CN114647985A - Training Method of Frontal Transit Prediction Model and Prediction Method of Frontal Transit - Google Patents

Abstract

The present invention provides a training method, a device and a storage medium for a forecasting model of frontal transit, and a method, a device and a storage medium for predicting the frontal transit. The training method of the frontal transit prediction model includes: establishing a neural network model; selecting a training sample set from the meteorological detection historical database, wherein the training sample set includes the meteorological data of a preset time period before the frontal transit occurs and the weather data without the frontal transit. Meteorological data; using the training sample set to train the neural network model to obtain a frontal transit prediction model, and the frontal transit prediction model is used to predict the frontal transit. In order to solve the technical problem that the frontal transit cannot be predicted in the prior art, the losses and risks caused by the wind shear caused by the frontal transit are prevented.

Description

Training Method of Frontal Transit Prediction Model and Prediction Method of Frontal Transit

technical field

The invention relates to the technical field of meteorological identification, in particular to a training method of a frontal transit prediction model and a frontal transit prediction method.

Background technique

Wind shear refers to the change of the wind speed vector or its components in the vertical direction or a certain horizontal direction. Wind shear is a vector value that reflects the change in wind speed and direction between the two points under study. In meteorology, low-level wind shear usually refers to wind shear below 600 meters above the ground. The formation of low-level wind shear requires certain weather background and environmental conditions. Thunderstorms, cumulonimbus clouds, tornadoes and other weather have strong convection, which can form strong vertical wind shear; strong downbursts reach the ground and spread to the surrounding gusts, which can form strong horizontal wind shear; The side meteorological elements are very different, and it is easy to produce strong wind shear.

Low-altitude wind shear has the characteristics of short life cycle, small scale and strong destructiveness, making it difficult to detect, warn and forecast, and is recognized as an "invisible killer" by meteorological departments of various countries. The six major types of weather processes that cause wind shear mainly include: frontal transit, land and sea wind, mountain wind, low-level jet, downburst, and turbulence. How to accurately detect the above weather processes is the key to improving the accuracy of wind shear early warning.

Front crossing refers to the phenomenon that a front passes through a certain place (such as an airport, etc.), and the meteorological elements on the ground change sharply at this time. For the meteorological change of the frontal transit, in the prior art, it is often judged by the meteorological data of the target area whether the weather of the frontal transit has occurred in the area. This method belongs to the post-event analysis. For some areas (such as airports), the delayed analysis cannot meet the actual forecasting needs. Therefore, a scheme that can predict the weather changes of the frontal transit in advance is urgently needed to prevent the frontal transit. Losses and risks from wind shear caused.

SUMMARY OF THE INVENTION

In view of this, the present invention provides a training method, device and storage medium for a frontal transit prediction model, and a frontal transit prediction method, device and storage medium, so as to solve the technical problem that the frontal transit cannot be predicted in the prior art.

In a first aspect, an embodiment of the present invention provides a training method for a frontal transit prediction model, including:

Build a neural network model;

A training sample set is selected from the meteorological detection historical database, wherein the training sample set includes the meteorological data of the preset time period before the frontal transit occurs and the meteorological data without the frontal transit;

The neural network model is trained by using the training sample set to obtain a front crossing prediction model, and the front crossing prediction model is used to predict front crossing.

In an optional embodiment, the building a neural network model includes:

Determine the output layer node of the neural network model according to the number of target categories of the front;

Determine the input layer node of the neural network model by using the forecast factor category in the selected meteorological detection historical database, wherein the forecast factor is meteorological change index data;

Determine the intermediate level nodes of the neural network model, and determine the initial weight set between the upper and lower level nodes.

In an optional implementation manner, if the intermediate layer node is a layer of hidden layer nodes, the initial weight set includes: the initial weight value from the input layer to the hidden layer, the initial weight value from the hidden layer to the hidden layer The initial weight value of the output layer.

In an optional embodiment, the selection of the training sample set from the meteorological detection historical database includes:

Obtain the surface observation data of the target area in the meteorological detection historical database;

The surface observation data of the target area is used as a training sample set.

In an optional embodiment, the neural network model is trained by using the training sample set to obtain a frontal transit prediction model, including:

Inputting the training sample set into the neural network model to obtain the corresponding current output set;

Calculate the error value of each output layer node between the current output set and the target expected output set;

Calculate the mean square error between the current output set and the target expected output set;

judging whether the mean square error is less than a preset threshold;

If the mean square error is not less than the preset threshold, it is determined that the current output set is invalid, and the weight set is modified by using the error value to obtain a modified weight set, and the training sample is returned to be executed. set input to the neural network model to obtain the corresponding current output set;

If the mean square error is less than the preset threshold, it is determined that the current output set is valid, and the corresponding weight set of the current frontal transit prediction model is determined as the target weight set to obtain the frontal transit prediction model.

