CN115857062A - Sub-season typhoon generation and prediction method based on multi-channel convolutional neural network - Google Patents

Sub-season typhoon generation and prediction method based on multi-channel convolutional neural network Download PDF

Info

Publication number
CN115857062A
CN115857062A CN202310174997.1A CN202310174997A CN115857062A CN 115857062 A CN115857062 A CN 115857062A CN 202310174997 A CN202310174997 A CN 202310174997A CN 115857062 A CN115857062 A CN 115857062A
Authority
CN
China
Prior art keywords
typhoon
data
model
season
neural network
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.)
Granted
Application number
CN202310174997.1A
Other languages
Chinese (zh)
Other versions
CN115857062B (en
Inventor
吕阳
智协飞
朱寿鹏
季焱
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 Institute Of Meteorological Science And Technology Innovation
Nanjing University of Information Science and Technology
Original Assignee
Nanjing Institute Of Meteorological Science And Technology Innovation
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 Institute Of Meteorological Science And Technology Innovation, Nanjing University of Information Science and Technology filed Critical Nanjing Institute Of Meteorological Science And Technology Innovation
Priority to CN202310174997.1A priority Critical patent/CN115857062B/en
Publication of CN115857062A publication Critical patent/CN115857062A/en
Application granted granted Critical
Publication of CN115857062B publication Critical patent/CN115857062B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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 discloses a method for generating and forecasting a sub-season typhoon based on a multichannel convolutional neural network, which comprises the following steps of: (1) Counting the generation frequency of typhoon cycle by cycle, carrying out data recombination on the typhoon frequency, extracting periodic signals with different time scales, and filtering redundant noise; (2) Constructing a mask field by diagnosing predictability sources of periodic signals of each time scale based on an information flow method; (3) Building a multi-channel convolutional neural network model, and developing and training the model based on a training set constructed by reanalysis data; (4) Performing transfer learning based on the acquired numerical model prediction data to obtain a final prediction model; (5) Substituting forecast data within preset time into the model to generate a typhoon generation forecast in a second season; the invention improves the typhoon generation and forecast skill in the next season; and redundant noise in the large-scale factor field is effectively filtered, so that the model forecasting effect is effectively improved.

