CN108196314B - Automatic northwest Pacific annular typhoon identification system - Google Patents

Automatic northwest Pacific annular typhoon identification system Download PDF

Info

Publication number
CN108196314B
CN108196314B CN201711407247.5A CN201711407247A CN108196314B CN 108196314 B CN108196314 B CN 108196314B CN 201711407247 A CN201711407247 A CN 201711407247A CN 108196314 B CN108196314 B CN 108196314B
Authority
CN
China
Prior art keywords
typhoon
annular
time
output
layer
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.)
Active
Application number
CN201711407247.5A
Other languages
Chinese (zh)
Other versions
CN108196314A (en
Inventor
任思婧
储可宽
谈哲敏
卓静仪
陈慧敏
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
Original Assignee
Nanjing University
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 filed Critical Nanjing University
Priority to CN201711407247.5A priority Critical patent/CN108196314B/en
Publication of CN108196314A publication Critical patent/CN108196314A/en
Application granted granted Critical
Publication of CN108196314B publication Critical patent/CN108196314B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01W—METEOROLOGY
    • G01W1/00—Meteorology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • 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
    • Y02A50/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE in human health protection, e.g. against extreme weather
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Environmental & Geological Engineering (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Ecology (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Data Mining & Analysis (AREA)
  • Environmental Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Atmospheric Sciences (AREA)
  • Image Analysis (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses an automatic identification system of the annular typhoon of the northwest Pacific, which is established for the first time, satellite data, environmental field and sea temperature field data of the typhoon to be judged are input into the automatic identification system, the system calculates a series of physical characteristics of the typhoon, and then warns the typhoon which accords with the characteristics of the annular typhoon of the northwest Pacific through pre-screening and linear discrimination or automatic identification of a BP neural network based on the characteristic parameters, thereby providing help for the forecast of the annular typhoon.

