JP7110004B2 - Processing device, processing method, and program - Google Patents

Description

The embodiments of the present invention relate to processing apparatuses, processing methods, and programs.

2. Description of the Related Art There is known an analysis device that determines the type of particles in the atmosphere using observation information observed by a weather radar, and estimates the amount of rainfall or snowfall based on the determination result and the observation information. However, in some cases, this analysis device cannot accurately derive information about snowfall.

JP 2010-181175 A

A problem to be solved by the present invention is to provide a processing device, a processing method, and a program capable of deriving information on snowfall with higher accuracy.

A processing device according to an embodiment has an acquisition unit and a derivation unit. The acquisition unit acquires observation information indicating the state of the atmosphere for each of a plurality of divided areas obtained by dividing an observation area in a distance direction, an azimuth direction, and a vertical direction. When observation information of a target segmented region selected from among segmented regions included in a group of segmented regions having common positions in the distance direction and the azimuth direction is input, the derivation unit calculates snowfall corresponding to the target segmented region. A second estimation model for outputting information indicating intensity, wherein when the observation information is input to the second estimation model, the second estimation model is learned to derive information indicating snowfall intensity associated with the observation information. By inputting the observation information acquired by the acquisition unit into the second estimation model having the parameters, information indicating the snowfall intensity of the target divided area is derived, and information based on the snowfall intensity of the target divided area is input, the first estimation model outputs information indicating the degree of snow cover on the ground corresponding to the target divided area, wherein when information based on the snowfall intensity is input to the first estimation model, the snowfall By inputting the information based on the snowfall intensity of the target divided area into the first estimation model having parameters learned so as to derive information indicating the degree of snow cover associated with the information based on the intensity, Information indicating the degree of snow cover on the ground is derived.

The figure which shows the structure of the processing system 1. FIG. 4 is a conceptual diagram of radio waves emitted by the radar device 10. FIG. 2 is a diagram showing configurations of a processing device 20, a display information generating device 60, and a learning device 80; FIG. The figure which shows an example of the process which acquires an optimal parameter. The figure for explaining the outline of an optimal parameter and an estimation model. 4 is a flowchart showing the flow of processing executed by the processing device 20; 4 is a diagram for explaining observation information D; FIG. The figure which shows conceptually the content of the process of step S118. The figure which shows conceptually the content of the process of step S122. The figure which shows an example of transition of the snow cover increase degree. FIG. 4 is a diagram showing an example of an image IM displayed on the display unit of the terminal device 100; 4 is a conceptual diagram of processing executed by the learning device 80. FIG.

Hereinafter, a processing apparatus, a processing method, and a program according to embodiments will be described with reference to the drawings.

FIG. 1 is a diagram showing the configuration of a processing system 1. As shown in FIG. A processing system 1 including a processing device 20 includes a radar device 10 , a processing device 20 , a display information generation device 60 and a learning device 80 . The processing device 20 and the radar device 10 or the learning device 80 communicate via a dedicated line or the like. The processing device 20 and the display information generating device 60 communicate via a network NW1 such as a LAN (Local Area Network). The display information generation device 50 also communicates with the terminal device 100 via a network NW2 such as a WAN (Wide Area Network). Note that the processing device 20 and the display information generating device 60 may be formed integrally.

The radar device 10 includes an antenna 12 and an antenna-side control device 14 . The radar device 10 is, for example, a weather radar that observes atmospheric conditions. Also, the antenna 12 is, for example, a phased array antenna that includes a plurality of antenna elements and whose directivity angle can be changed electronically. The radar device 10 rotates the antenna 12 in the horizontal direction while electronically scanning the elevation direction, thereby observing the state of the atmosphere in three dimensions. The antenna 12 transmits and receives radio waves under the control of the antenna-side control device 14 .

In the following description, the radar device 10 is described as being a dual polarized phased array radar device, but instead of this, it may be a parabolic radar device using dual polarized waves. A dual-polarization phased array radar system transmits and receives radio waves oscillating in the horizontal and vertical directions, and observes the state of the atmosphere based on the reception results.

The antenna-side control device 14 includes a radar-side control section 16 and an information storage section 18 . The radar-side control unit 16 causes the antenna 12 to transmit a fan beam with a wide elevation angle beam width. The radar-side control unit 16 performs DBF (Digital Beam Forming) processing. The radar-side control unit 16 receives the reflected radio waves returned from the transmitted fan beam hitting scattering objects such as snow particles at once, and simultaneously generates observation information (echo intensity) for a plurality of elevation angle ranges.

The information accumulation unit 18 accumulates observation information acquired by the radar-side control unit 16 and transmits the accumulated information to the processing device 20 .

Here, the characteristics of the dual polarization phased array radar system will be described. FIG. 2 is a conceptual diagram of radio waves emitted by the radar device 10. As shown in FIG. For example, the radar device 10 radiates into the atmosphere a horizontally polarized wave h as shown in the left diagram of FIG. 2 and a vertically polarized wave v as shown in the right diagram of FIG. Then, the radar device 10 acquires the reflected radio waves that are returned after the two types of radio waves that have been radiated strike particles in the atmosphere. The radar device 10 acquires information on the shape of particles in the atmosphere in the vertical direction and the shape in the horizontal direction by acquiring the reflected radio waves of the two types of radio waves.