Description

Sub-season typhoon generation and prediction method based on multi-channel convolutional neural network
Technical Field
The invention relates to the technical field of sub-season typhoon forecasting, in particular to a sub-season typhoon generation and forecasting method based on a multi-channel convolutional neural network.
Background
Typhoon is a disastrous weather system with extremely strong destructiveness, often brings fierce wind and rainstorm, and brings great threat to the life and property safety of people. Under the background, high-quality typhoon generation and forecast have important significance on disaster prevention and reduction. At present, typhoon generation and forecast for 1-14 days already shows high forecast skills, but the typhoon generation and forecast skills of the sub-seasonal scale (2 weeks-2 months) are very limited, so that the typhoon generation and forecast method is an important scientific and technical problem to be solved urgently.
The method is characterized in that typhoon is identified from a large-scale field predicted by a numerical mode, so that sub-season prediction of the typhoon is realized, but the scheme highly depends on the prediction level of the mode, and the numerical mode has low prediction skills under a long time scale, so that the current prediction skills of the scheme are generally low, and the method is difficult to provide guidance for disaster prevention and reduction work. A statistical model is also constructed on the basis of a linear relation between a large-scale factor field and typhoon, so that the generation and forecast of the typhoon in the sub-season are realized, but the scheme lacks the support of a power mechanism, and the forecasting skill is still very limited. In recent years, two schemes are combined to provide a hybrid statistical model, but the scheme is difficult to extract periodic signals of typhoon with multiple time scales, cannot capture nonlinear characteristics between typhoon and large-scale factor fields, and still has a large space for improving the forecasting skill.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a method for generating and forecasting the sub-season typhoon based on a multichannel convolutional neural network so as to solve the problems in the technical background.
The technical scheme is as follows: the invention relates to a method for generating and forecasting a sub-season typhoon based on a multichannel convolutional neural network, which comprises the following steps of:
(1) Counting the generation frequency of typhoons cycle by cycle, carrying out data recombination on the typhoon frequency based on singular spectrum analysis, extracting periodic signals with different time scales, and filtering redundant noise;
(2) Diagnosing the predictability source of each time scale periodic signal based on an information flow method, and constructing a mask field;
(3) Building a multi-channel convolutional neural network model, and developing and training the model based on a training set constructed by reanalysis data;
(4) Performing transfer learning based on the acquired numerical model prediction data to obtain a final prediction model;
(5) And substituting forecast data in preset time into the model to generate the typhoon generation forecast in the secondary season.
Further, the step (1) comprises the steps of:
(11) Counting typhoon generation frequency of the target area under 7 days one by one to form a typhoon generation frequency data set Y;
(12) Track matrix constructed based on typhoon generation frequency data set Y
Figure SMS_1
And singular value decomposition is carried out; trajectory matrix
Figure SMS_2
The calculation formula of (2) is as follows: />
Figure SMS_3
；
wherein ,
Figure SMS_4
for the i-th observation in the data set Y, <' > based on the data set>
Figure SMS_5
For the total number of samples in the data set Y, <' >>
Figure SMS_6
Is the window length, is greater or less than>
Figure SMS_7
As the number of vectors, it calculatesThe formula is as follows: .
Figure SMS_8
；
To the track matrix
Figure SMS_9
Singular value decomposition is carried out, and the calculation formula is as follows:
Figure SMS_10
；
wherein ,
Figure SMS_12
is the i-th characteristic value of the matrix S, <' >>
Figure SMS_15
Can be taken as the number of characteristic values>
Figure SMS_18
and />
Figure SMS_13
Small value of medium, is greater than or equal to>
Figure SMS_14
Is the orthonormal vector corresponding to the characteristic value, is greater than>
Figure SMS_17
Can be determined by>
Figure SMS_19
、/>
Figure SMS_11
、/>
Figure SMS_16
The calculation is carried out, and the specific calculation formula is as follows:
Figure SMS_20
；
(13) Data recombination of typhoon frequency data is completed through two steps of grouping and diagonal averaging, and periodic signals with different time scales are extracted.
Further, the step (13) comprises the steps of:
(131) Set subscripts
Figure SMS_21
Division into m disjoint subsets->
Figure SMS_22
Then, there are:
Figure SMS_23
；
(132) Converting each matrix in the formula (6) into a time sequence through diagonal average calculation, specifically: if one is
Figure SMS_24
Rank matrix->
Figure SMS_25
The element is->
Figure SMS_26
, wherein ,/>
Figure SMS_27
，
Order to
Figure SMS_28
，/>
Figure SMS_29
If is greater or greater>
Figure SMS_30
，/>
Figure SMS_31
Otherwise->
Figure SMS_32
Then carrying out diagonal averaging to complete the arrayRecombination, the calculation formula is as follows:
Figure SMS_33
；
after data recombination is completed, extracting periodic signals with multiple time scales in typhoon generation frequency:
Figure SMS_34
；
wherein ,
Figure SMS_35
for a sequence of original typhoon frequencies>
Figure SMS_36
Respectively representing periodic signals of different time scales.
Further, the step (2) comprises the following steps:
(21) Collecting sea temperature, radial wind, latitudinal wind and upward long wave radiation large-scale factor field, and performing 7-day-by-7-day sliding average processing on the data to construct the large-scale factor field
Figure SMS_37
、/>
Figure SMS_38
、/>
Figure SMS_39
、
Figure SMS_40
A data set is observed in which, among other things,SST、U、V、OLRrespectively indicate sea temperature, warp wind, weft wind and upward long-wave radiation>
Figure SMS_41
Longitude, latitude and time, respectively;
(22) Diagnosing the source of predictability of each time-scale periodic signal through an information flow method, wherein an information flow calculation formula is as follows:
Figure SMS_42
；
wherein ,
Figure SMS_45
information flow representing sequence 2 versus sequence 1, sequence 2 being taken from the large scale factor field:
Figure SMS_48
、/>
Figure SMS_52
、/>
Figure SMS_46
、/>
Figure SMS_50
the sequence 1 is taken from the periodic signals ^ on the time scales extracted from the typhoon-generating frequency>
Figure SMS_54
，/>
Figure SMS_56
Indicates that a sample is->
Figure SMS_43
and />
Figure SMS_47
Covariance in between, <' > based on the mean value of>
Figure SMS_51
Represents->
Figure SMS_55
and />
Figure SMS_44
Covariance in between, < >>
Figure SMS_49
Is->
Figure SMS_53
(ii) an euler front difference approximation;
(23) And respectively recording the positions passing the significance test in the large-scale factor field as 1 and the positions failing the test as 0 for the periodic signals of each time scale, and constructing the mask field of the periodic signals of each time scale.
Further, the step (3) comprises the following steps:
(31) Filtering the large-scale factor field based on the mask field, filtering redundant noise, and performing standardization processing;
(32) Building a multi-channel convolutional neural network, wherein the network structure mainly comprises a convolutional layer, a pooling layer, an expansion layer and a full-connection layer, and the model is expanded and trained based on a training set constructed by reanalysis data; the input data of the model is the predictability source of the large-scale factor field processed in the step (31), namely each time scale periodic signal, and the output data of the model is the actual typhoon generation frequency.
Further, the step (31) comprises the steps of:
(311) The filter calculation formula is as follows:
Figure SMS_57
；
wherein ,
Figure SMS_58
for the jth large-scale factor field corresponding to the ith time-scale periodic signal>
Figure SMS_59
Is its mask field, is>
Figure SMS_60
Is the updated field;
(312) The normalized process calculation formula is as follows:
Figure SMS_61
；
wherein ,
Figure SMS_62
represents->
Figure SMS_63
Is based on the mean value of (4)>
Figure SMS_64
Represents->
Figure SMS_65
Standard deviation of (d). />
Further, the step (32) comprises the steps of:
(321) Respectively performing convolution, pooling and expansion on the large-scale factor field, and finally connecting;
(322) And (5) carrying out training on the model based on a training set constructed by reanalysis data.
Further, the step (4) is: and (4) substituting the forecast data of the numerical mode for the large-scale factor field into the model trained in the step (3) for transfer learning to generate a final forecast model.
Further, the step (5) is: inputting the forecast data of the numerical mode in the future preset time for the large-scale factor field filtered by the mask field into the model trained in the step (4), and generating the typhoon generation forecast in the preset time.
Has the advantages that: compared with the prior art, the invention has the following remarkable advantages: the conflict between periodic signals with different time scales is avoided, the physical process can be better captured compared with the traditional scheme, and the model forecasting skill is improved; redundant noise in the large-scale factor field is effectively filtered, and the model forecasting effect is further effectively improved; the spatial characteristics of the large-scale factor field are fully considered, the nonlinear relation between typhoon and predictability sources under various time scales of typhoon is captured, and the model has high nonlinearity; the requirements of the deep learning model on the data volume are fully considered, the model is initialized based on reanalysis data, and then transfer learning is carried out based on mode data, so that the final model has strong robustness.
Drawings
FIG. 1 is a general flow chart of the present invention;
FIG. 2 is a schematic diagram of the model structure of the present invention.
Detailed description of the preferred embodiments
The technical scheme of the invention is further explained by combining the attached drawings.
As shown in fig. 1-2, an embodiment of the present invention provides a method for generating and forecasting a sub-seasonal typhoon based on a multi-channel convolutional neural network, including the following steps:
(1) Counting the generation frequency of typhoons cycle by cycle, carrying out data recombination on the typhoon frequency based on singular spectrum analysis, extracting periodic signals with different time scales, and filtering redundant noise, wherein the method comprises the following steps:
(11) Counting typhoon generation frequency of the target area under 7 days one by one to form a typhoon generation frequency data set Y;
(12) Track matrix constructed based on typhoon generation frequency data set Y
Figure SMS_66
And singular value decomposition is carried out; trajectory matrix
Figure SMS_67
The calculation formula of (2) is as follows:
Figure SMS_68
；
wherein ,
Figure SMS_69
for the i-th observation in the data set Y, <' > based on the data set>
Figure SMS_70
For the total number of samples in the data set Y, <' >>
Figure SMS_71
Is the window length, is greater or less than>
Figure SMS_72
The calculation formula is as follows: .
Figure SMS_73
；
To the track matrix
Figure SMS_74
