Turbulence prediction system and turbulence prediction method

Abstract

A turbulence prediction system and a turbulence prediction method that involve predicting a location where turbulence is likely to occur in a prediction area at a decision point, in which, based on meteorological data related to any meteorological parameter of a location where turbulence has occurred in the past, a plurality of turbulence prediction pattern data related to any meteorological parameter is created, for a prediction area at a decision point, decision-use meteorological data is created based on meteorological data related to any meteorological parameter, a calculation of a portion where the decision-use meteorological data and each of the plurality of turbulence prediction pattern data are highly similar is performed, and a place where the similarity is high is decided to be a location where turbulence is likely to occur in the prediction area.

Description

TECHNICAL FIELD

[0001] The present application relates to a turbulence prediction system and a turbulence prediction method, and particularly relates to a turbulence prediction system and a turbulence prediction method for navigation of an aircraft. BACKGROUND

[0002] When a turbulence is found on a predetermined route after the start of flight, it is necessary to make a judgment in the navigation before arrival at the destination and to change the route to avoid the turbulence, but there is a problem that the optimal route cannot be selected in the route change after the start of flight. Therefore, it is desired that the occurrence of turbulence on the route can be predicted in a short time with small calculation load and high accuracy.

[0003] PRIOR ART DOCUMENTS

[0004] PATENT DOCUMENTS

[0005] Patent Document 1: Japanese Patent Application Publication No. 2003-67900

[0006] Patent Document 2: Japanese Patent Application Publication No. 2009-192262 SUMMARY

[0007] PROBLEMS TO BE SOLVED BY THE INVENTION

[0008] Since the occurrence of turbulence is complexly involved with the topographical condition and the fluid condition of air, the mechanism of occurrence is complex, and a method of predicting the occurrence of turbulence by finding an exact solution based on a mathematical model on the physical behavior of fluid related to the occurrence of turbulence is limited. Therefore, in the existing method and system of predicting turbulence as disclosed in Patent Document 1 and Patent Document 2, for example, based on a model on the physical behavior of fluid related to the occurrence of turbulence, a parameter or index considered to contribute to the occurrence of turbulence is selected as a simple model of turbulence, and by calculating the parameter or index, turbulence is found.

[0009] In the methods disclosed in Patent Literature 1 and Patent Literature 2, parameters or indices calculated for the purpose of prediction are determined based on a mathematical model on the physical behavior of fluid in relation to the occurrence of turbulent flow. That is, the methods disclosed in Patent Literature 1 and Patent Literature 2 employ a process solved according to a mathematical model on the physical behavior of fluid in relation to the occurrence of turbulent flow. In general, if an accurate solution is to be precisely calculated according to a mathematical model on the behavior of fluid in relation to the occurrence of turbulent flow, the computational load is large and is not realistic, and therefore, a simple model in which the mathematical model on the behavior of fluid is simplified is used. Patent Literature 1 and Patent Literature 2 are technologies that intend to predict the occurrence of turbulent flow with this method. The method that takes an accurate solution as a simple model can affect the accuracy of the prediction of the occurrence of turbulent flow, and on the other hand, in the method that solves the analysis with a model close to the accurate solution, the problem of an increase in the computational load still remains. Thus, a system that predicts the occurrence of turbulent flow based on a new idea is desired.

[0010] Technical Solution for Solving the Problem

[0011] One technical solution of the present application that solves the above problem is a turbulent flow prediction system that has an arithmetic unit and a storage unit that stores a plurality of turbulent flow prediction pattern data related to an arbitrary meteorological parameter, and is a turbulent flow prediction system that predicts a place where the occurrence of turbulent flow is highly likely in a prediction area at a determination point (time point), the plurality of turbulent flow prediction pattern data related to the arbitrary meteorological parameter being data produced based on meteorological data related to the arbitrary meteorological parameter of a place where turbulent flow has occurred in the past, the arithmetic unit producing determination meteorological data based on meteorological data related to the arbitrary meteorological parameter for the prediction area at the determination point, the arithmetic unit performing calculation of a portion where the similarity of the determination meteorological data and each of the plurality of turbulent flow prediction pattern data is high, and determining a place where the similarity is high as a place where the occurrence of turbulent flow is highly likely in the prediction area.

[0012] Another aspect of the present application is a turbulence prediction method for calculating a degree of possibility of turbulence occurrence in a prediction area by a turbulence prediction system having an arithmetic unit and a storage unit, the turbulence prediction method including a learning process and a determination process, the learning process including a step of creating a plurality of turbulence prediction pattern data by the arithmetic unit based on meteorological data related to an arbitrary meteorological parameter of a past turbulence occurrence site and storing the data in the storage unit, and the determination process including a step of creating determination meteorological data by the arithmetic unit based on meteorological data related to the arbitrary meteorological parameter for the prediction area, and a step of calculating a degree of similarity of the determination meteorological data to each of the plurality of turbulence prediction pattern data by the arithmetic unit, and determining a place where the degree of similarity is high as a place where the degree of possibility of turbulence occurrence in an arbitrary region (area) in the prediction area is high.

[0013] Effects of the Invention

[0014] According to the turbulence prediction system and the turbulence prediction method of the present application, it is possible to predict the occurrence of turbulence with low calculation load and high accuracy based on data of the fact that turbulence has occurred without being based on a mathematical model related to the physical behavior of a fluid in which turbulence occurs. BRIEF DESCRIPTION OF DRAWINGS

[0015] FIG. 1A Fig. 1 shows a first turbulence prediction pattern principle of the present application.

[0016] FIG. 1B Fig. 2 shows a second turbulence prediction pattern principle of the present application.

[0017] FIG. 1C Fig. 3 shows a third turbulence prediction pattern principle of the present application.

[0018] FIG. 1D Fig. 4 shows a fourth turbulence prediction pattern principle of the present application.

[0019] FIG. 1E Fig. 5 shows a fifth turbulence prediction pattern principle of the present application.

[0020] FIG. 2A Fig. 6 is a diagram showing a representative configuration of a turbulence prediction system as an embodiment of the present application.

[0021] FIG. 2B Fig. 7 is a diagram showing a representative configuration of a turbulence prediction device of a turbulence prediction system as an embodiment of the present application.

[0022] FIG. 2C Fig. 8 is a diagram showing another representative configuration of a turbulence prediction system as an embodiment of the present application.

[0023] FIG. 2D FIG. 1 is a diagram showing the form of data used by the turbulence prediction system of the present application.

[0024] FIG. 3 FIG. 2 is a diagram schematically showing the learning process and the determination process of the turbulence prediction system 1 as an embodiment of the present application.

[0025] FIG. 4A FIG. 3 is a diagram showing a flowchart of the learning process in Embodiment 1 of the turbulence prediction system of the present application, i.e., an example using one meteorological parameter.

[0026] FIG. 4B FIG. 4 is a diagram schematically showing the process of making meteorological parameter discrete data in the learning process of Embodiment 1 of the turbulence prediction system of the present application.

[0027] FIG. 4C FIG. 5 is a diagram schematically showing the process of making turbulence data in the learning process of Embodiment 1 of the turbulence prediction system of the present application.

[0028] FIG. 4D FIG. 6 is a diagram schematically showing the process of making turbulence prediction pattern data in the learning process of Embodiment 1 of the turbulence prediction system of the present application.

[0029] FIG. 4E FIG. 7 is a diagram showing a flowchart of the determination process in Embodiment 1 of the turbulence prediction system of the present application, i.e., an example using one meteorological parameter.

[0030] FIG. 5A FIG. 8 is a diagram showing a flowchart of the learning process in Embodiment 2 of the turbulence prediction system of the present application, i.e., an example using a plurality of meteorological parameters.

[0031] FIG. 5B FIG. 9 is a diagram showing a flowchart of the determination process in Embodiment 2 of the turbulence prediction system of the present application, i.e., an example using a plurality of meteorological parameters.

[0032] FIG. 5C FIG. 10 is a diagram schematically showing the process of making one turbulence prediction pattern data in the learning process of Embodiment 2 of the turbulence prediction system of the present application.

[0033] FIG. 5D FIG. 11 is a diagram schematically showing the process of making one combination determination meteorological data in the determination process of Embodiment 2 of the turbulence prediction system of the present application.

[0034] FIG. 6A FIG. 12 is a diagram showing a flowchart of the learning process in Embodiment 3 of the turbulence prediction system of the present application, i.e., an example of making turbulence prediction pattern data based on a plurality of continuous discrete-time meteorological data.

[0035] FIG. 6B Fig. 3 is a flowchart showing a judgment process in the example 3 of the turbulence prediction system of the present application, i.e., in the example of making the judgment-use meteorological data based on a plurality of continuous discrete-time meteorological data.

[0036] FIG. 6C Fig. 4 is a diagram schematically showing a process of making one turbulence prediction pattern data in the learning process of the example 3 of the turbulence prediction system of the present application.

[0037] FIG. 6D Fig. 5 is a diagram schematically showing a process of making one combination judgment-use meteorological data in the judgment process of the example 3 of the turbulence prediction system of the present application. DETAILED DESCRIPTION

[0038] (Principle of turbulence prediction)

[0039] First, the principle of turbulence prediction in the turbulence prediction system and the turbulence prediction method of the present application will be described with reference to FIGS. 1A-1E Figs. 6 to 10. FIGS. 1A-1E Fig. 11 is a diagram showing each pattern of the principle of turbulence prediction, FIGS. 1A-1E respectively showing the 1st turbulence prediction pattern principle to the 5th turbulence prediction pattern principle.

[0040] The turbulence prediction system and the turbulence prediction method of the present application are a system and a method of predicting a place where turbulence is likely to occur in a prediction area at a judgment point. The turbulence prediction of the present application is characterized in that, in common to the 1st turbulence prediction pattern principle to the 5th turbulence prediction pattern principle FIGS. 1A-1E , data is made at a past point and a judgment point thereafter, and turbulence is predicted by comparing these data.

[0041] The past point (P0, P -1 , P -2 ...) is a time defined in a learning process in which data of a past fact where turbulence has occurred is stored as teacher (supervision) data. The past point is a point before the judgment point, and means a point where it is clear that turbulence has occurred. The point at which prediction of turbulence is made using the teacher data is the judgment point. In the 1st turbulence prediction pattern principle, the 2nd turbulence prediction pattern principle, and the 4th turbulence prediction pattern principle, since a weather forecast is used, the judgment point is the point of the future (F0, F -1 , F -2 ...) which is the point of the judgment point at which it is desired to make a judgment of turbulence prediction, and in the 3rd turbulence prediction pattern principle and the 5th turbulence prediction pattern principle, the judgment point becomes the present (C0, C -1 , C -2). The current time point is not the current time point in the real time flow of time, but is the time point at which the judgment of the prediction of the turbulence is desired. In the 3rd turbulence prediction mode principle and the 5th turbulence prediction mode principle, the weather forecast is not used, but the change tendency of the weather at the near judgment time point is used.

[0042] The prediction area is an arbitrary region in which the occurrence of the turbulence is desired to be predicted, and is a region such as the Japanese archipelago which can be selected as an arbitrary region in which the weather data on an arbitrary weather parameter and the data on the occurrence or nonoccurrence of the turbulence at the same time point can be obtained. The arbitrary weather parameter is a weather factor which can have an influence on the turbulence, and as representative examples, for example, the wind speed, the wind shear, and the like.

[0043] The present application calculates the similarity of the judgment weather data to the high-similarity portion of each of the plurality of turbulence prediction mode data as the teacher data, and judges the region of the high-similarity portion as a region in which the occurrence of the turbulence is highly likely in the prediction area at the judgment time point. The plurality of turbulence prediction mode data as the teacher data are respectively prepared in advance in the learning process. The judgment weather data are prepared in the judgment process at a certain judgment time point. On the basis of this, there are various variations in the preparation of the judgment weather data and the plurality of turbulence prediction mode data. In addition, there are various variations in the calculation of the high-similarity portion. These variations are the 1st turbulence prediction mode principle to the 5th turbulence prediction mode principle.

[0044] (1st Turbulence Prediction Mode Principle)

[0045] Reference FIG. 1A , the 1st turbulence prediction mode principle is outlined. The 1st turbulence prediction mode principle is a basic turbulence prediction mode. In the 1st turbulence prediction mode principle, the plurality of turbulence prediction mode data 111 are data on the weather related to an arbitrary weather parameter at the past time points (P0, P -1 , P -2 ) at which the turbulence occurred, which are prepared in advance in the learning process on the basis of the weather data on the places at which the turbulence occurred. The arbitrary weather parameter is one or a plurality of weather parameters. The number of the past time points and places at which the turbulence occurred is large, and therefore the number of the plurality of turbulence prediction mode data 111 is naturally large. The plurality of turbulence prediction mode data 111 are stored in the storage device 22 and are updated sequentially. In the judgment process, the judgment weather data 121 are prepared on the basis of the weather data on the arbitrary weather parameter for the entire region of the prediction area at the judgment time point. The judgment time point is set to the future (F0, F -1 , F -2…), the meteorological data is meteorological forecast data. The arbitrary meteorological parameter of the determination meteorological data and the plurality of turbulence prediction pattern data as the teacher data is common. The prediction area in which turbulence prediction is desired is generally a wide range, and thus, the range is divided and stored as a plurality of data in the storage device 22.