In an optional implementation manner, the weight set is modified by using the error value to obtain a modified weight set, including:

Using the error value of the output layer to calculate the error value of the hidden layer;

The weight is modified according to the error value of the hidden layer to obtain the modified weight value.

In a second aspect, an embodiment of the present invention provides a method for predicting frontal transit, including:

Obtain the meteorological detection data of the target area;

The meteorological detection data is input into the front transit prediction model trained by the training method for the front transit prediction model described in any embodiment of the first aspect, so as to obtain the prediction result of the front transit.

In a third aspect, an embodiment of the present invention provides a training device for a frontal transit prediction model, including:

Create modules for building neural network models;

The sampling module is used to select a training sample set from the meteorological detection historical database, wherein the training sample set includes the meteorological data of the preset time period before the frontal transit occurs and the meteorological data without the frontal transit;

The training module is used to train the neural network model by using the training sample set to obtain a frontal transit prediction model, and the frontal transit prediction model is used to predict the frontal transit.

In a fourth aspect, an embodiment of the present invention provides a forecasting device for frontal transit, including:

The acquisition module is used to acquire the meteorological detection data of the target area;

The prediction module is used for inputting the meteorological detection data into the front transit prediction model trained by the training method of the front transit prediction model according to any embodiment of the first aspect, so as to obtain the prediction result of the front transit.

In a fifth aspect, an embodiment of the present invention provides a computer device, including: a memory and a processor, the memory and the processor are connected in communication with each other, the memory stores computer instructions, and the processor executes The computer instructions are used to implement the training method of the frontal transit prediction model described in any one of the first aspect, or realize the frontal transit prediction method described in the second aspect.

A sixth aspect is a non-transitory computer-readable storage medium provided according to an embodiment of the present invention, where the non-transitory computer-readable storage medium stores computer instructions, and the first aspect is implemented when the computer instructions are executed by a processor Any one of the methods for training a frontal transit prediction model, or implementing the method for predicting frontal transits described in the second aspect.

The training method, device, and storage medium for a frontal transit prediction model provided by the embodiments of the present invention have at least the following beneficial effects:

The training method, device, and storage medium for a frontal transit prediction model provided by the embodiment of the present invention, and the frontal transit prediction method, device, and storage medium provided by the embodiment of the present invention can create a neural network model and obtain the data from the meteorological detection historical database by creating a neural network model. A training sample set is selected, and the training sample set includes the meteorological data of the preset time period before the frontal transit occurs and the meteorological data without the frontal transit; using the selected training sample set to train the created neural network model, the frontal transit prediction model is obtained, The frontal transit is predicted by using the frontal transit prediction model. In the problem of forecasting front crossing, the corresponding meteorological data has a large number of nonlinear relationships. By creating and training a neural network model, a front crossing forecasting model is obtained to predict front crossing, thereby improving the prediction of front crossing. Accuracy and enhanced forecasting ability for frontal transit. By accurately predicting the front crossing, the losses and risks caused by wind shear caused by front crossing can be prevented.

Description of drawings

In order to illustrate the specific embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the specific embodiments or the prior art. Obviously, the accompanying drawings in the following description The drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

1 is a flowchart of a specific example of a training method for a front transit prediction model in Embodiment 1 of the present invention;

2 is a flowchart of another specific example of the training method of the front transit prediction model in Embodiment 1 of the present invention;

3 is a schematic structural diagram of a specific example of a BP neural network model in Embodiment 1 of the present invention;

4 is a flowchart of another specific example of a training method for a frontal transit prediction model in Embodiment 1 of the present invention;

5 is a schematic diagram of the convergence of the mean square error with the training rounds of a specific example in Embodiment 1 of the present invention;

FIG. 6 is a flowchart of a specific example of a method for predicting frontal transit in Embodiment 2 of the present invention;

FIG. 7 is a sample diagram of a recognition result of a frontal transit prediction model of a specific example in Embodiment 2 of the present invention;

8 is a block diagram of a specific example of a training device for a front transit prediction model in Embodiment 3 of the present invention;

FIG. 9 is a block diagram of a specific example of a device for predicting front crossing in Embodiment 4 of the present invention;

FIG. 10 is a schematic structural diagram of a computer device according to a specific example in Embodiment 5 of the present invention.

