CN109164450A - A kind of downburst prediction technique based on Doppler Radar Data - Google Patents
A kind of downburst prediction technique based on Doppler Radar Data Download PDFInfo
- Publication number
- CN109164450A CN109164450A CN201811060986.6A CN201811060986A CN109164450A CN 109164450 A CN109164450 A CN 109164450A CN 201811060986 A CN201811060986 A CN 201811060986A CN 109164450 A CN109164450 A CN 109164450A
- Authority
- CN
- China
- Prior art keywords
- convection cell
- sequence
- collection
- disaster
- convergence
- 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
Links
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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The downburst prediction technique based on Doppler Radar Data that the invention discloses a kind of, it include: that the Storm Events collected 10 years report matched Doppler radar data, it identifies the convection cell at all moment, extracts static nature, obtain convection cell sequence；Convection cell sequence is matched with Storm Events；The moment occurs as time reference using disaster, convection cell sequence is split as several sample sequence segment collection；It extracts the feature set of sample sequence segment collection and carries out statistical significance inspection, find the notable feature collection of positive and negative sample characteristics collection；By principal component analysis, dimensionality reduction and decoupling are carried out to significant characteristics collection, validity feature, and then one sorter model of training is obtained, obtains a classifiers model；Combined application sorter model set, detection and forecast downburst.The present invention can detect the convection cell for generating downburst automatically, predict the time of origin and scene of downburst.
Description
Technical field
The present invention relates to the downburst hazard predictions in meteorology field, especially with Doppler radar data pair
Generate the method that the convection cell of downburst carries out automatic identification.
Background technique
Convective strong wind includes linearity strong wind, spout etc.；Linearity strong wind includes bow echo and downburst again.Its
In, downburst the ground such as airport and harbour harm clearly.In weather forecast field, Doppler radar is a kind of
The tool of effective monitoring Convective Storms can obtain the three-dimensional structure of Convective Storms using Doppler radar, while again
The particle radial motion speed being able to detect inside Convective Storms.By means of Doppler radar, it is able to detect under generation and hits
Some notable features existing for the convection cell flowed cruelly, the monomer core height such as uphang, middle layer radial direction convergence.
Common downburst automatic identifying method is to hit under being analyzed in several cases using Doppler radar cruelly
The some features for flowing convection cell, such as reflectivity Morphologic Characteristics and radial velocity structure feature.And it is based on these features
Devise the Prediction program of some downbursts.However the frame simply shows the application effect of several cases, does not give
The effect of algorithm actual motion out.
Automatic detection and prediction technique there are also a kind of downburst are by linear classifier, to downburst convection current list
Several Reflectivities and radial velocity characters of body have carried out combined test, have found some and downburst maximally related one
A little features.And a linear classifier is had trained using these features, it being capable of automatic identification and prediction downburst disaster；So
And this method is 0.40 in the discrimination between 0 to 40 kilometers of radar, the discrimination between 40 to 80 kilometers is
0.20。
The above-mentioned automatic identifying method of downburst in the prior art at least has the following disadvantages and deficiency:
The static nature that only only used convection cell, without considering that convection cell Character evolution is classified in convection cell
In effect.In addition, the discrimination to downburst is lower, and the downburst automatic identification based on Doppler radar and
There are improved spaces for prediction technique.
[bibliography]
[1]Johnson J T,Mackeen P L,Witt A,et al.The Storm Cell Identification
and Tracking Algorithm:An Enhanced WSR-88D Algorithm[J].Wea Forecasting,1998,
13(2):263-276.
[2]Smith T M,Elmore K L,Dulin S A.A Damaging Downburst Prediction and
Detection Algorithm for the WSR-88D[J].Weather&Forecasting,2004,19(2):240.
[3] Luo Hui, Zhang Jie, Zhu Keyun wait the early radar warning quantizating index of downburst to study [J] meteorology journal,
2015,73(5):853-867.
Summary of the invention
For the prior art, the downburst prediction technique based on Doppler Radar Data that the present invention provides a kind of is used
In the automatic Prediction downburst in meteorology, this prediction technique can detect that the convection current list of downburst will be generated automatically
Body carries out timely early warning to disaster.