[0046] In the determination process, the determination meteorological data O stored in the storage device 22 is compared with each of the plurality of turbulence prediction pattern data N stored in the storage device 22 to determine the similarity. The determination of the similarity can be performed by, for example, respectively taking the determination meteorological data O and each of the plurality of turbulence prediction pattern data N as image data and comparing the image data. Further, the similarity can be set as the presence or absence of a common feature. The feature can be calculated using various methods. For example, a feature amount of an image can be extracted, and the similarity can be calculated based on the tendency of the appearance thereof. In the first to third turbulence prediction pattern principles, a convolutional neural network (hereinafter, CNN) can also be used. In the CNN, without a process of recognizing and extracting a feature amount of an image, it is possible to calculate the presence or absence of a common feature.

[0047] The plurality of turbulence prediction pattern data 111 are data of meteorological parameters in which turbulence actually occurred in a certain region, but do not have a region attribute. For all of the determination meteorological data O, it is determined whether there is a common feature with each of the plurality of turbulence prediction pattern data N. In the determination meteorological data O, a region corresponding to a place in which there is a common feature with any of the plurality of turbulence prediction pattern data N means that there is a feature in which turbulence occurred in the past in the determination meteorological data O based on a meteorological forecast, and thus, the region corresponding to the place is determined to be a place in which turbulence is likely to occur in the prediction area.

[0048] (Second Turbulence Prediction Pattern Principle)

[0049] Reference FIG. 1B, the principle of the second turbulence prediction mode is explained. The principle of the second turbulence prediction mode is the principle of setting the arbitrary weather parameter in the creation of the plurality of turbulence prediction mode data 111 and the determination weather data 121 in the first turbulence prediction mode principle as a plurality of weather parameters. Other than this, it is the same as the first turbulence prediction mode principle. For example, as the plurality of weather parameters, the wind speed in the east-west direction, the wind speed in the north-south direction, the wind shear in the east-west direction, and the wind shear in the north-south direction are selected. Also, for each of the plurality of weather parameters, the plurality of turbulence prediction mode data 111 is created in the same manner as the first turbulence prediction mode principle. For example, in the case where four kinds of weather parameters are selected, four kinds of the plurality of turbulence prediction mode data 111 are created. The determination weather data 121 is also created for each of the plurality of weather parameters which is the same as the plurality of turbulence prediction mode data 111. Also, for each of the corresponding weather parameters, in the determination process, the determination weather data O stored in the storage device 22 is compared with each of the plurality of turbulence prediction mode data 111 stored in the storage device 22 to determine the similarity, and it is determined that the place where the turbulence is likely to occur in the prediction area is high.

[0050] In the similarity calculation, it is also possible to proceed in the same manner as the first turbulence prediction mode principle for each of the corresponding weather parameters. However, in the similarity calculation, the more the information amount of the data compared, the more the common feature can be highlighted, and the similarity calculation can be made easy. Therefore, it is also possible to further calculate the similarity as follows. The plurality of turbulence prediction mode data N in the plurality of turbulence prediction mode data 111 with respect to each of the plurality of weather parameters is arranged in a predetermined configuration with respect to each of the plurality of weather parameters as one combined turbulence prediction mode data N'. On the other hand, the determination weather data O with respect to each of the plurality of weather parameters is arranged in the same predetermined configuration as the configuration when creating one combined turbulence prediction mode data N' as one combined determination weather data O'. Also, it is possible to compare one combined turbulence prediction mode data N' with one combined determination weather data O' and calculate the similarity. By setting the data with respect to each of the plurality of weather parameters as combined data, the number of pixels in one image data can be increased, and the information amount of the image data can be increased.

[0051] (Third Turbulence Prediction Mode Principle)

[0052] Reference FIG. 1C, outlines the principle of the 3rd turbulence prediction model. The principle of the 3rd turbulence prediction model is different from the principle of the 1st turbulence prediction model using weather forecast in making the weather data for judgment, and the principle of the 2nd turbulence prediction model, and is a method not using weather forecast. First, one weather parameter is selected as an arbitrary weather parameter. Then, in making the weather data for judgment 121, not the weather data of the future time points as in weather forecast, but the weather data of each of the predetermined number of time points (C0, C -1 , C -2 , C -3 ) of the predetermined time interval from the present predetermined time interval as the judgment time point is made. That is, in the judgment process, the weather data for judgment 121 is made for the arbitrary weather parameter using the weather data of the predetermined number of time points of the predetermined time interval to make them a group. This data is made in the entire area of the prediction area and stored in the storage device 22. On the other hand, in the learning process, a plurality of turbulence prediction model data 111 is made in advance to be a group of weather data of the weather parameter in which turbulence actually occurred in a certain region, that is, from the weather data of one past time point (P0) in which turbulence occurred and the weather data of the weather parameter of the time points (P -1 , P -2 , P -3 ) of the predetermined time interval from the past from the time point, and stored in the storage device 22. This data is made in all areas of the divided area. The predetermined number of past time points (P0, P -1 , P -2 , P -3 ... P -i ) of the predetermined time interval selected when making the turbulence prediction model data 111 and the predetermined number of present time points (C0, C -1 , C -2 , C -3 ... C -i ) selected when making the weather data for judgment 121 are selected as the same number of time points of the same time interval. In the example of Fig. FIG. 1C , i = 3, and all are 4 time points.

[0053] In the similarity calculation, in the same manner as the principle of the second turbulence prediction mode, the time points of the turbulence prediction mode data N that constitute each group are arranged in a predetermined configuration as one combined turbulence prediction mode data N' for each of the plurality of groups of turbulence prediction mode data 111, and similarly, the time points of the determination meteorological data O that constitute the determination meteorological data 121 are arranged in a predetermined configuration as one combined determination meteorological data O'. Then, the similarity is calculated by comparing the one combined turbulence prediction mode data N' with the one combined determination meteorological data O'. The similarity calculation of the first turbulence prediction mode principle and the second turbulence prediction mode principle is based on the calculation of whether there is a characteristic portion by comparing the meteorological data at one time point based on the weather forecast, whereas in the similarity calculation of the third turbulence prediction mode principle, the characteristic portion of the one combined determination meteorological data O' is determined by comparing the meteorological data at a predetermined number of time points having a predetermined time interval, in the temporal change toward the turbulence occurrence. That is, in the similarity calculation of the third turbulence prediction mode principle, since the combined image has a characteristic of a static image and a common characteristic of a dynamic image of the temporal change in the vicinity at the time of determination in the comparison of the static image of the one combined turbulence prediction mode data N' and the one combined determination meteorological data O', it is possible to determine whether there is a characteristic portion of the process of the turbulence occurrence even in the comparison of the static image. In the case where the one combined determination meteorological data O' has a characteristic portion of the process of the turbulence occurrence in common with the one combined turbulence prediction mode data N', the place is determined as a place where the turbulence is likely to occur in the prediction area.

[0054] (Principle of the fourth turbulence prediction mode)

[0055] Reference FIG. 1D, the principle of the 4th turbulence prediction mode is outlined. In the aforementioned principle of the 2nd turbulence prediction mode, in the similarity calculation, the number of meteorological parameters is increased to a plurality of meteorological parameters to increase the amount of information of the image data to be compared. However, in the principle of the 2nd turbulence prediction mode, there is no comparison of elements that change over time as in the principle of the 3rd turbulence prediction mode. Thus, the principle of the 4th turbulence prediction mode is a method in which the aforementioned principle of the 2nd turbulence prediction mode is added with elements in chronological order. The principle of the 4th turbulence prediction mode, like the principle of the 2nd turbulence prediction mode, selects a plurality of meteorological parameters as the meteorological parameters. First, like the principle of the 2nd turbulence prediction mode, a plurality of turbulence prediction mode data 111 is created in the learning process for the plurality of meteorological parameters. In the determination process, the determination-use meteorological data 121 is also created for each of the same plurality of meteorological parameters as the plurality of turbulence prediction mode data 111. In the principle of the 4th turbulence prediction mode, in each of the learning process and the determination process, further, like the principle of the 2nd turbulence prediction mode, the plurality of turbulence prediction mode data N for each of the plurality of meteorological parameters in the plurality of turbulence prediction mode data 111 is arranged in a predetermined arrangement with respect to each of the plurality of meteorological parameters as one set of turbulence prediction mode data N', and the determination-use meteorological data O with respect to each of the plurality of meteorological parameters is arranged in the same predetermined arrangement as when creating one set of turbulence prediction mode data N' as one set of determination-use meteorological data O'.

[0056] Further, in the learning process, from the point in time at which the turbulence prediction mode data 111 is created, the same processing is performed for a predetermined number of past points in time (P0, P -1 , P -2 , P -3 …) with a predetermined time interval, and one set of turbulence prediction mode data N' is created for the predetermined number of past points in time (P0, P -1 , P -2 , P -3 …) with a predetermined time interval. In the determination process, one set of determination-use meteorological data O' is created at the same number of current points in time (C0, C -1 , C -2 , C -3 …) with a time interval that is the same as the predetermined number with a predetermined time interval selected at the point in time at which the turbulence prediction mode data 111 is created. In each of the learning process and the determination process, one set of turbulence prediction mode data N' is created only for the case where turbulence occurs, one set of determination-use meteorological data O' is created for the entire region of the divided region, and each is stored in the storage device 22.

[0057] The similarity calculation is performed by comparing the data of each current point (CO, C -1 , C -2 , C -3 …) of the combination decision meteorological data O' with the data of each past point (PO, P -1 , P -2 , P -3 …) of the combination turbulence prediction pattern data N'. At this time, in the similarity calculation, it is performed in such a manner that, in addition to finding whether there is a common feature based on the comparison of the static images of each point of each data, it is found whether there is a common feature based on the comparison of the temporal changes of each point (CO, C -1 , C -2 , C -3 …) of the combination decision meteorological data O' with the temporal changes of each past point (PO, P -1 , P -2 , P -3 …) of the combination turbulence prediction pattern data N'. The similarity calculation can apply, for example, a recurrent neural network (RNN) or the like.

[0058] (5th Turbulence Prediction Pattern Principle)

[0059] Referring to FIG. 1E , the 5th turbulence prediction pattern principle is outlined. The 5th turbulence prediction pattern principle is a method of utilizing the comparison of temporal changes in the similarity calculation as well as the aforementioned 4th turbulence prediction pattern principle. In addition, the turbulence prediction pattern data 111 and the decision meteorological data 121 are made in the same manner as the 3rd turbulence prediction pattern principle. That is, in the learning process, a plurality of turbulence prediction pattern data 111 are made in advance to be a group of meteorological data of meteorological parameters of the weather in which turbulence actually occurred in a certain region, that is, a group consisting of meteorological data of a past point (PO) in which turbulence occurred and meteorological parameters of points (P -1 , P -2 , P -3 ... P -i ) that trace back to the past at predetermined time intervals from the point, and are stored in the storage device 22. This data is made in all regions of the divided region. The decision meteorological data 121 is made from the meteorological data of each point of a predetermined number of points (CO, C -1 , C -2 , C -3 ) of the current predetermined time intervals as the decision points, so that they constitute a group. This data is made in the entire region of the prediction region and stored in the storage device 22. In FIG. 1EIn the example of i = 3, all of 4 time points are shown. In the 3rd turbulence prediction mode principle, a combined turbulence prediction mode data N' is made by combining the turbulence prediction mode data N made from the meteorological data of a predetermined number of past time points (P0, P -1 , P -2 , P -3 ) with a predetermined time interval, but in the 5th turbulence prediction mode principle, the turbulence prediction mode data N is not combined. Also, the determination meteorological data O made from the meteorological data of one time point of a predetermined number of time points (C0, C -1 , C -2 , C -3 ) with a predetermined time interval is not combined. The similarity determination is made by comparing the data of each current time point (C0, C -1 , C -2 , C -3 …) of the determination meteorological data O with the data of each past time point (P0, P -1 , P -2 , P -3 …) of the turbulence prediction mode data N, as in the 4th turbulence prediction mode principle. At this time, in the similarity calculation, in addition to finding whether there is a common feature portion based on the comparison of the static images of each time point of each data, whether there is a common feature portion is found by comparing the temporal changes of each time point (C0, C -1 , C -2 , C -3 …) of the determination meteorological data O with the temporal changes of each past time point (P0, P -1 , P -2 , P -3 …) of the turbulence prediction mode data N. As in the 4th embodiment, the similarity calculation can apply, for example, a recurrent neural network (RNN) or the like.

[0060] (Configuration of Turbulence Prediction System)

[0061] The above-mentioned 1st to 5th turbulence prediction mode principles can be implemented in the following embodiments. Hereinafter, the embodiments of the present application will be described in detail. For the embodiments of the present application, reference is made to FIGS. 2A-2D , and first, the overall configuration of the turbulence prediction system 1 will be described. FIG. 2A is a view showing a representative configuration of the turbulence prediction system 1. FIG. 2B is a view showing a representative configuration of the turbulence prediction device 2. FIG. 2C A representative example in which the turbulence prediction system 1 is configured as a cloud computing system is shown. FIG. 2DThe data division of the predicted turbulence in the turbulence prediction system 1 is shown. The turbulence prediction system 1 has an arithmetic unit that performs arithmetic processing and a storage unit as essential structures. For example, the turbulence prediction system 1 can be configured to have one or more turbulence prediction devices 2 that have a central processing device 21 as the arithmetic unit and a storage device 22 as the storage unit.