Description

Automatic identification system for northwest Pacific ocean annular typhoon
Technical Field
The invention belongs to the prediction identification technology of an atmospheric system, and particularly relates to an automatic identification and alarm system for northwest Pacific ocean annular typhoon.
Background
The annular typhoon has the characteristics of high strength, long maintenance time, high forecasting difficulty and the like. The northwest Pacific ocean annular typhoon generating area is concentrated and has high logging probability, which continuously causes great threat to east Asia area, but the northwest Pacific ocean annular typhoon generating area has not been paid sufficient attention at present. At present, the typhoon is predicted based on the miniature and environmental conditions of the east pacific and the Atlantic annular hurricanes, screening and identification are established, so that an objective identification algorithm of the pacific and the Atlantic annular hurricanes is established, and timely and quick response cannot be achieved in automatic identification and early warning and forecasting.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an automatic identification system for the western-north pacific annular typhoon, which is established for the first time and provides help for early warning and forecasting of the type of typhoon.
In order to solve the technical problems, the invention adopts the following technical scheme: an automatic identification system for northwest Pacific annular typhoon comprises the following steps:
step 1: analyzing and calculating physical quantity, firstly calling time data of occurrence of annular typhoon and time data of recent occurrence of non-annular typhoon in an ocean historical time period, and calculating characteristic data of the annular typhoon in an ocean area, wherein the characteristic data comprises the characteristic data of typhoon intensity, structure, environmental field and ocean temperature field;
step 2: calling all annular typhoon times of the past northwest Pacific and all non-annular typhoon times of the current year as training sets, obtaining calculation results of the characteristic data of the strength, structure, environmental field and sea temperature field of the typhoon of the training sets, and establishing an automatic pre-screening recognition mechanism: when the typhoon to be evaluated meets the condition of the formula (1), pre-screening is carried out, otherwise, screening is carried out,
Figure GDA0004131737820000011
in the formula, V max Maximum wind speed of typhoon, R c Is the average coldest radius, Δ T, of the typhoon circumference eye The coldest radius of the typhoon is at the maximum value of the average infrared bright temperature difference of the inner circumference of the typhoon, SHRD is the deep vertical wind shear, u200 is 200hPa environmental latitude wind, SST is the central ocean surface temperature of the typhoon, sigma c The standard deviation of the infrared bright temperature at the coldest radius of the typhoon is shown;
and step 3: simultaneously, carrying out linear discrimination on training sets of all annular typhoon times of the past northwest Pacific and all non-annular typhoon times of the current year by sigma c 、VAR、ΔT eye U200 and SST are used as linear judgment bases, the middle points of projections of the annular time training set and the non-annular training set in a one-dimensional space are used as classification standards, and the time falling to one side of the non-annular typhoon is screened out;
and 4, step 4: for typhoon times pre-screened in the steps 2 and 3, judging a back propagation neural network, and inputting a variable sigma c 、VAR、ΔT eye U200 and SST, multiplying the weight coefficient obtained by training, and screening out typhoon time which is output to be non-annular time through an excitation function;
and 5: and (3) checking whether the two continuous typhoon times pass through the steps 1-4, and judging that the current typhoon belongs to the annular typhoon when the two continuous typhoon times are annular typhoon times.
Furthermore, in the linear judgment of the step 3,
first, the [ σ ] of the cyclic time and the acyclic time is set by a training set c ,VAR,ΔT eye ,u 200 ,SST] T Form 5 Xn 1 And 5 xn 2 Matrix X of 1 And X 2 Wherein n is 1 And n 2 The number of the training set ring-shaped time and the number of the training set non-ring-shaped time are calculated, and the physical quantity time is averaged to obtain a matrix
Figure GDA0004131737820000021
And &>
Figure GDA0004131737820000022
Then, X is calculated 1 And X 2 Auto-covariance matrix S 1 And S 2 And passing through an autocovariance matrix S 1 And S 2 Calculating to obtain an intra-class walking matrix S,
Figure GDA0004131737820000023
then, a rotation matrix W is selected,
Figure GDA0004131737820000024
defining a decision midpoint Y 0 ，/>
Figure GDA0004131737820000025
Figure GDA0004131737820000026
Is rotated projection->
Figure GDA0004131737820000027
Figure GDA0004131737820000028
Is rotated projection->
Figure GDA0004131737820000029
By passingCalculating the physical quantity combination X of the times to be judged, and further obtaining a judgment value Y = W of the times to be judged T X, then determining the typhoon ring time by:
when Y is less than Y 0 And Y is 1 ＜Y 0 ＜Y 2 When, or Y > Y 0 And Y is 1 ＞Y 0 ＞Y 2 Judging the time to be a ring time;
when Y > Y 0 And Y is 1 ＜Y 0 ＜Y 2 When, or Y < Y 0 And Y is 1 ＞Y 0 ＞Y 2 Then, the time is determined to be non-annular time.
Further, in the process of establishing the BP neural network model in the step 4, the sigma is used c ,VAR,ΔT eye ,u 200 SST is used as an input variable of an input layer, the number of neurons of a hidden layer is set to be 40, and the output result of an output layer is binary output;
establishing BP neural network model, assuming input layer having
Figure GDA0004131737820000038