Detailed ways

The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. The indicated orientation or positional relationship is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the indicated device or element must have a specific orientation or a specific orientation. construction and operation, and therefore should not be construed as limiting the invention. Furthermore, the terms "first", "second", and "third" are used for descriptive purposes only and should not be construed to indicate or imply relative importance.

In the description of the present invention, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense, unless otherwise expressly specified and limited, for example, it may be a fixed connection or a detachable connection connection, or integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal connection of two components, which can be a wireless connection or a wired connection connect. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood in specific situations.

In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

Although the processes described below include a number of operations in a particular order, it should be expressly understood that the processes may include more or fewer operations, which may be performed sequentially or in parallel.

Front crossing refers to the phenomenon that a front passes through a certain place (such as an airport, a high-speed railway station, etc.), and the ground meteorological elements change rapidly at this time. For the meteorological change of the frontal transit, in the prior art, it is often judged by the meteorological data of the target area whether the weather of the frontal transit has occurred in the area. This method belongs to post-event analysis. For some areas (including but not limited to defining airports), taking airports as an example, the delayed analysis cannot meet the actual forecasting needs. In other words, the amount of meteorological elements corresponding to the frontal transit changes sharply, which poses a serious threat to the aircraft in the take-off or landing phase. If we only analyze whether frontal transit occurs in the airport area after the event, instead of predicting frontal transit in advance, we cannot effectively prevent frontal transit, which will cause huge losses, and there is also a high risk of aircraft landing and take-off.

Therefore, there is an urgent need for a scheme that can predict the weather changes of the frontal transit in advance, so as to prevent the losses and risks caused by the wind shear caused by the frontal transit.

Example 1

This embodiment provides a training method for a frontal transit prediction model, as shown in FIG. 1 , including the following steps:

Step S101, establishing a neural network model;

Step S102, selecting a training sample set from the meteorological detection historical database, wherein the training sample set includes meteorological data of a preset time period before frontal transit occurs and meteorological data without frontal transit;

Step S103 , using the training sample set to train the neural network model to obtain a frontal transit prediction model, and the frontal transit prediction model is used to predict the frontal transit.

In this embodiment, the established neural network model may be a BP neuron network model. The BP neuron network model is based on a multi-layer feedforward principle trained by an error back-propagation algorithm, and is currently widely used. Of course, the established neural network model may also be other neural network models, such as an RBF neuron network model. The meteorological detection history database may include meteorological detection history data corresponding to each region. When selecting the training sample set, it is necessary to select the meteorological data of the preset time period before the frontal transit and the meteorological data of no frontal transit from the meteorological detection historical database. The neural network model is trained using the selected training sample set, and the frontal transit prediction model is obtained. In practical applications, the meteorological detection data of the target area can be input into the trained frontal transit prediction model, so as to obtain the prediction result of frontal transit, and then predict the frontal transit.

In this embodiment, taking the application to an airport as an example, after establishing a BP neural network model, the historical meteorological detection data corresponding to the airport area is selected from the meteorological detection historical database as a training sample set. For example, if you select all the meteorological data within two years in the airport area, including the meteorological data of the preset time period before the frontal transit occurs and the meteorological data of the no frontal transit, the preset time period can be 0.5 hours. The BP neuron network model is trained, and the frontal transit prediction model is obtained. The meteorological detection data of the airport area is input into the frontal transit prediction model to obtain the prediction result of frontal transit, and then the frontal transit is predicted. In practical applications, it is necessary to flexibly choose to create a neural network model according to the needs of the actual application scenario, and select the corresponding training sample set for training to obtain a frontal transit prediction model to predict frontal transit, which is not limited in this application. .

In this embodiment, a neural network model is created, and a training sample set is selected from the meteorological detection historical database, and the training sample set includes the meteorological data of the preset time period before the frontal transit occurs and the meteorological data without the frontal transit; The created neural network model is trained with the training sample set of , and the frontal transit prediction model is obtained, so as to use the frontal transit prediction model to predict the frontal transit. In the problem of forecasting front crossing, the corresponding meteorological data has a large number of nonlinear relationships. By creating and training a neural network model, a front crossing forecasting model is obtained to predict front crossing, thereby improving the prediction of front crossing. Accuracy and enhanced forecasting ability for frontal transit. By accurately predicting the front crossing, the losses and risks caused by wind shear caused by front crossing can be prevented.