For example, raindrops in the atmosphere have different shapes depending on their size. Also, hail, hail, and snow have different shapes. For example, a large raindrop RD may be flat. The radar device 10 acquires information indicating the shape of the raindrops in the horizontal direction from the reflected radio waves with respect to the horizontal polarization, and obtains not only the strength (echo intensity) of the reflected radio waves but also the vertical direction of the raindrops from the reflected radio waves with the vertical polarization. Information indicating the shape can be obtained. The radar apparatus 10 can acquire information such as a reflection factor difference, an inter-polarization correlation coefficient, an inter-polarization phase difference change rate, and the like from the reflected radio wave. The reflectance factor difference is a parameter related to the aspect ratio of the particles. The polarization correlation coefficient is the correlation coefficient of the received power (received signal) of the horizontally polarized wave and the vertically polarized wave, which are the reflected radio waves. The inter-polarization phase difference change rate is a parameter related to the phase difference between the phase of the horizontally polarized wave and the phase of the vertically polarized wave, which are the reflected radio waves.

Also, when two kinds of radio waves pass through a particle having a flat shape or the like, a phase difference occurs between these radio waves. The radar device 10 can obtain inter-polarization phase differences, which are phase differences between the radar device 10 and the particles when they reciprocate with respect to each of the horizontal polarization and the vertical polarization. The inter-polarization phase difference change rate is the amount of change or differential amount of the inter-polarization phase difference in the distance direction.

FIG. 3 is a diagram showing configurations of the processing device 20, the display information generating device 60, and the learning device 80. As shown in FIG. The processing device 20 includes, for example, an acquisition unit 22, a setting unit 24, a processing unit 26, a first derivation unit 28, a first estimation unit 30, a second derivation unit 32, a second estimation unit 34, A risk derivation unit 36 , an information transmission unit 38 , and a storage unit 40 are provided. The acquisition unit 22, the setting unit 24, the processing unit 26, the first derivation unit 28, the first estimation unit 30, the second derivation unit 32, the second estimation unit 34, the risk derivation unit 36, and the information transmission unit 38 are implemented by the CPU ( It may be implemented by a processor such as a Central Processing Unit executing a program. In addition, these functional units may be realized by hardware such as LSI (Large Scale Integration), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array), GPU (Graphics Processing Unit), It may be realized by cooperation of software and hardware. Further, the above program may be stored in a storage device in advance, or may be stored in a detachable storage medium such as a DVD or CD-ROM, and the storage medium is installed in the drive device of the processing device 20. may be installed on the storage device by The storage unit 40 is implemented by a storage device such as a RAM (Random Access Memory), ROM (Read Only Memory), HDD (Hard Disk Drive), flash memory, or the like. Some or all of the processing unit 26, the first deriving unit 28, the first estimating unit 30, the second deriving unit 32, and the second estimating unit 34 are examples of the "deriving unit".

The acquisition unit 22 includes, for example, an observation information acquisition unit 22A and a learning information acquisition unit 22B. The observation information acquisition unit 22A acquires observation information 42 from the radar device 10 . The acquired observation information 42 is stored in the storage unit 40 . Note that the processing device 20 may be formed integrally with the radar device 10 . In this case, the acquisition unit 22 acquires the observation information 42 from the information storage unit 18 using a communication bus or the like in the device.

The learning information acquisition unit 22B acquires the estimated model 44 or the optimum parameters 46 from the learning device 80. FIG. The estimated model 44 and optimal parameters 46 are stored in the storage unit 40 . The estimation model 44 is a model used when deriving information about snowfall in the processing of the processing device 20 . Optimal parameters 46 are parameters that are applied to estimation model 44 .

The setting unit 24 determines to use the estimation model 44 for the target process, and sets the estimation model 44 for the determined process. The setting unit 24 also applies the optimum parameters 46 to the estimation model 44 used in the target process. Details will be described later.

The processing unit 26 performs processing for determining precipitation particles. The details of the processing of the processing unit 26 and the function units (first derivation unit 28, first estimation unit 30, second derivation unit 32, second estimation unit 34) described later will be described later.

The first derivation unit 28 derives the snowfall intensity for each divided area (mesh) obtained by dividing the observation area of the radar device 10 in the horizontal direction and the vertical direction. The first estimator 30 estimates the amount of snow on the ground using the snowfall intensity. The second derivation unit 32 derives the amount of snowfall in a predetermined number of meshes that are continuous in the vertical direction. The second estimator 34 estimates the amount of snow falling on the ground (increase in snow cover) based on the integrated amount derived by the second deriving unit 32 .

The risk deriving unit 36 derives risks related to snowfall and snow accumulation based on the processing results of the first estimating unit 30, the second estimating unit 34, and the like. The information transmission unit 38 transmits the processing result of the processing device 20 to the display information generation device 60 .

The display information generation device 60 includes a display information generation section 62 and a display information transmission section 64 . The display information generation unit 62 generates display information (image information) for displaying information including the processing result on the display unit of the terminal device 100 based on the processing result transmitted from the information transmission unit 38 . The display information transmission section 64 transmits the display information generated by the display information generation section 62 to the terminal device 100 .

The learning device 80 includes an acquisition section 82 , a learning section 84 , a provision section 86 and a storage section 90 . The acquisition unit 82 acquires learning data from, for example, an external terminal device. This learning data 92 is stored in the storage unit 90 . The external terminal device is a terminal device that communicates with the learning device 80 via a network, and is a terminal device that provides weather-related information.

The learning unit 84 performs machine learning or deep learning using the learning data to derive an estimation model or optimal parameters used for the estimation model. Also, the learning unit 84 may derive the estimation model (for example, function) itself instead of (or in addition to) deriving the parameters used in the estimation model. Details of the processing of the learning unit 84 will be described later. The providing unit 86 provides the processing device 20 with the optimum parameters or the estimated model derived by the learning unit 84 .

