Patent Application
20230022975
The present invention relates to an information processing device, a display device, an information processing method, and a program.
Priority is claimed on Japanese Patent Application No. 2020-020156, filed Feb. 7, 2020, the content of which is incorporated herein by reference.
Several hundred cases of lightning strikes occur annually in aircraft operations in Japan. Although a possibility that a lightning strike on an aircraft will directly lead to a serious accident is extremely low, several hundred million yen in costs are estimated to be incurred annually to repair damage to the outer fuselage of the aircraft and the like. Furthermore, because the inspection or emergency treatment for an aircraft that has been struck by lightning is time-consuming, even small-scale damage often leads to delays in the next flight. Furthermore, large-scale damage can lead to flight cancellations and have a significant impact on flight schedules.
Aircraft operations are broadly divided into a cruise phase and a takeoff/landing phase and individual weather information assistance technologies are used for each phase. As the weather information assistance for lightning in the cruise phase, information using a lightning monitoring system called a lightning detection network system (LIDEN) operated by the Japan Meteorological Agency (JMA) is widely used. Furthermore, the aircraft which is cruising are more likely to take evasive action and lightning strikes during the cruise phase are infrequent. On the other hand, it is estimated that more than 90 [%] of all lightning strikes occur during the takeoff/landing phase.
Technologies for quantitatively detecting a threat of lightning in relation to a lightning strike on such an aircraft are known (see, for example, Patent Literature 1 and 2).
Japanese Unexamined Patent Application, First Publication No. 2010-241412
Japanese Unexamined Patent Application, First Publication No. 2019-45403
However, it is difficult to predict a future threat of lightning capable of striking an aircraft in conventional technologies.
The present invention has been made in consideration of such circumstances and an objective of the present invention is to provide an information processing device, a display device, an information processing method, and a program capable of predicting a future threat of lightning.
According to an aspect of the present invention, there is provided an information processing device including: an acquirer configured to acquire at least one type of data within meteorological observation data indicating weather at a target location observed at each of a plurality of observation times and meteorological prediction data indicating the weather at the target location predicted using a meteorological prediction model; a deriver configured to derive an index value indicating a threat level of lightning of the target location at each of the plurality of observation times on the basis of at least one type of data within the meteorological observation data and the meteorological prediction data acquired by the acquirer; and a predictor configured to input the index value derived by the deriver to a model for outputting a future index value when a past or present index value is input and predict a threat of lightning of the target location at a future time later than the observation time on the basis of an output result of the model to which the index value has been input.
According to an aspect of the present invention, it is possible to predict a future threat of lightning.
Hereinafter, embodiments of an information processing device, a display device, an information processing method, and a program of the present invention will be described with reference to the drawings. In a case where the present application is translated from Japanese to English, as used throughout this disclosure, the singular forms “a,” “an,” and “the” include plural reference unless the context clearly dictates otherwise.
The meteorological observation device 10 observes weather at a target location using various sensors such as a meteorological radar and a radiosonde and generates data indicating an observation result (hereinafter referred to as meteorological observation data). Target locations are, for example, airports, areas near the airports, aircraft navigation routes, and areas near the aircraft navigation routes. The meteorological observation device 10 may be installed, for example, within or near a site of an airport where an aircraft takes off and lands, or may be mounted in the aircraft. Meteorological observation data includes physical quantities indicating atmospheric states such as an echo intensity (a precipitation intensity), a Doppler velocity (a wind speed), a wind direction, a temperature, and humidity. When the observation space is divided into a plurality of grid cells (also referred to as mesh cells), the physical quantity may be associated with each of the plurality of grid cells. For example, the grid may be divided into square grid cells at intervals of 5 [km] or 20 [km].
For example, the meteorological prediction device 20 predicts future weather at the target location from the meteorological observation data generated by the meteorological observation device 10 on the basis of the meteorological prediction model (also referred to as a numerical forecasting model) and generates data indicating a prediction result (hereinafter referred to as meteorological prediction data). The meteorological prediction data may include physical quantities whose types are the same as those of physical quantities included in the meteorological observation data such as an echo intensity (a precipitation intensity), a Doppler velocity (a wind speed), a wind direction, a temperature, and humidity. Like the meteorological observation data, the physical quantity in the meteorological prediction data may be associated with each of a plurality of grid cells into which the observation space is divided.