[0062] The turbulence prediction system 1 can further have a terminal 3, a transceiver 4, an aircraft 5, and a network 6 as needed. The central processing device 21 of the turbulence prediction device 2 as the arithmetic unit performs arithmetic processing and other instructions, and the storage device 22 as the storage unit stores data required for turbulence prediction. The turbulence prediction device 2 can be configured to have an output device 23 such as a display or a printer and a communication device 24. The communication device 24 is connected to the transceiver 4 such as an antenna and the network 6. The terminal 3 can be configured as part of the turbulence prediction device 2, and the terminal 3 and the transceiver 4 can be configured as part of the turbulence prediction device 2. The turbulence prediction device 2 is controlled by the terminal 3. The turbulence prediction device 2 can be configured as a personal computer, and on the other hand, the turbulence prediction device 2 can be configured as a host computer connected to the terminal 3 as a client. The output device 23 of the turbulence prediction device 2 can be configured as the terminal 3. The storage device 22, the output device 23, and the communication device 24 of the turbulence prediction device 2 can be configured in any form as internal or external devices. The output of the turbulence prediction device 2 is the turbulence occurrence probability calculated by the central processing device 21 of the turbulence prediction device 2, which will be described later.

[0063] The turbulence prediction device 2 of the turbulence prediction system 1 can be configured to have turbulence prediction devices 2a and 2b as a plurality of turbulence prediction devices. The turbulence prediction devices 2a and 2b are each the turbulence prediction device 2, and are configured as described above. The turbulence prediction devices 2a and 2b are host computers, and are connected to terminals 3a and 3b. The turbulence prediction devices 2a and 2b can be connected to the network 6. A terminal 3c is connected to the network 6 as a client, and is connected to either of the turbulence prediction devices 2a and 2b of the host computers via the network 6. However, the terminal 3c can not be a client of the turbulence prediction devices 2a and 2b, but can be configured as a turbulence prediction device 2c. The turbulence prediction devices 2a and 2b can each be connected to transceivers 4a and 4b. The aircraft 5 can communicate with the transceivers 4a and 4b. For example, the output of the turbulence prediction device 2a can be output by the terminal 3a as the output device 23 near the turbulence prediction device 2a, and of course, can be output remotely by the terminal 3c or in the aircraft 5 via the transceiver 4a.

[0064] The turbulence prediction system 1 can be configured as a cloud computing system. In this case, as shown in FIG. 1, the turbulence prediction system 1 can be configured to have a plurality of turbulence prediction devices 2a and 2b as a plurality of turbulence prediction devices, and a network 6. The turbulence prediction devices 2a and 2b are each the turbulence prediction device 2, and are configured as described above. The turbulence prediction devices 2a and 2b are host computers, and are connected to the network 6. A terminal 3c is connected to the network 6 as a client, and is connected to either of the turbulence prediction devices 2a and 2b of the host computers via the network 6. However, the terminal 3c can not be a client of the turbulence prediction devices 2a and 2b, but can be configured as a turbulence prediction device 2c. The turbulence prediction devices 2a and 2b can each be connected to transceivers 4a and 4b. The aircraft 5 can communicate with the transceivers 4a and 4b. For example, the output of the turbulence prediction device 2a can be output by the terminal 3a as the output device 23 near the turbulence prediction device 2a, and of course, can be output remotely by the terminal 3c or in the aircraft 5 via the transceiver 4a. FIG. 2CAs shown, any configuration is possible as long as the storage and processing units are not integrated into a single turbulence prediction device 2, but rather defined within the cloud 7. That is, it is not necessary to construct a central processing unit 21 (processing unit) as the processing unit and a storage unit 22 as the storage unit within the turbulence prediction device 2. Typically, the turbulence prediction system 1 is, for example, a cloud computing system connected to the cloud 7, which is connected to the processing unit and storage unit via a network. Furthermore, the turbulence prediction system 1 typically includes a terminal 3 connected to the cloud 7, and the output of the turbulence prediction system 1 is sent to the terminal 3. In the case of a cloud computing system, it is also possible to configure a transceiver 4 capable of communicating with the aircraft 5 to connect to the cloud 7 and provide the output of the turbulence prediction system 1 to the aircraft 5.

[0065] (Predicted area, wide-area region, and segmented divisions)

[0066] Here, refer to FIG. 2D This paper explains the concepts of prediction region X, wide-area region Y, and segmentation zone Z when digitizing meteorological data within the prediction area for turbulence occurrence in Turbulence Prediction System 1. FIG. 2D The apparatus shown illustrates turbulence prediction in region X. Region X can be selected as any region where meteorological data regarding any meteorological parameter, such as wind speed, and data on the presence or absence of turbulence at that same time point can be obtained. As long as meteorological data regarding any meteorological parameter and data on the presence or absence of turbulence at that same time point can be obtained, any region in the world can be freely selected as prediction region X. For example, in... FIG. 2DThe Japanese archipelago is assumed to be a prediction area X. In the turbulence prediction system 1, the prediction area X is equally divided into a plurality of wide-area regions Y. One wide-area region Y can be equally divided in any size desired to be suitable for data analysis of the prediction area X. For example, the wide-area region Y can be demarcated as a square with a side length of 320 km. However, the size is not restricted and can be freely selected. Further, each of the plurality of wide-area regions Y is equally divided into a plurality of partitioned areas Z. The size of one partitioned area Z can be equally divided in any size desired to be suitable for data analysis of the wide-area region Y. When similarity determination is performed using the image data described in the aforementioned turbulence prediction model principle, the partitioned area Z coincides with one pixel in the image when the prediction area X or the plurality of wide-area regions Y is captured as the image. That is, the value assigned to one partitioned area Z coincides with the value of the partitioned area that is a pixel constituting the prediction area X or the plurality of wide-area regions Y. For example, the wide-area region Y can be set as a square with a side length of 5 km, and equally divided in a manner that one side is divided into 64 parts. However, the size is not restricted and can be freely selected. It is preferable to select the number of partitioned areas and the position of each partitioned area in a manner that makes it easy to value each of the meteorological data. In other words, the entire area of the prediction area X will be divided into a plurality of partitioned areas Z. Further, a comparison region Y' having an area of the same size as the wide-area region Y can be freely selected within the range of the prediction area X as the partitioned area Z has in the wide-area region Y. Thus, the wide-area region Y and the comparison region Y' can be compared. At this time, comparison of the wide-area region Y and the comparison region Y' means comparison between the partitioned areas Z each having. The wide-area region Y is a region demarcated ad hoc in order to compare the comparison region Y' with the entire area of the prediction area X without omission, and essentially consists in comparing the partitioned areas Z constituting the comparison region Y' with all the partitioned areas Z constituting the prediction area X. Data obtained by discretizing and valuing an arbitrary meteorological parameter (e.g., wind speed, etc.) is assigned to the plurality of partitioned areas Z. Hereinafter, in the meteorological parameter discrete data, the turbulence data, the turbulence prediction model data, and the determination meteorological data, the plurality of wide-area regions Y obtained by dividing the prediction area X, the wide-area region Y, the comparison region Y', and the partitioned areas Z divided therefrom are used as a concept of a demarcated data region. As described later, in the turbulence prediction system 1, there are a learning process of making teacher data and a determination process for predicting occurrence of turbulence, but the partitioned areas Z must be common in the learning process and the determination process. In the learning process and the determination process, the wide-area region Y and the comparison region Y' need to be selected to have a predetermined size in common, and each need to be set to have partitioned areas Z divided in a common rule.

[0067] (Time of data used by turbulence prediction system)

[0068] Next, refer to FIG. 3 The function and method of predicting the occurrence of turbulence in turbulence prediction system 1 are explained. FIG. 3 This is a diagram illustrating the concept of predicting the occurrence of turbulence. First, refer to... FIG. 3 The concept of time used to predict the occurrence of turbulence in turbulence prediction system 1 is explained. The current time is denoted as C0, any future time later than C0 is denoted as F0, and any past time earlier than C0 is denoted as P0. Furthermore, P... -1 This represents consecutive points in time that are a predetermined time unit earlier than P0. -2 …P -(n-1) P -n This indicates that each time interval is selected as being associated with P0 and P. -1 The time intervals between them are the same, and the ratio of the predetermined time units is P. -1 The earliest discrete time. The predetermined time can be arbitrarily determined as 10 minutes, 30 minutes, 1 hour, etc., as an effective time interval for predicting turbulence occurrence. The predetermined number of n consecutive time points means the meteorological data at any given time (hereinafter, "t = 0"), the meteorological data before any given time point (predetermined time unit × 1) (hereinafter, "t = -1"), the meteorological data before any given time point (predetermined time unit × 2) (hereinafter, "t = -2"), the meteorological data before any given time point (predetermined time unit × 3) (hereinafter, "t = -3"), ..., the meteorological data before any given time point (predetermined time unit × n) (hereinafter, "t = -n"). For example, if the time unit of the predetermined time is determined to be 1 hour, it means the discrete time of any given moment (t = 0), 1 hour ago (t = -1), 2 hours ago (t = -2), and 3 hours ago (t = -3). Additionally, if the predetermined number of time points is set to 4 time points, then as in the current time point (C0, C... -1 C -2 C -3 ) and past time points (P0, P -1 P -2 P -3 This is how it's chosen. The predetermined time interval can be the same as the sampling time of the acquired meteorological data, or it can be a longer time than the sampling time of the acquired meteorological data. Similarly, C -1 It also represents a discrete time in the past that is a predetermined time unit earlier than C0. This predetermined time unit is determined according to P. -1 The time interval between P0 and P0 is selected using the same predetermined time unit. -2 …C -(n-1) C -n This indicates that the selected time intervals are associated with C0 and C. -1a ratio C of predetermined time units that are the same time interval between -1 Early past time.

[0069] (Turbulence prediction process)

[0070] Next, the concept of the process of turbulence prediction in the turbulence prediction system 1 will be described. FIG. 3 As described above, the turbulence prediction system 1 is provided with an arithmetic unit and a storage unit. The arithmetic unit executes the processing in the learning process 11 and the processing in the determination process 12. In the storage unit, there is stored a turbulence prediction pattern data group 111 formed of a plurality of turbulence prediction pattern data N. The turbulence prediction pattern data N is teacher data made in the learning process 11 and is used for judging the occurrence of turbulence in the determination process 12. The turbulence prediction pattern data N corresponds to the aforementioned contrast region Y'.

[0071] (Learning process)

[0072] First, in the learning process 11, the processing of the making process of the turbulence prediction pattern data group 111 and the arithmetic unit will be described. Unlike the judgment of the occurrence of turbulence, as a preparation for the judgment of the occurrence of turbulence, the turbulence prediction pattern data N constituting the turbulence prediction pattern data group 111 is made in advance using data at an arbitrary time in the past. Through the calculation of the meteorological parameter discrete data L of the meteorological parameter discrete data group 112, each turbulence prediction pattern data N of the turbulence prediction pattern data group 111 is made as follows. First, data of an arbitrary meteorological parameter is selected to make the turbulence prediction pattern data group 111 and the meteorological parameter discrete data group 112. The arbitrary meteorological parameter is, for example, wind speed, wind shear, air temperature, air temperature gradient, air density, air density gradient, cloud image grasped from infrared radiation brightness temperature, visible image, water vapor image, relative humidity, and the like. These arbitrary meteorological parameters can be selected singly or in combination. With regard to one or more arbitrary meteorological parameters selected, meteorological data at an arbitrary past time P0, P -1 , P -2 ... P -(n-1) , P -n is obtained as meteorological data. For example, in the case where wind speed is selected as the meteorological parameter, the values of the wind speed at the past P0, P -1 , P -2 ... P -(n-1) , P -n are obtained as meteorological data, and in the case where wind speed and wind shear are selected as the meteorological parameters, the values of the wind speed and the wind shear at the past P0, P -1 , P -2 ... P -(n-1) , P -nThe value of the wind speed and the value of the wind shear at the time are taken as the weather data. In order to obtain as much weather data as possible, the past time is selected to be as long as possible. The weather data is not particularly limited as long as it is weather data on weather parameters that is highly reliable, such as data obtained from a weather bureau.

[0073] The values of these weather data are demarcated and obtained as follows. First, a prediction area is selected. The selected prediction area is divided into a plurality of wide areas, and for each wide area, it is divided into a divisional area that constitutes the wide area. The prediction area can also be directly divided into a divisional area. These divisions are as described above with reference to FIG. 1. The prediction area and the wide area can be freely selected depending on the weather data obtained and the region in which it is desired to predict turbulence. For example, a region having a certain degree of width that becomes a large category, such as the entire Japan, a part of Japan, and the like, can be arbitrarily selected as the prediction area. Furthermore, a region that constitutes the prediction area that becomes a middle category can be selected as the wide area. In the case of the example of FIG. 1, the entire Japan is selected as the wide area. The wide area is divided into a plurality of divisional areas. Thus, for the wide area that is divided into a plurality of divisional areas, weather data at arbitrary past discrete points P0, P FIG. 2D FIG. 3 -1 -2 -(n-1) -n -1 -2 -(n-1) -n

[0074] ​​​​​​​​​​Unlike the creation of the discrete meteorological parameter data set 112, a turbulence data set 113 is obtained for a wide area divided into multiple partitions, containing turbulence data M with and without turbulence. The wide area in the turbulence data set 113 is the same as the wide area selected when creating the discrete meteorological parameter data set 112, or is selected as a wide area containing a common area. Furthermore, the multiple partitions divided from the wide area when creating the discrete meteorological parameter data set 112 are selected as the same partitions used when creating the discrete meteorological parameter data set 112. Moreover, the turbulence data set 113, containing the turbulence data M, is for any past discrete time point P0, P10, P20, P30 selected when creating the discrete meteorological parameter data set 112. -1 P -2 …P -(n-1) P -n The same arbitrary past discrete time points P0, P -1 P -2 …P -(n-1) P -n Obtained. For any discrete time points P0 and P1 in the past. -1 P -2 …P -(n-1) P -n Turbulence data M refers to data indicating the presence or absence of turbulence in various wide-area regions divided into multiple partitions. Turbulence data M is past observational data from aircraft, including data on the occurrence or absence of turbulence, such as data provided by meteorological operational support centers.