[Flowchart (Part 1)]
FIG. 4 is a diagram illustrating an example of a process of acquiring optimum parameters. In the example of FIG. 4, the optimum parameters are set, but when the estimation model derived by the learning device 80 is applied to the processing of the processing device 20, the same processing as this processing is performed.

First, the processing device 20 determines whether or not the timing for starting the processing relating to the amount of accumulated snow (the processing in FIG. 6, which will be described later) has arrived (step S10). When the timing for starting the processing related to the amount of snow cover is reached, the setting unit 24 instructs the learning device 80 to transmit the optimum parameters (step S12). Next, the setting unit 24 determines whether or not the learning information acquisition unit 22B has acquired the optimum parameters transmitted from the learning device 80 (step S14). When the optimum parameters are acquired, the setting unit 24 sets the acquired optimum parameters as the parameters of the estimation model using the optimum parameters (step S16). Thus, the processing of this flowchart ends.

In the above example, the processing device 20 acquires the optimal parameters by instructing transmission of the optimal parameters. Alternatively (or in addition), the processing device 20 may may be obtained. In this case, the learning device 80 transmits the optimum parameters to the processing device 20 at predetermined intervals.

FIG. 5 is a diagram for explaining an outline of optimum parameters and an estimation model. For example, the processing device 20 acquires the (A) estimation model (an example of the “fourth estimation model”) and the (B) to (D) optimum parameters from the learning device 80 . Note that the processing device 20 may acquire the estimation models (functions) corresponding to (B) to (D) from the learning device 80 .

(A) Membership function in determination of precipitation particles (step S104 in FIG. 6) (details will be described later)
(B) “a” and “b” of the formula (1) “Rs [mm/h]=a×Z ^b ” used for deriving the snowfall intensity (step S108 in FIG. 6)
(C) “c”, “d”, and “τ” of the formula (2) “SF=c×Ms ^d +τ×e” used for estimating snow cover (step S118 in FIG. 6)
(D) “τs” of the formula (3) “δSF=τs×f×VIS ^g− (1−τ)×e” used for estimating the amount of snow increase (step S122 in FIG. 6)
Equations (2), (1), and (3) are examples of the "first estimation model,""second estimation model," and "third estimation model," respectively.

[Flowchart (Part 2)]
FIG. 6 is a flow chart showing the flow of processing executed by the processing device 20. As shown in FIG. In the processing of this flowchart, the above-described estimation model (membership function) and all optimal parameters are used, but some optimal parameters may be used.

First, the observation information acquisition unit 22A determines whether or not the observation information D for one period accumulated in the information accumulation unit 18 of the radar device 10 has been acquired (step S100).

The observation information D is information indicating the echo intensity acquired by the radar device 10 for each mesh and observation period. Note that the observation information D is not limited to information indicating the echo intensity, and may be any information that can determine the type of particles, the shape of particles, the presence or absence of particles, and the like.

FIG. 7 is a diagram for explaining the observation information D. FIG. Hereinafter, description will be made using XYZ coordinates as necessary. Observation information D is information obtained by driving the antenna 12 of the radar device 10 in the horizontal direction from the start point to the end point of the driving. The observation information D acquired between the start point and the end point is observation information acquired in one cycle. The mesh M is a divided area divided by a predetermined width in each of the distance direction (Y direction) from the radar device 10, the azimuth direction (X direction), and the vertical direction (Z direction). In the illustrated example, the mesh M is approximated by a cube because the observation information D is at a position sufficiently far from the radar device 10 . A set of meshes M that are the same in the horizontal direction (distance direction and azimuth direction) and differ only in the vertical direction (altitude direction) will be referred to as a "mesh block MB". The mesh block MB is an example of "divided area group having a common horizontal position".

Returning to the description of FIG. When the observation information D for one period is acquired (or when the observation information D of the observation area to be observed is acquired), the processing unit 26 selects one mesh among the plurality of meshes M corresponding to the observation information D M is selected (step S102), and a process of discriminating precipitation particles is performed for the selected mesh M (step S104). In this discrimination, it is discriminated whether precipitation particles (for example, hail, hail, ice crystals, dry snow, wet snow, rain) or non-precipitation echoes (ground clutter) exist in the mesh M, and the discrimination result is Classified by three-dimensional mesh.

The processing unit 26 performs a process of discriminating precipitation particles using a known method. For example, the processing unit 26 applies the membership function acquired from the learning device 80 to fuzzy theory to discriminate precipitation particles. A membership function is a function that outputs a degree of belonging of an element x of a set X to a set A to which fuzzy logic is applied. A set A is a set expressing that an element x of a certain set X belongs to the set A with ambiguity, and is characterized by a membership function.

For example, the membership function is a function that outputs a value closer to "1" when the degree of belonging to set A is higher, and outputs a value closer to "0" when the degree of belonging to set A is lower. For example, in determining precipitation particles, the processing unit 26 associates ground observation information, high-rise observation information, or temperature and humidity information obtained from model information with meshes to be observed. The ground observation information, high-rise observation information, or temperature and humidity information may be uniformly associated with the meshes, or may be associated with different values for each mesh. Then, the processing unit 26 uses information associated with the mesh and fuzzy theory to determine precipitation particles included in the mesh.

Next, the processing unit 26 determines whether or not precipitation particles exist and the precipitation particles are snow (for example, dry snow or wet snow) based on the processing result of step S104 (step S106). If the precipitation particles are not snow, the processing unit 26 does not proceed to step S108 and thereafter, and selects the next (unselected) mesh M and performs precipitation particle discrimination (steps S102 and S104). Specifically, when the precipitation particles are not snow, the processing unit 26 determines whether or not the process of distinguishing the precipitation particles has been completed for all meshes M (step S112). If the process of discriminating precipitation particles has not been completed for all the meshes M, the process proceeds to step S102. If precipitation particle determination has been completed for all meshes, the process proceeds to step S114.