For example, the information processing device 100 may be installed within a site of an airport or may be mounted within an aircraft. The information processing device 100 acquires the meteorological observation data from the meteorological observation device 10 via the network NW and acquires the meteorological prediction data from the meteorological prediction device 20. The information processing device 100 predicts a future threat of lightning that may strike the aircraft on the basis of either one or both of the acquired meteorological observation data and the acquired meteorological prediction data. The information processing device 100 is an example of a “display device.”
Hereinafter, a configuration of the information processing device 100 will be described. The information processing device 100 may be a single device or a system in which a plurality of devices connected via the network NW operate in cooperation with each other. That is, the information processing device 100 may be implemented by a plurality of computers (processors) included in a system using distributed computing or cloud computing.
The communicator 102 includes, for example, a network interface card (NIC), a wireless communication module including a receiver and a transmitter, and the like. The communicator 102 communicates with the meteorological observation device 10, the meteorological prediction device 20, and other external devices via the network NW.
The display 104 is a user interface that displays various types of information. For example, the display 104 displays an image generated by the controller 110. The display 104 may display a graphical user interface (GUI) for receiving various types of input operations from users (for example, an airport staff member, a pilot, and the like). For example, the display 104 is a liquid crystal display (LCD), an organic electroluminescence (EL) display, or the like.
The controller 110 includes, for example, an acquirer 112, a deriver 114, a predictor 116, a model updater 118, and an output controller 120. The output controller 120 is an example of a “display controller.”
The components of the controller 110 are implemented by, for example, a processor such as a central processing unit (CPU) or a graphics processing unit (GPU) executing a program stored in the storage 130. Some or all of the components of the controller 110 may be implemented by hardware such as a large-scale integration (LSI) circuit, an application-specific integrated circuit (ASIC), or a field-programmable gate array (FPGA) or may be implemented by software and hardware in cooperation.
The storage 130 is implemented by, for example, a hard disc drive (HDD), a flash memory, an electrically erasable programmable read only memory (EEPROM), a read only memory (ROM), a random access memory (RAM), or the like. Various programs such as firmware and application programs are stored in the storage 130. In addition to the program to be referred to by the processor, the storage 130 stores threat level data D1, prediction model data D2, adjustment model data D3, and the like. These various types of data will be described below.
Hereinafter, a flow of a series of processing steps of the information processing device 100 will be described according to a flowchart.
First, the acquirer 112 acquires meteorological observation data from the meteorological observation device 10 and meteorological prediction data from the meteorological prediction device 20 via the communicator 102 (step S100). The acquirer 112 may acquire only one type of data within the meteorological observation data and the meteorological prediction data.
Subsequently, the deriver 114 derives an index value (hereinafter referred to as a threat level) indicating a threat level of lightning of a target location (near an airport or a navigation route) at an observation time on the basis of one or both types of data within the meteorological observation data and the meteorological prediction data acquired by the acquirer 112 (step S102). The observation time may be a time when the meteorological observation device 10 has observed the weather at the target location or may be a future time when the meteorological prediction device 20 has predicted the weather at the target location from the meteorological observation data.
For example, when the weather at the target location is observed at certain time A, the deriver 114 may derive a threat level of lightning of the target location at time A on the basis of one or both types of data within the meteorological observation data and the meteorological prediction data at time A.
The threat level is, for example, a discrete numerical value such as “small value (0.00 to 0.33)”, “medium value (0.33 to 0.66)”, and “large value (0.66 to 1.00).” The maximum value of the threat level is not limited to 1 and may be any value.
For example, the deriver 114 derives a probability that an aircraft landing at or passing over a target location will be struck by lightning, time loss (for example, operation delay time) or economic loss (for example, the level of damage to the aircraft, the cost of restoration, or the like) caused by the lightning strike as a threat level in the method described in Patent Literature 2. More specifically, the deriver 114 derives the probability of occurrence of a lightning strike, time loss, economic loss, or the like as the threat level of lightning using a certain function f(x). The threat level of lightning may be read as a level of an influence on human economic activities when an aircraft is struck by lightning.