Each neuron with hidden layers>
Figure GDA0004131737820000039
Neuron, output layer has d neurons, and the symbol is defined as follows:
input vector x = (x) 1 ,x 2 ,…x l )
Hidden layer output vector h = (h) 1 ,h 2 ,…h q )
Output layer output vector y = (y) 1 ,y 2 ,…y d )
The desired output vector is:
Figure GDA0004131737820000031
error function:
Figure GDA0004131737820000032
training a sample: (x) k ,y k ) K =1,2 … n, n is the total time number of the training set of the circular typhoon and the non-circular typhoon;
the excitation function is
Figure GDA0004131737820000033
Establishing BP neural network model according to training set, and using sigma c 、VAR、ΔT eye U200, SST as input variables x = (σ) of the input layer c ,VAR,ΔT eye ,u 200 SST), the number of hidden layer neurons is set to be 40, the output result of the output layer is binary output, the annular typhoon is 1, the non-annular typhoon is 0, and the learning step comprises:
(1) Calculating the output values of the hidden layer neurons and the output layer neurons:
hidden layer neuron output value:
Figure GDA0004131737820000034
/>
output layer neuron output values:
Figure GDA0004131737820000035
in the formula, W1 is a hidden layer weight, and W2 is an output layer weight; the lower corner mark i is the serial number of the hidden layer neuron, and the lower corner mark j is the serial number of the output layer neuron;
(2) Calculating the partial derivative of the error function to the weight of the output layer and transmitting the partial derivative to the weight of the hidden layer, and calculating the partial derivative of the error function to each weight of the hidden layer:
and (3) an output layer:
Figure GDA0004131737820000036
hiding the layer:
Figure GDA0004131737820000037
(3) Utilizing the gradient descent method and the partial derivative obtained in the step (2) to carry out weighting W1 'on the output layer' ij And hidden layer weight W2' ij And (3) correcting:
W1′ ij ＝W1 ij -η·g(W1 ij ),i＝1,2,…q,j＝1,2,…l (6)
W2′ ij ＝W2 ij -η·g(W2 ij ),i＝1,2,…d,j＝1,2,…q (7)
in the above formula, η is the learning rate;
(4) And calculating the corrected error function value, finishing training if the error function value reaches the preset precision, storing the hidden layer weight W1 and the output layer weight W2, determining the BP neural network model, and otherwise, selecting the next sample for learning, or selecting the next sample for learning.
Further, in the BP neural network model, the method for determining the cyclic time in step (4) is as follows:
the typhoon time to be judged is treated, and the sigma of the time is c 、VAR、ΔT eye And u200 and SST are input into a BP neural network model, hidden layer weight W1 and output layer weight W2 obtained by training are read, hidden layer neuron output values and output layer neuron output values obtained through an excitation function are respectively calculated according to the formula (2) and the formula (3), if the output layer neuron output values are 1, the cyclic time is judged, and if the output layer neuron output values are 0, the non-cyclic time is judged.
Has the advantages that: compared with the prior art, the invention has the following advantages: the invention establishes an automatic identification system for the northwest Pacific ocean ring typhoon for the first time, satellite data, environmental field and sea temperature field data of the typhoon to be judged are input into the automatic identification system, the system calculates to obtain a series of physical characteristics of the typhoon, then performs automatic identification of 'prescreening + linear discrimination (or BP neural network)' based on the characteristics, warns the typhoon which accords with the characteristics of the northwest Pacific ocean ring typhoon, and provides help for the forecast of the typhoon. When the method is applied to the analysis of all typhoons in 2000-2008, the hit rate of the recognition system reaches 100%, the accuracy can reach 91.3%, and the method can be comparable to the automatic recognition system of annular hurricanes in the east Pacific and Atlantic.
Drawings
Fig. 1 is a circumferential average coldest radius R exemplified by an annular typhoon WALT (1991.05.13.21) C Standard deviation sigma of infrared brightness temperature at the coldest radius C And the maximum value delta T of the average infrared bright temperature difference between the coldest radius and the inner circumference eye Calculating a schematic diagram;
FIG. 2 is a schematic diagram of a BP neural network model;
FIG. 3 is a schematic diagram of a BP neural network model according to the present invention;
FIG. 4 is a cloud image of an annular typhoon satellite according to the present invention;
FIG. 5 is a schematic illustration of a east Pacific and Atlantic annular hurricane generation area and landing conditions;
FIG. 6 is a schematic diagram of the annular typhoon generating area and landing status of the northwest Pacific ocean;
FIG. 7 is a schematic flow chart of an automatic identification system for annular typhoon in the northwest Pacific.
Detailed Description
The invention will be further elucidated with reference to the following description of an embodiment in conjunction with the accompanying drawing. It is to be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the invention, which is to be given the full breadth of the appended claims and any and all equivalent modifications within the scope of the present invention as defined by the following claims.
The invention establishes an automatic identification system for the northwest Pacific ocean ring typhoon for the first time, satellite data, environmental field and sea temperature field data of the typhoon to be judged are input into the automatic identification system, the system calculates to obtain a series of physical characteristics of the typhoon, then performs automatic identification of 'prescreening + linear discrimination (or BP neural network)' based on the characteristics, warns the typhoon which accords with the characteristics of the northwest Pacific ocean ring typhoon, and provides help for the forecast of the typhoon.