In an optional embodiment, referring to Fig. 2, a neural network model is established, including:

Step S1011, determining the output layer node of the neural network model according to the number of target categories of the front;

Step S1012, determining the input layer node of the neural network model by using the forecast factor category in the selected meteorological detection historical database, wherein the forecast factor is meteorological change index data;

Step S1013: Determine the intermediate level nodes of the neural network model, and determine the initial weight set between the upper and lower level nodes.

In this embodiment, the established neural network model is a BP neuron network model. Referring to Figure 3, the BP neural network model is to add several layers (one or more layers) of neurons between the input layer and the output layer. These neurons are called hidden units. There is no direct connection with the outside world, but the change of its state can affect the relationship between input and output. Each layer can have several nodes. Figure 3 shows a hidden unit layer.

In this embodiment, the output layer node of the neural network model is determined according to the number of target categories of the front. The number of nodes in the output layer is determined according to the number of target categories of the front. Exemplarily, the number of target categories of the front is determined to be two, one of which is used to represent the front, that is, the corresponding front index is 1, and the other is a non-front index. In the subsequent prediction and estimation stage, the closer the frontal index is to 1, the higher the possibility that the moment is considered to be a frontal transit. The input layer node of the neural network model is determined by using the selected forecast factor category in the meteorological detection historical database, wherein the forecast factor is meteorological change index data. Further, the selection of the forecast factor category can comprehensively consider the layout of meteorological detection and the change law of meteorological factors when the front passes. Optionally, the specifically selected predictor categories and their descriptions are shown in Table 1-1 below. After the input layer nodes of the neural network model and the output layer nodes of the neural network model are determined, the intermediate level nodes of the neural network model are further determined, wherein the determination of the intermediate level nodes includes the determination of the number of intermediate levels and each intermediate level. The corresponding number of nodes is determined, and then the initial weight set between the upper and lower level nodes is determined.

Table 1-1: Predictor categories and their descriptions

predictor illustrate wind speed Wind speed at each station at the current moment Time gradient of wind speed component The difference between the current time and the historical time wind speed component of each station Spatial Gradient of Wind Speed Component Difference of wind speed components between stations at the current moment spatial gradient of wind direction Wind direction difference between sites wind flux Wind vector flux through the polygon enclosed by all stations Wind field circulation Wind vector circulation along the polygon enclosed by all stations Turbulence factor Turbulent kinetic energy calculated from wind vectors at each station temperature time gradient The temperature difference between the current time and the historical time of each station Temperature Spatial Gradient temperature difference between sites Humidity time gradient The humidity difference between the current moment and the historical moment of each station Humidity Spatial Gradient Humidity difference between sites pressure time gradient The air pressure difference between the current time and the historical time of each station pressure spatial gradient Air pressure difference between sites

In this embodiment, the cover index is used as the output layer node of the neural network model to ensure the data reliability of the output layer node of the neural network model, thereby ensuring the reliability of the created neural network model. By determining the input layer nodes of the neural network model according to the predictor categories,

Selecting the predictor category that has the expected correlation with the output layer node of the neural network model and reducing the input amount of the output layer node of the neural network model can reduce the training difficulty of the neural network model and improve the training efficiency of the neural network model. This further improves the accuracy of frontal transit prediction and enhances the ability to predict frontal transit. By accurately predicting the front crossing, the losses and risks caused by wind shear caused by front crossing can be prevented.

In an optional implementation manner, as shown in FIG. 3 , if the intermediate layer node is a layer of hidden layer nodes, the initial weight set includes: the initial weight from the input layer to the hidden layer value, the initial weight value from the hidden layer to the output layer.

In this embodiment, a BP neural network model with a three-layer structure is adopted, the number of nodes in the input layer x (Input Layer) is N, the number of nodes in the hidden layer h (Hidden Layer) is H, and the number of nodes in the output layer y (Output Layer) The number is M.

Set the initial weight value of the BP neuron network model to a random small value.

Among them, the initial weight value from the input layer to the hidden layer is:

W _ij (0)(i=1,2,3,...,N; j=1,2,3,...,H);

The initial weights from the hidden layer to the output layer are:

W _jk (0) (j=1, 2, 3, ..., H; k=1, 2, 3, ..., M).

Among them, W _ij represents the weight value from the input layer i node to the hidden layer j node, and W _jk represents the weight value from the hidden layer j node to the output layer k node.