When the precipitation particles are snow, the first derivation unit 28 derives the snowfall intensity for each mesh M (step S108). For example, the first derivation unit 28 derives the snowfall intensity Rs using Equation (1) for the mesh M determined to have snow.
Rs ^[ mm/h]=a×Zb (1)

"Z" is the radar reflectance factor [mm ⁶ m ^-3 ]. The radar reflection factor Z is a parameter that varies according to the particle size of particles that reflect radio waves. The radar reflection factor Z is derived based on the received power when the radar device 10 receives radio waves and the distance from the particles that have reflected the radio waves to the radar device 10 .

Next, the first derivation unit 28 converts the snowfall intensity [mm/h] derived in step S108 to the amount of snow Ms [mm] for the mesh M determined to have snow, and calculates the amount of snow. It is derived (step S110). The first derivation unit 28 converts the snowfall intensity [mm/h] into the amount of accumulated snow Ms [mm] using a unit time (for example, a set time such as one hour) as an integration interval. After the process of step S110, if the precipitation particle discrimination process (step S104) has been completed for all the meshes M, the process proceeds to the next step. If not, the next mesh M is selected, and the selected mesh M is processed from step S104. Steps S116 and S118 and steps S120 and S122 described below are, for example, parallel processes.

Next, the first estimation unit 30 selects a mesh block MB to be processed (hereinafter referred to as a target mesh block) from among a plurality of mesh blocks MB to be observed (step S114), A target mesh to be processed in step S116 is selected (step S116). The mesh to be processed is the mesh M whose observation information is presumed to have been acquired without being masked by an obstacle such as a mountain. For example, numerical terrain data may be referenced to select target meshes. Also, if there is no mesh M determined to have snow in the target mesh block MB, selection of the target mesh M for that target mesh block MB may be skipped.

Next, the first estimator 30 estimates the amount of accumulated snow on the ground per unit time for each target mesh in the target mesh block MB (step S118). The first estimator 30 estimates the snowfall amount per unit time based on the snowfall amount Ms [mm] derived in step S110. For example, the first estimation unit 30 estimates the amount of snow cover per unit time using Equation (2).
SF=c×Ms ^d +τ×e (2)

"SF" is the amount of accumulated snow at the end of the integration (when the unit time has elapsed). The first term (“c×Ms ^d ”) on the right side of Equation (2) indicates that the amount of snowfall derived in step S110 is converted to the actual amount of snow cover. 'c' or 'd' is the optimum parameter.

The second term (“τ×e”) on the right side indicates the amount of accumulated snow at the start point of integration (start point of unit time). In other words, “e” is the amount of accumulated snow observed or derived at a certain time (for example, at the start of integration) in the target mesh. The amount of snow "e" may be uniformly given the same value to all target meshes, or may be given as a continuous distribution on a two-dimensional plane (for example, on the XY plane).

"τ" is a time constant that expresses the amount of increase or decrease in snow cover. have That is, "τ" is a time constant for converting the derived snow amount "e" into the snow amount at the start of the next integration (snow amount after decreasing with the lapse of time). For example, "τ" is a numerical value of 1 or less and is a time constant indicating a tendency such as gradual decrease. Also, the time constant "τ" may be uniformly given to all the meshes M, or may be given as a continuous distribution on a two-dimensional plane.

FIG. 8 is a diagram conceptually showing the contents of the processing in step S118. The illustrated example shows the relationship between the integrated amount corresponding to the first term and the second term of Expression (2) and the time. Moreover, the vertical axis of FIG. 8 indicates the amount of accumulated snow. For example, it is assumed that the integrated amount "e" exists at the initial time T (integration start time). This amount of snow "e" is the amount of snow observed or derived at time T. In this case, at time T+1, according to equation (2), the integrated amount of "e × τ" (the integrated amount that is the result of e decreasing from time T to T+1) and "(c × Ms ^d ) ” (accumulated amount of snow falling from time T to T+1) is added (“e#”). At time T+2, the accumulated amount of snow is the sum of the accumulated amount "e#×τ" and the accumulated amount of "(c×Ms ^d )#" (the accumulated amount of snow falling from time T to T+1). At this time, for example, if there is a newly observed or derived "e" at time T+2, at time T+3, the amount of snow cover may be estimated using time T+2 as the initial time.

As described above, the processing device 20 can derive (estimate) the amount of accumulated snow per unit time on the ground corresponding to the target mesh using the estimation model such as Equation (2) and the optimum parameters.

Next, the second derivation unit 32 derives the vertically integrated amount of snowfall in the target mesh block MB (step S120). The second derivation unit 32 vertically integrates the amount of accumulated snow Ms [mm] for each mesh M derived in step S110. The section to be accumulated in the vertical direction may be all the meshes M in the mesh block MB, or may be the meshes M up to 3 km below, for example, focusing on the amount of snow near the ground. The amount of snowfall accumulated in the vertical direction is called VIS (Vertical Integrated Snow). The "VIS" is taken as the potential amount of snowfall in the sky above the cloud.

Next, the second estimating unit 34 estimates the amount of snow falling on the ground per unit time in the target mesh block MB (increase in snow cover: δSF) based on the "VIS" derived in step S120 ( step S122). The second estimator 34 derives the amount of increased snow cover using, for example, Equation (3).
δSF=τs×f×VIS ^g− (1−τ)×e (3)

The first term (“τs×f×VIS ^g ”) on the right side of Equation (3) is a term for converting the amount of snow in the sky per unit time. In other words, the first term on the right-hand side can be rephrased as the amount of snowfall (amount equivalent to accumulated snow) in which the total amount of latent snow in the sky is likely to pass near the ground and accumulate. "f" and "g" are predetermined coefficients. “τs” is a time constant indicating the amount of increase per time. "VIS" is converted to snow equivalent by "τs", "f" and "g".

