CN114594443A - Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network - Google Patents

Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network Download PDF

Info

Publication number
CN114594443A
CN114594443A CN202210286562.1A CN202210286562A CN114594443A CN 114594443 A CN114594443 A CN 114594443A CN 202210286562 A CN202210286562 A CN 202210286562A CN 114594443 A CN114594443 A CN 114594443A
Authority
CN
China
Prior art keywords
radar echo
self
layer
causal
attention mechanism
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210286562.1A
Other languages
Chinese (zh)
Inventor
方巍
薛琼莹
沈亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Information Science and Technology
Original Assignee
Nanjing University of Information Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Information Science and Technology filed Critical Nanjing University of Information Science and Technology
Priority to CN202210286562.1A priority Critical patent/CN114594443A/en
Publication of CN114594443A publication Critical patent/CN114594443A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/95—Radar or analogous systems specially adapted for specific applications for meteorological use
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

The invention provides a meteorological radar echo extrapolation method and system based on a self-attention mechanism and a prediction recurrent neural network, wherein the method comprises the following steps: acquiring a meteorological radar echo image; preprocessing the acquired meteorological radar echo image to acquire gray data of the meteorological radar echo image; obtaining a Self-attention PredRNN + + network model based on a Self-attention mechanism and PredRNN + +; and obtaining a weather radar echo extrapolation image sequence through a weather radar echo image sequence based on a Self-attention PredRNN + + network model. The invention can extrapolate a radar echo image sequence of 2 hours in the future according to the existing radar echo image.