Further, after the initial weight set is determined, the signal forward transmission processing principle of the BP neuron network model is determined. For example, take a sample X ₁ (x ₁ ,x ₂ ,x ₃ ,x ₄ ,…x _N ) and specify its expected output D ₁ (d ₁ ,d ₂ ,d ₃ ,d ₄ ,…d _M ), where 0≤d _k ≤1, (k=1, 2, 3,...M), where d _k represents the expected output value of the k node in the output layer.

Calculate the actual output of the BP neuron network model:

Among them, y _k is the actual output value of the k node in the output layer, h _j is the actual output value of the j node in the hidden layer, x _i is the actual input value of the i node in the input layer, and W _ij represents the input layer from the i node to the The weight value of the hidden layer j node, W _jk represents the weight value of the hidden layer j node to the output layer k node; θ _i represents the weight offset coefficient corresponding to the input layer i node, θ _j represents the hidden layer j node corresponding Weight offset factor.

In this embodiment, by setting the middle layer node as a layer of hidden layer nodes, and only the middle layer node of one layer is a layer of hidden layer nodes, the training difficulty of the neuron network model can be reduced, and the training of the neuron network model can be improved. efficiency. This further improves the accuracy of frontal transit prediction and enhances the ability to predict frontal transit. By accurately predicting the front crossing, the losses and risks caused by wind shear caused by front crossing can be prevented.

In an optional embodiment, the selection of the training sample set from the meteorological detection historical database includes:

Obtain the surface observation data of the target area in the meteorological detection historical database;

The surface observation data of the target area is used as a training sample set.

In this embodiment, the surface observation data of the target area in the meteorological detection historical database is obtained, and the surface observation data of the target area may include the surface meteorological data monitored by the automatic meteorological observation system, such as the monitored surface meteorological data of the AWOS airport meteorological automatic observation system. Surface meteorological data. The surface observation data of the target area is used as the training sample set.

For example, select the meteorological detection historical data corresponding to the airport area from the meteorological detection historical database as the training sample set, then obtain the monitored surface meteorological data corresponding to the airport's AWOS meteorological automatic observation system in the meteorological detection historical database, and determine the training sample. set.

The surface observation data of the target area is used as the training sample set to ensure the richness and rationality of the corresponding training cases in the training sample set. Furthermore, the training difficulty of the neuron network model is further reduced, the training efficiency of the neuron network model is improved, the accuracy of the forecasting of the frontal transit is improved, and the forecasting ability of the frontal transit is enhanced. It is beneficial to accurately predict the front crossing, and then prevent the losses and risks caused by the wind shear caused by the front crossing.

In an optional embodiment, as shown in FIG. 4 , the neural network model is trained by using the training sample set to obtain a frontal transit prediction model, including:

Step S1031, inputting the training sample set into the neural network model to obtain a corresponding current output set;

Step S1032, calculating the error value of each output layer node in the current output set and the target expected output set;

Step S1033, calculating the mean square error between the current output set and the target expected output set;

Step S1034, judging whether the mean square error is less than a preset threshold;

Step S1035: If the mean square error is not less than the preset threshold, it is determined that the current output set is invalid, and the weight set is modified by using the error value to obtain a modified weight set, and the execution is returned to execute the set of weights. The training sample set is input into the neural network model, and the step of obtaining the corresponding current output set;

Step S1036: If the mean square error is less than the preset threshold, determine that the current output set is valid, and determine the corresponding weight set of the current frontal transit prediction model as the target weight set, and obtain the frontal transit prediction model.

In this embodiment, the training sample set is input into the neural network model to obtain the corresponding current output set, and then the corresponding mean square error is calculated according to the error value of each output layer node between the current output set and the target expected output set, If the mean square error does not meet the convergence requirement, then use the error value of each output layer node between the current output set and the target expected output set to modify the weight set repeatedly until the mean square error does not meet the convergence requirement, that is, when the mean square error converges to After the corresponding threshold, the training of the neural network model is completed, and the frontal transit prediction model is obtained. Illustratively, as shown in FIG. 5 , the mean square error decreases rapidly at the beginning, and then the rate of error decrease slows down and gradually stabilizes. According to the change of the mean square error, it also indirectly reflects that there is a high correlation between the input layer nodes of the selected neural network model and the output layer expectations, which does not cause trouble to the training of the neural network model, and the training efficiency of the neural network model. It is relatively high, and the final convergence result of the mean square error also has a relatively ideal accuracy.