[0075] Furthermore, among the various turbulence data M constituting the turbulence data group 113, if, at the same time point as any past discrete time point, there is a segmented region where turbulence occurred among multiple segmented regions divided from a wide geographical area, the segmented region where turbulence occurred and the segmented regions adjacent to and surrounding the segmented region where turbulence occurred are designated as the turbulence occurrence region data region. Alternatively, the value "1" can be assigned to the segmented region where turbulence occurred as a value indicating that turbulence occurred, and "0" can be assigned to other segmented regions where turbulence did not occur as a value indicating that turbulence did not occur. The segmented region with "1" and the segmented regions surrounding it with "0" are combined to designate the turbulence occurrence region data region. Moreover, the discrete meteorological parameter data L of the discrete meteorological parameter data group 112 corresponding to each segmented region in the turbulence occurrence region data region is extracted and designated as turbulence prediction model data N. For example, FIG. 3 For example, in P -nIn the turbulence data group 113 in the wide area at the time, in the case where a turbulence-occurring divided region exists in the southeast of the island of Sado, the divided region is assigned "1" as a turbulence-occurring divided region. The divided regions other than this are assigned "0".

[0076] Further, the divided regions belonging to the region centered on the divided region assigned "1" are assigned as turbulence-occurring region data regions. The turbulence-occurring region data regions correspond to the contrast regions Y' in the explanation of the above-mentioned reference FIG. 2D . That is, the turbulence-occurring region data regions are centered on the regions where turbulence occurs, and become contrast regions Y' selected as being centered on the turbulence-occurring divided region and having divided regions where turbulence does not occur around them. The turbulence-occurring region data regions become the size of the turbulence prediction pattern data N which is the teacher data as explained below. In the turbulence prediction system 1, as described later, the comparison judgment meteorological data O and the teacher data turbulence prediction pattern data N are compared, and therefore, the turbulence-occurring region data regions as the contrast regions Y' and the turbulence prediction pattern data N are selected so as to have a predetermined size as the size common to the comparison judgment meteorological data O described later.

[0077] Further, the portion of the divided regions corresponding to the turbulence-occurring region data regions is cut out from the meteorological parameter discrete data group 112 at the times P -n , and is assigned as the turbulence prediction pattern data N. The turbulence prediction pattern data N is also obtained and assigned at other past discrete times P -1 , P -2 ... P -(n-1) , P -n . The data group of the turbulence prediction pattern data N is assigned as the turbulence prediction pattern data group 111, and is stored in the storage section of the turbulence prediction system 1. The turbulence prediction pattern data group 111 is preferably updated and stored in the storage section for a predetermined period. The obtained turbulence prediction pattern data group 111 can be said to be past data on the behavior of an arbitrary meteorological parameter in the vicinity of a place where turbulence occurs. That is, the turbulence prediction pattern data group 111 can be said to be a characteristic pattern showing the behavior of a meteorological parameter having a high possibility of turbulence occurrence. By setting the periods P0, P -1 , P -2 ... P -(n-1) , P -n to long periods, it is possible to prepare a turbulence prediction pattern data group 111 from a large number of turbulence prediction pattern data N in a long past period, and to improve the accuracy of the prediction of turbulence occurrence.

[0078] (Judgment process)

[0079] After the explanation of the learning process 11, the determination process 12 is explained. In the determination process 12, the determination weather data O which is a region of the same size as the predetermined size selected when the turbulence prediction pattern data N is made in the learning process 11 is used. As described above, the determination weather data O has the division area divided when the turbulence prediction pattern data N is made. In addition, the weather parameter must be selected as the same weather parameter as the weather parameter selected when the turbulence prediction pattern data N is made in the learning process 11. That is, in the case where the wind speed and the wind shear are selected as the weather parameter as the weather parameter discrete data L in the learning process 11, the wind speed and the wind shear must also be selected as the weather parameter in the determination process 12. The wide area region in the determination process 12 is generally preferably set as the same wide area region as that used in the learning process 11. The determination weather data O is generally selected as a region having the same predetermined size as the turbulence prediction pattern data N constituting the turbulence prediction pattern data group 111. That is, with respect to the determination weather data O, the determination weather data O is selected as a region having the same predetermined size as the turbulence prediction pattern data N constituting the turbulence prediction pattern data group 111. In the determination process 12, the determination weather data O is calculated and used. With respect to the determination weather data O, a large number of determination weather data O is made to correspond to the above-mentioned reference area Y' with respect to the turbulence prediction pattern data N constituting the turbulence prediction pattern data group 111. That is, the determination weather data O is calculated and used in the determination process 12. In the determination process 12, the determination weather data O is calculated and used. With respect to the determination weather data O, a large number of determination weather data O is made to correspond to the above-mentioned reference area Y' with respect to the turbulence prediction pattern data N constituting the turbulence prediction pattern data group 111. That is, the determination weather data O is calculated and used in the determination process 12. FIG. 2D With respect to the aforementioned example explained, the determination weather data O corresponds to the wide area region Y and the turbulence prediction pattern data N corresponds to the reference area Y'. Even if the determination weather data O and the turbulence prediction pattern data N are different in size, the determination weather data O at least needs to be a region having a larger region than the turbulence prediction pattern data N constituting the turbulence prediction pattern data group 111.

[0080] In the determination process 12, the determination weather data O is calculated and used. With respect to the determination weather data O, a large number of determination weather data O is made to correspond to the above-mentioned reference area Y' with respect to the turbulence prediction pattern data N constituting the turbulence prediction pattern data group 111. That is, the determination weather data O is calculated and used in the determination process 12. FIG. 2DThe wide-area region of the entire region of the prediction region X illustrated is a wide-area region, and is configured as the determination-use meteorological data group 121. In one example, the determination-use meteorological data O is, for example, meteorological forecast data at an arbitrary future time point F0 later than the current time point C0, which is predicted at the current time point C0. The meteorological data of the wide-area region relating to the meteorological parameter is obtained from the predicted meteorological forecast data at the arbitrary future time point F0, the meteorological parameter is numerically converted and allocated in each of the divided regions configuring the wide-area region, and the determination-use meteorological data O is produced in the same manner as the production method of the meteorological parameter discrete data L. Since the determination-use meteorological data O produced from the meteorological forecast data at the arbitrary future time point F0 relating to the meteorological parameter is a meteorological forecast of the wide-area region, as the determination-use meteorological data O, the determination of the future turbulent flow prediction can be directly performed. Further, in order to improve the determination accuracy of the turbulent flow prediction, a plurality of meteorological parameters can be used. In the case where a plurality of meteorological parameters is used, for each of the plurality of meteorological parameters, the determination-use meteorological data O is produced in the same manner as the production of the determination-use meteorological data O explained so far in the example of one meteorological parameter, and is combined in a predetermined arrangement into one as the combined determination-use meteorological data O'. The combined determination-use meteorological data O' is also a plurality of meteorological forecasts of the wide-area region produced from the meteorological forecast data at the arbitrary future time point F0 relating to the plurality of meteorological parameters. For each of the plurality of meteorological parameters, the determination-use meteorological data O is produced in the same manner as the production of the determination-use meteorological data O explained so far in the example of one meteorological parameter, and is combined in a predetermined arrangement into one as the combined determination-use meteorological data O'. In the case where a plurality of meteorological parameters is used to produce the combined determination-use meteorological data O', in the production of the turbulent flow prediction pattern data N, the same plurality of meteorological parameters as those used when the combined determination-use meteorological data O' is produced is also required to be selected, for each of the plurality of meteorological parameters, the turbulent flow prediction pattern data N is produced, and they are combined in a predetermined arrangement into one in the same manner as in the case of the production of the combined determination-use meteorological data O', thereby producing the combined turbulent flow prediction pattern data N as the teacher data.

[0081] In another example, the determination-use meteorological data O is obtained from, for example, meteorological data at the current time point C0 and the most recent consecutive time points C -1 , C -2 …C -(n-1) , C -n , sampled at a predetermined discrete interval from the current time point C0 and past from the time point. The number n of the most recent consecutive time points can be arbitrarily selected. In addition, C0, C -1 , C -2 …C -(n-1) , C -nThe discrete intervals of each of the points are kept at a time interval of a predetermined time unit. The predetermined time unit is the same as the time unit selected when the weather parameter discrete data L is created. For example, if four points including the present are used, it means that the weather data at points C0, C -1 , C -2 , C -3 is used to obtain the determination weather data O. In each of the weather data at points C0, C -1 , C -2 , C -3 , each of the weather data at each of the points is obtained in the same way as the weather parameter discrete data L is created. The weather data of the wide area related to the weather parameter is obtained, and the weather parameter is numerically processed and allocated in each of the divided areas that constitute the wide area to create the determination weather data O. Further, the present point C0 does not mean the present in the strict sense, but means the latest point at which the weather data required to create the determination weather data O for determination can be used. In the case where the determination weather data O is created using the most recent consecutive points C0, C -1 , C -2 ... C -(n-1) , C -n , for example, in the case where the determination weather data O is created using each of the weather data at points C0, C -1 , C -2 , C -3 , a plurality of determination weather data O will be created. Therefore, in this case, the determination weather data O' is set as one combination in which a predetermined configuration is combined as one. This is the same as in the case where one combination determination weather data O' is created from a plurality of weather parameters. In the case where the combination determination weather data O' is created from the weather data at a plurality of points, in the creation of the turbulence prediction pattern data N, the turbulence prediction pattern data N is also created from the weather data at a plurality of consecutive points in the same way as in the creation of the combination determination weather data O', and they are combined as one in the same predetermined configuration as in the case where the combination determination weather data O' is created, so that the combination turbulence prediction pattern data N as the teacher data is created.

[0082] The operation section calculates the degree of similarity between the determination weather data O and the turbulence prediction pattern data N included in the turbulence prediction pattern data set 111 in the determination process 12. In the case where the determination weather data O is made based on the weather forecast data for an arbitrary future time point F0 later than C0, the degree of similarity between the determination weather data O and the plurality of turbulence prediction pattern data N included in the turbulence prediction pattern data set 111 means the possibility of the weather pattern in which turbulence occurs existing at the future time point F0. In the case where the determination weather data O is made based on the weather data for the most recent consecutive time point with C0 as a reference, the degree of similarity between the determination weather data O and the plurality of turbulence prediction pattern data N included in the turbulence prediction pattern data set 111 means the possibility of the weather pattern in which turbulence occurs existing and developing up to the current time point C0. That is, the degree of similarity between the determination weather data O and the turbulence prediction pattern data N becomes the degree of the possibility of turbulence occurring. Thus, the degree of the possibility of turbulence occurring can be calculated from the degree of similarity between the determination weather data O and the turbulence prediction pattern data N.

[0083] The calculation of the similarity of the determination weather data O and the turbulence forecast pattern data N is performed by comparing all the turbulence forecast pattern data N included in the turbulence forecast pattern data group 111 with the determination weather data O for the entire region of the forecast region. Specifically, for each of the plurality of determination weather data O corresponding to the entire region of the forecast region, the divided areas constituting it are compared with the divided areas of all the turbulence forecast pattern data N included in the turbulence forecast pattern data group 111. In this comparison, the divided areas constituting the determination weather data O and the divided areas constituting the turbulence forecast pattern data N are compared while being shifted one by one. The comparison at this time is performed in such a manner that the determination weather data O and the turbulence forecast pattern data N are each captured as image data, and for each image, the values of the divided areas coinciding with the pixels of the image are compared. In the case where there is a divided area similar to the turbulence forecast pattern data N in a certain determination weather data O of the determination weather data group 121, it means that the turbulence is likely to occur in the region corresponding to the divided area. Further, in the case where the divided areas around a certain region are similar to a plurality of turbulence forecast pattern data N in the turbulence forecast pattern data group 111, it means that the turbulence is particularly likely to occur. Here, the similarity simply means the similarity of the values of the weather data assigned to each divided area of the turbulence forecast pattern data N and the distribution of the values in the divided area, and the values of the weather data assigned to each divided area of the determination weather data O and the distribution of the values in the divided area. The regional attribute of the original weather data of the turbulence forecast pattern data N is not taken into account. That is, a certain turbulence forecast pattern data N is data produced with a certain region in Hokkaido as the center of the region where the turbulence occurred, and even the determination weather data O of Kyushu, as long as the similarity is recognized in the divided areas constituting each data, it can be predicted that the turbulence will occur in the region of the determination weather data O of Kyushu where the similarity is high. Further, the turbulence forecast pattern data N included in the turbulence forecast pattern data group 111 can be classified into a certain degree of categories according to the arrangement and tendency of the values of the weather parameters of the divided areas thereof. In the categories, there are categories including a large number of turbulence forecast pattern data N and categories including a small number of turbulence forecast pattern data N. At this time, when the similarity calculation of the determination weather data O and the turbulence forecast pattern data N is performed, the similarity can also be calculated in such a manner that a large weight is given to the frequency in the case where the category including a large number of turbulence forecast pattern data N is included in the determination weather data O, and a small weight is given to the frequency in the case where the category including a small number of turbulence forecast pattern data N is included in the determination weather data O. The similarity corresponds to the region where the turbulence occurs and the degree of the possibility of the turbulence. The calculation of the similarity can be performed in various methods.