The second term on the right side of Equation (3) indicates the amount of accumulated snow at the start of observation. In other words, “e” is the amount of accumulated snow in each mesh M at the start of observation. The second term “(1−τ)×e” on the right side indicates the difference between the derived snow amount “e” and the snow amount from the time of derivation to the next observation start time. In other words, it indicates the amount of snow that has decreased at the next observation start point from the point at which the amount of snow "e" was derived. When the value of the first term on the right side is 0, the second term on the right side can be used to express the decrease in snow cover.

Note that the second estimating unit 34 assumes that snow particles uniformly exist in the vertical region related to the integrated amount derived in step S120, for example, and that snow particles falling on the ground per unit time Derive the amount of increase in snow cover. Further, the second estimating unit 34 may derive the amount of increase in the amount of snow that falls on the ground per unit time based on a predetermined model, for example. For example, the predetermined model is a model from which the amount of snow falling for each hour is derived.

FIG. 9 is a diagram conceptually showing the contents of the processing in step S122. The illustrated example shows the relationship between the integrated amount corresponding to the first term and the second term of Expression (3), the amount of increase in snow cover, and time. Moreover, the vertical axis of FIG. 9 indicates the amount of accumulated snow. For example, the net cumulative increase (δSF) from time T# to T#+1 is "the amount of snow accumulated from the sky from time T# to T#+1, which is the amount of simple increase" (τs x f x VIS ^g )*”−the amount of decrease in the existing snow cover from time T# to T#+1 “(1−τ)×e”.

For example, at time T#, it is assumed that an integrated amount "τs×f×VIS ^g (=e)" exists. In this case, at time T#+1, if the time constant of equation (3) is set to less than 1, the amount of accumulated snow decreases by "(1−τ)×e" according to equation (3). In this state, between the time T# and the time T#+1, when the amount of snow "(τs×f×VIS ^g )*" falls, from "(τs×f×VIS ^g )*" according to the equation (3) The amount of accumulated snow from which "(1-τ)×e" has been subtracted becomes the net increased amount of accumulated snow (δSF). In other words, the net amount of increase in snow cover is the amount obtained by subtracting the amount of decrease in already-covered snow from the amount of snow that has fallen from above in a certain period. For example, if the amount of snow that has already fallen is greater than the amount of snow that has fallen from above, the net amount of snow increase will be a negative value, and the phenomenon of the amount of snow can be expressed.

FIG. 10 is a diagram showing an example of changes in the degree of increase in snow cover. The vertical axis in FIG. 10 indicates the degree of increase in snow cover, and the horizontal axis indicates time. The degree of increase in snow cover varies from moment to moment depending on weather conditions. For example, at time Tx, the user, the administrator of the processing system 1, or the like can estimate the transition of the future increase in snow cover based on the increase in snow cover near time Tx (for example, the tendency of the change in the increase in snow cover). can. For example, as shown in the figure, if the degree of increase in snow cover tends to decrease before time Tx, the administrator can estimate that the amount of snow cover will tend to decrease in the future as indicated by the transition line Tr in the figure. . Note that the processing device 20 may estimate the transition of the future snow cover increase degree based on the snow cover increase degree around the time Tx.

Returning to the description of FIG. After the processing of steps S118 and S122, the processing unit 26 determines whether or not the processing of steps S116 to S122 has been completed for all mesh blocks MB to be observed (step S124). If the processes of steps S116 to S122 have not been completed for all mesh blocks MB, the process returns to step S114, an unselected mesh block MB is selected, and the processes of steps S116 to S122 are executed.

If the processes of steps S116 to S122 have been completed for all mesh blocks MB, the risk derivation unit 36 derives the risk of snow cover based on the above-described process results (step S126). For example, the risk deriving unit 36 derives the snow risk based on the amount of snowfall, the amount of snowfall, and the prediction of an increase in the amount of snowfall. For example, the risk derivation unit 36 can calculate the most recent (several 10 minutes ahead) snowfall risk by focusing on the amount of snowfall and snow cover, and the medium- to long-term (several hours ahead) snow cover risk using the prediction of the amount of snowfall and increase in snow cover. Risk may be derived.

Further, the risk derivation unit 36 may generate information on the snow risk that is suitable for the user from the snow risk derived for each mesh M, taking into account the specificity of the area. For example, the risk deriving unit 36 assigns a weight assigned to each region, such as around a railway line, around a main road, around a disaster emergency road, etc., to the snow risk corresponding to the region, and Risk may be derived.

Further, the risk derivation unit 36 may make a risk judgment by combining the forecast data of the numerical weather model and the processing result as materials for further long-term snow accumulation risk judgment. For example, the risk derivation unit 36 quantifies the information obtained from the numerical weather model, the integrated risk, etc., derives the integrated risk by adding, subtracting, multiplying, and dividing the digitized numerical value, and derives the derived integrated risk. A final integrated risk may be derived by comparing with a threshold value. In addition, the risk derivation unit 36 quantifies the information obtained from the numerical weather model and the accumulated risk, etc., and applies the quantified values to the probability prediction model and the learning model to derive the snow risk. may

Next, the information transmission unit 38 transmits the analysis data to the display information generation device 60 (step S128). The analysis data includes snowfall amount, snow cover amount, prediction of increase in snow cover amount, snow cover risk, and the like. This completes the processing of one routine in this flow chart.