In this embodiment, the corresponding current output set is obtained by inputting the training sample set into the neural network model, and the corresponding mean square is obtained according to the error value of each output layer node in the obtained current output set and the target expected output set. Error, take the comparison result of the mean square error and the preset threshold as the condition of whether to perform repeated training, and use the error value of each output layer node between the current output set and the target expected output set to correct the weight set until the mean square error is less than Preset threshold. The training efficiency of the neural network model is improved, the accuracy of the forecasting of the frontal transit is improved, and the forecasting ability of the frontal transit is enhanced. It is beneficial to accurately predict the front crossing, and then prevent the losses and risks caused by the wind shear caused by the front crossing.

In an optional implementation manner, the weight set is modified by using the error value to obtain a modified weight set, including:

Using the error value of the output layer to calculate the error value of the hidden layer;

The weight is modified according to the error value of the hidden layer to obtain the modified weight value.

In this embodiment, the input values of the corresponding training samples in the training sample set are loaded into the neural network model, the corresponding output values are obtained, and the output values are compared with the expected output values corresponding to the samples. When the output values are inconsistent with the expected values , the weight set needs to be modified.

For the output layer, the error value of the output layer is:

δ _Ok =y _k (1-y _k )(d _k -y _k )

Using the error value of the output layer to calculate the error value of the hidden layer, the error value of the hidden layer is:

Among them, δ _Ok represents the error value of node k in the output layer, d _k represents the expected output value of node k in the output layer; y _k represents the actual output value of node k in the output layer; δ _Hj represents the error value of node j in the hidden layer, h _j represents the actual output value of the j node in the hidden layer; w _jk represents the weight value from the hidden layer j node to the output layer k node;

The weight is corrected according to the error value of the hidden layer, and the corrected weight value is obtained. When correcting the weights in turn from the output layer to the input layer, the formula for changing the weights from the input layer to the hidden layer is:

w _ij (t+1)=w _ij (t)+αδ _Hj x _i +β[w _ij (t)-w _ij (t-1)]

Among them, α is called the learning parameter, which controls the search step size, and generally takes 0.2≤α≤0.5; β is called the momentum parameter, which controls the smoothness, and generally takes 0≤β≤1; t is the current number of iterations.

This completes a round of training of the connection weights between the sample X ₁ and the expected output D ₁ , repeating the second and third steps to complete the expected output D ₂ corresponding to the samples X ₂ , X ₃ ,...,X _N , One round of training of D ₃ ,...,D _N pairs of connection weights.

小于一个阈值S。Repeat the above steps, using the input value of the selected sample data and the expected output to repeatedly train the weight data until the mean square error

less than a threshold S.

The training sample set is input to the neural network model by correcting the weights in turn from the output layer to the input layer, which improves the training efficiency of the neural network model, improves the accuracy of frontal transit prediction, and enhances the frontal transit prediction ability. . It is beneficial to accurately predict the front crossing, and then prevent the losses and risks caused by the wind shear caused by the front crossing.

Example 2

This embodiment provides a method for predicting frontal transit, as shown in FIG. 6 , including the following steps:

Step S601, obtaining meteorological detection data of the target area;

Step S602, inputting the meteorological detection data into the frontal transit prediction model trained by the training method of the frontal transit prediction model, so as to obtain the prediction result of the frontal transit, wherein the training method of the frontal transit prediction model is the one in Embodiment 1. The training method for a frontal transit prediction model according to any one of the embodiments.

In this embodiment, the meteorological detection data of the target area is obtained, and the meteorological detection data of the target detection area can be real-time detection data, such as the meteorological data monitored by the AOWS airport meteorological automatic observation system, and the meteorological detection data is input into the front. The training method of the transit prediction model is trained in the front transit prediction model obtained by training, so as to obtain the prediction result of the front transit. Figure 7 shows a sample of the recognition results based on the frontal transit prediction model. Exemplarily, use the meteorological data monitored by the AWOS Airport Meteorological Automatic Observation System of an airport for one year to simulate and test the front transit prediction model obtained by training, and use the results obtained from the aerial report records of the aircraft combined with the manual identification as the standard comparison ( Since the air report is only issued when there are aircraft taking off and landing, and there is a lack of frontal transit records that occurred during the absence of aircraft taking off and landing, the manual identification method is used to supplement it. The results are shown in Table 1-2.

Table 1-1: Comparison of the recognition results of the frontal transit prediction model and the aerial report combined with the manual recognition results

For the 51 frontal transit weather processes identified by aerial reports combined with manual identification, the frontal transit prediction model identified 36 times, and the detection rate (Probability of Detection, POD) reached 70.6%; among the 61 frontal transits identified by the frontal transit prediction model, There were 26 errors in identification, and the false alarm rate (FAR) was 42.6%. According to the simulation test results, it can be seen that the prediction model of frontal transit has a higher accuracy in predicting frontal transit.