[0084] In addition, the calculation of the similarity of the determination weather data O and the turbulence prediction pattern data N can also be performed by a CNN. As the CNN, various methods of the CNN other than the method of the general CNN can also be adopted. In the CNN, for example, the turbulence prediction pattern data N included in the turbulence prediction pattern data group can be executed by the CNN as a convolution layer. In the case of the deep learning based on the CNN, for the determination, the process thereof is not particularly recognized, but the feature part of the similar image is output. Thereby, for all the divided areas of the determination weather data O corresponding to all the areas of the prediction area as described above, and the divided areas constituting the plurality of turbulence prediction pattern data N in the turbulence prediction pattern data group 111, the comparison explained so far can be executed as a black box. The output of the CNN becomes the determination weather data O similar to the turbulence prediction pattern data N. In the output, there are a case where a large number of turbulence prediction pattern data N is included and a case where a small number of turbulence prediction pattern data N is included. The case where a large number of turbulence prediction pattern data N is included means that the degree of the possibility of the occurrence of the turbulence is high.

[0085]

EMBODIMENT

[0086] Next, the turbulence prediction system 1 as an embodiment to which the embodiment explained so far is concretely applied will be explained. First, the turbulence prediction processing of the turbulence prediction system 1 of Embodiment 1 will be explained.

[0087] (Embodiment 1)

[0088] Reference FIGS. 4A-4E The turbulence prediction processing of Embodiment 1 based on the turbulence prediction system 1 of the aforementioned embodiment will be explained. Embodiment 1 corresponds to the 1st turbulence prediction pattern principle of the embodiment. The turbulence prediction system 1 of Embodiment 1 is executed by the arithmetic unit. The turbulence prediction processing executed by the arithmetic unit includes a learning process and a determination process. FIG. 4A A flowchart showing the learning process of the turbulence prediction processing of Embodiment 1, FIG. 4E A flowchart showing the determination process of the turbulence prediction processing of Embodiment 1. FIGS. 4B-4D is a graph showing the determination weather data O and the turbulence prediction pattern data N in the case where the determination weather data O is similar to the turbulence prediction pattern data N. FIG. 3corresponding to the learning process 11 of the present embodiment, i.e., the concept of the process of creating the turbulence prediction pattern data. Hereinafter, the turbulence prediction processing performed by the central processing device 21 of the turbulence prediction device 2 as an example of the operation section of the turbulence prediction system will be described. In the learning process of the turbulence prediction device 2 of the present embodiment, past meteorological data and observation data actually observed by an aircraft including the presence or absence of past turbulence are used. The past observation data actually observed by the aircraft includes data of the occurrence site of the turbulence. In the determination process, meteorological forecast data is used. As the meteorological forecast data, for example, meteorological forecast data including arbitrary meteorological parameters provided by a group such as a meteorological agency is used, and is meteorological data that predicts the future weather. As the arbitrary meteorological parameters, various kinds of meteorological parameters can be used. In the present specification, each of the various kinds of meteorological parameters is denoted as a meteorological parameter α, β, γ,.... The main examples of each of the meteorological parameters α, β, γ,.... are as follows. These are merely main examples, and other than these, arbitrary meteorological parameters can be selected.

[0089] (1) Meteorological parameter α = absolute value of vertical wind shear

[0090] (2) Meteorological parameter β = vertical temperature gradient

[0091] (3) Meteorological parameter γ = vertical air density gradient

[0092] (4) Meteorological parameter δ = satellite photograph image (infrared radiation brightness temperature, visual image, water vapor image, etc.)

[0093] Reference FIG. 4A The creation of the turbulence prediction pattern data N as the teacher data in the learning process of the turbulence prediction processing of the present embodiment will be described. In the turbulence prediction processing, a prediction region is determined as a region in which turbulence occurrence prediction is desired. Based on the reference FIG. 2DAs explained above, the prediction area is divided into a wide area and a segmented region. The definitions of the prediction area, the wide area, and the segmented region follow the definitions explained previously. The central processing unit 21 of the turbulence prediction device 2 performs the following learning process. First, a meteorological parameter is selected from meteorological parameters α, β, γ, and δ. Meteorological parameters are factors that have certain effects on turbulence, and parameters known to be effective in turbulence prediction are selected. It is known that the above-mentioned meteorological parameters α, β, γ, and δ are effective in turbulence prediction as meteorological parameters, so any one of them can be used. The selected meteorological parameter is used as the evaluation meteorological parameter. Hereinafter, the explanation will focus on the meteorological parameter α as the evaluation meteorological parameter. Furthermore, past meteorological data K (S101) regarding the evaluation meteorological parameter is obtained within any wide area. "Past" means a discrete time point P0 that is the reference point before the current time point C0. Regarding the current time point C0, the past time point P0, and the time point P past from P0. -1 P -2 …P -(n-1) P -n As described in the above embodiments. That is, in this process, discrete time points P0, P1, P2, P3, P4, P5, P6, P7, P8, P9, P1, P1, P2, P1, P2, P1, P2, P3, P4, P5, P1, P2, P1, P2, P3, P4, P5, P1, P2, P1, P2, P3, P4, P5, P1, P2, P1, P2, P1, P2, P2, P3, P4, P1, P2 ... -1 P -2 …P -(n-1) P -n And obtain P0, P for each past point in time. -1 P -2 …P -(n-1) P -n Meteorological data. The predetermined time interval, such as 10 minutes, 30 minutes, 1 hour, etc., can be arbitrarily determined as an effective time interval for predicting turbulence occurrence. This becomes discrete time points P0, P... -1 P -2 …P -(n-1) P -n The predetermined time interval can be the same as the sampling time of the obtained meteorological data, or it can be a longer time than the sampling time of the obtained meteorological data.

[0094] On one hand, the process of acquiring meteorological data K is performed (S101); on the other hand, the process of acquiring past time points P0, P10, P20, P30, P40, P50, P60, P70, P80, P90, P10, P110, P120, P130, P140, P150, P160, P170, P180, P190, P100, P1 -1 P -2 …P -(n-1) P -nTurbulence data (S102). Procedures S101 and S102 can be executed either first, simultaneously, or alternately. Regarding past meteorological data, as much past meteorological data as possible is preferred, provided the data is reliable, and progressively updated meteorological data should be used.

[0095] In terms of computational mathematical meaning for turbulence prediction processing, the predetermined wide-area region is an arbitrary rectangular region. Dividing this region into an arbitrary number of grid elements along the longitude (east-west) direction and an arbitrary number of grid elements along the latitude (north-south) direction results in each grid element corresponding to a segmentation zone. For example, regarding... FIGS. 4B-4D This is just one example. Any division can be defined as the i-th division in the east-west direction and the j-th division in the north-south direction, Lij (i = 1 to h, j = 1 to h). For example, h is 64. The aforementioned meteorological parameters α, β, γ, and δ in this division Lij are defined as αij, βij, γij, and δij (i = 1 to h, j = 1 to h).

[0096] For the past discrete time points P0 and P1 obtained in the previous process (S101), -1 P -2 …P -(n-1) P -n The following meteorological data, categorized by any wide geographical area, such as FIG. 4B As shown, a wide area is divided into segments of predetermined size Lij (i = 1 to h, j = 1 to h). Typically, this is divided into squares, but rectangles are also possible. The size of the wide area and the size of each segment are set such that the segments within the wide area are of the same size. Furthermore, based on the obtained past discrete time points P0, P... -1 P -2 …P -(n-1) P -nThe weather data is used to create a weather parameter discrete data set 112 made up of weather parameter discrete data L obtained by numerically converting and assigning the evaluation weather parameter in each of the divided sections (S103). In the case of the present embodiment, in each of the divided sections (each mesh element), the absolute value of the vertical wind shear is assigned as the evaluation weather parameter. With respect to each of the divided sections Lij (i = 1 ~ h, j = 1 ~ h) that make up the weather parameter discrete data L, if the wide area is considered as one image, each of the divided sections becomes a so-called calculation mesh element, and also corresponds to a pixel of the image. That is, by numerically converting and assigning the evaluation weather parameter to the divided sections Lij (i = 1 ~ h, j = 1 ~ h), each mesh element will be created as digital values and two-dimensional pseudo image data. Further, the pseudo image data means image data that is not actually taken, and is image data obtained by assigning numerical values. However, the pseudo image data is used as image data, and can also be used only as a collection of numerical data of the evaluation weather parameter of the divided sections Lij (i = 1 ~ h, j = 1 ~ h). As image data, for example, the numerical values of each of the mesh elements of the divided sections Lij (i = 1 ~ h, j = 1 ~ h) can be transformed into numerical data in the range of 0 to 1 by non-dimensionally normalizing the numerical values of each of the mesh elements of the divided sections Lij (i = 1 ~ h, j = 1 ~ h) with the maximum value of the evaluation weather parameter of the divided sections Lij (i = 1 ~ h, j = 1 ~ h), and the numerical values can be displayed as monochrome image data of black and white binary gradation. This operation is performed for all of the prediction areas, so that numerical values are assigned to all of the divided sections that make up the prediction areas.

[0097] The weather data is used to create a weather parameter discrete data set 112 made up of weather parameter discrete data L obtained by numerically converting and assigning the evaluation weather parameter in each of the divided sections (S103). In the case of the present embodiment, in each of the divided sections (each mesh element), the absolute value of the vertical wind shear is assigned as the evaluation weather parameter. With respect to each of the divided sections Lij (i = 1 ~ h, j = 1 ~ h) that make up the weather parameter discrete data L, if the wide area is considered as one image, each of the divided sections becomes a so-called calculation mesh element, and also corresponds to a pixel of the image. That is, by numerically converting and assigning the evaluation weather parameter to the divided sections Lij (i = 1 ~ h, j = 1 ~ h), each mesh element will be created as digital values and two-dimensional pseudo image data. Further, the pseudo image data means image data that is not actually taken, and is image data obtained by assigning numerical values. However, the pseudo image data is used as image data, and can also be used only as a collection of numerical data of the evaluation weather parameter of the divided sections Lij (i = 1 ~ h, j = 1 ~ h). As image data, for example, the numerical values of each of the mesh elements of the divided sections Lij (i = 1 ~ h, j = 1 ~ h) can be transformed into numerical data in the range of 0 to 1 by non-dimensionally normalizing the numerical values of each of the mesh elements of the divided sections Lij (i = 1 ~ h, j = 1 ~ h) with the maximum value of the evaluation weather parameter of the divided sections Lij (i = 1 ~ h, j = 1 ~ h), and the numerical values can be displayed as monochrome image data of black and white binary gradation. This operation is performed for all of the prediction areas, so that numerical values are assigned to all of the divided sections that make up the prediction areas. -1 -2 -(n-1) -n ​​​After the process (S102) of the turbulence data, the presence or absence of turbulence is numerically processed and assigned to the divided areas Lij (i = 1 ~ h, j = 1 ~ h) to produce a turbulence data group 113 made of a plurality of turbulence data M. For example, the turbulence data M is produced by assigning "1" to the divided areas where turbulence occurs and "0" to the divided areas where turbulence does not occur (S104). Further, the divided areas where "1" is assigned are centered to define a region that becomes a quadrangle of the same size as the wide-area region as a turbulence occurrence region data area (S105). Further, the divided areas corresponding to the turbulence occurrence region data area are cut out from the meteorological parameter discrete data L and defined as turbulence prediction pattern data N (S106). The turbulence prediction pattern data N is produced as such, that is, the turbulence prediction pattern data N is two-dimensional data indicating the behavior of the meteorological parameters at the point where turbulence occurs and the position and its surroundings. In the case where the wide-area region of the macro-classification is divided into a plurality of areas and the divided areas Lij are calculated for each wide-area region, the processes S101 to S106 are repeated for the entire area of the wide-area region of the macro-classification. The outputs in the processes S101 to S106 are stored in the storage section of the turbulence prediction system 1 as needed. In the case where the turbulence prediction system 1 is provided with the turbulence prediction device 2, the outputs are stored in the storage device 22.

[0098] Next, the processes from the process S101 of acquiring the meteorological data K to the process S106 of acquiring the turbulence prediction pattern data N explained so far will be specifically explained with reference to FIG. 4B and FIG. 4C . The meteorological data of the wide-area region at an arbitrary point P0 to P -n in the past is acquired. For the acquired meteorological data, the absolute value of the vertical wind shear as the meteorological parameter a is discretized and the discretized value is assigned to each of the divided areas Lij (i = 1 ~ h, j = 1 ~ h) divided as such. This is the meteorological parameter discrete data L as shown in FIG. 4B . For all the prediction areas, the value of the meteorological parameter a is assigned to the divided areas. On the other hand, the turbulence data of the prediction area at the same point as this is acquired. For the divided areas of the entire area of the prediction area, "1" is assigned to the place where turbulence occurs and "0" is assigned to the place where turbulence does not occur as shown in FIG. 4C , and the meteorological parameter discrete data L is produced. For example, in FIG. 4CIn the present embodiment, a value "1" is assigned to the divisional region in the Hokuriku region where the turbulence occurs, and a value "0" indicating that no turbulence occurs is assigned to the region other than the above. Further, the divisional region is selected so as to become a quadrangle of a predetermined size, and the region is set as the turbulence occurrence region data region. Further, the divisional region corresponding to the turbulence occurrence region data region is cut out from the meteorological parameter discrete data Lij (i = 1 ~ h, j = 1 ~ h). That is, the divisional region of a region of a predetermined size centered on the region where the turbulence occurs and to which the value "1" is assigned as shown in FIG. 4C is cut out from the meteorological parameter discrete data, and is defined as the turbulence prediction pattern data N as shown in FIG. 4D . That is, the meteorological parameter discrete data of the divisional region indicates the local behavior of the meteorological parameter in the region where the turbulence occurs and the periphery thereof, and means that the turbulence is highly likely to occur.