A display information generation unit 62 of the display information generation device 60 generates display information based on the processing result transmitted from the information transmission unit 38 . Then, the display information transmitting section 64 transmits the display information generated by the display information generating section 62 to the terminal device 100 . The terminal device 100 acquires the display information and causes the display section of the terminal device 100 to display the image IM according to the acquired display information. FIG. 11 is a diagram showing an example of the image IM displayed on the display unit of the terminal device 100. As shown in FIG. The image IM includes information indicating areas with high, medium, and low snow risks, and information such as warnings about snow risks.

[Processing performed by the learning unit]
FIG. 12 is a conceptual diagram of processing executed by the learning device 80. As shown in FIG. First, the acquiring unit 82 of the learning device 80 acquires learning data (1). The learning data is a combination of observation information acquired by the radar device 10 (or another radar device) and one or more of the following information (A) to (D), for example. (A) is information about snowfall observed by ground observation equipment, (B) is information about snowfall provided on networks such as SNS (for example, information about snowfall posted), (C) is an in-vehicle camera or Snow cover information acquired by an in-vehicle radar, (D) is snow cover information provided by companies and local governments. Also, the learning data may be weather model data. Weather model data includes virtually generated observation information, snowfall intensity, amount of snow cover, amount of increase in snow cover, etc. derived by applying this observation information to a model.

The learning unit 84 performs machine learning or deep learning using learning data, and when observation information is input, derives an estimation model for deriving the degree of snowfall or an optimum parameter applied to the estimation model (2). Then, the providing unit 86 provides the derived estimation model or optimal parameters to the processing device 20 (3).

For example, the learning unit 84 changes the parameters in the membership functions, formulas (1), (2), and (3), and when the observation information of the radar device 10 is input, the membership functions, formula (1) , (2) and (3) are applied to derive an estimation model or optimal parameters such that the output information approaches the actually observed information (teaching data). The learning unit 84 derives an estimation model or optimal parameters using models such as neural networks, regression models, and support vector machines, for example. Further, the learning unit 84 may derive the optimum parameters by addition, subtraction, multiplication, and division without being limited to the above method.

[summary]
According to the present embodiment, by using data obtained from a dual-polarized phased-array weather radar that is observed at high frequency without gaps, spatio-temporally continuous snowfall and snowfall amounts can be obtained at intervals of several tens of seconds. It is possible to estimate the distribution and time change with high frequency and accuracy, and determine the risk of snow cover. Furthermore, by generating risk information by combining observational data of snow risk, land information, route information, and traffic information, it is possible to quickly and accurately notify and provide human and economic risks due to heavy snow. will be able to Specifically, by providing information through application software for the general public, SNS, etc., it is expected to help alert the general public and urge them to evacuate. In addition, it is expected that the system will effectively contribute to speeding up responses to heavy snowfall by local governments, road administrators, railway operation administrators, and bus operation administrators.

[Comparative example]
Here, the information obtained by the weather radar of the single-polarized wave radar cannot distinguish precipitation particles, and the radar cannot distinguish whether the particles falling on the ground are rain or snow. In the dual-polarized weather radar that appeared later, a technology was developed to distinguish precipitation particles using the polarization parameters obtained from observations. information has become available.

However, in order to improve the accuracy of these particle discrimination results, the main method has been to compare the information obtained by directly observing the inside of the cloud with the particle discrimination results analyzed from the information obtained by radar, and to adjust the parameters. However, since this method has been verified with a limited number of cases, it is unlikely that it always takes the optimal parameters for meteorological phenomena that change from moment to moment and have rich regional characteristics, especially for snowfall phenomena here. hard to say

Focusing on the amount of snowfall or amount of accumulated snow, in general, information at a specific point (such as an AMeDAS point) is provided only as spatially sparse information, and sufficient information is not available for traffic and civic life. It is hard to say that it is. In other words, there has not yet been established a means for continuously and two-dimensionally obtaining the amount of snowfall or amount of snow cover that changes from moment to moment.

Focus on the observation method of weather radar. The data obtained by the parabolic weather radar is a collection of two-dimensional data and is sparse in the vertical direction. Therefore, it is difficult to analyze with high accuracy when obtaining the physical quantity in the cloud, especially the quantity integrated in the vertical direction. In addition, it takes 4 to 10 minutes, for example, to obtain radar data necessary for three-dimensional analysis of precipitation clouds and precipitation systems. It was difficult to capture

Heavy snowfall, which causes disasters and causes economic loss, often increases the amount of snowfall in a short period of time, and it is important to predict the amount of snowfall in consideration of regional characteristics rather than precipitation. However, for the reasons described above, there is a problem that it is difficult to provide useful snowfall/snow cover information (that is, spatio-temporally continuous and highly accurate information).

The processing device 20 of the present embodiment effectively utilizes polarization parameter observation data that is obtained by a dual-polarization phased array weather radar and is observed without gaps at high frequency, and ground information that can be obtained by various means. Therefore, it is possible to provide spatio-temporally continuous and highly accurate snowfall and snow cover information based on radar information. This can be expected to improve the accuracy of judgments on disaster risks and risks that hinder economic activities.

For example, the processing unit 20 may derive information regarding snowfall for runways, railroad areas, and the like. As a result, the administrator who manages traffic conditions can recognize the possibility of snowfall that affects take-offs and landings of aircraft, railroad operations, and the like. Then, the administrator can adjust the flight schedules of airplanes and railroads in advance.