Singular value decomposition is carried out, and the calculation formula is as follows: />
Figure SMS_75
；
wherein ,
Figure SMS_77
for the i-th characteristic value of the matrix S>
Figure SMS_81
Can be taken as the number of characteristic values>
Figure SMS_83
and />
Figure SMS_78
Small value of medium, is greater than or equal to>
Figure SMS_79
Is the orthonormal vector to which the characteristic value corresponds>
Figure SMS_82
Can be determined by>
Figure SMS_84
、/>
Figure SMS_76
、/>
Figure SMS_80
The calculation is carried out, and the specific calculation formula is as follows:
Figure SMS_85
；
(13) The data recombination of typhoon frequency data is completed through two steps of grouping and diagonal averaging, and periodic signals with different time scales are extracted, and the method comprises the following steps:
(131) Set subscripts
Figure SMS_86
Divide into m disjoint subsets>
Figure SMS_87
Then, there are:
Figure SMS_88
；
(132) Converting each matrix in the formula (6) into a time sequence through diagonal average calculation, specifically: if one is
Figure SMS_89
Matrix of order>
Figure SMS_90
The element is->
Figure SMS_91
, wherein ,/>
Figure SMS_92
，
Order to
Figure SMS_93
，/>
Figure SMS_94
If is greater or greater>
Figure SMS_95
，/>
Figure SMS_96
Otherwise->
Figure SMS_97
Then carrying out diagonal averaging to finish the array recombination, wherein the calculation formula is as follows:
Figure SMS_98
；
after data recombination is completed, extracting periodic signals with multiple time scales in typhoon generation frequency:
Figure SMS_99
；
wherein ,
Figure SMS_100
for a sequence of original typhoon frequencies>
Figure SMS_101
Respectively representing periodic signals of different time scales.
(2) Diagnosing the predictability source of each time scale periodic signal based on an information flow method, and constructing a mask field, wherein the method comprises the following steps:
(21) Collecting sea temperature, radial wind, latitudinal wind and upward long wave radiation large-scale factor field, performing 7-day-by-7-day sliding average processing on the data, and constructing to obtain
Figure SMS_102
、/>
Figure SMS_103
、/>
Figure SMS_104
、/>
Figure SMS_105
A data set is observed in which, among other things,SST、U、V、OLRrespectively indicates the sea temperature, the warp wind, the weft wind and the upward long-wave radiation, and is used for>
Figure SMS_106
Longitude, latitude and time, respectively;
(22) Diagnosing the source of predictability of each time-scale periodic signal through an information flow method, wherein an information flow calculation formula is as follows:
Figure SMS_107
；
wherein ,
Figure SMS_109
information flow representing sequence 2 versus sequence 1, sequence 2 being taken from the large scale factor field:
Figure SMS_114
、/>
Figure SMS_118
、/>
Figure SMS_111
、/>
Figure SMS_115
the sequence 1 is taken from the periodic signals ^ on the time scales extracted from the typhoon-generating frequency>
Figure SMS_119
，/>
Figure SMS_121
Indicates that a sample is->
Figure SMS_108
and />
Figure SMS_113
Covariance in between, < >>
Figure SMS_117
Represents->
Figure SMS_120
and />
Figure SMS_110
Covariance in between, < >>
Figure SMS_112
Is->
Figure SMS_116
Of EuropeThe difference before pulling is approximate;
(23) And respectively recording the positions passing the significance test in the large-scale factor field as 1 and the positions failing the test as 0 for the periodic signals of each time scale, and constructing the mask field of the periodic signals of each time scale.
(3) Building a multi-channel convolution neural network model, and developing and training the model based on a training set constructed by reanalysis data, wherein the method comprises the following steps:
(31) Filtering the large-scale factor field based on the mask field, filtering redundant noise, and performing standardization treatment, comprising the following steps:
(311) The filter calculation formula is as follows:
Figure SMS_122
；
wherein ,
Figure SMS_123
for the jth large-scale factor field corresponding to the ith time-scale periodic signal>
Figure SMS_124
Is its mask field, is>
Figure SMS_125
Is the updated field;
(312) The normalized process calculation formula is as follows:
Figure SMS_126
；
wherein ,
Figure SMS_127
represents->
Figure SMS_128
Is based on the mean value of (4)>
Figure SMS_129
Represents->
Figure SMS_130
Standard deviation of (d).
(32) Building a multi-channel convolutional neural network, wherein the network structure mainly comprises a convolutional layer, a pooling layer, an expansion layer and a full-connection layer, and the model is expanded and trained based on a training set constructed by reanalysis data; the input data of the model is the predictability source of the large-scale factor field processed in the step (31), namely each time scale periodic signal, and the output data of the model is the generation frequency of the actual typhoon, and the method comprises the following steps:
(321) Respectively performing convolution, pooling and expansion on the large-scale factor field, and finally connecting;
(322) And (5) carrying out training on the model based on a training set constructed by reanalysis data.
(4) Developing transfer learning based on the collected numerical model prediction data to obtain a final prediction model, which specifically comprises the following steps: and (4) substituting the forecast data of the numerical mode for the large-scale factor field into the model trained in the step (3) for transfer learning to generate a final forecast model.
(5) Substituting forecast data in preset time into the model to generate a sub-season typhoon generation forecast, which specifically comprises the following steps: inputting the forecast data of the numerical mode in the future preset time for the large-scale factor field filtered by the mask field into the model trained in the step (4), and generating the typhoon generation forecast in the preset time.