[0099] Next, the determination process 12 of the turbulence prediction processing of the embodiment 1 will be described with reference to FIG. 4E . The operation section (the central processing device 21 in the case where the turbulence prediction device 2 is provided) of the turbulence prediction system 1 acquires the meteorological forecast data of an arbitrary wide area (S111). The definition of the prediction region and the divisional region and the definition of the meteorological data are the same as described above. The meteorological forecast data is the meteorological data on the meteorological parameter a at the future time point F0 later than the current time point CO. The meteorological parameter must be common in the learning process 11 and the determination process 12. The relationship of the future time point F0 later than the current time point CO is as described in the description of the foregoing embodiment. Hereinafter, the same as in the learning process 11, the case where the turbulence prediction device 2 is provided will be described. The meteorological forecast data in an arbitrary wide area can be acquired from the communication device 24, for example, or can be acquired in the form of being stored in the storage device 22 by other means.

[0100] When the meteorological parameter discrete data group 112 is created for each wide area in all the wide areas of the prediction region from the meteorological forecast data acquired in the previous process (S111), the region of a quadrangle of the same size as the region of the meteorological parameter discrete data L is defined as the region of the determination-use meteorological data O and the divisional region constituting the region, and the evaluation meteorological parameter is numerically evaluated for each divisional region of the region (S112). Here, as described above, the meteorological parameter a is selected as the evaluation meteorological parameter in the learning process 11, and thus the meteorological parameter a, that is, the absolute value of the vertical wind shear in the description of the present embodiment is numerically evaluated. It is preferable that the acquisition of the meteorological forecast data is performed at each predetermined time set as necessary for the update of the current time point CO and the future time point F0. The determination-use meteorological data O is stored in the storage section of the turbulence prediction system 1, that is, the storage device 22 in the case where the turbulence prediction device 2 is provided, as necessary.

[0101] The operation section or central processing device 21 reads the determination weather data O, which is produced by numerically converting the weather parameter a in the previous step (S112), and the turbulence prediction pattern data N, which is produced in the learning process, from the storage section or storage device 22. Then, the read determination weather data O and turbulence prediction pattern data N are compared to determine the degree of similarity (S113). The operation section or central processing device 21 determines whether or not there is a divisional area of the determination weather data O that is similar to any of the turbulence prediction pattern data N included in the turbulence prediction pattern data group 111 in each divisional area of the determination weather data O. That is, in this comparison, since each divisional area of the determination weather data O will have a divisional area that is highly similar to the turbulence prediction pattern data N and a divisional area that is less similar to the turbulence prediction pattern data N, the degree of similarity is calculated as the degree of similarity and determined. As a result, in the divisional area with a high degree of similarity, it means that the possibility of turbulence occurrence is high, and in the divisional area with a low degree of similarity, it means that the possibility of turbulence occurrence is low. For example, in the case where the degree of similarity to any of the turbulence prediction pattern data N included in the turbulence prediction pattern data group 111 is higher than that of other divisional areas in a divisional area of a certain wide area, it is determined that the possibility of turbulence occurrence in that area is high. The degree of similarity of the determination weather data O to the turbulence prediction pattern data N can be selected in various ways in any divisional area of the determination weather data O.

[0102] For example, the determination weather data O can also be applied to the size of the correlation coefficient r of the numerical values of the weather parameters of the divisional areas corresponding to each other at that time, as the degree of similarity of the turbulence prediction pattern data N. That is, the average of all the divisional areas of the turbulence prediction pattern data N is calculated, and the deviation of the numerical values of the weather data in each divisional area of the turbulence prediction pattern data N from the average is calculated. On the other hand, the average of the divisional areas of the determination weather data O of the position corresponding to the turbulence prediction pattern data N is calculated, and the deviation of the numerical values of the weather data in each divisional area of the determination weather data O of the position corresponding to the turbulence prediction pattern data N from the average is calculated. Thus, the product of the two deviations can be calculated, and the correlation coefficient r calculated by dividing the product by the product of the square roots of the squares of the respective deviations is taken as the degree of similarity. In addition, it is not necessarily limited to this method. For example, the degree of the frequency of occurrence of the turbulence prediction pattern data N in the turbulence prediction pattern data group 111 included in the determination weather data O can be calculated as a score value. The definition and calculation method of the score value can be freely selected, and the calculation method is not limited. As long as the score value is defined as a ratio as the degree of the frequency of occurrence to produce a predetermined distribution, the degree of the frequency of occurrence of the turbulence prediction pattern data N in the turbulence prediction pattern data group 111 included in the determination weather data O can be taken as the degree of the possibility of turbulence occurrence (S114).

[0103] Further, in the aforementioned process of determining the degree of similarity (S114), a method of deep learning based on CNN can be used to automatically calculate the degree of frequency of occurrence of the turbulence prediction pattern data N in the turbulence prediction pattern data group 111 included in the determination meteorological data O, and determine the degree of similarity of both. Although the determination meteorological data O is a two-dimensional image with the north-south direction as the first axis and the east-west direction as the second axis, a three-dimensional data set is created with the time axis of the determination meteorological data O made for each meteorological forecast data with the future time point F0 as the reference as the third time axis. Here, all the turbulence prediction pattern data N in the turbulence prediction pattern data group 111 are executed as a convolution layer (convolution layer) to perform CNN. The detailed execution method of CNN is based on the method of general CNN. Since the output value of CNN is based on a specific image element cut out from past meteorological data at a place where turbulence occurred, the output value of CNN can be used as the turbulence occurrence probability (S114).

[0104] By repeating the processes (S111 to S114) described above at predetermined time intervals corresponding to the timing of acquiring meteorological forecast data within an arbitrary wide area, a turbulence occurrence probability that is updated at time intervals can be obtained. The timing of acquiring meteorological forecast data can also be freely set at predetermined time intervals. It is also possible to freely set the timing of acquiring meteorological forecast data at predetermined time intervals within the process of acquiring meteorological forecast data (S111), to collect a certain amount of meteorological forecast data at one time, and it is also possible to acquire one meteorological forecast data at one time in communication in the communication device 24 and store it in the storage device 22, and thereafter not proceed to the next process (S112), but continue the process of acquiring meteorological forecast data (S111) until a certain amount of meteorological forecast data is accumulated, and execute the processes (S112 to S114) at the timing when a certain amount of meteorological forecast data is collected. The determination process 12 includes all of these.

[0105] In the present application represented by Embodiment 1, one meteorological parameter is focused on, turbulence prediction pattern data N is created from past meteorological data when turbulence occurred with respect to the meteorological parameter, on the other hand, determination meteorological data O is created from meteorological forecast data at the future time point F0, and the degree of similarity is calculated by comparing the image data of both, the degree of possibility of turbulence occurrence can be calculated, and the effect that turbulence occurrence can be predicted based on the fact / state that turbulence occurred in the past in a simple method is achieved.

[0106] (Embodiment 2)

[0107] Reference FIGS. 5A-5DThe turbulence prediction processing of Embodiment 2 of the turbulence prediction system 1 will be described. Embodiment 2 corresponds to the principle of the second turbulence prediction mode in the embodiment. In Embodiment 2, the turbulence prediction processing, including the learning process 11 and the decision process 12, is performed in the same manner as in Embodiment 1. The concepts of the learning process 11 and the decision process 12 are as explained in the embodiment, and are similar to... FIG. 3 The concepts shown are the same. In Example 2, similarly to Example 1, past meteorological data and past observation data actually observed by the aircraft are used in the learning process 11, and meteorological forecast data is used in the decision process 12. The difference between Example 2 and Example 1 is that in Example 1, only one meteorological parameter α, β, γ, ... is selected as the evaluation meteorological parameter in both the learning process 11 and the decision process 12, but in Example 2, multiple meteorological parameters α, β, γ, ... are selected as evaluation meteorological parameters and combined for use. That is, the method is as follows: in order to make more accurate predictions by identifying similarity from multiple perspectives, the amount of information for determining similarity is increased by using multiple meteorological parameters of different types to create turbulence prediction model data N and decision meteorological data O. Here, FIG. 5A The flowchart illustrates the learning process 11 of the turbulence prediction processing in Example 2. FIG. 5B The flowchart illustrates the determination process 12 of the turbulence prediction processing in Example 2. FIG. 5C This represents the concept of creating a combined turbulence prediction model data N' from multiple evaluation meteorological parameters as a single turbulence prediction model data N'. FIG. 5D This describes the concept of creating a combined set of meteorological data O for judging multiple evaluation meteorological parameters as a single set of judgment meteorological data O'. Hereinafter, Example 2 will be described focusing on the parts that differ from Example 1, while the parts that are the same as in Example 1 will be omitted. The main examples of the meteorological parameters α, β, γ, ... in Example 2 are as follows. These are merely main examples; other meteorological parameters can also be selected. Two or more meteorological parameters from the following meteorological parameters α, β, γ, ... are selected as evaluation meteorological parameters and combined for use.

[0108] (1) Meteorological parameter αEW = East-west component of vertical wind shear

[0109] (2) Meteorological parameter αNS = north-south component of vertical wind shear

[0110] (3) Meteorological parameter β = vertical temperature gradient

[0111] (4) Meteorological parameter γ = vertical air density gradient

[0112] (5) Weather parameter δ = cloud image of satellite photograph (infrared radiation brightness temperature, visual image, water vapor image)

[0113] (6) Weather parameter ε = relative humidity

[0114] (7) Weather parameter ζEW= east-west component of wind speed

[0115] (8) Weather parameter ζNS= north-south component of wind speed

[0116] (9) Weather parameter η = air temperature (absolute value)

[0117] (10) Weather parameter θ = air density (absolute value)

[0118] As a representative example of the evaluation weather parameters of Embodiment 2, for example, four of (1) weather parameter αEW (east-west component of vertical wind shear), (2) weather parameter αNS (north-south component of vertical wind shear), (7) weather parameter ζEW (east-west component of wind speed), and (8) weather parameter ζNS (north-south component of wind speed) are selected. Hereinafter, a case where these four weather parameters are selected as the evaluation weather parameters will be described. As the selection and combination of weather parameters, the number of weather parameters is selected such that an even number of pseudo weather image data can be created. By even number, it is meant that a quadrangular image data is created when the image data is combined to be one image data. Thus, in Embodiment 2, the weather data created for a plurality of weather parameters is combined, unlike Embodiment 1.

[0119] FIG. 5A is a flowchart of the creation of one combination of turbulence prediction pattern data N' as teacher data in the learning process 11 of the turbulence prediction process in Embodiment 2. In Embodiment 2, the same processes as the processes S101 to S106 ( FIG. 4A ) of Embodiment 1 are performed as processes S201 to S206 ( FIG. 5A ) for each of a plurality of evaluation weather parameters. The processes S201 to S206 ( FIG. 5A ) are the same as Embodiment 1, and thus the description is omitted. By the process 206, for each of a plurality of evaluation weather parameters, turbulence prediction pattern data N is created. In a case where the east-west component of vertical wind shear, the north-south component of vertical wind shear, the east-west component of wind speed, and the north-south component of wind speed are selected as the weather parameters, for each of them, an even number (in this explanation, four two-dimensional image data) of turbulence prediction pattern data N is created as in FIG. 5C .

[0120] The four turbulence prediction model data N corresponding to the four evaluation meteorological parameters, produced in the previous step (S206), are arranged and combined to create a combined turbulence prediction model data N' (S206) as the teacher data. For example, as FIG. 5C As shown, imagine dividing a completed combined turbulence prediction model data N' into four equal parts, arranging them as follows: (a) the east-west component of wind speed in the upper left, (b) the north-south component of wind speed in the upper right, (c) the east-west component of vertical wind shear in the lower left, and (d) the north-south component of vertical wind shear in the lower right. A pseudo-meteorological image data is then stored as teacher data for a predetermined time in the storage unit or storage device 22 (S207). This process is repeated over the entire area of ​​any wide-area region. Similar to Example 1, steps S201 to 207 are repeated over the entire area of ​​the predetermined wide-area region (the entire segmented wide-area region). The method for creating the turbulence model data for each meteorological parameter constituting the combined turbulence prediction model data N' is the same as in Example 1.

[0121] Next, the determination process 12 of Example 2 will be explained. FIG. 5B The flowchart illustrates the determination process 12 of the turbulence prediction processing in Example 2. In the determination process of Example 2, since the teacher data is based on multiple meteorological parameters, the same meteorological parameters used in the learning process 11 need to be selected in determination process 12 to create a combined determination meteorological data O' in the same arrangement as the teacher data. This is the difference between Example 2 and Example 1. Otherwise, Example 2 is the same as Example 1. Conversely, the method for creating the determination meteorological data for each meteorological parameter that constitutes the combined determination meteorological data O' is the same as in Example 1.