Description

Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network
Technical Field
The invention relates to a meteorological radar echo extrapolation method and system based on a self-attention mechanism and a prediction recurrent neural network, and belongs to the technical field of radar echo extrapolation short-term prediction.
Background
For short-term forecasting, two methods, namely an extrapolation technology based on radar echo and a numerical weather forecasting mode, mainly exist at present, but both have respective problems, such as insufficient spatial information, difficulty in capturing characteristics of a deep network, various constructed model parameters, low accuracy of radar echo extrapolation and the like.
Deep learning is essentially a deep neural network that stacks multiple hidden layers and maps the original features to a high-dimensional feature space through nonlinear transformation. Such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long-Short Term Memory Networks (LSTM).
However, the existing RNN, CNN or LSTM deformation model is difficult to capture a long-time image, and lacks a structure capable of learning global spatiotemporal correlation information from an input radar image, so that it is difficult to capture the features of a deep network.
Therefore, the application provides a meteorological radar echo extrapolation method and system based on a self-attention mechanism and a prediction recurrent neural network, so as to solve the technical problems in the field at the present stage.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides a meteorological radar echo extrapolation method and system based on an attention mechanism and a prediction recurrent neural network, which can extrapolate a radar echo image sequence for 2 hours in the future according to the existing radar echo image.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
in one aspect, the invention provides a meteorological radar echo extrapolation method based on a self-attention mechanism and a prediction recurrent neural network, which comprises the following steps:
acquiring a meteorological radar echo image;
preprocessing the acquired meteorological radar echo image to acquire gray data of the meteorological radar echo image;
obtaining a Self-attention PredRNN + + network model based on a Self-attention mechanism and PredRNN + +;
and obtaining a weather radar echo extrapolation image sequence through a weather radar echo image sequence based on a Self-attention PredRNN + + network model.
Further, the preprocessing the acquired weather radar echo image and the obtaining of the gray scale data of the weather radar echo image include:
denoising the acquired meteorological radar echo image based on a three-dimensional block matching algorithm;
and normalizing the denoised meteorological radar echo image and converting the denoised meteorological radar echo image into gray data.
Further, the gradation value range of the gradation data is [0,255 ].
Further, the obtaining of the Self-annotation PredRNN + + network model based on the Self-annotation mechanism and PredRNN + +, includes:
obtaining a layer A Self-attention mechanism, a layer A GHU and a 4 layer A cause LSTM unit;
dividing the obtained Self-attention mechanism, GHU and Causal LSTM units into A group cell units;
sequentially connecting four layers of Causal LSTM units of each group of cell units in series, and connecting the output end of the first layer of Causal LSTM unit with a layer of GHU in a communication manner, wherein the output end of the GHU is connected with a layer of Self-anchorage mechanism in a communication manner, and the output end of the Self-anchorage mechanism is connected with the second layer of Causal LSTM unit in a communication manner;
and sequentially cascading all groups of cell units, and enabling the output end of the fourth layer of Causal LSTM unit of the previous step of long cell unit to be in communication connection with the first layer of Causal LSTM unit of the next step of long cell unit, wherein the first layer of Causal LSTM unit of the first step of long cell unit receives gray data of a weather radar echo image, and the fourth layer of Causal LSTM unit of the A-step of long cell unit outputs a weather radar echo extrapolation image sequence.
Further, a is 3.
Further, the obtaining of the weather radar echo extrapolation image sequence based on the Self-anchorage PredRNN + + network model through the weather radar echo image sequence includes:
initializing model parameters of a Self-attribution PredRNN + + model;
inputting gray data of a weather radar echo image into a first layer of vehicular LSTM unit of the first step length cell unit;
updating the space-time memory and target sequence of model parameters by utilizing the gray data of the meteorological radar echo image based on the first layer of vehicular LSTM unit;
updating the hidden state of the meteorological radar echo image by using a target sequence updated by a first layer of practical LSTM unit based on GHU;
updating a target sequence of a meteorological radar echo image by using a hidden state updated by a GHU based on a Self-attention mechanism;
sequentially updating the target sequence by using the target sequence updated by a Self-attention mechanism based on the second-fourth layer Causal LSTM unit;
based on the sequential cascade of all groups of cell units, the first layer of Causal LSTM unit of the cell unit with the current step updates a space-time memory and a target sequence by using a target sequence updated by the fourth layer of Causal LSTM unit of the cell unit with the previous step, and model parameters and gray data updated by the cell unit with the previous step;
and updating the target sequence by using the target sequence updated by the third layer of Causal LSTM unit of the A-step cell unit based on the fourth layer of Causal LSTM unit of the A-step cell unit, and obtaining a weather radar echo extrapolation image sequence.
In another aspect, the present invention provides a weather radar echo extrapolation system based on a self-attentive mechanism and a predictive recurrent neural network, including:
the acquisition template is used for acquiring a meteorological radar echo image;
the preprocessing template is used for preprocessing the acquired meteorological radar echo image to obtain gray data of the meteorological radar echo image;
a model construction template used for obtaining a Self-attention PredRNN + + network model based on a Self-attention mechanism and PredRNN + +;
and the output template is used for obtaining a weather radar echo extrapolation image sequence through the weather radar echo image sequence based on the Self-anchorage PredRNN + + network model.
Compared with the prior art, the invention has the following beneficial effects:
the invention can extrapolate a radar echo image sequence of 2 hours in the future according to the existing radar echo image; the lightweight Self-attention mechanism not only can effectively control the number of model parameters, but also can extract global attention, effectively captures the characteristics of a deep network according to the structure of global space-time correlation information, and improves the accuracy of an extrapolation image sequence.
Drawings
FIG. 1 is a flow chart of an embodiment of a meteorological radar echo extrapolation method based on a self-attention mechanism and a predictive recurrent neural network according to the present invention;
FIG. 2 is a schematic structural diagram of a Causal LSTM unit according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of an embodiment of the self-attack mechanism and GHU of the present invention;
FIG. 4 is a schematic structural diagram illustrating an embodiment of the self-association mechanism of the present invention;
FIG. 5 is a schematic structural diagram illustrating an embodiment of a Self-association PredRNN + + network model according to the present invention;
FIG. 6 is an extrapolated image sequence of an embodiment of the weather radar echo extrapolation method based on the self-attention mechanism and the prediction recurrent neural network according to the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
Example 1
The embodiment provides a meteorological radar echo extrapolation method based on a self-attention mechanism and a prediction recurrent neural network, as shown in fig. 1, comprising the following steps:
s1, acquiring a weather radar echo image;
s2, preprocessing the acquired weather radar echo image, obtaining the gray scale data of the weather radar echo image,
in the application, based on a three-dimensional block matching algorithm, denoising an acquired meteorological radar echo image, namely removing Gaussian white noise, reserving image detail textures, and carrying out slicing processing on the image detail textures so that the image has good edge characteristics; and then, normalizing the denoised meteorological radar echo image, and converting the denoised meteorological radar echo image into gray data, wherein the gray data comprises the sequence characteristics of the meteorological radar echo image, and the range of the gray value is [0,255 ].
S3, obtaining a Self-attention (Self-attention) mechanism and a prediction recurrent neural network (PredRNN + +), and obtaining a Self-attention (Self-attention) PredRNN + + network model;
s4 is based on the Self-attention PredRNN + + network model, and a weather radar echo extrapolation image sequence is obtained through the weather radar echo image sequence.
The invention can extrapolate a radar echo image sequence of 2 hours in the future according to the existing radar echo image; the lightweight Self-attention mechanism not only can effectively control the number of model parameters, but also can extract global attention, effectively captures the characteristics of a deep network according to the structure of global space-time correlation information, and improves the accuracy of an extrapolation image sequence.
Example 2
On the basis of embodiment 1, this embodiment describes in detail a method for constructing a Self-anchorage PredRNN + + network model.
Based on a Self-attention mechanism and PredRNN + +, obtaining a Self-attention PredRNN + + network model comprises the following steps:
s31, acquiring a layer A Self-attribute mechanism, a layer A Gradient Highway Unit (GHU) and a 4 layer A Long Short-Term Memory network (Causal Long Short-Term Memory, Causal LSTM) Unit; in application, A is 3.
S32, dividing the obtained Self-attention mechanism, GHU and Causal LSTM units into A group cell units;
s33 As shown in FIG. 5, the four layers of Causal LSTM units of each group of cell units are connected in series in sequence, the output gate of the first layer of Causal LSTM unit is connected with a layer of GHU in a communication mode, the output end of the GHU is connected with a layer of Self-attention mechanism in a communication mode, the output end of the Self-attention mechanism is connected with the second layer of Causal LSTM unit in a communication mode, and in the application, the Self-attention mechanism can capture space-time memory of meteorological elements and accelerate convergence speed.
S34 as shown in FIG. 5, the cell units are cascaded in sequence, and the output end of the fourth layer of Causal LSTM unit of the previous long cell unit is connected with the first layer of Causal LSTM unit of the next long cell unit in a communication manner, wherein the first layer of Causal LSTM unit of the first long cell unit receives the gray data of the weather radar echo image, and the fourth layer of Causal LSTM unit of the A-th long cell unit outputs a weather radar echo extrapolation image sequence.
Firstly, dividing gray data of a meteorological radar echo image into a training set and a testing set, and inputting the training set into a Self-attention PredRNN + + network model in the same batch to train the Self-attention PredRNN + + network model. The training set of the present embodiment includes Batchsize samples.
In application, based on the effective prediction numerical range of the radar reflectivity factor of each pixel point of the meteorological radar echo image being [0,75], the model weight of the Self-attribution PredRNN + + network model is adjusted according to the following formula:
Figure BDA0003560169480000051
where w (x) is the model weight and x is the effective prediction value of the radar reflectivity factor.
The prediction capability of the Self-attention PredRNN + + network model for a strong echo region can be enhanced by adjusting the model weight.
And then comparing the weather radar echo extrapolation image sequence output by the Self-attention PredRNN + + network model with an actual weather radar echo extrapolation image sequence, calculating the model error of the Self-attention PredRNN + + network model by using a loss function and a back propagation algorithm, and adjusting the model parameters according to the model error.
In application, the loss function includes a mean square error function and a mean absolute error function.
Figure BDA0003560169480000052
Figure BDA0003560169480000053
Where MSE is the mean square error, MAE is the mean absolute error, X (i ') is the standard value, Y (i') is the model identification value, and n is the total number of samples.
And then inputting the test sets into the trained Self-attention PredRNN + + network model in the same batch to obtain a weather radar echo extrapolation image sequence, comparing the weather radar echo extrapolation image sequence output by the trained Self-attention PredRNN + + network model with an actual weather radar echo extrapolation image sequence, judging the accuracy of the Self-attention PredRNN + + network model through a preset evaluation index, and determining the optimal Self-attention PredRNN + + network model.