In order to detect the convection cell for generating downburst automatically, time of origin and the spot of downburst are predicted
Point carries out timely early warning to disaster, reduces economic loss and casualties, proposed by the present invention a kind of based on Doppler radar
The downburst prediction technique of data, comprising the following steps:
Step 1: the Storm Events collected 10 years report matched Doppler radar data；
Step 2: from the radar data of each Storm Events, being identified using storm core identification and method for tracing
The convection cell at all moment extracts the static nature of convection cell, and is tracked to obtain convection cell sequence to convection cell
Column；
The static characteristic includes Reflectivity and velocity characteristic；
The Reflectivity includes core monomer height, monomer maximum reflectivity, monomer maximum reflectivity height, bottom height
With rise；
Velocity jump point is detected on radar radial velocity image, when velocity jump value is greater than a given threshold value,
Then the velocity jump point is convergency point, and adjacent convergency point is merged, and the conjunction by constantly executing adjacent position convergency point
And operate, the convergency point belt-like zone inside convection cell is found, which is known as belt of convergency；Traverse all spokes
Chalaza calculates positive speed point n in its eight neighborhood+, positive speed summationAnd maximum positive speedNegative velocity points n-, bear speed
Spend summationAnd maximum negative velocityThe then convergence value of the convergency pointThe convergence of the convergency point is average
ValueAccording to convergency point coordinate each in belt of convergency with least square method fitting a straight line, obtained by calculating
Fitting a straight line and radar projection straight line angle theta, be the new of the convergency point by the A/cos θ of convergency point each in belt of convergency
Convergence value；
Obtain the velocity characteristic of the convection cell by statistics, including middle layer convergence maximum value, middle layer convergence minimum value,
Middle layer convergence average value maximum value, middle layer convergence average value minimum value, convergence maximum value and convergence maximum layer average value, convergence
Point sum, convergence maximum layer height, convergence maximum layer points, convergence maximum value height, convergence minimum value height, convergence average value
Maximum value height and convergence average value minimum value height；
Step 3: convection cell sequence is matched with Storm Events, disaster is had recorded in Storm Events report
When and where, if disaster occur the moment, a convection cell sequence with a distance from disaster spot be less than 10km, then
Think that the convection cell sequence is related to disaster report；If having multiple convection cell sequences and storm disaster simultaneously
Distance is less than 10km, then it is assumed that wherein related with disaster report apart from nearest convection cell sequence to storm disaster；
To all convection cell sequences relevant to disaster report, according to convection cell sequence in the report of relevant disaster
The difference of disaster is divided into:
Convection cell sequence relevant to downburst,
Convection cell sequence relevant to hail,
Convection cell sequence relevant to heavy rain；
Step 4: the moment occurs as time reference using disaster, convection cell sequence is split as several sample sequence pieces
Section collection；Step is:
It is defined as follows two time parameters: Timing Advance first: point, biased forwards on the basis of at the time of disaster occurs
The body total number be Timing Advance；The value range of Timing Advance is that 1~3 individual is swept；Sequence fragment length: in the convection current
The length of the segment intercepted in sequence monomer, the value range of sequence fragment length are that 3~6 individuals are swept；It is different by combination
Timing Advance and sequence fragment length, the convection cell sequence of all Disasters Types is split as 12 sequence fragment collection；
When splitting convection cell sequence, at the time of disaster occurs on the basis of point, first biased forwards Timing Advance ginseng
The determining body total number of number, will be divided into 1~6 segment according to the difference of convection cell sequence length according to sequence fragment length；
Each segment all has two time parameters of Timing Advance and sequence fragment length, by the sequence with same time parameter
Segment is stored in a set, to form sample sequence segment collection；
The sample sequence segment collection that disaster is downburst is denoted as positive sample sequence fragment collection, by disaster for hail and cruelly
The relevant sample sequence segment collection of rain is denoted as negative sample sequence fragment collection；
Step 5: extracting the feature set of sample sequence segment collection: calculating the time difference of the static nature of sample sequence segment
Divide information characteristics sequence, time difference information characteristics sequence refers to the static state of each of sample sequence segment collection convection cell
Change rate of the feature within some period, for each static nature f of convection cell, in moment t, the convection cell
Time difference informationWherein,It is static nature f in t+dtThe value at moment,
It is static nature f in t-dtThe value at moment；dtValue range be 3-9min；If the convection cell is in affiliated sample sequence piece
The both ends of section, then 0 filling of missing values；The feature of sample sequence segment is all static natures and time difference information in segment
The maximum value and minimum value of feature describe sample sequence segment with a feature vector；Obtain all positive and negative sample sequence pieces
The feature set of i.e. extractable sample sequence segment collection after the sequence fragment feature of section collection；Have in the feature set of sample sequence segment collection
There is time parameter identical with corresponding sequence segment collection；
Step 6: using each feature of the positive and negative sample characteristics collection with same time parameter as one group of input, respectively
Statistical significance inspection is carried out, using all features of level of significance α < 0.05 as the positive and negative sample characteristics under the time parameter
The notable feature collection of collection；
Step 7: by principal component analysis, the conspicuousness of all positive and negative sample characteristics collection with same time parameter
Feature set carries out dimensionality reduction and decoupling, retains 85% validity feature, as having for the positive and negative sample characteristics collection under the time parameter
Imitate feature；
Step 8: according to the validity feature of the positive and negative sample characteristics collection with same time parameter, one classifier of training
Model obtains a classifiers model；Combined application sorter model set, detection and forecast downburst.
Compared with prior art, the beneficial effect of the technical scheme provided by the present invention is that: pass through and improve original convection cell
The calculation method of radial convergence strength characteristic joined convection cell sequence on the basis of the static nature of convection cell
Time difference feature generates the convection cell and non-downburst convection cell of downburst using these feature differentiations, realizes
Downburst convection cell automatic identifying method.In addition, passing through the piece that the sequence of convection cell is split as to different length
Duan Jihe, using the set training convection cell of individual segment, so that relevant temporal information will be attached in classifier,
Enable the result of classifier to the predicted time of downburst out.Timely early warning is carried out to disaster, reduces economic damage
It becomes estranged casualties；And pass through the experimental verification validity of this method.
Detailed description of the invention
Fig. 1 is the recognition result of a convection cell on radar reflectivity image；
Fig. 2 a is the original radar radial velocity map in zone of convergence；
Fig. 2 b is to close the method that velocity amplitude is modified to middle layer radial spoke to illustrate schematic diagram；
Fig. 3 matches schematic diagram with disaster report for monomer track；
Fig. 4 a is the sample of 0.5 degree of elevation angle reflectivity sequence fragment of convection cell sequence；
Fig. 4 b is the convection cell sequence fragment sample after splitting；
Fig. 5 a is the example of 1 conspicuousness of feature in analytical sequence segment；
Fig. 5 b is the example of 2 conspicuousness of feature in analytical sequence segment；
Fig. 6 is that convection cell sequence fragment inputs classifier methods schematic diagram after splitting.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is described in further detail.
A kind of downburst prediction technique based on Doppler Radar Data proposed by the present invention, comprising the following steps:
Step 1: collect 10 years Storm Events (including downburst and hail Rainfall Disaster) report and it is matched
Doppler radar data.
Step 2: identification, tracking convection cell in all radar datas, and extract convection cell feature；Use wind
Sudden and violent core identification and method for tracing identify the convection cell at all moment, mention from the radar data of each Storm Events
The static nature of convection cell is taken, and convection cell is tracked to obtain convection cell sequence, here the static state of convection cell
Feature refers to the feature for not considering temporal information during calculating；
Wherein, the static characteristic includes Reflectivity and velocity characteristic；
The Reflectivity includes core monomer height, monomer maximum reflectivity, monomer maximum reflectivity height, bottom height
With rise, as shown in table 1.
Velocity jump point is detected on radar radial velocity image, when velocity jump value is greater than a given threshold value,
Then the velocity jump point is convergency point, and adjacent convergency point is merged, and the conjunction by constantly executing adjacent position convergency point
And operate, the convergency point belt-like zone inside convection cell is found, which is known as belt of convergency；Traverse all spokes
Chalaza calculates positive speed point n in its eight neighborhood+, positive speed summationAnd maximum positive speedNegative velocity points n-, bear speed
Spend summationAnd maximum negative velocityThe then convergence value of the convergency pointThe convergence of the convergency point is average
ValueAccording to convergency point coordinate each in belt of convergency with least square method fitting a straight line, obtained by calculating
Fitting a straight line and radar projection straight line angle theta, be the new of the convergency point by the A/cos θ of convergency point each in belt of convergency
Convergence value；
Obtain the velocity characteristic of the convection cell by statistics, including middle layer convergence maximum value, middle layer convergence minimum value,
Middle layer convergence average value maximum value, middle layer convergence average value minimum value, convergence maximum value and convergence maximum layer average value, convergence
Point sum, convergence maximum layer height, convergence maximum layer points, convergence maximum value height, convergence minimum value height, convergence average value
Maximum value height and convergence average value minimum value height, as shown in table 1.
Table 1
Step 3: needing the convection cell sequence identified in step 2 by these convection cell sequences and storm
Event report record establishes association.Convection cell sequence is matched with Storm Events, has recorded calamity in Storm Events report
The when and where that evil occurs, if the moment occurs in disaster, a convection cell sequence is less than with a distance from disaster spot
10km, then it is assumed that the convection cell sequence is related to disaster report；If having multiple convection cell sequences and a wind simultaneously
The distance of sudden and violent disaster is less than 10km, then it is assumed that wherein reports phase apart from nearest convection cell sequence and the disaster with storm disaster
It closes；
To all convection cell sequences relevant to disaster report, by convection cell sequence samples according to its relevant disaster
Weather pattern classification is divided into according to convection cell sequence to the difference of disaster in the report of relevant disaster:
Convection cell sequence relevant to downburst,
Convection cell sequence relevant to hail,
Convection cell sequence relevant to heavy rain.
Step 4: the moment occurs as time reference using disaster, convection cell sequence is split as several sample sequence pieces
Section collection；Specific step is as follows:
It is defined as follows two time parameters: Timing Advance first: point, biased forwards on the basis of at the time of disaster occurs
The body total number be Timing Advance；The value range of Timing Advance is that 1~3 individual is swept；Sequence fragment length: in the convection current
The length of the segment intercepted in sequence monomer, the value range of sequence fragment length are that 3~6 individuals are swept；It is different by combination
Timing Advance and sequence fragment length, the convection cell sequence of all Disasters Types is split as 12 sequence fragment collection；
When splitting convection cell sequence, at the time of disaster occurs on the basis of point, first biased forwards Timing Advance ginseng
The determining body total number of number, will be divided into 1~6 segment according to the difference of convection cell sequence length according to sequence fragment length；
Each segment all has two time parameters of Timing Advance and sequence fragment length, by the sequence with same time parameter
Segment is stored in a set, to form sample sequence segment collection；
The sample sequence segment collection that disaster is downburst is denoted as positive sample sequence fragment collection, by disaster for hail and cruelly
The relevant sample sequence segment collection of rain is denoted as negative sample sequence fragment collection.
Step 5: extracting the feature set of sample sequence segment collection: calculating the time difference of the static nature of sample sequence segment
Divide information characteristics sequence, time difference information characteristics sequence refers to the static state of each of sample sequence segment collection convection cell
Change rate of the feature within some period, for each static nature f of convection cell, in moment t, the convection cell
Time difference informationWherein,It is static nature f in t+dtThe value at moment,
It is static nature f in t-dtThe value at moment；dtValue range be 3-9min；If the convection cell is in affiliated sample sequence piece
The both ends of section, then 0 filling of missing values；The feature of sample sequence segment is all static natures and time difference information in segment
The maximum value and minimum value of feature describe sample sequence segment with a feature vector；Obtain all positive and negative sample sequence pieces
The feature set of i.e. extractable sample sequence segment collection after the sequence fragment feature of section collection；Have in the feature set of sample sequence segment collection
There is time parameter identical with corresponding sequence segment collection.
Step 6: using each feature of the positive and negative sample characteristics collection with same time parameter as one group of input, respectively
Statistical significance inspection is carried out, using all features of level of significance α < 0.05 as the positive and negative sample characteristics under the time parameter
The notable feature collection of collection.
Step 7: by principal component analysis, the conspicuousness of all positive and negative sample characteristics collection with same time parameter
Feature set carries out dimensionality reduction and decoupling, retains 85% validity feature, as having for the positive and negative sample characteristics collection under the time parameter
Imitate feature.
Step 8: according to the validity feature of the positive and negative sample characteristics collection with same time parameter, in each sequence fragment
On collection, one sorter model of training obtains a classifiers model；Combined application sorter model set, detection and forecast
Downburst.
Experimental example: for downburst to be identified and forecast in meteorology, steps are as follows:
1) the radar historical data of downburst and other types hazard weather is collected；
The information about downburst and the historical sample data of other hazard weather types being collected into are as follows: all samples
The initial time of data and termination time are respectively 2008-2017.By analysis storm disaster report, altogether 594 are had found
Item includes that the Disaster Event report of downburst and 6289 record with the incoherent disaster of downburst.By down-sampled, from
594 records have been picked out in the sample of non-downburst disaster.
It is reported and is recorded according to the disaster of downburst and non-downburst, have found 1254 altogether from radar database
Radar scanning data, between radar data between be divided into 3 minutes.One disaster records corresponding one group of radar data sequence, referred to as
One case, the average length of case are that 63 individuals are swept.
2) it identifies, the convection cell in tracking case, and extracts convection cell feature；
For all cases collected in step 1), identified using storm core identification and method for tracing (SCIT algorithm),
Track the convection cell in case.Multiple convection cells are usually had in one case.By SCIT method in all cases
On, it detected 12316 convection cell sequences altogether, the mean time of sequence a length of 45 minutes.Have 9.8 in average each case
A convection cell sequence.
The recognition result of a convection cell on radar reflectivity image is given in Fig. 1.Wherein, corresponding pair of every a line
A moment of monomer is flowed, each arranges a corresponding elevation angle, is the bottom Fig. 1 monomer at the time of related to downburst.Figure
The feature that upper convection cell extracts at the downburst relevant instant moment are as follows: monomer core height is 4.31km, and maximum reflectivity is
78dbz, maximum reflectivity height are 3.77km, a height of 1.35km in bottom, are risen as 9.41km, and convergence maximum value is 22m/s, convergence
Maximum value height is 5.12km, and convergence average value maximum value is 15.08m/s, and convergence average value minimum value is 6.14m/s, convergence
Minimum value is 12m/s, and convergence minimum value height is 3.84km, convergence maximum value and be 268m/s, and average value maximum value height is
5.15km, average value minimum value height are 2.66km, and convergency point sum is 17, and convergence maximum value average value is 15.76m/s, spoke
Closing layer line extreme length is 7.These features include Reflectivity, radial velocity characters.
The modification method of convection cell inner radial convergence field intensity is innovative point of the invention, is illustrated in figure using one
Figure come illustrate how amendment convection cell inside radial convergence field intensity value.
Fig. 2 a is radial velocity structure chart of the convection cell on 0.5 ° of elevation angle.In the convergence that radar ray direction finding arrives
Band is shown in figure 2b, illustrates straight line fitting result, the directions of rays of radar, convection cell of the belt of convergency in Fig. 2 b simultaneously
The direction of motion and radar directions of rays and belt of convergency normal vector angle.Convection cell inner radial convergence field in Fig. 2 a
Intensity is 42m/s.After being corrected by speed, obtained convergence field intensity is 60m/s.
3) by convection cell sequence samples according to its relevant hazard weather classification of type；
For the convection cell sequence identified in step 2), need to report these convection cell sequences and Storm Events
Record establishes association.One exemplary diagram is as shown in figure 3, which show the position of a downburst Storm Events, Yi Ji
The track of convection cell target and convection cell near this moment all downbursts.Wherein with the downburst disaster phase
The convection cell sequence of pass shares 3, only related to establishing apart from nearest sequence 1.
After each disaster record is all related apart from nearest convection cell sequence foundation to one, obtain altogether
551 convection cell sequences relevant to downburst storm disaster event report, 208 are reported with hail Rainfall Disaster event
The relevant sequence of related convection cell and 6289 report associated convection cell sequence without any Disaster Event,
Realize the classification of the monomer to downburst disaster.
4) moment is occurred as time reference using disaster, convection cell sequence is split as to the set of several sequence fragments；
The sample of a fractionation convection cell sequence fragment in the present embodiment is given in Fig. 4 a, wherein convection cell
One co-exists in the segment at 12 moment, the radar reflectivity figure at 0.5 degree of elevation angle is shown in Fig. 4 a, at the time of related to downburst
It is marked with wire frame, Fig. 4 b gives the convection cell sequence fragment sample after splitting.Convection cell sequence shown in Fig. 4 a
Segment is split for 6 segments altogether.Using three digit word marks, it occurs relative to disaster each segment in the sequence
The position at moment.
To all right in convection cell sequence relevant to downburst and convection cell sequence relevant with hail heavy rain
It flows after sequence monomer carries out segment fractionation, one has been obtained 50 set of segments, each not phase of the number of segment in each set
Together.551 segments are contained in maximum set, and 51 segments are contained in the smallest set.Number of fragments is flat in set
Mean value is 242.4.
5) feature of abstraction sequence segment；
The feature vector of one sample sequence segment of acquisition is illustrated how in table 2 using a simple example.Table 2 is opened up
The case where three features change over time in the convection cell sequence is shown, by seeking in each feature evolution data most
It is worth greatly, minimum value, the data of three features can be using a 6 dimensional vectors description in table 2: (80,77,3.75,3.61,
11.15,9.08), realize that the characteristic variation curve data of sequence fragment are converted into a feature vector.In this way, exist
Each sequence fragment is described with the feature vector that a length is 136 in practical operation.
Table 2
| 01 point 24 minutes | 01 point 28 minutes | 01 point 33 minutes | 01 point 37 minutes | |
| Maximum reflection value | 77 | 80 | 80 | 77 |
| Maximum reflection value height | 3.61 | 2.46 | 2.49 | 3.75 |
| Echo high | 9.08 | 11.15 | 9,14 | 9,36 |
6) it by significance test, finds out each sequence fragment and concentrates, the specific characteristic of downburst convection cell；
Analyze the example that positive and negative sample characteristics concentrate each feature significance.Fig. 5 a and Fig. 5 b are respectively shown at one
Sequence fragment collection closes, and the curve in figure with dot is positive sample characteristics, without the sample characteristics that are negative of dot, feature 1 and spy
Distribution situation of the sign 2 in positive sample and negative sample.By calculating, the level of significance α of feature 1 is on positive negative sample
0.000827, illustrate that this feature can effectively distinguish the downburst in the sequence and non-downburst segment；In positive negative sample
The level of significance α of upper feature 2 is 0.104, illustrates that this feature can not effectively distinguish the downburst in the sequence and non-
Downburst segment.Therefore, in this set of segments, feature 2, keeping characteristics 1 are deleted.
By carrying out above-mentioned analysis to each of positive and negative sample characteristics collection feature, retain significance less than 0.05
The validity feature concentrated as positive and negative sample characteristics of feature.After screening, on all positive and negative sample characteristics collection, put down
Eliminate 98.5 invalid features.
7) by principal component analysis, Feature Dimension Reduction and the decoupling of downburst convection cell are concentrated to each sequence fragment,
Retain 85% validity feature, the validity feature as the positive and negative sample characteristics collection under the time parameter.
The significant characteristics of one positive and negative sample characteristics collection with same time parameter are carried out with the knot of principal component analysis
Fruit, as shown in table 3, wherein first row is the number of principal component, and the first behavior characteristic characterization removes the lower master of contribution margin in table
Ingredient component is achieved that the dimensionality reduction of feature, can describe the sequence fragment by 3 features.It, can by the data in table 3
To know that the feature of every principal component is constituted situation.The significant characteristics of all positive and negative sample characteristics collection are all used above
After principal component analytical method carries out dimensionality reduction, the average characteristics number for describing sequence is reduced to 3 from 12.
Table 3
8) on each sequence fragment collection, one sorter model of training obtains a classifiers model；Combined application point
Class device model set, on-line real-time measuremen and forecast downburst.
For the supporting vector machine model of Gaussian kernel, trained purpose finds optimal the sorter model used in invention
Two model parameters.By taking sequence fragment set 3-2-2 as an example, positive and negative sample set that sequence fragment collection is closed with the ratio of 4:1,
It is divided into training set and test set, extracts input of each principal component component of validity feature on training set and test set as classifier
Vector；It later, is stepping with 0.05, training 100 times finds C value between 0-2 and gamma value makes classification accuracy rate highest,
Two optimal model parameters obtained are respectively as follows: C=1.45, gamma=0.05.
The result of classifier training further includes the testing evaluation value of classifier in addition to the parameter of classifier.Collected by verifying
It closes, obtains the accuracy rate of classifier；It is closed in test set, obtains the error rate of classifier, be with sequence fragment set 3-2-2
Sample data, verifying collection are combined into sequence fragment set 3-2-2, and test set is remaining all sequences set of segments in addition to 3-2-2,
The accuracy rate and error rate of the classifier of acquisition are respectively 0.7 and 0.3.The classifier meets performance indicator requirement, and (accuracy rate is big
In 0.6, error rate is less than 0.5), so retaining.
In the above manner, all having trained a classifier on all sequence fragments, and test classifier
The classifier for being wherein unsatisfactory for performance indicator requirement is removed, finally only remains 16 classifiers, corresponding sequence sample by performance
This segment is respectively as follows:
3_1_1,3_2_1,4_2_1,4_3_1,5_1_1,5_2_1,5_3_1,6_2_1,6_3_1,3_3_2,4_2_2,5_
1_2,5_2_2,6_1_2,5_1_3,3_2_4。
One integrated application sorter model is measured in real time and forecasts that the sample of downburst is as shown in Figure 6: in Fig. 6
Each sequence fragment be sequentially inputted in classifier corresponding with its length, for example, length be 3 sequence, be input to volume
Number for 3-1-1 classification wherein, illustrate the relationship of sequence fragment and classifier in the case in right end, while also showing point
The output result of class device.In this case, exporting altogether there are three classifier in three fragment lengths is Yes, wherein classifier
The Performance Score highest of 3-1-1, only retains the recognition result of the classifier.According to the recognition result of the classifier, calculate i.e.
It is after 1 individual is swept, to issue the early warning of downburst disaster by the time that downburst occurs.
Meteorology is used in order to verify a kind of downburst prediction technique based on Doppler Radar Data provided by the invention
Identify and predict that the feasibility of downburst carries out following test experiments in:
In test experiments altogether apply 16 classifiers for downburst disaster automatic identification and prediction, to this 16
The performance of a classifier is individually evaluated and tested.The data of test process application are the set of segments of convection cell sequence.It is each
The performance of a classifier is assessed by accuracy rate and rate of false alarm.
First accuracy rate tested for assessing each classifier.It is evaluated and tested in an experiment using cross-validation method point
The accuracy of class device.In a cross-validation process, need the training sample (including positive sample and negative sample) of classifier,
Proportionally 4:1 is divided into two parts, is trained with the sample that accounting is 4/5 to classifier, the test sample for being 1/5 with accounting
The accuracy of classifier.
The accuracy evaluation index of classifier is critical success index (CSI), its calculation formula is CSI=X/ (X+Y+Z),
Wherein X is the number that positive sample is divided into positive sample, and Y is the number that negative sample is classified as to positive sample, and Z is that positive sample is classified
For the number of negative sample.The variation range of critical success index CSI is 0 to 1, when the value of critical success index is 1, classification
The performance of device is best.
Appeal cross-validation process 100 times is repeated, counts the average value of the critical success index CSI of acquisition, then with more
The measurement of accuracy rate of the average value for the critical success index that secondary test obtains as the classifier.
The accuracy rate score value for 16 classifiers applied in this test experiments is given in table 4.It can from table 4
Out, the range of the accuracy rate score value of all classifiers illustrates that these classifiers can be distinguished effectively between 0.6 to 0.8
The segment of downburst convection cell and non-downburst convection cell.
Table 4
| Classifier | 3_1_1 | 3_2_1 | 4_2_1 | 4_3_1 | 5_1_1 | 5_2_1 |
| Accuracy | 0.69 | 0.70 | 0.73 | 0.70 | 0.73 | 0.68 |
| Classifier | 5_3_1 | 6_2_1 | 6_3_1 | 3_3_2 | 4_2_2 | 5_1_2 |
| Accuracy | 0.71 | 0.73 | 0.72 | 0.68 | 0.73 | 0.76 |
| Classifier | 5_2_2 | 6_1_2 | 5_1_3 | 3_2_4 | ||
| Accuracy | 0.74 | 0.76 | 0.73 | 0.71 |
Second rate of false alarm tested for assessing each classifier.In this test, trained for one point
Class device uses the data in other sequences set of segments as the test data of the classifier, tests the classifier for other
The calculation formula of the rate of false alarm FAR, FAR of sequence fragment data are as follows: FAR=A1/A1+B1, wherein A1 is to mark test data
For the number of positive sample, B1 is the number that test data is labeled as to negative sample.For a classifier, rate of false alarm is lower,
Performance is better.
The rate of false alarm score value for 18 classifiers applied in this test experiments is given in table 5.It can from table 5
Out, below the 0.5 of the rate of false alarm score value of all classifiers.Illustrate that these classifiers can effectively identify downburst
The sequence fragment in different time periods of convection cell.
Table 5
| Classifier | 3_1_1 | 3_2_1 | 4_2_1 | 4_3_1 | 5_1_1 | 5_2_1 |
| Error rate | 0.48 | 0.49 | 0.47 | 0.44 | 0.49 | 0.46 |
| Classifier | 5_3_1 | 6_2_1 | 6_3_1 | 3_3_2 | 4_2_2 | 5_1_2 |
| Error rate | 0.45 | 0.21 | 0.36 | 0.44 | 0.46 | 0.46 |
| Classifier | 5_2_2 | 6_1_2 | 5_1_3 | 3_2_4 | ||
| Error rate | 0.38 | 0.25 | 0.07 | 0.42 |
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (1)
1. a kind of downburst prediction technique based on Doppler Radar Data, which comprises the following steps:
Step 1: the Storm Events collected 10 years report matched Doppler radar data；
Step 2: from the radar data of each Storm Events, being identified all using storm core identification and method for tracing
The convection cell at moment extracts the static nature of convection cell, and is tracked to obtain convection cell sequence to convection cell；
The static characteristic includes Reflectivity and velocity characteristic；
The Reflectivity includes core monomer height, monomer maximum reflectivity, monomer maximum reflectivity height, bottom is high and pushes up
It is high；
Velocity jump point is detected on radar radial velocity image, when velocity jump value is greater than a given threshold value, then should
Velocity jump point is convergency point, and adjacent convergency point is merged, and the merging behaviour by constantly executing adjacent position convergency point
Make, finds the convergency point belt-like zone inside convection cell, which is known as belt of convergency；Traverse all convergence
Point calculates positive speed point n in its eight neighborhood+, positive speed summationAnd maximum positive speedNegative velocity points n-, negative velocity
SummationAnd maximum negative velocityThe then convergence value of the convergency pointThe convergence average value of the convergency pointAccording to convergency point coordinate each in belt of convergency with least square method fitting a straight line, calculate obtained
The A/cos θ of convergency point each in belt of convergency is the new spoke of the convergency point by the angle theta of fitting a straight line and radar projection straight line
Conjunction value；
The velocity characteristic of the convection cell, including middle layer convergence maximum value, middle layer convergence minimum value, middle layer are obtained by statistics
Convergence average value maximum value, middle layer convergence average value minimum value, convergence maximum value and convergence maximum layer average value, convergency point are total
Number, convergence maximum layer height, convergence maximum layer points, convergence maximum value height, convergence minimum value height, convergence average value are maximum
Value height and convergence average value minimum value height；
Step 3: convection cell sequence is matched with Storm Events, when having recorded disaster generation in Storm Events are reported
Between and place, if disaster occur the moment, a convection cell sequence with a distance from disaster spot be less than 10km, then it is assumed that
The convection cell sequence is related to disaster report；If there is multiple convection cell sequences at a distance from a storm disaster simultaneously
Less than 10km, then it is assumed that wherein related with disaster report apart from nearest convection cell sequence to storm disaster；
To all convection cell sequences relevant to disaster report, according to convection cell sequence and disaster in the report of relevant disaster
Difference be divided into:
Convection cell sequence relevant to downburst,
Convection cell sequence relevant to hail,
Convection cell sequence relevant to heavy rain；
Step 4: the moment occurs as time reference using disaster, convection cell sequence is split as several sample sequence segment collection；
Step is:
It is defined as follows two time parameters: Timing Advance first: point on the basis of at the time of disaster occurs, the body of biased forwards
The total number is Timing Advance；The value range of Timing Advance is that 1~3 individual is swept；Sequence fragment length: in the convection cell
The length of the segment intercepted in sequence, the value range of sequence fragment length are that 3~6 individuals are swept；When by combining different
Between lead and sequence fragment length, the convection cell sequence of all Disasters Types is split as 12 sequence fragment collection；
When splitting convection cell sequence, at the time of disaster occurs on the basis of point, first biased forwards Timing Advance parameter is true
The fixed body total number will be divided into 1~6 segment according to the difference of convection cell sequence length according to sequence fragment length；It is each
A segment all has two time parameters of Timing Advance and sequence fragment length, by the sequence fragment with same time parameter
It is stored in a set, to form sample sequence segment collection；
The sample sequence segment collection that disaster is downburst is denoted as positive sample sequence fragment collection, is hail and heavy rain phase by disaster
The sample sequence segment collection of pass is denoted as negative sample sequence fragment collection；
Step 5: extracting the feature set of sample sequence segment collection: calculating the time difference letter of the static nature of sample sequence segment
Characteristic sequence is ceased, time difference information characteristics sequence refers to the static nature of each of sample sequence segment collection convection cell
Change rate within some period, for each static nature f of convection cell, in moment t, the convection cell when
Between difference informationWherein,It is static nature f in t+dtThe value at moment,It is quiet
F is in t-d for state featuretThe value at moment；dtValue range be 3-9min；If the convection cell is in affiliated sample sequence segment
Both ends, then 0 filling of missing values；The feature of sample sequence segment is all static natures and time difference information characteristics in segment
Maximum value and minimum value, i.e., sample sequence segment is described with a feature vector；Obtain all positive and negative sample sequence segment collection
Sequence fragment feature after i.e. extractable sample sequence segment collection feature set；In the feature set of sample sequence segment collection have with
The identical time parameter of corresponding sequence segment collection；
Step 6: being carried out respectively using each feature of the positive and negative sample characteristics collection with same time parameter as one group of input
Statistical significance is examined, using all features of level of significance α < 0.05 as the positive and negative sample characteristics collection under the time parameter
Notable feature collection；
Step 7: by principal component analysis, the significant characteristics of all positive and negative sample characteristics collection with same time parameter
Collection carries out dimensionality reduction and decoupling, retains 85% validity feature, effective spy as the positive and negative sample characteristics collection under the time parameter
Sign；
Step 8: according to the validity feature of the positive and negative sample characteristics collection with same time parameter, one sorter model of training,
Obtain a classifiers model；Combined application sorter model set, detection and forecast downburst.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811060986.6A CN109164450B (en) | 2018-09-12 | 2018-09-12 | Downburst prediction method based on Doppler radar data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811060986.6A CN109164450B (en) | 2018-09-12 | 2018-09-12 | Downburst prediction method based on Doppler radar data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109164450A true CN109164450A (en) | 2019-01-08 |
| CN109164450B CN109164450B (en) | 2022-11-22 |
Family
ID=64894743
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811060986.6A Active CN109164450B (en) | 2018-09-12 | 2018-09-12 | Downburst prediction method based on Doppler radar data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109164450B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109917394A (en) * | 2019-03-13 | 2019-06-21 | 南京信息工程大学 | A kind of short based on weather radar faces intelligent Extrapolation method |
| CN111310565A (en) * | 2020-01-15 | 2020-06-19 | 同济大学 | Pedestrian detection method fusing image static characteristics and long-time and short-time motion characteristics |
| CN112946655A (en) * | 2021-01-27 | 2021-06-11 | 江苏省气象科学研究所 | Downburst flow identification method and device based on near-surface divergence field and storage medium |
| CN113447931A (en) * | 2021-06-10 | 2021-09-28 | 天津大学 | Short-time strong precipitation identification method based on Doppler radar data |
| CN115542326A (en) * | 2022-09-14 | 2022-12-30 | 江苏省气象台 | Isolated downburst identification method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102662172A (en) * | 2012-03-29 | 2012-09-12 | 天津大学 | Stormy cloud cluster extrapolation method based on Doppler radar reflectivity image |
| CN104237890A (en) * | 2014-09-03 | 2014-12-24 | 天津大学 | Recognition and forecast method for rainstorm caused by train effect |
| CN107229084A (en) * | 2017-06-08 | 2017-10-03 | 天津大学 | A kind of automatic identification, tracks and predicts contracurrent system mesh calibration method |
| CN107526083A (en) * | 2017-10-18 | 2017-12-29 | 国网新疆电力公司电力科学研究院 | A kind of strong convection wind scale Forecasting Methodology based on weather radar data |
| CN108020840A (en) * | 2017-11-20 | 2018-05-11 | 天津大学 | A kind of Hail Cloud By Using Weather EARLY RECOGNITION method based on Doppler radar data |
| CN108254750A (en) * | 2017-11-24 | 2018-07-06 | 南京信息工程大学 | A kind of downburst intelligent recognition method for early warning based on Radar Data |
| CN108459323A (en) * | 2018-01-31 | 2018-08-28 | 天津大学 | Strong convergence field identification based on Doppler radar data and speed correction method |
| CN108474841A (en) * | 2015-04-20 | 2018-08-31 | 瑞思迈传感器技术有限公司 | Detection and identification by characteristic signal to the mankind |
-
2018
- 2018-09-12 CN CN201811060986.6A patent/CN109164450B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102662172A (en) * | 2012-03-29 | 2012-09-12 | 天津大学 | Stormy cloud cluster extrapolation method based on Doppler radar reflectivity image |
| CN104237890A (en) * | 2014-09-03 | 2014-12-24 | 天津大学 | Recognition and forecast method for rainstorm caused by train effect |
| CN108474841A (en) * | 2015-04-20 | 2018-08-31 | 瑞思迈传感器技术有限公司 | Detection and identification by characteristic signal to the mankind |
| CN107229084A (en) * | 2017-06-08 | 2017-10-03 | 天津大学 | A kind of automatic identification, tracks and predicts contracurrent system mesh calibration method |
| CN107526083A (en) * | 2017-10-18 | 2017-12-29 | 国网新疆电力公司电力科学研究院 | A kind of strong convection wind scale Forecasting Methodology based on weather radar data |
| CN108020840A (en) * | 2017-11-20 | 2018-05-11 | 天津大学 | A kind of Hail Cloud By Using Weather EARLY RECOGNITION method based on Doppler radar data |
| CN108254750A (en) * | 2017-11-24 | 2018-07-06 | 南京信息工程大学 | A kind of downburst intelligent recognition method for early warning based on Radar Data |
| CN108459323A (en) * | 2018-01-31 | 2018-08-28 | 天津大学 | Strong convergence field identification based on Doppler radar data and speed correction method |
Non-Patent Citations (2)
| Title |
|---|
| 李茜 等: "华北地区夏季一次致雹强风暴的分析", 《内蒙古气象》 * |
| 熊小伏 等: "基于天气雷达数据的强对流天气下输电线风偏放电预警方法", 《电力自动化设备》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109917394A (en) * | 2019-03-13 | 2019-06-21 | 南京信息工程大学 | A kind of short based on weather radar faces intelligent Extrapolation method |
| CN109917394B (en) * | 2019-03-13 | 2022-12-23 | 南京信息工程大学 | Short-term intelligent extrapolation method based on weather radar |
| CN111310565A (en) * | 2020-01-15 | 2020-06-19 | 同济大学 | Pedestrian detection method fusing image static characteristics and long-time and short-time motion characteristics |
| CN112946655A (en) * | 2021-01-27 | 2021-06-11 | 江苏省气象科学研究所 | Downburst flow identification method and device based on near-surface divergence field and storage medium |
| CN112946655B (en) * | 2021-01-27 | 2022-05-13 | 江苏省气象科学研究所 | Downburst flow identification method and device based on near-surface divergence field and storage medium |
| CN113447931A (en) * | 2021-06-10 | 2021-09-28 | 天津大学 | Short-time strong precipitation identification method based on Doppler radar data |
| CN115542326A (en) * | 2022-09-14 | 2022-12-30 | 江苏省气象台 | Isolated downburst identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109164450B (en) | 2022-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109164450A (en) | A kind of downburst prediction technique based on Doppler Radar Data | |
| CN106980113A (en) | Article detection device and object detecting method | |
| CN102842045B (en) | A kind of pedestrian detection method based on assemblage characteristic | |
| AU2012384518B2 (en) | Infrared image based early detection of oil spills in water | |
| Lin et al. | ASCAT wind quality control near rain | |
| CN107330372A (en) | A kind of crowd density based on video and the analysis method of unusual checking system | |
| CN106951879A (en) | Multiple features fusion vehicle checking method based on camera and millimetre-wave radar | |
| CN107273852A (en) | Escalator floor plates object and passenger behavior detection algorithm based on machine vision | |
| CN103353988B (en) | Allos SAR scene Feature Correspondence Algorithm performance estimating method | |
| CN103871077B (en) | A kind of extraction method of key frame in road vehicles monitoring video | |
| CN104237890B (en) | The heavy rain identification that one is caused by " train effect " and forecasting procedure | |
| CN101975940A (en) | Segmentation combination-based adaptive constant false alarm rate target detection method for SAR image | |
| KR101255736B1 (en) | Method for classifying meteorological/non-meteorological echoes using single polarization radars | |
| CN105046259A (en) | Coronal mass ejection (CME) detection method based on multi-feature fusion | |
| KR102185225B1 (en) | Method for detecting sinkhole using deep learning and data association and sinkhole detecting system using it | |
| CN111880242B (en) | Method for arranging strong wind monitoring points along high-speed rail | |
| CN113689392A (en) | Railway fastener defect detection method and device | |
| KR101221793B1 (en) | Method for tracking reflectivity cells associated with severe weather | |
| Cromwell et al. | Lidar cloud detection with fully convolutional networks | |
| Chen et al. | High-speed image velocimetry system for rainfall measurement | |
| CN117517868A (en) | Lightning stroke fault matching method and system based on distributed measurement data | |
| CN107945183A (en) | A kind of combination improves the quick watershed segmentation methods for merging algorithm | |
| CN117111181A (en) | Short-time strong precipitation probability prediction method and system | |
| James et al. | Classification of lidar sensor contaminations with deep neural networks | |
| CN105761504B (en) | Speed method for real-time measurement based on the collection of non-homogeneous video frame image |
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 |