[0122] In the determination process 12 of Example 2, from step S211, which obtains weather forecast data for any wide-area region regarding each of the plurality of evaluation meteorological parameters selected in learning process 11, to step S212, which generates determination meteorological data O for each of the plurality of evaluation meteorological parameters, the process is the same as in Example 1, and therefore the description is omitted. Through step 212, determination meteorological data O is generated for each of the plurality of evaluation meteorological parameters. In the description of learning process 11, four meteorological parameters were selected: the east-west component of vertical wind shear, the north-south component of vertical wind shear, the east-west component of wind speed, and the north-south component of wind speed. Therefore, for each of them, as... FIG. 5D That's how you make a judgment using meteorological data O.

[0123] The four pieces of the evaluation-weather-parameter-corresponding meteorological data O produced in the previous step (S212) are arranged in combination to produce one piece of the combined evaluation-weather-parameter-corresponding meteorological data O' (S213). For example, as shown in FIG. 5D the evaluation-weather-parameter-corresponding meteorological data O is arranged in a predetermined configuration to produce one piece of the combined evaluation-weather-parameter-corresponding meteorological data O' in the same configuration as the turbulence prediction pattern data. That is, the evaluation-weather-parameter-corresponding meteorological data O is arranged in the configuration in which the upper left is the east-west component of the (a) wind speed, the upper right is the north-south component of the (b) wind speed, the lower left is the east-west component of the (c) vertical wind shear, and the lower right is the north-south component of the (d) vertical wind shear to produce one piece of the combined evaluation-weather-parameter-corresponding meteorological data O'. The produced one piece of the combined evaluation-weather-parameter-corresponding meteorological data O' is stored in the storage section or the storage device 22. This process is repeatedly performed for the entire area of the arbitrary wide area.

[0124] Next, one piece of the combined turbulence prediction pattern data N' is read from the storage section or the storage device 22 as the teacher data. Then, the read one piece of the combined turbulence prediction pattern data N' is compared with the one piece of the combined evaluation-weather-parameter-corresponding meteorological data O', and the similarity is calculated (S215) as the degree of the possibility of the occurrence of the turbulence (S215). The steps S214 and S215 are the same as the steps S113 and S114 of the embodiment 1, and thus the description is omitted. In the comparison between the one piece of the combined turbulence prediction pattern data N' and the one piece of the combined evaluation-weather-parameter-corresponding meteorological data O' for all the areas of the prediction area, each of the turbulence prediction pattern data N constituting the one piece of the combined turbulence prediction pattern data N' is compared with each of the evaluation-weather-parameter-corresponding meteorological data O constituting the one piece of the combined evaluation-weather-parameter-corresponding meteorological data O on the computer. At this point, each of the comparisons is the same as the embodiment 1. In the comparison between each of the turbulence prediction pattern data N constituting the one piece of the combined turbulence prediction pattern data N' and each of the evaluation-weather-parameter-corresponding meteorological data O constituting the one piece of the combined evaluation-weather-parameter-corresponding meteorological data O, the comparison is performed while the division regions are staggered one by one in the respective comparisons.

[0125] In the case of the embodiment 2, since a plurality of evaluation-weather-parameters can be selected, a plurality of weather parameters related to the complex mechanism of the turbulence can be used as the evaluation-weather-parameters, and the degree of the possibility of the occurrence of the turbulence calculated in the determination can be more accurate than the embodiment 1.

[0126] (Embodiment 3)

[0127] Next, reference is made to FIGS. 6A-6D and FIG. 6BThe turbulence prediction process of Example 3 will be described. Example 3 corresponds to the 3rd turbulence prediction mode principle of the embodiment. Example 3 also performs the turbulence prediction process including the learning process 11 and the determination process 12 in the same manner as Example 1 and Example 2. FIG. 6A A flowchart showing the learning process 11 of the turbulence prediction process of Example 3 is shown in FIG. 12. FIG. 6B A flowchart showing the determination process 12 of the turbulence prediction process of Example 3 is shown in FIG. 13. In Example 3, the common point is that, in the same manner as Example 1 and Example 2, past meteorological data and past observed data actually observed by the aircraft are used in the learning process to create turbulence prediction mode data. On the other hand, Example 3 differs from Example 1 and Example 2 in that, in the determination process, meteorological forecast data is not used. In Example 3, current meteorological data and just-past meteorological data are used instead of meteorological forecast data. That is, Example 1 and Example 2 are methods in which predicted meteorological data is image-processed to compare the degree of similarity, but Example 3 is a method in which current meteorological data and just-past meteorological data are used to create determination-use meteorological data O, and the degree of similarity with turbulence prediction mode data N is compared to predict the occurrence of turbulence based on the change trend in the time series.

[0128] In addition, in Example 3, the same as in Example 1 and different from Example 2 is that one meteorological parameter is selected. The main examples of the meteorological parameters α, β, γ,... in Example 3 are as follows. These are main examples, and other than these, one of the meteorological parameters α, β, γ,... can be selected. One of the meteorological parameters α, β, γ,... shown below is used.

[0129] (1) Meteorological parameter α = absolute value of vertical wind shear

[0130] (2) Meteorological parameter β = vertical temperature gradient

[0131] (3) Meteorological parameter γ = vertical air density gradient

[0132] (4) Meteorological parameter δ = cloud image of satellite photograph (infrared radiation brightness temperature, visual image, water vapor image)

[0133] In Example 2, a plurality of evaluation meteorological parameters are selected, and for each of the plurality of evaluation meteorological parameters, one turbulence prediction mode data N is created by combining turbulence prediction mode data N. On the other hand, in Example 3, one meteorological parameter is selected, and for the current point Co and the nearest consecutive plurality of points C -1 , C -2 ... C -(n-1) , C-n The weather data, i.e., the determination weather data O, is combined to produce one set of combined determination weather data O'. As a representative example of the evaluation weather parameter in Embodiment 3, for example, a cloud image of a satellite photograph of the weather parameter δ, for example, an infrared radiation brightness temperature, a visible image, a water vapor image, etc.

[0134] FIG. 6A is a flowchart of the process of producing the turbulence prediction pattern data N as the teacher data in the learning process 11 of the turbulence prediction process of Embodiment 3. The processes 301 to 306 in Embodiment 3 are substantially the same as the processes 101 to 106 of Embodiment 1. The processes 301 to 306 in Embodiment 3 differ from the processes 101 to 106 of Embodiment 1 in that a plurality of turbulence prediction pattern data N is produced from a plurality of weather data at past discrete time points P0, P -1 , P -2 ... P -(n-1) , P -n The number of past discrete time points P0, P -1 , P -2 ... P -(n-1) , P -n needs to be the same as the number of the current time point CO and the number of the past nearest consecutive time points C -1 , C -2 ... C -(n-1) , C -n selected at the time of producing the determination weather data O. Here, as an example, four time points P0, P -1 , P -2 , P -3 are selected at the time of producing the turbulence prediction pattern data N, and four time points CO, C -1 , C -2 , C -3 are selected including the current time point CO at the time of producing the determination weather data O, and are set to the same number of time points. Here, as an example, the consecutive time unit is set to four hours, and the predetermined time unit is set to one hour. In the processes of the learning process 11, the weather data at each time point of the past arbitrary time point P0 (t=0), the time point P -1 one hour earlier than the time point, the time point P -2 two hours earlier than the time point, and the time point P -3 three hours earlier than the time point are acquired, and the processes 301 to 306 are executed.

[0135] i.e., from the process S301 to the process S306, until the weather data at the four discrete time points P0, P-1 P -2 P -3 The process of generating turbulence prediction model data N at each point in time is the same as steps S101 to S106 of Example 1 already described. As past meteorological data, if, for example, a cloud image of water vapor from a satellite photograph is selected as the meteorological parameter, it means that a cloud image of water vapor corresponding to that point in time must be obtained. In this case, the definition of numericalization also includes, for cloud images (water vapor) of consecutive time units, including, with arrows indicating the direction vectors of movement of any meteorological parameter element in each segmentation that constitutes a computationally valid grid.

[0136] In process S306, at the past four discrete time points P0, P... -1 P -2 P -3 The turbulence prediction model data N at various time points is arranged and combined according to a predetermined configuration. For example, imagine dividing a completed combined turbulence prediction model data N' into four equal parts, arranging it as follows: top left (a) t=0, top right (b) t=-1, bottom left (c) t=-2, and bottom right (d) t=-3. Additionally, this data is stored in the storage unit or storage device 22 (S307) as needed. This predetermined configuration is not particularly limited, but it must be easy to monitor changes over time and must be compatible with the consecutive time points C0, C... -1 C -2 C -3 The same configuration is used when multiple decisions are made using meteorological data O to create a combined decision using meteorological data O'. This process S301 to S307 is repeatedly executed over the entire area of ​​any wide-area region.

[0137] FIG. 6B The flowchart illustrates the determination process 12 of the turbulence prediction processing in Example 3. In the determination process 12 of Example 3, consecutive time points C0, C... -1 C -2 C -3 Meteorological data for any wide area at multiple points in time (S311). Here, since, as an example, in steps S301 to S306, a predetermined time unit was set to 1 hour to create turbulence prediction model data N for the past 4 hours, similarly, the predetermined time unit is set to 1 hour to obtain current meteorological data for a continuous 4-hour period to create decision-making meteorological data O. That is, meteorological data at the current time C0 (t=0) and time C10 at a time 1 hour earlier than the current time are obtained. -1 Meteorological data (t=-1), time point C 2 hours earlier than the present. -2weather data at time point C (t = -2) and the weather data 3 hours earlier than the current time point C -3 weather data at each time point of the weather data at time point C (t = -2) (S311). Then, the evaluation weather parameters in the divided areas obtained by dividing the arbitrary wide area are numerically processed based on the obtained weather forecast data, and the decision weather data O is created (S312). In the decision process 12, the divided areas constituting the wide area in the decision weather data O must be the same as the divided areas constituting the wide area in the turbulence prediction pattern data N in the learning process 11.

[0138] The continuous multiple time points C0, C -1 , C -2 , C -3 points are combined to create one combined pseudo weather time-series image data (S313). The arrangement of the multiple decision weather data O to create one combined decision weather data O' needs to be arranged in the same order (position) as the arrangement in the learning process 11 when one turbulence prediction pattern data is created by the process 307. That is, according to the example here, as shown in FIG. 6, the multiple decision weather data O are arranged to create one combined decision weather data O' in the arrangement of the upper left (a) t = 0, the upper right (b) t = -1, the lower left (c) t = -2, and the lower right (d) t = -3 (S313). FIG. 6D

[0139] Next, one combined turbulence prediction pattern data N' as the teacher data is read from the storage section, storage device 22. Then, the read one combined turbulence prediction pattern data N' is compared with the one combined decision weather data O', and the similarity is calculated (S314), which is the degree of the possibility of the occurrence of turbulence (S315). The processes S314 and S315 are the same as in the example 1 and the example 2, and thus the description is omitted. The comparison method is the same as in the example 2, and the divided areas constituting the turbulence prediction pattern data N at each time point of the one combined turbulence prediction pattern data N' are compared with the divided areas constituting the decision weather data O at each time point of the one combined decision weather data O'. The respective comparisons are the same as in the example 1.

[0140] ​In Embodiment 3, one combined decision-making weather data O' is made based on weather data at the time from t = -3, which is traced back from the current t = 0, and compared with one combined turbulence prediction pattern data N'. Thus, the comparison of the two data and the similarity thereof in Embodiment 3 are different from Embodiments 1 and 2 in that the comparison is not between past data. Thus, with respect to the turbulence occurrence probability, in the case where the portion at t = -3 is similar to the past image in which turbulence occurred and the portion at t = 0 is less similar to the past image in which turbulence occurred, it can be determined that the turbulence occurrence probability is low. Conversely, in the case where the portion at t = 0 is similar to the past image in which turbulence occurred and the portion at t = -3 is less similar to the past image in which turbulence occurred, it can be determined that the turbulence occurrence probability is high. Thus, for example, if the difference obtained by subtracting the similarity at t = -3 from the similarity at t = 0 is positive, there is a possibility that turbulence will occur, and the average of the similarities at the respective points of time from t = 0 to t = -3 at that time can be used as the turbulence occurrence probability. In addition, the turbulence occurrence probability can be defined by weighting in addition to this.

[0141] In the case of Embodiment 3, the method of calculating the similarity to the past image in which turbulence occurred based on past weather data in chronological order is employed, and thus the trend of the temporal change in which turbulence occurs is included in the direct calculation, so that the effect of being able to grasp the occurrence of turbulence from the weather data at the current time without being based on weather forecasts is obtained.

[0142] (Embodiment 4)

[0143] Next, the turbulence prediction processing of Embodiment 4 will be described. Embodiment 4 corresponds to the 4th turbulence prediction pattern principle of the embodiment. Embodiment 4 makes one combined turbulence prediction pattern data N' as teacher data in the learning process 11 in the same method as Embodiment 2. The selection of arbitrary weather data can be applied to Embodiment 2. In Embodiment 4, further, when making the one combined turbulence prediction pattern data N', the past point P0 at which the making is included is taken as a starting point, and one combined turbulence prediction pattern data N' is also made for a predetermined number of past points (P0, P -1 , P -2 , P -3 ) traced back in the future. The series of one combined turbulence prediction pattern data N' of the past points (P0, P -1 , P -2 , P -3 ) is taken as teacher data. On the other hand, in the decision-making process, one combined decision-making weather data O' is also made as teacher data in the decision-making process 12 from weather forecast data in the same method as Embodiment 2. At this time, one combined decision-making weather data O' is made for a predetermined number of future points (F0, F -1, F -2 , F -3 ), and a series of the combination determination weather data O' is also created. Also, in the determination process, it is determined whether or not a feature of the series of the combination turbulence prediction pattern data N' exists in the series of the combination determination weather data O'. In Embodiment 2, it is determined whether or not the common feature exists by the comparison of the static images of the points in time, but in Embodiment 4, in addition to the existence of the common feature based on the comparison of the static images of the points in time, the existence of the common feature under the dynamic change with the passage of time of the predetermined number of points in time is also added to the determination. While the convolutional neural network (CNN) can be applied with respect to Embodiment 2, the recurrent neural network (RNN) can be applied with respect to Embodiment 4.