In this implementation manner, the frontal transit prediction is performed by using the trained frontal transit prediction model, so as to realize the ability to predict the frontal transit. It is beneficial to accurately predict the front crossing, and then prevent the losses and risks caused by the wind shear caused by the front crossing.

Example 3

This embodiment provides a training device for a frontal transit prediction model. This embodiment is described by applying the training device for the frontal transit prediction model to the training method for the frontal transit prediction model described in the above-mentioned Embodiment 1. As shown in Figure 8, the training device of the frontal transit prediction model includes at least the following modules:

A creation module 81 is used to establish a neural network model;

The sampling module 82 is configured to select a training sample set from the meteorological detection historical database, wherein the training sample set includes the meteorological data of the preset time period before the frontal transit occurs and the meteorological data without the frontal transit;

The training module 83 is configured to use the training sample set to train the neural network model to obtain a frontal transit prediction model, and the frontal transit prediction model is used to predict the frontal transit.

The training device for the frontal transit prediction model provided by the embodiment of the present application can be used for the training method of the frontal transit prediction model as performed in the above embodiment 1. For relevant details, refer to the above method embodiments, and the implementation principles and technical effects thereof are similar. Repeat.

It should be noted that: when the training device for the frontal transit prediction model provided in the above-mentioned embodiment is to perform the training of the frontal transit prediction model, only the division of the above-mentioned functional modules is used as an example for illustration. The function allocation is completed by different function modules, that is, the internal structure of the training device of the frontal transit prediction model is divided into different function modules, so as to complete all or part of the functions described above. In addition, the training device of the frontal transit prediction model provided in the above embodiment and the embodiment of the training method of the frontal transit prediction model belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

Example 4

This embodiment provides an apparatus for predicting frontal transit, and this embodiment is described by applying the frontal transit prediction apparatus to the method for predicting frontal transit described in Embodiment 2 above. As shown in Figure 9, the forecasting device for the frontal transit at least includes the following modules:

an acquisition module 91, used for acquiring meteorological detection data of the target area;

The prediction module 92 is configured to input the meteorological detection data into the frontal transit prediction model trained by the training method of the frontal transit prediction model, so as to obtain the prediction result of the frontal transit, wherein the training method of the frontal transit prediction model is to implement The training method of the frontal transit prediction model according to any one of the embodiments in Example 1.

The device for predicting front crossing provided by this embodiment of the present application can be used in the method for predicting front crossing performed in the above embodiment 2. For relevant details, refer to the above method embodiments, and the implementation principles and technical effects thereof are similar, and are not repeated here.

It should be noted that: when the forecasting device for frontal transit provided in the above-mentioned embodiment performs the forecasting of frontal crossing, only the division of the above-mentioned functional modules is used as an example for illustration. The functional modules of the frontal transit are completed, that is, the internal structure of the forecasting device for frontal transit is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the device for predicting front crossing provided by the above embodiments and the method for predicting front crossing belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

Example 5

Referring to FIG. 10 , an embodiment of the present invention further provides a computer device, which may be a desktop computer, a notebook computer, a palmtop computer, a cloud server, and other computer devices. The computer device may include, but is not limited to, a processor and a memory, where the processor and memory may be connected by a bus or otherwise.

The processor may be a central processing unit (Central Processing Unit, CPU) or other general-purpose processors, a digital signal processor (Digital Signal Processor, DSP), a graphics processor (GraphicsProcessing Unit, GPU), an embedded neural network processor (Neural-network Processing Unit, NPU) or other dedicated deep learning co-processors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices , discrete gate or transistor logic devices, discrete hardware components and other chips, or a combination of the above types of chips.

As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs and modules, such as program instructions/modules corresponding to the methods in the foregoing method embodiments. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, ie, implements the methods in the above method embodiments.

The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system and an application program required by at least one function; the storage data area may store data created by the processor, and the like. Additionally, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory may optionally include memory located remotely from the processor, such remote memory being connectable to the processor via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The one or more modules are stored in the memory, and when executed by the processor, perform the methods in the above method embodiments.

Embodiments of the present invention further provide a non-transitory computer-readable storage medium, where the non-transitory computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions can execute the methods in the foregoing method embodiments. Wherein, the non-transitory computer-readable storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a flash memory (FlashMemory) , Hard Disk Drive (Hard Disk Drive, abbreviation: HDD) or Solid-State Drive (Solid-State Drive, SSD), etc.; the non-transitory computer-readable storage medium may also include a combination of the above-mentioned types of memories.