Example 3
On the basis of embodiment 1 or 2, this embodiment describes in detail a method for obtaining a weather radar echo extrapolation image sequence.
The method for obtaining the meteorological radar echo extrapolation image sequence based on the Self-attribution PredRNN + + network model through the meteorological radar echo image sequence comprises the following steps:
s41 initializes the model parameters of the Self-attribution PredRNN + + model, wherein in the application, the model parameters include hidden state, target sequence, time memory and space memory.
S42, inputting the gray data of the weather radar echo image into the first layer of Causal LSTM unit of the first step length cell unit, and in application, an input gate of the first layer of Causal LSTM unit receives the gray data.
S43 As shown in FIG. 2, more nonlinear layers are added to the Causal LSTM units in each layer to amplify each signal and effectively capture the characteristics of the deep network and the burst changes in the short term. Based on the first layer of Causal LSTM units, the time-space memory and the target sequence of the model parameters are updated by utilizing the gray data of the meteorological radar echo image.
In application, the spatiotemporal memory includes temporal memory and spatial memory. The first layer of Causal LSTM unit updates the time memory through the time editor:
Figure BDA0003560169480000061
Figure BDA0003560169480000062
wherein subscript is step length, superscript is specific hidden layer in stacked Causal LSTM network, C is time memory, X is sequence feature, H is target sequence, f is time forgetting gate, i is time input gate, g is time input modulation gate, W is time input modulation gate1For the convolution filter, tanh is the activation function, σ is the sigmoid function in the activation function, and [ ] is the convolution operation, which is the Hadamard product.
The first layer of Causal LSTM unit updates the space memory and the target sequence through the space editor:
Figure BDA0003560169480000063
Figure BDA0003560169480000064
Figure BDA0003560169480000065
Figure BDA0003560169480000066
wherein f ' is a space forgetting gate, i ' is a space input gate, g ' is a space input modulation gate, otIs an output gate, W2～W5Respectively, convolution filters, and M is spatial memory.
S44 As shown in FIG. 3, based on GHU, the target sequence updated by the first layer of Causal LSTM units is used for updating the hidden state of the weather radar echo image:
Figure BDA0003560169480000067
Figure BDA0003560169480000068
Zt＝Ut⊙Pt+(1-Ut)⊙Zt-1
wherein P is the input of the conversion, U is the switch gate, Z is the hidden state, Wpx、Wpz、Wsx、WszRespectively GHU filters.
S45 as shown in FIG. 4, updating the target sequence of the weather radar echo image by using the hidden state updated by the GHU based on the Self-attention mechanism;
V＝Zt*Wv
K＝Zt*Wk
Q＝Zt*Wq
Figure BDA0003560169480000071
wherein the content of the first and second substances,
Figure BDA0003560169480000072
for the currently updated target sequence, Wv、Wk、WqHidden weights for hidden states, V, Q, K are hidden weight matrices for hidden states, and d is the vector dimension of Q.
S46 sequentially updates the target sequence based on the second-fourth layer cause LSTM units using the target sequence updated by the Self-attention mechanism.
S47 is cascaded in turn based on each group of cell units, and the first layer of Causal LSTM unit of the cell unit of the current step updates the space-time memory (time memory and space memory) and the target sequence by using the target sequence updated by the fourth layer of Causal LSTM unit of the cell unit of the previous step and the model parameters and gray scale data updated by the cell unit of the previous step.
S48, based on the fourth layer of Causal LSTM units of the A-th step size cell units, updating the target sequence by using the target sequence updated by the third layer of Causal LSTM units of the A-th step size cell units, and obtaining a weather radar echo extrapolation image sequence.
Example 4
In this embodiment, the pixel value of the weather radar echo image is 500 × 500, the coverage time is 3 hours, the time length of inputting the gray data is 1 hour, the interval is 6 minutes, the time length of the target sequence is 2 hours, the interval is 12 minutes, and 20 time periods are total. And (4) carrying out processing such as turning, mirroring and the like on the gray data of the meteorological radar echo image to expand the number of data sets.
Training a model:
inputting the gray data of the meteorological radar echo image into a primarily constructed Self-anchorage PredRNN + + network model to obtain a meteorological radar echo extrapolation image sequence, calculating an average loss function value of each frame of the meteorological radar echo extrapolation image, and adjusting the model parameters to reduce the loss function value.
The training parameters include: the initial learning rate is 10-4, the learning rate penalty factor is 0.5, the batch size is set to 30 and the number of iterations is 30.
Acquiring a meteorological radar echo extrapolation image:
inputting the gray data of 10 weather radar echo images into a trained and optimized Self-attention PredRNN + + network model, and obtaining 10 groups of weather radar echo extrapolation image sequences for 2 hours in the future.
As shown in fig. 6, the positions, the profiles, the distributions and the changes of the echoes in the weather radar echo extrapolated image obtained by the present embodiment are very similar to those of the actual weather radar echo image, so that the accuracy of the weather radar echo extrapolated image sequence obtained by the present embodiment is high.
Example 5
The embodiment provides a weather radar echo extrapolation system based on a self-attention mechanism and a prediction recurrent neural network, which comprises:
the acquisition template is used for acquiring a meteorological radar echo image;
the preprocessing template is used for preprocessing the acquired meteorological radar echo image to obtain gray data of the meteorological radar echo image;
a model construction template used for obtaining a Self-attention PredRNN + + network model based on a Self-attention mechanism and PredRNN + +;
and the output template is used for obtaining a weather radar echo extrapolation image sequence through the weather radar echo image sequence based on the Self-anchorage PredRNN + + network model.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application 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, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (7)