[0144] (Embodiment 5)

[0145] Next, the turbulence prediction processing of Embodiment 5 is explained. Embodiment 5 corresponds to the 5th turbulence prediction pattern principle of the embodiments. Embodiment 5 is a method in which Embodiment 1 and Embodiment 3 are mixed. First, in the same manner as Embodiment 1, the turbulence prediction pattern data N as the teacher data is created in the learning process 11. The selection of the arbitrary weather data can apply Embodiment 1. In Embodiment 5, further, at the time of creating the turbulence prediction pattern data N, the past point P0 of the creation is included and is used as the starting point, and the turbulence prediction pattern data N of the predetermined number of past points in time (P0, P -1 , P -2 , P -3 ) is also created. The series of the turbulence prediction pattern data N of the past points in time (P0, P -1 , P -2 , P -3 ) is used as the teacher data.

[0146] On the other hand, in the determination process, in the same manner as Embodiment 3, the determination weather data O as the teacher data is created from the weather data of the current point C0 in the determination process 12. At this time, the current points (C0, C -1 , C -2 , C -3Similarly, a series of meteorological data O for judgment are generated. Furthermore, during the judgment process, it is determined whether a series of turbulence prediction model data N features exist within the series of meteorological data O. In Embodiments 1 and 3, the existence of common features is determined by comparing static images at a single point in time. However, in Embodiment 5, in addition to determining the existence of common features based on the comparison of static images at various points in time, the existence of common features under dynamic changes over time at a predetermined number of points in time is also included in the judgment. Compared to Embodiments 1 and 3, which can apply convolutional neural networks (CNNs), Embodiment 5 can apply recurrent neural networks (RNNs).

[0147] Label Explanation

[0148] 1 Turbulence prediction system; 2, 2a, 2b Turbulence prediction devices; 3, 3a, 3b terminals; 4, 4a, 4b transceiver devices (antennas); 5 Aircraft; 6 Network; 7 Cloud; 11 Learning process; 12 Decision process; 21 Central processing unit; 22 Storage device; 23 Output device; 24 Communication device.

Claims

1. A turbulence prediction system comprising a computation unit and a storage unit storing multiple turbulence prediction model data related to arbitrary meteorological parameters, for predicting locations within a prediction area at a given time point where turbulence is highly likely to occur. The multiple turbulence prediction model data related to the arbitrary meteorological parameter are factual data of turbulence occurrence, generated based on meteorological data related to the arbitrary meteorological parameter at past turbulence locations, rather than on mathematical models. The calculation unit generates judgment meteorological data for the prediction area at the judgment time point based on meteorological data related to the arbitrary meteorological parameters. The calculation unit calculates the high similarity between the meteorological data used for judgment and each of the plurality of turbulence prediction model data, and determines the high similarity area as a location with a high probability of turbulence occurring within the prediction area.

2. The turbulence prediction system according to claim 1, The arbitrary meteorological parameter can be multiple meteorological parameters. The multiple turbulence prediction model data are generated for each of the multiple meteorological parameters. The determination is made using meteorological data for each of the plurality of meteorological parameters. The calculation of the highly similar portion is performed by comparing the meteorological data used for the determination with the data from the multiple turbulence prediction models for each of the multiple meteorological parameters.

3. The turbulence prediction system according to claim 2, The plurality of turbulence prediction model data for each of the plurality of meteorological parameters are arranged in a predetermined configuration for each of the plurality of meteorological parameters as a combined turbulence prediction model data, and the determination meteorological data for each of the plurality of meteorological parameters are arranged in the same configuration as the predetermined configuration as a combined determination meteorological data.

4. The turbulence prediction system according to claim 1, The multiple turbulence prediction model data are prepared as multiple groups, each consisting of data from a predetermined number of time points at a predetermined time interval. The meteorological data used for determining any meteorological parameter is generated from data at a predetermined number of time points within the predetermined time interval. The calculation of the high similarity portion is based on the comparison of the characteristics of time-varying changes at a predetermined number of time points in the comparison of the meteorological data used for determination with the data of each of the multiple groups constituting the multiple turbulence prediction model data.

5. The turbulence prediction system according to claim 1, The predicted region is divided into multiple partitions. The multiple turbulence prediction model data related to the arbitrary meteorological parameter are generated by quantifying and assigning the meteorological data related to the arbitrary meteorological parameter at any discrete time point in the past to each of the multiple segmented regions belonging to a predetermined-size region centered on a location where turbulence occurred at any discrete time point in the past. The meteorological data used for the determination is generated by obtaining meteorological data related to the arbitrary meteorological parameters for the prediction area at the determination time point, quantifying and allocating the data within each of the multiple segmented divisions.

6. The turbulence prediction system according to claim 5, The meteorological data used for the determination is prepared into a region of the same size as the predetermined size and has multiple subdivisions. The calculation of the high similarity portion is based on a comparison between the meteorological data for judgment and the multiple turbulence prediction model data, while offsetting the segmentation of each turbulence prediction model data constituting the multiple turbulence prediction model data from the segmentation of the meteorological data for judgment, one by one.

7. The turbulence prediction system according to claim 5, The arbitrary meteorological parameter can be multiple meteorological parameters. The multiple turbulence prediction model data are generated for each of the multiple meteorological parameters. This involves arranging the multiple turbulence prediction model data generated for each of the multiple meteorological parameters in a predetermined configuration and combining them into a single combined turbulence prediction model data set. The determination meteorological data is generated for each of the plurality of meteorological parameters. The determination meteorological data generated for each of the plurality of meteorological parameters is combined into a single combined determination meteorological data by arranging the plurality of determination meteorological data generated for each of the plurality of meteorological parameters in the same configuration as the predetermined configuration.

8. The turbulence prediction system according to claim 7, The similarity calculation involves comparing the turbulence prediction model data constituting the plurality of combined turbulence prediction model data with respect to each of the plurality of meteorological parameters, and the decision meteorological data constituting the plurality of decision meteorological data with respect to each of the plurality of meteorological parameters, with each corresponding meteorological parameter being offset from the corresponding segmentation.

9. The turbulence prediction system according to claim 5, The multiple turbulence prediction model data are prepared as multiple groups, each consisting of data from a predetermined number of time points at a predetermined time interval. The meteorological data used for determining a meteorological parameter is generated from data at a predetermined number of time points within the predetermined time interval. The calculation of the high similarity portion is based on the comparison of the characteristics of time-varying changes at a predetermined number of time points in the comparison of the meteorological data used for determination with the data of each of the multiple groups constituting the multiple turbulence prediction model data.

10. The turbulence prediction system according to any one of claims 1 to 9, The meteorological data used for the determination is meteorological forecast data.

11. The turbulence prediction system according to claim 1, The arbitrary meteorological parameter is a single meteorological parameter. The multiple turbulence prediction model data are generated by quantifying and allocating past meteorological data for any given meteorological parameter at any discrete time point in the past and at a predetermined number of discrete time points consecutively at predetermined discrete intervals from the given discrete time point into a single combined turbulence prediction model data set, which is produced by arranging these multiple turbulence prediction model data sets in a predetermined configuration into a single set of multiple combined turbulence prediction model data sets within multiple segmented regions of a predetermined size area centered on the location where turbulence occurred. The meteorological data for judgment is a combined meteorological data for judgment, which is a combination of multiple meteorological data for judgment based on meteorological data of each discrete time point in the predetermined number of discrete time points continuously traced from the judgment time point at the predetermined discrete interval, in the same configuration as the predetermined configuration. The calculation of the highly similar portion is performed by using the meteorological data for determination as the combined meteorological data for determination and by using each of the multiple turbulence prediction model data as each of the multiple combined turbulence prediction model data.

12. The turbulence prediction system according to claim 11, The similarity calculation involves comparing the turbulence prediction model data constituting the plurality of combined turbulence prediction model data with respect to each of the plurality of meteorological parameters, and the decision meteorological data constituting the plurality of decision meteorological data with respect to each of the plurality of meteorological parameters, with each corresponding meteorological parameter being offset from the corresponding segmentation.

13. The turbulence prediction system according to claim 1, The arbitrary meteorological parameter is a single meteorological parameter. The multiple turbulence prediction model data are generated by creating multiple data sets, each consisting of a segmented region of a predetermined size centered on the location where turbulence occurred. These data sets represent past meteorological data for any given meteorological parameter at any discrete time point in the past, and past meteorological data for each of a predetermined number of discrete time points consecutively at predetermined discrete intervals from that arbitrary discrete time point. The data are then quantified and assigned to each of the multiple segmented regions. The meteorological data used for the determination is data composed of multiple data sets created for the arbitrary meteorological parameter based on meteorological data from each of the predetermined number of discrete time points continuously traced back from the determination time point at the predetermined discrete intervals. The calculation of the high similarity portion is based on the comparison of the time-varying characteristics of the multiple time points in the comparison between the meteorological data used for determination and the data of each of the multiple groups constituting the multiple turbulence prediction model data.

14. The turbulence prediction system according to claim 1, The multiple turbulence prediction model data refers to a large amount of turbulence prediction model data obtained over a long period of time. The calculation of the highly similar portion is based on the degree to which at least a portion of the large number of turbulence prediction models are included in the meteorological data used for determination.

15. The turbulence prediction system according to claim 1, The multiple turbulence prediction model data refers to a large amount of turbulence prediction model data obtained over a long period of time. The calculation of the highly similar parts is performed by using the large amount of turbulence prediction pattern data as a convolutional neural network layer.

16. The turbulence prediction system according to claim 4, The calculation of the highly similar parts is performed by a recurrent neural network.

17. The turbulence prediction system according to claim 1, The turbulence prediction system includes a turbulence prediction device connected to a network, and the turbulence prediction device includes the storage unit and the computing unit.

18. The turbulence prediction system according to claim 1, The turbulence prediction system is a cloud computing system that is connected to the computing unit and the storage unit via a network.

19. The turbulence prediction system according to claim 17 or 18, The turbulence prediction system has a terminal connected to the network. The output of the turbulence prediction system is sent to the terminal.

20. The turbulence prediction system according to claim 17 or 18, The turbulence prediction system includes a communication unit. The turbulence prediction system is equipped with an aircraft. The communication unit is connected to the network and sends the output of the prediction results of the occurrence of turbulence to the aircraft.

21. A turbulence prediction method, which calculates the probability of turbulence occurring within a prediction region at a determination time point using a turbulence prediction system equipped with a computing unit and a storage unit. The turbulence prediction method includes a learning process and a decision process. The learning process includes: The process of generating factual data on turbulence occurrence, i.e., multiple turbulence prediction model data, from meteorological data related to arbitrary meteorological parameters at past turbulence locations, without relying on mathematical models, and storing this data in the storage unit. The determination process includes: The process by which the computing unit generates judgment meteorological data for the prediction area based on meteorological data related to the arbitrary meteorological parameter; and The process involves the calculation of the similarity between the meteorological data used for determination and each of the plurality of turbulence prediction model data by the calculation unit, and determining the areas with high similarity as areas with a high probability of turbulence occurring in any region of the prediction area.

22. The turbulence prediction method according to claim 21, The arbitrary meteorological parameter can be multiple meteorological parameters. The multiple turbulence prediction model data are generated for each of the multiple meteorological parameters. The determination is made using meteorological data for each of the plurality of meteorological parameters. The calculation of the high similarity portion is performed by comparing the meteorological data used for the determination with the data from the multiple turbulence prediction models for each of the multiple meteorological parameters.

23. The turbulence prediction method according to claim 22, The learning process includes a step of arranging the multiple turbulence prediction model data for each of the multiple meteorological parameters in a predetermined configuration for each of the multiple meteorological parameters to form a combined turbulence prediction model data. In the determination process, the determination meteorological data for each of the plurality of meteorological parameters are arranged in the same configuration as the predetermined configuration to form a combined determination meteorological data step.

24. The turbulence prediction method according to any one of claims 21 to 23, The multiple turbulence prediction model data are prepared as multiple groups, each consisting of data from a predetermined number of time points at a predetermined time interval. The meteorological data used for determining any meteorological parameter is generated from data at a predetermined number of time points within the predetermined time interval. The calculation of the high similarity portion is based on the comparison of the characteristics of time-varying changes at a predetermined number of time points in the comparison of the meteorological data used for determination with the data of each of the multiple groups constituting the multiple turbulence prediction model data.

25. The turbulence prediction method according to claim 21, In the turbulence prediction method, The predicted region is divided into multiple partitions. The multiple turbulence prediction model data related to the arbitrary meteorological parameter are generated by quantifying and assigning the meteorological data related to the arbitrary meteorological parameter at any discrete time point in the past to each of the multiple segmented regions belonging to a predetermined-size region centered on a location where turbulence occurred at any discrete time point in the past. The meteorological data used for the determination is generated by obtaining meteorological data related to the arbitrary meteorological parameters for the prediction area at the determination time point, quantifying and allocating the data within each of the multiple segmented divisions.

Images