The method is applied to the analysis of all typhoons in 2000-2008, the hit rate of the identification system reaches 100%, the accuracy reaches 91.3%, and the method can be comparable to the automatic identification system of the annular hurricanes of the east Pacific and the Atlantic.
The invention calculates the physical quantities of typhoon intensity, structure, environmental temperature field, wind field, sea temperature field and the like based on satellite data of the northwest Pacific ocean typhoon and environmental field and sea temperature field data, and warns the typhoon which accords with the annular typhoon characteristic of the northwest Pacific ocean by the identification program of 'prescreening + linear discrimination' (or 'prescreening + BP neural network'). The method comprises the following specific steps:
1. analysis and calculation of physical quantities
Firstly, the typhoon position, the maximum wind speed (vmax) and the infrared brightness temperature are directly obtained from the typhoon satellite data. The infrared bright temperature data are distributed by taking longitude and latitude as horizontal and vertical coordinates, linear interpolation is respectively carried out twice along the longitude and latitude directions to obtain data under a polar coordinate system, and an infrared bright temperature sequence and a standard deviation (VAR) of the sequence are further calculated, wherein the infrared bright temperature sequence is averaged on the circumference of each radius from the center of the typhoon to the position 560km away from the center of the typhoon. Selecting the circumferential average coldest radius (R) C ) Analyzing typhoon eye size, standard deviation (sigma) of infrared bright temperature at coldest radius C ) The eye wall was analyzed for uniformity and maximum of the mean infrared bright temperature difference (Δ T) at the coldest radius from its inner circumference eye ) The convection intensity of the eye wall was analyzed. Table 1-1 shows the physical quantities, abbreviations and their calculation methods (satellite data)
TABLE 1-1
Figure GDA0004131737820000051
/>
Figure GDA0004131737820000061
Secondly, according to the environmental field data, the temperature (t 200) of the upper part (200 hPa) of the troposphere of the typhoon peripheral average (200-800 km) is calculated, and the high-layer temperature field characteristic is obtained. Similarly, calculating the characteristics (u 200, v200, u500, v500, u850, v 850) of the warp and weft wind fields of the 200-800km average troposphere (200 hPa), the middle troposphere (500 hPa) and the lower troposphere (850 hPa), carrying out vector subtraction on the wind fields of the upper layer, the lower layer, the middle layer and the lower layer to obtain 200-850hPa deep vertical wind Shear (SHRD) and 500-850hPa shallow vertical wind shear (SHRS), and obtaining the environmental shear characteristics.
And finally, interpolating the ocean surface temperature data in the ocean temperature field data to the typhoon center position by adopting a cubic spline interpolation method to obtain the ocean surface temperature (SST) at the typhoon center.
Tables 1 to 2
Figure GDA0004131737820000062
TABLE 1-2 summary of physical quantities, abbreviations and calculation methods (environmental field and sea temperature field data)
2. Principle and application of identification procedure
Table 2-1 northwest pacific 1990-2009 physical quantity calculations (satellite data) for non-cyclic times with intensities greater than 80knots, all values of the variables with increasing tilt indicate significant differences (α = 0.05) between the mean values of the two sets of data.
TABLE 2-1
Figure GDA0004131737820000071
And (2) calculating and analyzing typical characteristics (2-2) of the intensity, the structure, the environmental field and the sea temperature field of the annular typhoon of the northwest pacific according to the step (1) by taking all annular typhoon times in 1999 to 2009 of the northwest pacific and all non-annular typhoon times in 2009 as training sets.
Tables 2 to 2
Figure GDA0004131737820000072
Based on the calculation results of the training set, a recognition program is established as follows:
first, pre-screening criteria were set (tables 2-3): vmax, R C ,ΔT eye The normal distribution was not met, so the three screen-out criteria were set to be slightly less than the minimum value (MIN 0.95) for the ring typhoon. The screening conditions for SHRD and u200 are less than the maximum value of all the cycles and greater than the minimum value of all the cycles, and those which do not meet the conditions are screened out. Finally, the SST satisfies normal distribution, so that the time outside the plus and minus triple standard deviation range of the mean value of the annular typhoon SST is screened out.
Tables 2 to 3
Figure GDA0004131737820000081
The pre-screening can almost guarantee 100% hit rate, but also can result in higher null report rate. Therefore, linear discrimination is applied once more to the time passed through the prescreening.
Linear discrimination is a simple linear logistic regression method, and its basic idea is to project a sample of n-dimensional space onto a straight line in a specific direction to form a one-dimensional space. In this direction, the projections of the sample on the straight line may be separated best. In the present invention, uncorrelated σ is selected C ，VAR，ΔT eye U200 SST is used as a criterion for establishing linear discrimination, the middle points of projections of the annular time training set and the non-annular time training set on a 1-dimensional space are selected as a classification standard, and the time falling to one side of the non-annular typhoon is screened out.