1. A meteorological radar echo extrapolation method based on a self-attention mechanism and a prediction recurrent neural network is characterized by comprising the following steps:
acquiring a meteorological radar echo image;
preprocessing the acquired meteorological radar echo image to acquire gray data of the meteorological radar echo image;
obtaining a Self-attention PredRNN + + network model based on a Self-attention mechanism and PredRNN + +;
and obtaining a weather radar echo extrapolation image sequence through a weather radar echo image sequence based on a Self-attention PredRNN + + network model.
2. The method for meteorological radar echo extrapolation based on self-attention mechanism and predictive recurrent neural network according to claim 1, wherein the preprocessing the acquired meteorological radar echo image to obtain the gray scale data of the meteorological radar echo image comprises:
denoising the acquired meteorological radar echo image based on a three-dimensional block matching algorithm;
and normalizing the denoised meteorological radar echo image and converting the denoised meteorological radar echo image into gray data.
3. The weather radar echo extrapolation method based on the self-attentive mechanism and the predictive recurrent neural network as claimed in claim 2, wherein the gray scale value range of the gray scale data is [0,255 ].
4. The weather radar echo extrapolation method based on Self-attention mechanism and predictive recurrent neural network as claimed in claim 1, wherein said obtaining a Self-attention mechanism and PredRNN + + network model based on a Self-attention mechanism and PredRNN + +, comprises:
obtaining a layer A Self-attention mechanism, a layer A GHU and a 4 layer A vehicular LSTM unit;
dividing the obtained Self-attention mechanism, GHU and Causal LSTM units into A group cell units;
sequentially connecting four layers of Causal LSTM units of each group of cell units in series, and connecting the output end of the first layer of Causal LSTM unit with a layer of GHU in a communication manner, wherein the output end of the GHU is connected with a layer of Self-anchorage mechanism in a communication manner, and the output end of the Self-anchorage mechanism is connected with the second layer of Causal LSTM unit in a communication manner;
and sequentially cascading all groups of cell units, and enabling the output end of the fourth layer of Causal LSTM unit of the previous step of long cell unit to be in communication connection with the first layer of Causal LSTM unit of the next step of long cell unit, wherein the first layer of Causal LSTM unit of the first step of long cell unit receives gray data of a weather radar echo image, and the fourth layer of Causal LSTM unit of the A-step of long cell unit outputs a weather radar echo extrapolation image sequence.
5. The weather radar echo extrapolation method based on the self-attention mechanism and the predictive recurrent neural network as claimed in claim 4, wherein: the A is 3.
6. The method for weather radar echo extrapolation based on Self-attention mechanism and prediction recurrent neural network as claimed in claim 4, wherein said obtaining a weather radar echo extrapolation image sequence based on a Self-attention mechanism and prediction recurrent neural network model by using a weather radar echo image sequence comprises:
initializing model parameters of a Self-attribution PredRNN + + model;
inputting gray data of a meteorological radar echo image into a first layer of Causal LSTM unit of a first step length cell unit;
based on the first layer of Causal LSTM units, updating the space-time memory and target sequence of the model parameters by using the gray data of the meteorological radar echo image;
updating the hidden state of the meteorological radar echo image by using a target sequence updated by a first layer of practical LSTM unit based on GHU;
updating a target sequence of a meteorological radar echo image by using a hidden state updated by a GHU based on a Self-attention mechanism;
based on the second-fourth layer Causal LSTM units, sequentially updating the target sequence by using the target sequence updated by the Self-attention mechanism;
based on the sequential cascade of all groups of cell units, the first layer of Causal LSTM unit of the cell unit with the current step size updates the space-time memory and the target sequence by using the target sequence updated by the fourth layer of Causal LSTM unit of the cell unit with the previous step size and the model parameter and the gray data updated by the cell unit with the previous step size;
and updating the target sequence by using the target sequence updated by the third layer of Causal LSTM unit of the A-step cell unit based on the fourth layer of Causal LSTM unit of the A-step cell unit, and obtaining a weather radar echo extrapolation image sequence.
7. A meteorological radar echo extrapolation system based on a self-attention mechanism and a prediction recurrent neural network, comprising:
the acquisition template is used for acquiring a meteorological radar echo image;
the preprocessing template is used for preprocessing the acquired meteorological radar echo image to obtain gray data of the meteorological radar echo image;
a model construction template used for obtaining a Self-attention PredRNN + + network model based on a Self-attention mechanism and PredRNN + +;
and the output template is used for obtaining a weather radar echo extrapolation image sequence through the weather radar echo image sequence based on the Self-anchorage PredRNN + + network model.
CN202210286562.1A 2022-03-23 2022-03-23 Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network Pending CN114594443A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210286562.1A CN114594443A (en) 2022-03-23 2022-03-23 Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210286562.1A CN114594443A (en) 2022-03-23 2022-03-23 Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network