As will be appreciated by those skilled in the art, embodiments of the present invention may be provided as methods, apparatuses, computer devices, or non-transitory computer-readable storage media, all of which may involve or include a computer program product.

Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The technical features of the above-described embodiments can be combined arbitrarily. In order to simplify the description, all possible combinations of the technical features in the above-described embodiments are not described. However, as long as there is no contradiction between the combinations of these technical features, All should be regarded as the scope described in this specification.

Obviously, the above-mentioned embodiments only represent several embodiments of the present application, and the descriptions thereof are relatively specific and detailed, but should not be construed as a limitation on the scope of the invention patent. It should be noted that, for those skilled in the art, changes or modifications in other different forms can also be made on the basis of the above description without departing from the concept of the present application. There is no need and cannot be exhaustive of all implementations here. And the obvious changes or changes derived from this are still within the protection scope of the present invention. Therefore, the scope of protection of the patent of the present application shall be subject to the appended claims.

Claims (10)

1. a kind of training method of frontal transit prediction model, is characterized in that, comprises: Build a neural network model; A training sample set is selected from the meteorological detection historical database, wherein the training sample set includes the meteorological data of the preset time period before the frontal transit occurs and the meteorological data without the frontal transit; The neural network model is trained by using the training sample set to obtain a front crossing prediction model, and the front crossing prediction model is used to predict front crossing. 2. training method according to claim 1, is characterized in that, described establishing neural network model, comprises: Determine the output layer node of the neural network model according to the number of target categories of the front; Determine the input layer node of the neural network model by using the forecast factor category in the selected meteorological detection historical database, wherein the forecast factor is meteorological change index data; Determine the intermediate level nodes of the neural network model, and determine the initial weight set between the upper and lower level nodes. 3 . The training method according to claim 2 , wherein if the intermediate layer node is a layer of hidden layer nodes, the initial weight set comprises: the initial value from the input layer to the hidden layer. 4 . Weight value, initial weight value from hidden layer to output layer. 4. training method according to claim 1, is characterized in that, described selecting training sample set from meteorological detection historical database, comprises: Obtain the surface observation data of the target area in the meteorological detection historical database; The surface observation data of the target area is used as a training sample set. 5. The training method according to claim 2, wherein the neural network model is trained by using the training sample set to obtain a frontal transit prediction model, comprising: Inputting the training sample set into the neural network model to obtain the corresponding current output set; Calculate the error value of each output layer node between the current output set and the target expected output set; Calculate the mean square error between the current output set and the target expected output set; judging whether the mean square error is less than a preset threshold; If the mean square error is not less than the preset threshold, it is determined that the current output set is invalid, and the weight set is modified by using the error value to obtain a modified weight set, and the training sample is returned to be executed. set input to the neural network model to obtain the corresponding current output set; If the mean square error is less than the preset threshold, it is determined that the current output set is valid, and the corresponding weight set of the current frontal transit prediction model is determined as the target weight set to obtain the frontal transit prediction model. 6. The training method according to claim 5, wherein the weight set is modified by using the error value to obtain a modified weight set, comprising: Using the error value of the output layer to calculate the error value of the hidden layer; The weight is modified according to the error value of the hidden layer to obtain the modified weight value. 7. A method for predicting frontal transit, comprising: Obtain the meteorological detection data of the target area; The meteorological detection data is input into the frontal transit prediction model trained by the training method of the frontal transit prediction model according to any one of claims 1-6, so as to obtain the prediction result of the frontal transit. 8. A training device for a frontal transit prediction model, characterized in that it comprises: Create modules for building neural network models; The sampling module is used to select a training sample set from the meteorological detection historical database, wherein the training sample set includes the meteorological data of the preset time period before the frontal transit occurs and the meteorological data without the frontal transit; The training module is used to train the neural network model by using the training sample set to obtain a frontal transit prediction model, and the frontal transit prediction model is used to predict the frontal transit. 9. A forecasting device for frontal transit, characterized in that it comprises: The acquisition module is used to acquire the meteorological detection data of the target area; The forecasting module is used for inputting the meteorological detection data into the front crossing forecasting model trained by the training method of the front crossing forecasting model according to any one of claims 1-6, so as to obtain the forecasting result of front crossing. 10. A non-transitory computer-readable storage medium, characterized in that, the non-transitory computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, any one of claims 1-6 is implemented. A method for training the frontal transit prediction model, or implementing the frontal transit prediction method as claimed in claim 7.

Images