According to the embodiment described above, the acquisition unit 22 acquires the observation information indicating the state of the atmosphere for each of a plurality of divided regions obtained by dividing the observation region in the horizontal direction and the vertical direction, and the divided regions having the same horizontal position. A first estimation model for outputting information indicating a degree of snow cover on the ground corresponding to a target divided area when observation information of a target divided area selected from among the divided areas included in the area group is input, the first estimation model comprising: Observation information acquired by an acquisition unit 22 is input to a first estimation model having parameters learned so that information indicating the degree of snow cover associated with the observation information is derived when the observation information is input to the first estimation model. By inputting the derivation units (26, 28, 30, 32) for deriving information indicating the degree of snow cover on the ground, it is possible to derive information on snowfall with higher accuracy.

While several embodiments of the invention have been described, these embodiments have been presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, as well as the scope of the invention described in the claims and equivalents thereof.

1 processing system 10 radar device 20 processing device 22 acquisition unit 24 setting unit 26 processing unit 28 first derivation unit 30 first estimation unit 32 second derivation unit , 34 Second estimation unit 36 Risk derivation unit 38 Information transmission unit 40 Storage unit 42 Observation information 44 Estimation model 46 Optimal parameter 60 Display information generation device 62 Display Information generation unit 64 Display information transmission unit 80 Learning device 82 Acquisition unit 84 Learning unit 86 Providing unit 92 Learning data 100 Terminal device

Claims (10)