Publications (1)

Publication Number Publication Date
CN114594443A true CN114594443A (en) 2022-06-07

Family

ID=81819975

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210286562.1A Pending CN114594443A (en) 2022-03-23 2022-03-23 Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network

Country Status (1)

Country Link
CN (1) CN114594443A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116953653A (en) * 2023-09-19 2023-10-27 成都远望科技有限责任公司 Networking echo extrapolation method based on multiband weather radar

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116953653A (en) * 2023-09-19 2023-10-27 成都远望科技有限责任公司 Networking echo extrapolation method based on multiband weather radar
CN116953653B (en) * 2023-09-19 2023-12-26 成都远望科技有限责任公司 Networking echo extrapolation method based on multiband weather radar

Similar Documents

Publication Publication Date Title
CN112446419B (en) Attention mechanism-based space-time neural network radar echo extrapolation prediction method
CN111951207B (en) Image quality enhancement method based on deep reinforcement learning and semantic loss
CN113688723A (en) Infrared image pedestrian target detection method based on improved YOLOv5
CN110148088B (en) Image processing method, image rain removing method, device, terminal and medium
CN112560918B (en) Dish identification method based on improved YOLO v3
CN112183742B (en) Neural network hybrid quantization method based on progressive quantization and Hessian information
CN111062410B (en) Star information bridge weather prediction method based on deep learning
CN112766062B (en) Human behavior identification method based on double-current deep neural network
CN113870335A (en) Monocular depth estimation method based on multi-scale feature fusion
CN110930378A (en) Emphysema image processing method and system based on low data demand
CN111145102A (en) Synthetic aperture radar image denoising method based on convolutional neural network
CN114594443A (en) Meteorological radar echo extrapolation method and system based on self-attention mechanism and prediction recurrent neural network
CN114862699B (en) Face repairing method, device and storage medium based on generation countermeasure network
CN110781755A (en) Intelligent identification and metering method and system for construction steel bars
CN112488125B (en) Reconstruction method and system based on high-speed visual diagnosis and BP neural network
CN113341419A (en) Weather extrapolation method and system based on VAN-ConvLSTM
CN116416212B (en) Training method of road surface damage detection neural network and road surface damage detection neural network
CN117665825B (en) Radar echo extrapolation prediction method, system and storage medium
CN116524199B (en) Image rain removing method and device based on PReNet progressive network
CN113449611B (en) Helmet recognition intelligent monitoring system based on YOLO network compression algorithm
CN112926619B (en) High-precision underwater laser target recognition system
CN116152699B (en) Real-time moving target detection method for hydropower plant video monitoring system
CN116071825B (en) Action behavior recognition method, system, electronic equipment and storage medium
CN111382761B (en) CNN-based detector, image detection method and terminal
CN112465862A (en) Visual target tracking method based on cross-domain deep convolutional neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination