological block. A coupling block that processes the output values from both of the position blocks to produce an intermediate value.

A weather forecasting method according to one aspect of the present disclosure is executed by a weather forecasting system including at least one processor. The weather forecast method includes a step of generating forecast data indicating future weather at a target point by applying input data indicating past weather at each of a plurality of observation points to a neural network. The input data includes a plurality of data sets corresponding to a plurality of observation points. One data set corresponding to one observation point includes time series meteorological data at the one observation point, the geographical position of the one observation point, and the geographical position of the target point. The neural network includes a plurality of point blocks that correspond to a plurality of data sets and are independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate prediction data. Each of the plurality of point blocks includes a meteorological block corresponding to time-series meteorological data, a position block corresponding to both geographical positions of an observation point and a target point and independent of the meteorological block, the meteorological block and the meteorological block. A coupling block that processes the output values from both of the position blocks to produce an intermediate value.

A meteorological prediction program according to an aspect of the present disclosure includes a step of generating prediction data indicating future weather at a target point by supplying input data indicating past weather at each of a plurality of observation points to a neural network. Let the computer run. The input data includes a plurality of data sets corresponding to a plurality of observation points. One data set corresponding to one observation point includes time series meteorological data at the one observation point, the geographical position of the one observation point, and the geographical position of the target point. The neural network includes a plurality of point blocks that correspond to a plurality of data sets and are independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate prediction data. Each of the plurality of point blocks includes a meteorological block corresponding to time-series meteorological data, a position block corresponding to both geographical positions of an observation point and a target point and independent of the meteorological block, the meteorological block and the meteorological block. A coupling block that processes the output values from both of the position blocks to produce an intermediate value.

In such an aspect, input data including not only the past weather at a plurality of observation points but also the geographical position of each observation point and the geographical position of the target point is given to the neural network. Since the geographical position between the observation point and the target point is considered when predicting the weather at the target point, it becomes possible to predict the weather at any point with high accuracy.

In the weather prediction system according to another aspect, the target point may be different from any of the plurality of observation points. In this case, the meteorological observation device can be used to predict the meteorology at a point where no meteorological observation is made.

In the weather prediction system according to another aspect, the target point may be located within a geographical range defined by connecting two or more outermost observation points among the plurality of observation points. By setting the target point so as to be surrounded by a plurality of observation points, the weather at the target point is predicted based on the surrounding time series meteorological data. Therefore, it becomes possible to predict the weather at the target point with higher accuracy.

[Example]
Examples will be specifically described below, but the present disclosure is not limited thereto.

As the time series meteorological data, actual measurement values obtained from an ultra high density meteorological observation system called POTEKA (registered trademark) were used. The meteorological elements constituting the time series meteorological data are temperature, humidity, sea level pressure, wind (u, v), and rain. Time series meteorological data was prepared so that the changes in meteorological values could be shown at 10-minute intervals.

One of the plurality of observation points 91 shown in FIG. 1 was selected as a target point, which is located in the center. Further, the remaining observation points 91 are selected as the observation points for predicting the weather of the target point. Then, the weather forecasting system 1 was used to obtain a process of predicting the temperature of the target point one hour later, and the actual temperature of the target point one hour later was measured. The prediction and measurement were repeated so that the predicted and measured temperature values were obtained at 10-minute intervals in a given time span. FIG. 7 shows the result of the prediction and measurement. The solid line shows the measured value, and the broken line shows the predicted value.

[Modification]
In the above, it explained in detail based on the embodiment in this indication. However, the present disclosure is not limited to the above embodiment. The present disclosure can be variously modified without departing from the gist thereof.

The processing procedure of the weather forecasting method executed by at least one processor is not limited to the example in the above embodiment. For example, some of the steps described above may be omitted, or the steps may be performed in a different order. Further, any two or more of the steps described above may be combined, or some of the steps may be modified or deleted. Alternatively, other steps may be executed in addition to the above steps.

When comparing the magnitude relationship of two numerical values in a weather forecast system, either of the two criteria "greater than or equal to" and "greater than" may be used, and the two criteria "less than or equal to" and "less than" may be used. Either of these may be used. The selection of such a criterion does not change the technical significance of the process of comparing the magnitude relationship of two numerical values.

According to one aspect of the present disclosure, it is possible to predict the weather at an arbitrary point with high accuracy.

1 Meteorological Prediction System 10 First System 11 Learning Unit 20 Second System 21 Prediction Unit 30 Neural Networks 31, 31a, 31b, 31c Point Block 32 Assembly Block 33 Meteorological Block 34 Position Block 35 Coupling Block 91 Observation Point 92 Geographical Range 100 computer 101 processor 110 weather forecast program

Claims (5)

With at least one processor,
The at least one processor, by inputting the input data indicating the past weather at each of the plurality of observation points to the neural network, to generate prediction data indicating the future weather at the target point,
The input data includes a plurality of data sets corresponding to the plurality of observation points,
One said data set corresponding to one said observation point includes time series meteorological data at said one observation point, geographical position of said one observation point, and geographical position of said target point,
The neural network includes a plurality of point blocks corresponding to the plurality of data sets and independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate the prediction data,
Each of the plurality of point blocks, a weather block corresponding to the time series meteorological data, a position block corresponding to the geographical position of both the observation point and the target point and independent of the weather block, A coupling block for processing output values from both the weather block and the position block to produce the intermediate value,
Weather forecast system. The target point is different from any of the plurality of observation points,
The weather forecasting system according to claim 1. The target point is located within a geographical range defined by connecting two or more outermost observation points among the plurality of observation points,
The weather prediction system according to claim 1. A weather forecasting method executed by a weather forecasting system comprising at least one processor,
By providing the neural network with input data indicating past weather at each of the plurality of observation points, including a step of generating prediction data indicating future weather at the target point,
The input data includes a plurality of data sets corresponding to the plurality of observation points,
One of the data sets corresponding to one of the observation points includes time-series weather data at the one observation point, a geographical position of the one observation point, and a geographical position of the target point,
The neural network includes a plurality of point blocks corresponding to the plurality of data sets and independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate the prediction data,
Each of the plurality of point blocks, a weather block corresponding to the time-series weather data, and a position block corresponding to the geographical position of both the observation point and the target point and independent of the weather block, A coupling block for processing output values from both the weather block and the position block to produce the intermediate value.
Weather forecast method. By giving input data indicating past weather at each of a plurality of observation points to the neural network, causing the computer to execute the step of generating prediction data indicating future weather at the target point,
The input data includes a plurality of data sets corresponding to the plurality of observation points,
One said data set corresponding to one said observation point includes time series meteorological data at said one observation point, geographical position of said one observation point, and geographical position of said target point,
The neural network includes a plurality of point blocks corresponding to the plurality of data sets and independent of each other, and an assembly block that processes a plurality of intermediate values from the plurality of point blocks to generate the prediction data,
Each of the plurality of point blocks, a weather block corresponding to the time series meteorological data, a position block corresponding to the geographical position of both the observation point and the target point and independent of the weather block, A coupling block for processing output values from both the weather block and the position block to produce the intermediate value,
Weather forecast program.

Images