Claims (9)

1. A method for generating and forecasting a sub-season typhoon based on a multi-channel convolutional neural network is characterized by comprising the following steps:
(1) Counting the generation frequency of typhoons cycle by cycle, carrying out data recombination on the typhoon frequency based on singular spectrum analysis, extracting periodic signals with different time scales, and filtering redundant noise;
(2) Diagnosing the predictability source of each time scale periodic signal based on an information flow method, and constructing a mask field;
(3) Building a multi-channel convolutional neural network model, and developing and training the model based on a training set constructed by reanalysis data;
(4) Performing transfer learning based on the acquired numerical model prediction data to obtain a final prediction model;
(5) And substituting forecast data in preset time into the model to generate the typhoon generation forecast in the secondary season.
2. The method for generating and forecasting the sub-season typhoon based on the multichannel convolutional neural network as claimed in claim 1, wherein the step (1) comprises the following steps:
(11) Counting typhoon generation frequency of the target area under 7 days one by one to form a typhoon generation frequency data set Y;
(12) Track matrix constructed based on typhoon generation frequency data set Y
Figure QLYQS_1
Singular value decomposition is carried out; track matrix>
Figure QLYQS_2
The calculation formula of (2) is as follows:
Figure QLYQS_3
；
wherein ,
Figure QLYQS_4
for the i-th observation in the data set Y, <' > based on the data set>
Figure QLYQS_5
For the number of total samples in the data set Y, for>
Figure QLYQS_6
Is the window length, is greater or less than>
Figure QLYQS_7
The calculation formula is as follows:
Figure QLYQS_8
；
to the track matrix
Figure QLYQS_9
Singular value decomposition is carried out, and the calculation formula is as follows:
Figure QLYQS_10
；
wherein ,
Figure QLYQS_12
is the i-th characteristic value of the matrix S, <' >>
Figure QLYQS_16
Can be taken as the number of characteristic values>
Figure QLYQS_18
and />
Figure QLYQS_13
Small value in (iv), in>
Figure QLYQS_14
Is the orthonormal vector corresponding to the characteristic value, is greater than>
Figure QLYQS_17
Can be determined by>
Figure QLYQS_19
、/>
Figure QLYQS_11
、/>
Figure QLYQS_15
The calculation is carried out, and the specific calculation formula is as follows:
Figure QLYQS_20
；
(13) Data recombination of typhoon frequency data is completed through two steps of grouping and diagonal averaging, and periodic signals with different time scales are extracted.
3. The method for generating and forecasting of sub-season typhoon based on multichannel convolutional neural network as claimed in claim 2, wherein said step (13) comprises the following steps:
(131) Set the subscripts
Figure QLYQS_21
Division into m disjoint subsets->
Figure QLYQS_22
Then, there are:
Figure QLYQS_23
；
(132) Converting each matrix in the formula (6) into a time sequence through diagonal average calculation, specifically: if one is
Figure QLYQS_24
Matrix of order>
Figure QLYQS_25
The element is->
Figure QLYQS_26
, wherein ,/>
Figure QLYQS_27
，
Order to
Figure QLYQS_28
，/>
Figure QLYQS_29
If is greater or greater>
Figure QLYQS_30
，/>
Figure QLYQS_31
Otherwise>
Figure QLYQS_32
Then, diagonal averaging is carried out to complete the recombination of the arrays, and the calculation formula is as follows:
Figure QLYQS_33
；
after data recombination is completed, extracting periodic signals with multiple time scales in typhoon generation frequency:
Figure QLYQS_34
；
wherein ,
Figure QLYQS_35
for a sequence of frequent times of the original typhoon>
Figure QLYQS_36
Respectively representing periodic signals of different time scales.
4. The method for generating and forecasting the sub-season typhoon based on the multichannel convolutional neural network as claimed in claim 1, wherein the step (2) comprises the following steps:
(21) Collecting sea temperature, radial wind, latitudinal wind and upward long wave radiation large-scale factor field, and performing 7-day-by-7-day sliding average processing on the data to construct the large-scale factor field
Figure QLYQS_37
、/>
Figure QLYQS_38
、/>
Figure QLYQS_39
、
Figure QLYQS_40
A data set is observed in which, among other things,SST、U、V、OLRrespectively indicate sea temperature, warp wind, weft wind and upward long-wave radiation>