When each physical quantity included in the meteorological observation data and the meteorological prediction data is input as an explanatory variable x, the function f(x) is a function for outputting the threat level of lightning (a probability of occurrence of the lightning strike, time loss, economic loss, or the like) as an objective variable, and may be, for example, a linear function such as f(x)=a1x1+a2x2+ . . . a1 and a2 are weighting coefficients. The function f(x) may include a bias component. The weight coefficient and the bias component may be decided on using a least-squares method on the basis of, for example, the meteorological observation data and meteorological prediction data observed when an aircraft has been actually struck by lightning and the time loss and economic loss caused by the lightning strike. The function f(x) may be implemented by a neural network as described in Patent Literature 2.
As described above, the deriver 114 inputs a multidimensional vector or tensor having each physical quantity as an element included in the meteorological observation data and the meteorological prediction data as an explanatory variable x to the function f(x) and derives a value output by the function f as the threat level of lightning. When the observation space including the target location is divided into a plurality of grid cells, the deriver 114 derives the threat level of lightning of each grid cell from physical quantities associated with the plurality of grid cells. The deriver 114 causes the storage 130 to store the threat level derived for each grid cell as threat level data D1.
The flowchart of
When the predictor 116 determines that the number of times the threat level has been derived has not reached the prescribed number of times, the process returns to S100. Thereby, the meteorological observation data and/or the meteorological prediction data at the observation time are sequentially acquired until the number of times the threat level has been derived reaches the prescribed number of times and the threat level of lightning is derived from the meteorological observation data and/or the meteorological prediction data each time.
When the predictor 116 determines that the number of times the threat level has been derived has reached the prescribed number of times, the predictor 116 predicts a threat level of lightning of the target location at a future time from a plurality of threat levels derived for the prescribed number of times (step S106). The future time mentioned here is a time later than any observation time of meteorological observation data or the like acquired for the prescribed number of times to derive the threat level of lightning. For example, when the meteorological observation data and the meteorological prediction data are sequentially acquired at times t1, t2, t3, . . . , t10, the future time may be future time t11 later than time t10, subsequent time t12, or the like.
For example, the predictor 116 predicts the threat level of lightning of a target location at a future time from a plurality of threat levels derived for the prescribed number of times using a prediction model MDL1 defined by the prediction model data D2.
[Equation. 1]
I
t+1
(i)
=a
(i)
I
t
(i)
+b
(i) (1)
In Eq. (1), It(i) denotes a threat level of certain grid cell i at time t and It+1(i) denotes a threat level of grid cell i after time t+1. a(i) denotes a weighting coefficient of grid cell i and b(i) denotes a bias component of grid cell i.
The autoregressive model for all grid cells can be expressed by Eqs. (2) and (3). Vectors are assumed to be expressed by symbols (→).
[Equation. 2]
I
t+1
=AI
t
+b (2)
[Equation. 3]
I
t=[It(0),It(1), . . . ,It(N)]T (3)
In Eq. (3), N denotes the total number of grid cells. A(→) denotes a matrix that is a set of weighting coefficients a of grid cells G0 to GN, and b(→) denotes a matrix that is a set of bias components b of grid cells G0 to GN. For example, the prediction model data D2 may define a polynomial function of an autoregressive model, the matrices A(→) and b(→), which are sets of coefficients of the function, and the like as the prediction model MDL1.
In the example of
The flowchart of
The screen illustrated in
The screen illustrated in
The screen illustrated in
On the other hand, in the present embodiment, as in the screen illustrated in
The flowchart of
When the end condition is not satisfied, the acquirer 112 acquires meteorological observation data of the weather observed by the meteorological observation device 10 in the observation corresponding to the time when the threat level of lightning has been predicted by the predictor 116 (hereinafter referred to as a prediction time) and meteorological prediction data of the weather predicted by the meteorological prediction device 20 using the meteorological observation data (step S112).
Subsequently, the deriver 114 derives a threat level of the target location at the prediction time on the basis of one or both types of data within the meteorological observation data and the meteorological prediction data at the prediction time acquired by the acquirer 112 (step S114).
Subsequently, the model updater 118 updates the prediction model MDL1 on the basis of at least one type of data within meteorological observation data and meteorological observation data at the prediction time acquired by the acquirer 112 in the processing of S114 and at least one type of data within meteorological observation data and meteorological observation data at a past observation time earlier than the prediction time acquired by the acquirer 112 in the processing of S100 (step S116). The model updater 118 returns the process to S106 and iteratively updates the prediction model MDL1 until the end condition is satisfied.
For example, the model updater 118 calculates a difference between the threat level predicted by the predictor 116 in the processing of S106 and the threat level derived by the deriver 114 in the processing of S114.
When the calculated difference is greater than or equal to a threshold value (when it can be considered that the prediction of the predictor 116 is wrong), the model updater 118 updates (decides on) a matrix A(→) of a weighting coefficient and a matrix b(→) of a bias component, which are parameters of the prediction model MDL1, using an adjustment model MDL2 defined by the adjustment model data D3. When the prediction model MDL1 is updated, the model updater 118 rewrites the prediction model data D2 of the storage 130 to a data in which the updated prediction model MDL1 is redefined.
Such an adjustment model MDL2 may be implemented by, for example, various models such as a neural network, a support vector machine, regularized regression, a random forest, and Gaussian process regression. Hereinafter, as an example, the adjustment model MDL2 will be described as being implemented by a neural network.
When the adjustment model MDL2 is implemented by the neural network, the adjustment model data D3 includes, for example, various types of information such as concatenation information indicating how units included in each of a plurality of layers constituting the neural network are concatenated to each other and a concatenation coefficient assigned to a data input/output between the concatenated units.
The concatenation information includes, for example, information of the number of units included in each layer, information for designating a type of unit to which each unit is concatenated, an activation function of each unit, a gate provided between the units in the hidden layer, and the like. The activation function may be, for example, a rectified linear unit (ReLU) function, a sigmoid function, a step function, another function, or the like. The gate selectively passes or weights data transmitted between the units, for example, in accordance with a value (for example, 1 or 0) returned by the activation function. The concatenation coefficient includes, for example, a weight given to the output data when data is output from a unit of a certain layer to a unit of a deeper layer in a hidden layer of a neural network. Also, the concatenation coefficient may include a bias component peculiar to each layer and the like.
It is assumed that the adjustment model MDL2 is sufficiently trained on the basis of, for example, training data. The training data is a data set in which a correct parameter to be output by the adjustment model MDL2 as a correct label (also referred to as a target) is associated with the input data such as the meteorological observation data and/or the meteorological observation data.
For example, the adjustment model MDL2 may be trained by decreasing the weights of the meteorological observation data and/or the meteorological observation data at the older observation time among the plurality of pieces of meteorological observation data and/or the plurality of pieces of meteorological observation data included as input data in the training data and increasing the weights of the meteorological observation data and/or the meteorological observation data at the newer observation time.
In this way, when the prediction model MDL1 outputs the future threat level and it can be considered that the prediction of the prediction model MDL1 is wrong from the output result, a process of updating the parameters of the prediction model MDL1 using the adjustment model MDL2 that has been sufficiently trained is iterated. Thereby, it is possible to accurately predict the future threat of lightning at the target location while sequentially adapting the prediction model MDL1 to an ever-changing meteorological environment.
In the example of
As illustrated in
As illustrated in
According to the embodiment described above, the information processing device 100 acquires meteorological observation data and meteorological prediction data at a plurality of observation times. The information processing device 100 derives a threat level of lightning of a target location at each of a plurality of observation times on the basis of at least one type of data within the acquired meteorological observation data and meteorological prediction data. When a past or present threat level is input, the information processing device 100 inputs the derived threat level to the prediction model MDL1 that outputs a future threat level, and predicts a threat of lightning of a target location at a future time later than an observation time on the basis of an output result of the prediction model MDL1. Thereby, aircraft pilots and the like can take evasive action on the basis of a result of predicting the threat of lightning and reduce economic loss and time loss due to the lightning strike.
Further, according to the above-described embodiment, the information processing device 100 updates the parameters of the prediction model MDL1 on the basis of at least one type of data within the meteorological observation data and the meteorological prediction data acquired after the time when the threat level of lightning is predicted and therefore it is possible to accurately predict the future threat of lightning at the target location while sequentially adapting the prediction model MDL1 to an ever-changing meteorological environment.
Hereinafter, modified examples of the above-described embodiment will be described. For example, when there are a plurality of target locations for which lightning threat prediction is required, the above-described adjustment model MDL2 may be generated for each target location.
While modes for carrying out the present invention have been described using embodiments, the present invention is not limited to such embodiments in any way and various modifications and replacements can be added without departing from the scope of the present invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2020-020156 | Feb 2020 | JP | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2021/003245 | 1/29/2021 | WO |