an acquisition unit that acquires observation information indicating the state of the atmosphere for each of a plurality of divided areas obtained by dividing an observation area in a distance direction, an azimuth direction, and a vertical direction;
When observation information of a target segmented region selected from among segmented regions included in a segmented region group having a common position in the distance direction and the azimuth direction is input, information indicating snowfall intensity corresponding to the target segmented region and has parameters learned so that information indicating snowfall intensity associated with the observation information is derived when the observation information is input to the second estimation model Deriving information indicating the snowfall intensity of the target divided area by inputting the observation information acquired by the acquisition unit into the second estimation model,
A first estimation model that, when information based on the snowfall intensity of the target divided area is input, outputs information indicating the degree of snow accumulation on the ground corresponding to the target divided area, wherein the snowfall intensity is applied to the first estimation model When the information based on the snowfall intensity is input, the first estimation model having parameters learned so as to derive information indicating the degree of snowfall associated with the information based on the snowfall intensity is added to the snowfall intensity of the target divided area a derivation unit that derives information indicating the degree of snow cover on the ground by inputting information based on
A processing device comprising: an acquisition unit that acquires observation information indicating atmospheric conditions for each of a plurality of divided areas obtained by dividing an observation area in a distance direction, an azimuth direction, and a vertical direction;
A predetermined number of vertically continuous divided areas are selected from divided areas included in a group of divided areas having common positions in the distance direction and the azimuth direction , and observation information of the predetermined number of target divided areas is input. Then, a third estimation model that outputs information indicating the degree of snow cover increase on the ground corresponding to the target divided area, wherein when the observation information is input to the third estimation model, the observation information is associated with By inputting the observation information acquired by the acquisition unit into the third estimation model having parameters learned so as to derive information indicating the degree of snow increase on the ground, the degree of snow increase on the ground is calculated. a derivation unit for deriving information indicating
A processing device comprising: an acquisition unit that acquires observation information indicating the state of the atmosphere for each of a plurality of divided areas obtained by dividing an observation area in a distance direction, an azimuth direction, and a vertical direction;
A predetermined number of vertically continuous divided areas are selected from divided areas included in a group of divided areas having common positions in the distance direction and the azimuth direction , and observation information of the predetermined number of target divided areas is input. a second estimation model for outputting information indicating the snowfall intensity corresponding to the target divided area, wherein when the observation information is input to the second estimation model, the snowfall intensity associated with the observation information Deriving information indicating the snowfall intensity of the target divided area by inputting the observation information obtained by the obtaining unit into the second estimation model having parameters learned so as to derive information indicating ,
A third estimation model for outputting information indicating a degree of snowfall increase on the ground corresponding to the target divided area when information based on the snowfall intensity of the target divided area is input, wherein the snowfall intensity is input to the third estimation model When the information based on the intensity is input, the third estimation model having parameters learned so as to derive information indicating the degree of snow accumulation increase on the ground associated with the information based on the snowfall intensity, the target division a derivation unit for deriving information indicating the degree of snow accumulation increase on the ground by inputting information based on the snowfall intensity of the area ;
A processing device comprising: The derivation unit is a fourth estimation model that outputs information indicating a type of particles contained in the target divided area when observation information of the target divided area is input, wherein the observation information is added to the fourth estimation model. By inputting the observation information acquired by the acquisition unit into the fourth estimation model trained so that information indicating the type of particle associated with the observation information is derived when is input, deriving information indicating the type of particles contained in the target divided area, and deriving information indicating the snowfall intensity of the target divided area when the derived type of particles is snow;
4. The processing apparatus according to claim 1 or 3 . Observation information indicating the state of the atmosphere for each of a plurality of divided areas obtained by dividing the observation area in the distance direction, the azimuth direction, and the vertical direction, and the degree of snow cover on the ground associated with each of the divided areas are associated with each other. an acquisition unit that acquires learning data or virtually generated learning data in which the observation information and the degree of snow cover on the ground are associated with each other;
A second estimation model that outputs information indicating snowfall intensity corresponding to a divided area corresponding to the observation information when the observation information is input, wherein the observation information of the learning data acquired by the acquisition unit is the Learning parameters to be given to the second estimation model so that information indicating the snowfall intensity associated with the divided regions corresponding to the observation information of the learning data is derived when input to the second estimation model ,
A first estimation model for outputting information indicating the degree of snow cover on the ground associated with the divided area corresponding to the information based on the snowfall intensity when the information based on the snowfall intensity is input, wherein the When the information based on the snowfall intensity obtained from the observation information is input to the first estimation model, it is associated with the divided regions corresponding to the information based on the snowfall intensity obtained from the observation information of the learning data. a learning unit that learns the parameters given to the first estimation model so as to derive information indicating the degree of snow cover on the ground;
By inputting the observation information into the second estimation model, information indicating the snowfall intensity associated with the divided area corresponding to the observation information is derived, and the information based on the snowfall intensity is transferred to the first estimation model. a providing unit that provides at least the parameter learned by the learning unit to a derivation unit that derives information indicating the degree of snow cover on the ground associated with the divided area corresponding to the snowfall intensity by inputting to
A processing device comprising: Observation information indicating the state of the atmosphere for each of a plurality of sub-areas obtained by dividing the observation area in the distance direction, the azimuth direction, and the vertical direction, and the degree of snow cover increase on the ground associated with each of the sub-areas are associated with each other. an acquisition unit that acquires learning data that has been obtained, or virtually generated learning data in which the observation information and the degree of increase in snow cover are associated with each other;
A third estimation model that, when the observation information is input, outputs information indicating the degree of snow cover increase on the ground associated with the divided area corresponding to the observation information, wherein the learning data acquired by the acquisition unit When the observation information is input to the third estimation model, the third estimation model is adapted to derive information indicating a degree of snow cover increase on the ground associated with the divided areas associated with the observation information of the learning data. 3 a learning unit that learns the parameters given to the estimation model;
By inputting the observation information into the third estimation model, a derivation unit that derives information indicating the degree of snow cover increase on the ground associated with the divided area corresponding to the observation information is learned by at least the learning unit. a provider that provides parameters for
A processing device comprising: the computer
Acquire observation information that indicates the state of the atmosphere for each of the multiple divided areas in which the observation area is divided in the distance direction, the azimuth direction, and the vertical direction,
When observation information of a target segmented region selected from among segmented regions included in a segmented region group having a common position in the distance direction and the azimuth direction is input, information indicating snowfall intensity corresponding to the target segmented region and has parameters learned so that information indicating snowfall intensity associated with the observation information is derived when the observation information is input to the second estimation model Deriving information indicating the snowfall intensity of the target divided area by inputting the acquired observation information into the second estimation model,
A first estimation model that, when information based on the snowfall intensity of the target divided area is input, outputs information indicating the degree of snow accumulation on the ground corresponding to the target divided area, wherein the snowfall intensity is applied to the first estimation model When the information based on the snowfall intensity is input, the first estimation model having parameters learned so as to derive information indicating the degree of snowfall associated with the information based on the snowfall intensity is added to the snowfall intensity of the target divided area Deriving information indicating the degree of snow cover on the ground by inputting information based on
Processing method. to the computer,
Acquiring observation information indicating the state of the atmosphere for each of a plurality of divided areas in which the observation area is divided in the distance direction, the azimuth direction, and the vertical direction,
When observation information of a target segmented region selected from among segmented regions included in a segmented region group having a common position in the distance direction and the azimuth direction is input, information indicating snowfall intensity corresponding to the target segmented region and has parameters learned so that information indicating snowfall intensity associated with the observation information is derived when the observation information is input to the second estimation model Deriving information indicating the snowfall intensity of the target divided area by inputting the acquired observation information into the second estimation model,
A first estimation model that, when information based on the snowfall intensity of the target divided area is input, outputs information indicating the degree of snow accumulation on the ground corresponding to the target divided area, wherein the snowfall intensity is applied to the first estimation model When the information based on the snowfall intensity is input, the first estimation model having parameters learned so as to derive information indicating the degree of snowfall associated with the information based on the snowfall intensity is added to the snowfall intensity of the target divided area Deriving information indicating the degree of snow cover on the ground by inputting information based on
program. the computer
Acquire observation information that indicates the state of the atmosphere for each of the multiple divided areas in which the observation area is divided in the distance direction, the azimuth direction, and the vertical direction,
A predetermined number of vertically continuous divided areas are selected from divided areas included in a group of divided areas having common positions in the distance direction and the azimuth direction , and observation information of the predetermined number of target divided areas is input. Then, a third estimation model that outputs information indicating the degree of snow cover increase on the ground corresponding to the target divided area, wherein when the observation information is input to the third estimation model, the observation information is associated with Deriving information indicating the degree of increase in snow cover by inputting the acquired observation information into the third estimation model having parameters learned so as to derive information indicating the degree of increase in snow cover on the ground.
Processing method. the computer
Acquire observation information that indicates the state of the atmosphere for each of the multiple divided areas in which the observation area is divided in the distance direction, the azimuth direction, and the vertical direction,
A predetermined number of vertically continuous divided areas are selected from divided areas included in a group of divided areas having common positions in the distance direction and the azimuth direction , and observation information of the predetermined number of target divided areas is input. a second estimation model for outputting information indicating the snowfall intensity corresponding to the target divided area, wherein when the observation information is input to the second estimation model, the snowfall intensity associated with the observation information Deriving information indicating the snowfall intensity of the target divided area by inputting the acquired observation information into the second estimation model having parameters learned so as to derive information indicating
A third estimation model for outputting information indicating a degree of snowfall increase on the ground corresponding to the target divided area when information based on the snowfall intensity of the target divided area is input, wherein the snowfall intensity is input to the third estimation model When the information based on the intensity is input, the third estimation model having parameters learned so as to derive information indicating the degree of snow accumulation increase on the ground associated with the information based on the snowfall intensity, the target division Deriving information indicating the degree of increase in snow cover by inputting information based on the snowfall intensity of the region ;
Processing method.

Images