In the actual procedure, each cyclic time [ σ ] of the training set C ，VAR，ΔT eye ，u200，SST] T Form 5 Xn 1 Matrix X of 1 (wherein n is 1 For training the number of times of the set's ring), the physical quantities are averaged over time (i.e., matrix X) 1 Each line is averaged) to
Figure GDA0004131737820000082
Similarly, the physical quantities of each non-annular order form a matrix X 2 The mean value of which constitutes the matrix->
Figure GDA0004131737820000083
Respectively calculate X 1 、X 2 Auto-covariance matrix S 1 、S 2 Calculate->
Figure GDA0004131737820000084
Select the rotation matrix->
Figure GDA0004131737820000085
Determination of midpoint
Figure GDA0004131737820000086
And & ->
Figure GDA0004131737820000087
Falls on Y 0 On both sides of the base.
For a physical quantity combination X of a time to be determined, Y = W is calculated T And (4) X. If Y and Y are 1 On the same side (i.e. all greater than or all less than Y) 0 ) If the time belongs to the annular time, linear judgment is carried out; if Y and Y are 2 If they fall on the same side, they are not cyclic, and they do not pass through the linear discrimination.
In order to try to further improve the screening accuracy, a BP neural network algorithm is adopted to replace linear discrimination.
The BP neural network is a multi-layer feedforward neural network (as shown in figure 2), and the main characteristics of the network are signal forward transmission and error backward transmission. In forward conduction, the input signal is processed layer by layer from the input layer through the hidden layer to the output layer. The neuronal state of each layer only affects the neuronal state of the next layer. If the output layer does not reach the expected output, the reverse operation is carried out, and the network weight and the threshold are adjusted according to the prediction error, so that the predicted output of the BP neural network continuously approaches the expected output. The BP neural network can be viewed as a nonlinear mapping of inputs to outputs, and the model can approximate a continuous function of arbitrary complexity with arbitrary precision.
The BP network standard learning algorithm assumes that an input layer has
Figure GDA00041317378200000912
Each neuron with hidden layers>
Figure GDA00041317378200000910
Neuron and output layer have d neurons, and the symbol is defined as follows:
inputting a vector:
Figure GDA00041317378200000913
hidden layer output vector:
Figure GDA00041317378200000911
output layer output vector y = (y) 1 ，y 2 ，…，y d )
The desired output vector is:
Figure GDA0004131737820000091
error function:
Figure GDA0004131737820000092
training a sample: (x) k ，y k ) K =1,2, … n; (n is the total time of the training set for the cyclic and acyclic typhoons)
The hidden layer excitation function is defined as f, and the common excitation function is in the form of sigmoid function
Figure GDA0004131737820000093
Or tanh function f (x) =2sigmoid (2 x) -1, and the like.
The learning step of the BP neural network is:
(1) Calculating the output values of the hidden layer neurons and the output layer neurons:
hiding the layer:
Figure GDA0004131737820000094
and (3) an output layer:
Figure GDA0004131737820000095
(2) Calculating the partial derivative of the error function to the weight of the output layer and transmitting the partial derivative to the weight of the hidden layer, and calculating the partial derivative of the error function to each weight of the hidden layer:
and (3) an output layer:
Figure GDA0004131737820000096
Figure GDA0004131737820000097
hiding the layer:
Figure GDA0004131737820000098
Figure GDA0004131737820000099
(3) And (3) correcting the output layer weight and the hidden layer weight by utilizing a gradient descent method and the partial derivative obtained in the step (2):
W1，2 ik ＝W ik -η·g(W ik ) And eta is the learning rate.
(4) And calculating the corrected error function value, ending the training if the error function value reaches the preset precision, and otherwise, selecting the next sample for learning.
The BP neural network adopted by the invention adopts a single hidden layer structure with only one middle layer, and the activation function is a sigmoid function. Fig. 3 gives a schematic diagram of the BP neural network. The input layer is determined by 5 independent variables obtained by pre-screening, namely the input variable x = [ sigma ] C ，VAR，ΔT eye ，u200，SST](ii) a Setting the number of hidden layer neurons as 40; the output layer is binary output, namely, cyclic time (1) or acyclic time (0). And finally, obtaining a BP neural network discrimination model based on the training set.
In the automatic identification process, all the time-frequency calculation results of typhoon to be judged in the step (1) are screened one by one through a pre-screening and linear discrimination (or pre-screening and BP neural network) identification program, when two continuous time-frequency screening is carried out, namely the two continuous time-frequency screening is considered to be an annular time-frequency, the automatic identification system outputs the current typhoon to belong to the annular typhoon, otherwise, the current typhoon is considered to belong to the non-annular typhoon and is not output. If the typhoon has data loss at a certain time, but the two times before and after the data loss are considered as the annular times, the automatic identification system also outputs the current typhoon as the annular typhoon.
FIG. 4 is a schematic illustration of the cloud image of the North West Pacific annular typhoon satellite (Chu and Tan, 2014) described in the present invention. It can be seen that such typhoons are mainly characterized by a clear ring-like structure, generally highly axisymmetric, with a larger typhoon eye than a normal typhoon, an almost uniform deep convection loop surrounding the typhoon eye, and a lack of spiral rain zones around the periphery of the deep convection loop. Furthermore, knaff et al, (2003) indicate that such typhoons are generated only under specific environmental conditions. The invention provides a theoretical basis for establishing the northwest Pacific ocean annular typhoon automatic identification system based on satellite data and environmental field and sea temperature field data.