Figure QLYQS_41
Longitude, latitude and time, respectively;
(22) Diagnosing the source of predictability of each time-scale periodic signal through an information flow method, wherein an information flow calculation formula is as follows:
Figure QLYQS_42
；
wherein ,
Figure QLYQS_44
information flow representing sequence 2 versus sequence 1, sequence 2 being taken from the large scale factor field: />
Figure QLYQS_47
、
Figure QLYQS_51
、/>
Figure QLYQS_45
、/>
Figure QLYQS_49
The sequence 1 is taken from the periodic signals ^ on the time scales extracted from the typhoon-generating frequency>
Figure QLYQS_53
，/>
Figure QLYQS_56
Indicates that a sample is->
Figure QLYQS_43
and />
Figure QLYQS_48
Covariance in between, < >>
Figure QLYQS_52
Represents->
Figure QLYQS_55
and />
Figure QLYQS_46
Covariance in between, < >>
Figure QLYQS_50
Is->
Figure QLYQS_54
(ii) an euler front difference approximation;
(23) And respectively recording the positions passing the significance test in the large-scale factor field as 1 and the positions failing the test as 0 for the periodic signals of each time scale, and constructing the mask field of the periodic signals of each time scale.
5. The method for generating and forecasting the sub-season typhoon based on the multichannel convolutional neural network as claimed in claim 1, wherein the step (3) comprises the following steps:
(31) Filtering sea temperature, radial wind, latitudinal wind and upward long-wave radiation large-scale factor fields based on a mask field, filtering redundant noise, and carrying out standardization treatment;
(32) Building a multi-channel convolutional neural network, wherein the network structure mainly comprises a convolutional layer, a pooling layer, an expansion layer and a full-connection layer, and the model is expanded and trained based on a training set constructed by reanalysis data; the input data of the model is the predictability source of the large-scale factor field processed in the step (31), namely each time scale periodic signal, and the output data of the model is the actual typhoon generation frequency.
6. The method for generating and forecasting of sub-season typhoon based on multichannel convolutional neural network as claimed in claim 5, characterized in that said step (31) comprises the following steps:
(311) The filter calculation formula is as follows:
Figure QLYQS_57
；
wherein ,
Figure QLYQS_58
for the jth large-scale factor field corresponding to the ith time-scale periodic signal>
Figure QLYQS_59
Is its mask field, is>
Figure QLYQS_60
Is the updated field;
(312) The normalized process calculation formula is as follows:
Figure QLYQS_61
；
wherein ,
Figure QLYQS_62
represents->
Figure QLYQS_63
Is based on the mean value of (4)>
Figure QLYQS_64
Represents->
Figure QLYQS_65
Standard deviation of (2).
7. The method of claim 5, wherein the step (32) comprises the steps of:
(321) Respectively carrying out convolution, pooling and expansion on the large-scale factor field, and finally connecting;
(322) And (5) carrying out model development training on the training set constructed based on the reanalysis data.
8. The method for generating and forecasting the sub-season typhoon based on the multichannel convolutional neural network as claimed in claim 1, wherein the step (4) is as follows: and (4) substituting the forecast data of the numerical mode for the large-scale factor field into the model trained in the step (3) for transfer learning to generate a final forecast model.
9. The method for generating and forecasting the sub-season typhoon based on the multichannel convolutional neural network as claimed in claim 1, wherein the step (5) is as follows: inputting the forecast data of the numerical mode in the future preset time for the large-scale factor field filtered by the mask field into the model trained in the step (4), and generating the typhoon generation forecast in the preset time.
CN202310174997.1A 2023-02-28 2023-02-28 Sub-season typhoon generation forecasting method based on multichannel convolutional neural network Active CN115857062B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310174997.1A CN115857062B (en) 2023-02-28 2023-02-28 Sub-season typhoon generation forecasting method based on multichannel convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310174997.1A CN115857062B (en) 2023-02-28 2023-02-28 Sub-season typhoon generation forecasting method based on multichannel convolutional neural network

Publications (2)

Publication Number Publication Date
CN115857062A true CN115857062A (en) 2023-03-28
CN115857062B CN115857062B (en) 2023-06-13

Family

ID=85659322

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310174997.1A Active CN115857062B (en) 2023-02-28 2023-02-28 Sub-season typhoon generation forecasting method based on multichannel convolutional neural network

Country Status (1)

Country Link
CN (1) CN115857062B (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583653A (en) * 2018-11-30 2019-04-05 南京信息工程大学 The extended peroid forecasting procedure of NORTHWESTERN PACIFIC TYPHOON based on statistical model
US20200348448A1 (en) * 2019-04-30 2020-11-05 ClimateAI, Inc. Methods and systems for climate forecasting using artificial neural networks
CN112419259A (en) * 2020-11-19 2021-02-26 中国石油大学(华东) Typhoon image automatic detection system based on convolutional neural network, computer equipment and storage medium
CN112785034A (en) * 2020-12-22 2021-05-11 河海大学 Typhoon path forecasting method, system, medium and terminal based on fusion neural network
CN113807583A (en) * 2021-09-18 2021-12-17 南京信息工程大学 Low-frequency increment space-time coupling-based power statistics combined sub-season prediction method
CN115271181A (en) * 2022-07-06 2022-11-01 同济大学 Typhoon probability forecasting intelligent method and device based on multi-mode data fusion
CN115630742A (en) * 2022-10-31 2023-01-20 山东大学 Weather prediction method and system based on self-supervision pre-training

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583653A (en) * 2018-11-30 2019-04-05 南京信息工程大学 The extended peroid forecasting procedure of NORTHWESTERN PACIFIC TYPHOON based on statistical model
US20200348448A1 (en) * 2019-04-30 2020-11-05 ClimateAI, Inc. Methods and systems for climate forecasting using artificial neural networks
CN112419259A (en) * 2020-11-19 2021-02-26 中国石油大学(华东) Typhoon image automatic detection system based on convolutional neural network, computer equipment and storage medium
CN112785034A (en) * 2020-12-22 2021-05-11 河海大学 Typhoon path forecasting method, system, medium and terminal based on fusion neural network
CN113807583A (en) * 2021-09-18 2021-12-17 南京信息工程大学 Low-frequency increment space-time coupling-based power statistics combined sub-season prediction method
CN115271181A (en) * 2022-07-06 2022-11-01 同济大学 Typhoon probability forecasting intelligent method and device based on multi-mode data fusion
CN115630742A (en) * 2022-10-31 2023-01-20 山东大学 Weather prediction method and system based on self-supervision pre-training

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
郑宗生 等: "基于迁移学习及气象卫星云图的台风等级分类研究" *

Also Published As

Publication number Publication date
CN115857062B (en) 2023-06-13

Similar Documents

Publication Publication Date Title
CN109214592B (en) Multi-model-fused deep learning air quality prediction method
CN112364779B (en) Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion
CN109635928B (en) Voltage sag reason identification method based on deep learning model fusion
CN111080032A (en) Load prediction method based on Transformer structure
CN114092832B (en) High-resolution remote sensing image classification method based on parallel hybrid convolutional network
CN113205226A (en) Photovoltaic power prediction method combining attention mechanism and error correction
CN110991690A (en) Multi-time wind speed prediction method based on deep convolutional neural network
Zhang et al. Multi-head attention-based probabilistic CNN-BiLSTM for day-ahead wind speed forecasting
CN107463993A (en) Medium-and Long-Term Runoff Forecasting method based on mutual information core principle component analysis Elman networks
CN111369057A (en) Air quality prediction optimization method and system based on deep learning
Li et al. A new hybrid model FPA-SVM considering cointegration for particular matter concentration forecasting: a case study of Kunming and Yuxi, China
CN113469427A (en) Convolution LSTM-based photovoltaic power station day-ahead power generation amount prediction method and system
CN111580151A (en) SSNet model-based earthquake event time-of-arrival identification method
CN111242355A (en) Photovoltaic probability prediction method and system based on Bayesian neural network
CN112285376A (en) Wind speed prediction method based on CNN-LSTM
Tang et al. A new hybrid forecasting model based on dual series decomposition with long-term short-term memory
CN117006654A (en) Air conditioner load control system and method based on edge calculation
CN114818579A (en) Analog circuit fault diagnosis method based on one-dimensional convolution long-short term memory network
CN111596292A (en) Radar target identification method based on importance network and bidirectional stacking recurrent neural network
CN115801152B (en) WiFi action recognition method based on hierarchical transformer model
CN117076931A (en) Time sequence data prediction method and system based on conditional diffusion model
CN116187540B (en) Wind power station ultra-short-term power prediction method based on space-time deviation correction
CN115857062A (en) Sub-season typhoon generation and prediction method based on multi-channel convolutional neural network
CN116307206A (en) Natural gas flow prediction method based on segmented graph convolution and time attention mechanism
CN116030304A (en) Cross-domain remote sensing image migration resisting method based on weighted discrimination and multiple classifiers

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
GR01 Patent grant
GR01 Patent grant