Fig. 5 and 6 are areas of eastern and Atlantic circular hurricanes and northwestern circular typhoons generation and landing conditions (Chu and Tan, 2014), respectively. The annular typhoon has the characteristics of high intensity, long maintenance time, high forecasting difficulty and the like, but compared with annular hurricanes of the east pacific and the Atlantic, the annular typhoon generating area is far away from the continent and has low landing probability, and the annular typhoon generating area of the northwest pacific is close to the continent and has high landing probability (wherein all generated in the R1 area land), thereby continuously causing great threat to the east Asia region.
The examples of the invention are as follows:
the method is applied to the identification of all typhoons in northwest Pacific ocean in 2000-2008. The typhoon satellite data entered is from the latest hurricane satellite records project and observations from the international satellite cloud climate plan B1 geostationary satellite (HURSAT-B1). Environmental field data was from the national environmental forecast center and the national atmospheric research center at 6 hour intervals of the global re-analysis data (NCEP-NCAR analysis). Sea temperature data is derived from Reynolds ocean surface temperature reanalysis data (Reynolds's SST reanalysis).
Fig. 7 is a schematic flow chart of an automatic northwest pacific annular typhoon identification system of the present invention, which specifically includes:
1. physical quantity calculation:
and directly obtaining the current secondary typhoon position, the maximum wind speed (vmax) and the infrared brightness temperature from the typhoon satellite data. The infrared brightness temperature data distribution takes longitude and latitude as horizontal and vertical coordinates, after linear interpolation is respectively carried out on the infrared brightness temperature data along the longitude and latitude directions for two times, data under a polar coordinate system are obtained, and further an infrared brightness temperature sequence and a standard deviation (VAR) of the sequence are calculated, wherein the infrared brightness temperature sequence is averaged on the circumference of each radius from the center of the typhoon to a position 560km away from the center of the typhoon. Selecting the circumferential average coldest radius (R) C ) Analyzing typhoon eye size, standard deviation (sigma) of infrared bright temperature at coldest radius C ) Analysis eye wallOf the maximum temperature difference (Δ T) between the coldest radius and its inner circumference eye ) The convection intensity of the eye wall was analyzed.
And (4) calculating the temperature (t 200) of the upper part (200 hPa) of the troposphere (200-800 km) of the peripheral average of the current typhoon according to the environmental field data to obtain the high-rise temperature field characteristic. Similarly, calculating the characteristics (u 200, v200, u500, v500, u800, v 800) of the warp and weft wind fields of the 200-800km average troposphere (200 hPa), the middle troposphere (500 hPa) and the lower troposphere (800 hPa), and performing vector subtraction on the wind fields of the upper layer, the lower layer, the middle layer and the lower layer to obtain 200-850hPa deep vertical wind Shear (SHRD) and 500-850hPa shallow vertical wind shear (SHRS), thereby obtaining the environmental shear characteristics.
And interpolating the ocean surface temperature data in the sea temperature field data to the center position of the typhoon by adopting a cubic spline interpolation method to obtain the ocean surface temperature (SST) at the center of the typhoon.
2. Screening of the identification program:
using the current time vmax, R C ,ΔT eye SHRD, u200, SST is subjected to a prescreening (predi.m) and the time of passage is reused by σ C ，VAR，ΔT eye U200, SST performs linear discrimination (lda.m) or BP neural network recognition (predict.m).
When two continuous times pass the screening, namely the two continuous times are considered as the annular times, the automatic identification system outputs the current typhoon to belong to the annular typhoon, otherwise, the current typhoon is considered as the non-annular typhoon and is not output.
If the typhoon has data loss at a certain time, but the two times before and after the data loss are considered as the annular times, the automatic identification system also outputs the current typhoon as the annular typhoon. If the typhoon passes the screening continuously for multiple times, the automatic identification system will warn the typhoon for multiple times until all the time calculation of the typhoon is completed.
Tests show that if two groups of automatic identification systems are combined, namely the pre-screening and linear discrimination and the pre-screening and BP neural network consider that the current typhoon belongs to the annular typhoon, the current typhoon is identified as the annular typhoon, and the accuracy can be further improved.
The recognition results were finally obtained as follows (tables 2 to 4):
Figure GDA0004131737820000121
TABLE 2-4 northwest Pacific ocean annular typhoon automatic identification program identification results in 2000-2008
It can be seen that when the method is applied to the analysis of all typhoons in 2000-2008, the hit rate of the recognition system reaches 100%, the accuracy can reach 91.3%, and the method can be comparable to the automatic recognition system of annular hurricanes in the east Pacific and Atlantic.

Claims (2)

1. An automatic identification system for northwest Pacific ocean annular typhoon is characterized by comprising the following steps:
step 1: analyzing and calculating physical quantity, firstly calling time data of annular typhoon occurrence and recently occurring non-annular typhoon time data in an ocean historical time period, and calculating annular typhoon characteristic data of the ocean area, wherein the characteristic data comprises typhoon intensity, structure, environmental field and ocean temperature field characteristic data;
and 2, step: calling all annular typhoon times of the past northwest Pacific and all non-annular typhoon times of the current year as training sets, obtaining calculation results of the characteristic data of the strength, structure, environmental field and sea temperature field of the typhoon of the training sets, and establishing an automatic pre-screening recognition mechanism: when the typhoon to be evaluated meets the condition of the formula (1), pre-screening is carried out, otherwise, screening is carried out,
Figure FDA0004131737800000011
in the formula, V max Maximum wind speed of typhoon, R c Is the average coldest radius, Δ T, of the typhoon circumference eye The coldest radius of the typhoon is at the maximum value of the average infrared bright temperature difference of the inner circumference of the typhoon, SHRD is the deep vertical wind shear, u200 is 200hPa environmental latitude wind, SST is the central ocean surface temperature of the typhoon, sigma c Is the standard deviation of infrared bright temperature at the coldest radius of the typhoon;
and step 3: simultaneously, carrying out linear discrimination on training sets of all annular typhoon times of the past northwest Pacific and all non-annular typhoon times of the current year by sigma c 、VAR、ΔT eye U200 and SST are used as linear judgment bases, the middle points of projections of the annular time training set and the non-annular training set in a one-dimensional space are used as classification standards, and the time falling to one side of the non-annular typhoon is screened out; the VAR represents the standard deviation of a circumferential average infrared light temperature sequence;
and 4, step 4: for typhoon times pre-screened in the steps 2 and 3, judging a back propagation neural network, and inputting a variable sigma c 、VAR、ΔT eye U200 and SST, multiplying the weight coefficient obtained by training, and screening out typhoon time which is output to be non-annular time through an excitation function;
and 5: checking whether two continuous typhoon times pass through the steps 1-4, and judging that the current typhoon belongs to the annular typhoon when two continuous typhoon times are annular;
in the step 3 of the linear discrimination,
first, the [ σ ] of the cyclic time and the acyclic time in the training set c ,VAR,ΔT eye ,u 200 ,SST] T Form 5 Xn 1 And 5 xn 2 Matrix X of 1 And X 2 Wherein n is 1 And n 2 The number of the training set ring-shaped time and the number of the training set non-ring-shaped time are calculated, and the physical quantity time is averaged to obtain a matrix
Figure FDA0004131737800000021
And &>
Figure FDA0004131737800000022
Then, X is calculated 1 And X 2 Auto-covariance matrix S 1 And S 2 And by means of an autocovariance matrix S 1 And S 2 Calculating to obtain an intra-class walking matrix S,
Figure FDA0004131737800000023
then, a rotation matrix W is selected,
Figure FDA00041317378000000211
defining a decision midpoint Y 0 ，/>
Figure FDA0004131737800000024
Is rotated projection->
Figure FDA0004131737800000025
Rotating projection of->
Figure FDA0004131737800000026
Calculating a physical quantity combination X of the times to be determined, and further obtaining a determination value Y = W of the times to be determined T X, then determining the typhoon ring time by:
when Y is less than Y 0 And Y is 1 ＜Y 0 ＜Y 2 When, or Y > Y 0 And Y is 1 ＞Y 0 ＞Y 2 Judging the time to be a ring time;
when Y > Y 0 And Y is 1 ＜Y 0 ＜Y 2 When, or Y < Y 0 And Y is 1 ＞Y 0 ＞Y 2 Judging the time to be non-annular time;
in the process of establishing the BP neural network model in the step 4, sigma is used c ,VAR,ΔT eye ,u 200 SST is used as an input variable of an input layer, the number of neurons of a hidden layer is set to be 40, and the output result of an output layer is binary output;
establishing BP neural network model, assuming that there are l neurons in input layer and one neuron in hidden layer
Figure FDA00041317378000000210
Neuron, output layer has d neurons, and the symbol is defined as follows:
input vector x = (x) 1 ,x 2 ,x l )
Hidden layer output vector h = (h) 1 ,h 2 ,h q )
Output layer output vector y = (y) 1 ,y 2 ,y d )
The desired output vector is:
Figure FDA0004131737800000027
error function:
Figure FDA0004131737800000028
training a sample: (x) k ,y k ) K =1,2n, n is the total time number of the training set of the circular typhoon and the non-circular typhoon;
the excitation function is
Figure FDA0004131737800000029
Establishing a BP neural network model according to the training set by sigma c 、VAR、ΔT eye U200, SST as input variables x = (sigma) of input layer c ,VAR,ΔT eye ,u 200 SST), the number of hidden layer neurons is set to be 40, the output result of the output layer is binary output, the annular typhoon is 1, the non-annular typhoon is 0, and the learning step comprises:
(1) Computing the output values of hidden and output layer neurons:
hidden layer neuron output value:
Figure FDA0004131737800000031
output layer neuron output values:
Figure FDA0004131737800000032
in the formula, W1 is a hidden layer weight, and W2 is an output layer weight; the lower corner mark i is the serial number of the hidden layer neuron, and the lower corner mark j is the serial number of the output layer neuron;
(2) Calculating partial derivatives of the error function to the weights of the output layer and transmitting the partial derivatives to the hidden layer, and calculating the partial derivatives of the error function to the weights of the hidden layer:
an output layer:
Figure FDA0004131737800000033
hiding the layer:
Figure FDA0004131737800000034
(3) Utilizing a gradient descent method and the partial derivative obtained in the step (2) to carry out output layer weight W1' ij And hidden layer weight W2' ij And (5) correcting:
W1′ ij ＝W1 ij -η·g(W1 ij ),i＝1,2,…q,j＝1,2,…l (6)
W2′ ij ＝W2 ij -η·g(W2 ij ),i＝1,2,…d,j＝1,2,…q (7)
in the above formula, η is the learning rate;
(4) And calculating the corrected error function value, finishing training if the error function value reaches the preset precision, storing the hidden layer weight W1 and the output layer weight W2, determining the BP neural network model, and otherwise, selecting the next sample for learning, or selecting the next sample for learning.
2. The northwest pacific ring typhoon automatic identification system according to claim 1, wherein: in the BP neural network model, the method for determining the cyclic time in the step (4) is as follows:
the typhoon time to be judged is treated, and the sigma of the time is c 、VAR、ΔT eye And u200 and SST are input into a BP neural network model, hidden layer weight W1 and output layer weight W2 obtained by training are read, hidden layer neuron output values and output layer neuron output values obtained through an excitation function are respectively calculated according to the formula (2) and the formula (3), if the output layer neuron output values are 1, the cyclic time is judged, and if the output layer neuron output values are 0, the non-cyclic time is judged.
CN201711407247.5A 2017-12-22 2017-12-22 Automatic northwest Pacific annular typhoon identification system Active CN108196314B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711407247.5A CN108196314B (en) 2017-12-22 2017-12-22 Automatic northwest Pacific annular typhoon identification system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711407247.5A CN108196314B (en) 2017-12-22 2017-12-22 Automatic northwest Pacific annular typhoon identification system

Publications (2)

Publication Number Publication Date
CN108196314A CN108196314A (en) 2018-06-22
CN108196314B true CN108196314B (en) 2023-04-18

Family

ID=62583533

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711407247.5A Active CN108196314B (en) 2017-12-22 2017-12-22 Automatic northwest Pacific annular typhoon identification system

Country Status (1)

Country Link
CN (1) CN108196314B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109583653B (en) * 2018-11-30 2022-06-21 南京信息工程大学 Northwest Pacific ocean typhoon extension period forecasting method based on statistical model
CN109583095B (en) * 2018-11-30 2023-04-28 南京信息工程大学 North Pacific typhoon extension period forecasting method based on hybrid statistical power model
CN109993282B (en) * 2019-03-28 2020-06-12 北京航空航天大学 Typhoon wave and range prediction method
CN110852169B (en) * 2019-10-12 2023-02-03 中国气象局上海台风研究所(上海市气象科学研究所) Method for identifying typhoon maximum wind speed radius based on airborne data
CN114740550B (en) * 2022-06-14 2022-08-26 广东海洋大学 Intelligent recognition early warning method and system for continuous storm events
CN117036983B (en) * 2023-10-08 2024-01-30 中国海洋大学 Typhoon center positioning method based on physical reinforcement deep learning

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2407044C1 (en) * 2009-09-30 2010-12-20 Павел Николаевич Манташьян Method of lowering temperature of ocean surface water in area of possible occurrence of typhoon
CN103955009A (en) * 2014-04-25 2014-07-30 宁波市气象台 Method for extracting typhoon objective forecast information from numerical forecasting product
CN104036462A (en) * 2014-06-06 2014-09-10 南京大学 Method to estimate intensity of tropical cyclone across the Northwest Pacific Ocean by using geostationary meteorological satellite image
CN107103173A (en) * 2016-10-31 2017-08-29 陈柏宇 A kind of Design Wave projectional technique for embodying the influence of the factor of typhoon three

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
RU2407044C1 (en) * 2009-09-30 2010-12-20 Павел Николаевич Манташьян Method of lowering temperature of ocean surface water in area of possible occurrence of typhoon
CN103955009A (en) * 2014-04-25 2014-07-30 宁波市气象台 Method for extracting typhoon objective forecast information from numerical forecasting product
CN104036462A (en) * 2014-06-06 2014-09-10 南京大学 Method to estimate intensity of tropical cyclone across the Northwest Pacific Ocean by using geostationary meteorological satellite image
CN107103173A (en) * 2016-10-31 2017-08-29 陈柏宇 A kind of Design Wave projectional technique for embodying the influence of the factor of typhoon three

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
聂高臻 等.台风"云娜"(2004)的数值模拟:眼墙对流与环境风切变的关系.《南京大学学报（自然科学）》.2010,第46卷(第3期),第317-327页. *

Also Published As

Publication number Publication date
CN108196314A (en) 2018-06-22

Similar Documents

Publication Publication Date Title
CN108196314B (en) Automatic northwest Pacific annular typhoon identification system
Giffard-Roisin et al. Tropical cyclone track forecasting using fused deep learning from aligned reanalysis data
CN107180530B (en) A kind of road network trend prediction method based on depth space-time convolution loop network
Wang et al. Tropical cyclone intensity estimation from geostationary satellite imagery using deep convolutional neural networks
CN109917394B (en) Short-term intelligent extrapolation method based on weather radar
Liu et al. Chaotic oscillatory-based neural network for wind shear and turbulence forecast with LiDAR data
CN109086540A (en) A kind of method and device constructing Forecasting of Tropical Cyclone model
Valverde et al. Neural network and fuzzy logic statistical downscaling of atmospheric circulation-type specific weather pattern for rainfall forecasting
CN110322075A (en) A kind of scenic spot passenger flow forecast method and system based on hybrid optimization RBF neural
CN111723523B (en) Estuary surplus water level prediction method based on cascade neural network
Liu et al. Short-term daily prediction of sea ice concentration based on deep learning of gradient loss function
Qin et al. Trajectory prediction based on long short-term memory network and Kalman filter using hurricanes as an example
Wang et al. A conformal regressor with random forests for tropical cyclone intensity estimation
Lababede et al. Mosques smart domes system using machine learning algorithms
CN109829583A (en) Mountain fire Risk Forecast Method based on probability programming technique
CN108510097B (en) Power transmission line corridor snow disaster monitoring method and system based on satellite remote sensing
JPH0949884A (en) Predicting method for local weather using neural network
CN116524230A (en) Ship type classification and identification method based on priori knowledge base
Kwong et al. An artificial neural network with chaotic oscillator for wind shear alerting
Valach et al. Geoeffectiveness of XRA events associated with RSP II and/or RSP IV estimated using the artificial neural network
CN115345245A (en) Tropical cyclone rapid reinforcement forecasting method based on random forest and transfer learning
Bari et al. LVP conditions at Mohamed V airport, Morocco: local characteristics and prediction using neural networks
Farooq et al. A hybrid time series forecasting model for disturbance storm time index using a competitive brain emotional neural network and neo-fuzzy neurons
Zhang et al. A three-dimensional convolutional-recurrent network for convective storm nowcasting
CN107231668A (en) Transmission mechanism model under forest fire based on gray scale association analysis

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