CN109164450B - Downburst prediction method based on Doppler radar data - Google Patents
Downburst prediction method based on Doppler radar data Download PDFInfo
- Publication number
- CN109164450B CN109164450B CN201811060986.6A CN201811060986A CN109164450B CN 109164450 B CN109164450 B CN 109164450B CN 201811060986 A CN201811060986 A CN 201811060986A CN 109164450 B CN109164450 B CN 109164450B
- Authority
- CN
- China
- Prior art keywords
- convection
- convergence
- sequence
- monomer
- time
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/95—Radar or analogous systems specially adapted for specific applications for meteorological use
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention discloses a downburst prediction method based on Doppler radar data, which comprises the following steps: collecting Doppler weather radar data matched with a storm event report of 10 years, identifying convection monomers at all moments, and extracting static characteristics to obtain a convection monomer sequence; matching the convective monomer sequence to a storm event; taking the disaster occurrence time as a time reference point, splitting the convection monomer sequence into a plurality of sample sequence fragment sets; extracting feature sets of the sample sequence fragment set, performing statistical significance test, and searching significant feature sets of positive and negative sample feature sets; performing dimensionality reduction and decoupling on the significance characteristic set through principal component analysis to obtain effective characteristics, and further training a classifier model to obtain a group of classifier models; and (4) combining and applying a classifier model set, and detecting and forecasting downburst. The method can automatically detect the convection monomer generating the downburst and predict the occurrence time and the occurrence place of the downburst.
Description
Technical Field
The invention relates to downburst disaster prediction in the field of meteorology, in particular to a method for automatically identifying a convection monomer generating downburst by using Doppler weather radar data.
Background
Convection strong winds include straight strong winds, tornadoes, and the like; the linear high winds in turn include bow-shaped echoes and downbursts. Among them, the damage of downburst is very obvious in airports, ports and the like. In the field of weather forecast, a Doppler weather radar is an effective tool for monitoring convection storms, and by means of the Doppler weather radar, a three-dimensional structure of the convection storm can be obtained, and meanwhile, the radial movement speed of particles in the convection storm can be detected. With the aid of doppler weather radar, it is possible to detect some significant features of the presence of convective cells producing downburst flows, such as the height of the highly suspended cell nuclei, the mid-layer radial convergence.
The common automatic downburst identification method is to analyze some characteristics of the downburst convection monomer in several cases, such as reflectivity morphological structural characteristics and radial velocity structural characteristics, by using a Doppler weather radar. And a prediction program of downburst is designed based on the characteristics. However, the framework only gives the application effect of a few cases, and does not give the effect of the actual operation of the algorithm.
Still another automatic downburst detection and prediction method is to combine and test several reflectivity characteristics and radial velocity characteristics of the downburst convection monomer by a linear classifier to find out some characteristics most relevant to the downburst. A linear classifier is trained by utilizing the characteristics, so that the downburst disaster can be automatically identified and predicted; however, this method has a recognition rate of 0.40 between 0 and 40 km from the radar and a recognition rate of 0.20 between 40 and 80 km.
The automatic downburst identification method in the prior art at least has the following defects and shortcomings:
only the static characteristics of the convective cell are used without considering the role of the convective cell characteristic evolution in convective cell classification. In addition, the downburst identification rate is low, and there is room for improvement in the automatic downburst identification and prediction method based on the doppler weather radar.
[ reference documents ]
[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] Raohui, zhangjie, jurkun, et al research on radar warning quantitative indicators of downburst [ J ] Meteorological report 2015,73 (5): 853-867.
Disclosure of Invention
Aiming at the prior art, the invention provides a downburst prediction method based on Doppler radar data, which is used for automatically predicting downbursts in meteorology.
In order to automatically detect a convection monomer generating downburst, predict the occurrence time and the occurrence place of the downburst, perform early warning on disasters in time and reduce economic loss and casualties, the invention provides a downburst prediction method based on Doppler radar data, which comprises the following steps:
step one, collecting Doppler weather radar data matched with a storm event report of 10 years;
identifying convection monomers at all moments from radar data of each storm event by using a storm core identification and tracking method, extracting static characteristics of the convection monomers, and tracking the convection monomers to obtain a convection monomer sequence;
the static characteristics include reflectivity characteristics and speed characteristics;
the reflectivity characteristics comprise monomer core height, monomer maximum reflectivity height, bottom height and top height;
detecting a speed jump point on a radar radial speed image, when a speed jump value is larger than a given threshold value, taking the speed jump point as a convergence point, merging adjacent convergence points, and finding a convergence point strip-shaped region in a convection monomer by continuously executing merging operation of the convergence points at adjacent positions, wherein the convergence point strip-shaped region is called a convergence strip; traversing all the convergent points, and calculating a positive speed point n in eight neighborhoods of the convergent points + Sum of positive speeds
And maximum positive velocity
Number of negative velocity points n - Negative sum of velocities
And maximum negative velocity
The sum of the points
Of the convergence point sum of averageValue of
Fitting a straight line by a least square method according to the coordinates of each convergence point in the convergence zone, calculating an included angle theta between the obtained fitted straight line and a radar projection straight line, and taking the A/cos theta of each convergence point in the convergence zone as a new convergence value of the convergence point;
obtaining the speed characteristics of the convection monomer through statistics, wherein the speed characteristics comprise a middle layer convergence maximum value, a middle layer convergence minimum value, a middle layer convergence average value maximum value, a middle layer convergence average value minimum value, a convergence maximum value sum, a convergence maximum layer average value, a convergence point total number, a convergence maximum layer height, a convergence maximum layer point number, a convergence maximum value height, a convergence minimum value height, a convergence average value maximum value height and a convergence average value minimum value height;
matching the convection monomer sequences with storm events, recording the time and place of disaster occurrence in a storm event report, and if the distance between one convection monomer sequence and the disaster occurrence place is less than 10km at the disaster occurrence time, determining that the convection monomer sequence is related to the disaster report; if the distance between a plurality of convection monomer sequences and a storm disaster is less than 10km, regarding the convection monomer sequence closest to the storm disaster as being related to the disaster report;
for all convection cell sequences related to disaster reports, the method is divided into the following steps according to the difference between the convection cell sequences and the disasters in the related disaster reports:
convective monomer sequences associated with downburst,
a convective monomer sequence associated with hail,
convective monomer sequences associated with heavy rains;
step four, taking the disaster occurrence time as a time reference point, splitting the convection current monomer sequence into a plurality of sample sequence fragment sets; the method comprises the following steps:
the following two time parameters are first defined: the time advance: taking the time when the disaster happens as a reference point, and taking the number of body sweeps which are shifted forwards as the time lead; the value range of the time lead is 1-3 individual sweeps; length of sequence fragment: the length of the intercepted segment in the convection monomer sequence is in the range of 3-6 individual sweeps; splitting convection monomer sequences of all disaster types into 12 sequence fragment sets by combining different time leads and sequence fragment lengths;
when the convection monomer sequence is split, the moment of occurrence of a disaster is taken as a reference point, the sweep number determined by the time lead parameter is shifted forwards, and the convection monomer sequence is divided into 1-6 segments according to the length of the sequence segments according to the difference of the lengths of the convection monomer sequence; each fragment has two time parameters of time advance and sequence fragment length, and sequence fragments with the same time parameters are stored in a set, so that a sample sequence fragment set is formed;
recording a sample sequence fragment set with a disaster as downburst as a positive sample sequence fragment set, and recording a sample sequence fragment set related to the disaster as hail and rainstorm as a negative sample sequence fragment set;
step five, extracting a feature set of the sample sequence fragment set: calculating a time difference information characteristic sequence of the static characteristics of the sample sequence segment, wherein the time difference information characteristic sequence refers to the change rate of the static characteristics of each convection single body in the sample sequence segment set in a certain time period, and for each static characteristic f of the convection single body, at the moment t, the time difference information of the convection single body
Wherein the content of the first and second substances,
for static features f at t + d t The value of the time of day is,
for static features f at t-d t A value of a time of day; d is a radical of t The value range of (A) is 3-9min; if the convection current monomer is positioned at the two ends of the sample sequence fragment, filling the missing value with 0; the sample sequence segment is characterized by the maximum value and the maximum value of all static characteristics and time difference information characteristics on the segmentSmall values, i.e. describing the sample sequence fragment with one feature vector; after sequence fragment characteristics of all positive and negative sample sequence fragment sets are obtained, the characteristic set of the sample sequence fragment set can be extracted; the characteristic set of the sample sequence fragment set has the same time parameter with the corresponding sequence fragment set;
taking each feature of the positive and negative sample feature sets with the same time parameter as a group of input, respectively carrying out statistical significance test, and taking all features with significance level alpha less than 0.05 as the significant feature sets of the positive and negative sample feature sets under the time parameter;
seventhly, performing dimensionality reduction and decoupling on the significance feature sets of all positive and negative sample feature sets with the same time parameter through principal component analysis, and reserving 85% of effective features as the effective features of the positive and negative sample feature sets under the time parameter;
step eight, training a classifier model according to the effective characteristics of the positive and negative sample characteristic sets with the same time parameter to obtain a group of classifier models; and (4) combining and applying a classifier model set, and detecting and forecasting downburst.
Compared with the prior art, the technical scheme provided by the invention has the beneficial effects that: the method for automatically identifying the downburst convection monomer is realized by improving the calculation method of the radial convergence strength characteristic of the original convection monomer, adding the time difference characteristic of a convection monomer sequence on the basis of the static characteristic of the convection monomer, and distinguishing the convection monomer generating downburst and the non-downburst convection monomer by utilizing the characteristics. In addition, the sequence of the convection current monomer is divided into the fragment sets with different lengths, and the convection current monomer is trained by using the single fragment set, so that the relevant time information is added to the classifier, and the result of the classifier can predict the time of the downburst. The disaster is early warned in time, so that the economic loss and the casualties are reduced; and the effectiveness of the method is verified through experiments.
Drawings
FIG. 1 is a result of identifying a convective cell on a radar reflectance image;
FIG. 2a is a radial velocity diagram of an original radar in a convergence region;
FIG. 2b is a schematic diagram illustrating a method for correcting a radial convergence velocity value of an intermediate layer;
FIG. 3 is a schematic diagram of matching of a single trajectory with a disaster report;
FIG. 4a is a sample of a convective cell sequence 0.5 degree elevation reflectivity sequence segment;
FIG. 4b is a sample convective monomer sequence fragment after splitting;
FIG. 5a is an example of the significance of feature 1 in an analysis sequence fragment;
FIG. 5b is an example of the significance of feature 2 in an analyzed sequence fragment;
FIG. 6 is a schematic diagram of a method for inputting a stream monomer sequence fragment into a classifier after splitting.
Detailed Description
To make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
The invention provides a downburst prediction method based on Doppler radar data, which comprises the following steps:
step one, collecting a report of storm events (including downburst and hail storm disasters) for 10 years and Doppler weather radar data matched with the report.
Identifying and tracking convection monomers in all radar data, and extracting characteristics of the convection monomers; identifying convection monomers at all moments from radar data of each storm event by using a storm core identification and tracking method, extracting static characteristics of the convection monomers, and tracking the convection monomers to obtain a convection monomer sequence, wherein the static characteristics of the convection monomers refer to characteristics without considering time information in the calculation process;
wherein the static characteristics include a reflectivity signature and a velocity signature;
the reflectance characteristics included a monomer core height, a monomer maximum reflectance height, a bottom height, and a top height, as shown in table 1.
At radar radial velocityDetecting a speed jump point on an image, when a speed jump value is larger than a given threshold value, taking the speed jump point as a convergence point, merging adjacent convergence points, and finding a convergence point band-shaped region in a convection monomer by continuously executing the merging operation of the convergence points at adjacent positions, wherein the convergence point band-shaped region is called a convergence band; traversing all the convergence points, and calculating a positive speed point n in eight neighborhoods of the convergence points + Sum of positive speeds
And maximum positive velocity
Number of negative velocity points n - Negative sum of velocities
And maximum negative velocity
The summation value of the summation point
The sum average of the sum
Fitting a straight line by a least square method according to the coordinates of each convergence point in the convergence zone, calculating an included angle theta between the obtained fitted straight line and a radar projection straight line, and taking the A/cos theta of each convergence point in the convergence zone as a new convergence value of the convergence point;
the velocity characteristics of the convection monomer are obtained through statistics, and the velocity characteristics comprise a middle layer convergence maximum value, a middle layer convergence minimum value, a middle layer convergence average value maximum value, a middle layer convergence average value minimum value, a convergence maximum value sum, a convergence maximum layer average value, a convergence point total number, a convergence maximum layer height, a convergence maximum layer point number, a convergence maximum value height, a convergence minimum value height, a convergence average value maximum value height and a convergence average value minimum value height, which are shown in table 1.
TABLE 1
And step three, for the convection monomer sequences identified in the step two, establishing association between the convection monomer sequences and storm event report records. Matching the convection monomer sequences with storm events, recording the time and place of disaster occurrence in a storm event report, and if the distance between one convection monomer sequence and the disaster occurrence place is less than 10km at the disaster occurrence time, determining that the convection monomer sequence is related to the disaster report; if the distance between a plurality of convection monomer sequences and a storm disaster is less than 10km, regarding the convection monomer sequence closest to the storm disaster to be related to the disaster report;
for all convection monomer sequences related to the disaster report, classifying the convection monomer sequence samples according to the related disaster weather types, namely, classifying the convection monomer sequence samples according to the difference between the convection monomer sequences and the disasters in the related disaster report:
convective monomer sequences associated with downburst,
a convective monomer sequence associated with hail,
convection monomer sequences associated with heavy rains.
Step four, taking the disaster occurrence time as a time reference point, splitting the convection monomer sequence into a plurality of sample sequence fragment sets; the method comprises the following specific steps:
the following two time parameters are first defined: time advance: taking the time when the disaster happens as a reference point, and taking the sweep number which is shifted forwards as the time lead; the value range of the time lead is 1-3 individual sweeps; length of sequence fragment: the length of the segment intercepted from the convection monomer sequence is in the range of 3-6 individual sweeps; splitting convection monomer sequences of all disaster types into 12 sequence fragment sets by combining different time leads and sequence fragment lengths;
when the convection monomer sequence is split, the moment of occurrence of a disaster is taken as a reference point, the sweep number determined by the time lead parameter is shifted forwards, and the convection monomer sequence is divided into 1-6 segments according to the length of the sequence segments according to the difference of the lengths of the convection monomer sequence; each fragment has two time parameters of time advance and sequence fragment length, and sequence fragments with the same time parameters are stored in a set, so that a sample sequence fragment set is formed;
and recording a sample sequence fragment set with the disaster as downburst as a positive sample sequence fragment set, and recording a sample sequence fragment set with the disaster as hail and rainstorm related as a negative sample sequence fragment set.
Step five, extracting a feature set of the sample sequence fragment set: calculating a time difference information characteristic sequence of the static characteristics of the sample sequence segment, wherein the time difference information characteristic sequence refers to the change rate of the static characteristics of each convection single body in the sample sequence segment set in a certain time period, and for each static characteristic f of the convection single body, at the moment t, the time difference information of the convection single body
Wherein the content of the first and second substances,
for static features f at t + d t The value of the time of day is,
for static features f at t-d t A value of a time of day; d is a radical of t The value range of (A) is 3-9min; if the convection current monomer is positioned at the two ends of the sample sequence fragment, filling the missing value with 0; the characteristics of the sample sequence segment are the maximum value and the minimum value of all static characteristics and time difference information characteristics on the segment, namely, one characteristic vector is used for describing the sample sequence segment; after sequence fragment characteristics of all positive and negative sample sequence fragment sets are obtained, the characteristic set of the sample sequence fragment set can be extracted; feature set of sample sequence fragment setHaving the same temporal parameters as the corresponding set of sequence fragments.
And step six, taking each feature of the positive and negative sample feature sets with the same time parameter as a group of input, respectively carrying out statistical significance test, and taking all features with significance level alpha <0.05 as the significant feature set of the positive and negative sample feature sets under the time parameter.
And seventhly, reducing and decoupling the significance characteristic sets of the positive and negative sample characteristic sets with the same time parameter through principal component analysis, and reserving 85% of effective characteristics as the effective characteristics of the positive and negative sample characteristic sets under the time parameter.
Step eight, training a classifier model on each sequence fragment set according to the effective characteristics of the positive and negative sample feature sets with the same time parameters to obtain a group of classifier models; and (4) combining and applying a classifier model set, and detecting and forecasting downburst.
Experimental example: the method is used for identifying and forecasting downburst in meteorology, and comprises the following steps:
1) Collecting radar history data of downburst and other types of disaster weather;
the information collected about historical sample data for downburst and other disaster weather types is: the start time and end time of all sample data were 2008-2017, respectively. By analyzing storm disaster reports, a total of 594 disaster event reports containing downburst and 6289 disaster records unrelated to downburst were found. By downsampling, 594 records were picked from a sample of non-downburst disasters.
According to the disaster report records of downburst and non-downburst, 1254 pieces of radar scanning data are found from the radar database, and the interval between the radar data is 3 minutes. One disaster record corresponds to one set of radar data sequences, called a case, which has an average length of 63 individual sweeps.
2) Identifying and tracking the convection monomers in the case, and extracting the characteristics of the convection monomers;
identifying and tracking convection monomers in the cases by utilizing a storm core identification and tracking method (SCIT algorithm) for all the cases collected in the step 1). In one case there will typically be multiple convective monomers. In all cases, a total of 12316 convective monomer sequences were detected by the SCIT method, with an average duration of 45 minutes. There were on average 9.8 convective monomer sequences in each case.
The recognition result of a single flow on the radar reflectivity image is given in fig. 1. Wherein each row corresponds to a time of the convection cell, each column corresponds to an elevation angle, and the time associated with downburst is the lowermost cell of fig. 1. The characteristics of the convection monomer on the graph extracted at the moment related to the downburst are as follows: the height of the monomer core is 4.31km, the maximum reflectivity is 78dbz, the height of the maximum reflectivity is 3.77km, the height of the bottom is 1.35km, the height of the top is 9.41km, the maximum value of the sum of the summations is 22m/s, the height of the maximum value of the sum of the summations is 5.12km, the maximum value of the summations is 15.08m/s, the minimum value of the summations is 6.14m/s, the minimum value of the summations is 12m/s, the height of the summations is 3.84, the sum of the summations is 268m/s, the height of the maximum value of the averages is 5.15km, the height of the average value of the minimums is 2.66km, the total number of the summations is 17, the average value of the summations is 15.76m/s, and the longest length of the summations is 7. These characteristics include reflectivity characteristics, radial velocity characteristics.
The method for correcting the radial convergence field strength inside the convection monomer is the innovation point of the invention, and a schematic diagram is adopted in the figure to explain how to correct the radial convergence field strength inside the convection monomer.
Fig. 2a is a graph of the radial velocity profile of the convective cell at 0.5 ° elevation. The convergence zone found in the radar ray direction is shown in fig. 2b, and fig. 2b shows the straight line fitting result of the convergence zone, the radar ray direction, the motion direction of the convection cell, and the included angle between the radar ray direction and the normal vector of the convergence zone. Convection cell interior in FIG. 2a the radial field strength is 42m/s. After the velocity correction, the resultant radial field strength was 60m/s.
3) Classifying the convection monomer sequence samples according to the related disaster weather types;
for the convection monomer sequences identified in step 2), it is necessary to correlate these convection monomer sequences with the storm event report records. An exemplary diagram is shown in fig. 3, where the location of a downburst storm event is shown, along with the convective cell targets and the convective cell trajectories in the vicinity of all the downbursts at that time. The total number of convection monomer sequences related to the downburst disaster is 3, and only the sequences 1 closest to the downburst disaster are related.
After each disaster record is related to a convection monomer sequence which is closest to the disaster record, 551 convection monomer sequences related to downdraft storm disaster event reports, 208 convection monomer sequences related to hail storm disaster event reports and 6289 convection monomer sequences without any disaster event reports related to the hail storm disaster event reports are obtained in total, and classification of the downdraft storm disaster monomers is achieved.
4) Dividing the convection monomer sequence into a set of a plurality of sequence segments by taking the disaster occurrence time as a time reference point;
a sample of splitting the convection cell sequence segment in this embodiment is shown in fig. 4a, in which convection cells coexist in a segment at 12 moments, a radar reflectivity map at an elevation angle of 0.5 degrees is shown in fig. 4a, moments related to downburst are marked by line frames, and a sample of splitting the convection cell sequence segment is shown in fig. 4 b. The convective monomer sequence fragment shown in fig. 4a is split into 6 fragments in total. Each segment is marked with a three-digit number in its position in the sequence relative to the time of disaster occurrence.
After all convection monomer sequences in the convection monomer sequences related to downburst and the convection monomer sequences related to hail storm are subjected to fragment splitting, 50 fragment sets are obtained in total, and the number of the fragments in each set is different. The largest set contains 551 fragments, and the smallest set contains 51 fragments. The average number of fragments in the set was 242.4.
5) Extracting the characteristics of the sequence fragments;
a simple example is used in table 2 to illustrate how to obtain a feature vector for a sample sequence fragment. Table 2 shows the time variation of the three features in the convection monomer sequence, and by finding the maximum value and the minimum value in the evolution data of each feature, the data of the three features in table 2 can be described by using a 6-dimensional vector: (80, 77,3.75,3.61,11.15, 9.08) to convert the characteristic curve data of the sequence segment into a characteristic vector. In this way, each sequence segment is described in actual operation by a length 136 eigenvector.
TABLE 2
| 01 point and 24 points | 01 point 28 point | Point 01 and point 33 | Point 01 and point 37 | |
| Maximum reflection value | 77 | 80 | 80 | 77 |
| Height of maximum reflection value | 3.61 | 2.46 | 2.49 | 3.75 |
| Echo peak height | 9.08 | 11.15 | 9,14 | 9,36 |
6) Finding out the unique characteristics of each sequence fragment set, downburst convection monomers through significance test;
an example of the significance of each feature in the positive and negative sample feature sets is analyzed. Fig. 5a and 5b show the distribution of features 1 and 2 on positive and negative samples, respectively, on a collection of sequence segments, where the curve with dots is the positive sample feature and the curve without dots is the negative sample feature. By calculation, the significance level alpha of the feature 1 on the positive and negative samples is 0.000827, which shows that the feature can effectively distinguish downburst and non-downburst fragments on the sequence; the significance level α of feature 2 on positive and negative samples was 0.104, indicating that the feature was not able to effectively distinguish downburst from non-downburst segments on the sequence. Thus, on this fragment set, feature 2 is deleted, and feature 1 is retained.
By performing the above analysis on each feature in the positive and negative sample feature sets, a feature having a significance level of less than 0.05 is retained as a valid feature in the positive and negative sample feature sets. After screening, 98.5 invalid features were removed on average across all sets of positive and negative sample features.
7) Through principal component analysis, the characteristics of downburst convection monomers in each sequence segment set are reduced and decoupled, and 85% of effective characteristics are reserved and used as the effective characteristics of a positive and negative sample characteristic set under the time parameter.
The result of principal component analysis on the salient features of a positive and negative sample feature set with the same time parameters is shown in table 3, wherein the first column is the number of the principal component, the first behavior feature description, the dimension reduction of the feature is realized by removing the principal component with lower contribution value in the table, and the sequence segment can be described by 3 features. From the data in table 3, the characteristic configuration of each principal component can be known. After the significant features of all the positive and negative sample feature sets are subjected to dimensionality reduction by adopting the principal component analysis method, the average feature number of the description sequence is reduced from 12 to 3.
TABLE 3
8) Training a classifier model on each sequence fragment set to obtain a group of classifier models; and (4) combining and applying a classifier model set, and detecting and forecasting downburst flow in real time on line.
The classifier model adopted in the method is a support vector machine model of a Gaussian kernel, and the purpose of training is to find two optimal model parameters. Taking the sequence fragment set 3-2-2 as an example, dividing a positive sample set and a negative sample set on the sequence fragment set into a training set and a testing set according to the proportion of 4; then, training 100 times by taking 0.05 as a step, searching a C value and a gamma value between 0 and 2 to ensure that the classification accuracy is highest, and respectively obtaining two optimal model parameters: c =1.45, gamma =0.05.
The result of the classifier training includes the test score value of the classifier in addition to the parameters of the classifier. Obtaining the accuracy of the classifier through the verification set; and obtaining the error rate of the classifier on the test set, taking the sequence fragment set 3-2-2 as sample data, taking the verification set as the sequence fragment set 3-2-2, taking the test set as all the other sequence fragment sets except 3-2-2, and obtaining the accuracy rate and the error rate of the classifier which are respectively 0.7 and 0.3. The classifier meets the performance index requirements (the accuracy rate is more than 0.6, and the error rate is less than 0.5), so that the classifier is reserved.
Through the above manner, one classifier is trained on all sequence segments, the performance of the classifier is tested, the classifiers which do not meet the performance index requirements are removed, only 16 classifiers are finally reserved, and the corresponding sequence sample segments are respectively:
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。
an example of a comprehensive application classifier model for real-time detection and prediction of downburst is shown in FIG. 6: each sequence segment in fig. 6 is sequentially input into a classifier corresponding to the length of the sequence segment, for example, a sequence with the length of 3 is input into a classification with the number of 3-1-1, wherein the relationship between the sequence segment and the classifier in this case is shown at the right end, and the output result of the classifier is also shown. In this case, there are three classifiers with output Yes in total among the three segment lengths, where the performance score of classifier 3-1-1 is the highest, and only the recognition result of that classifier is retained. And calculating the time of the impending downburst as 1 individual sweep according to the recognition result of the classifier, and then issuing the early warning of the downburst disaster.
In order to verify the feasibility of the downburst prediction method based on Doppler radar data, which is provided by the invention, for identifying and predicting downburst in meteorology, the following test experiments are carried out:
in the test experiment, 16 classifiers are applied to automatic identification and prediction of downburst disasters, and the performances of the 16 classifiers are evaluated independently. The data applied by the evaluation process is a set of fragments of the flow monomer sequence. The performance of each classifier is evaluated by accuracy and false alarm rate.
The first test is used to evaluate the accuracy of each classifier. In the experiment, a cross-validation method is adopted to evaluate the accuracy of the classifier. In the process of one-time cross validation, training samples (including positive samples and negative samples) of the classifier are divided into two parts according to the ratio of 4.
The accuracy evaluation index of the classifier is a Critical Success Index (CSI) calculated by the formula CSI = X/(X + Y + Z), where X is the number of positive samples classified into positive samples, Y is the number of negative samples classified into positive samples, and Z is the number of positive samples classified into negative samples. The critical success index CSI varies from 0 to 1, and when the value of the critical success index is 1, the performance of the classifier is the best.
Repeating the above cross-validation process for 100 times, then counting the average value of the obtained critical success indexes CSI, and taking the average value of the critical success indexes obtained by multiple tests as the measure of the accuracy of the classifier.
The accuracy score values for the 16 classifiers used in this test experiment are given in table 4. As can be seen in table 4, the accuracy score values for all classifiers ranged from 0.6 to 0.8, indicating that these classifiers are effective in distinguishing between downburst convection monomers and segments of non-downburst convection monomers.
TABLE 4
| Sorting device | 3_1_1 | 3_2_1 | 4_2_1 | 4_3_1 | 5_1_1 | 5_2_1 |
| Accuracy rate | 0.69 | 0.70 | 0.73 | 0.70 | 0.73 | 0.68 |
| Sorting device | 5_3_1 | 6_2_1 | 6_3_1 | 3_3_2 | 4_2_2 | 5_1_2 |
| Accuracy rate | 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 rate | 0.74 | 0.76 | 0.73 | 0.71 |
The second test is used to evaluate the false positive rate of each classifier. In this test, for a trained classifier, data on other sequence fragment sets are used as test data of the classifier, and the false alarm rate FAR of the classifier on other sequence fragment data is tested, and the calculation formula of FAR is as follows: FAR = A1/A1+ B1, where A1 is the number of samples that mark the test data as positive and B1 is the number of samples that mark the test data as negative. For a classifier, the lower the false alarm rate, the better the performance.
The false positive score values for the 18 classifiers used in this test experiment are given in table 5. As can be seen from table 5, the false alarm rate scores for all classifiers are below 0.5. The classifiers can effectively identify sequence segments of different time periods of downburst convection monomers.
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 |
| Sorting device | 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 |
| Sorting device | 5_2_2 | 6_1_2 | 5_1_3 | 3_2_4 | ||
| Error rate | 0.38 | 0.25 | 0.07 | 0.42 |
Those skilled in the art will appreciate that the drawings are only schematic illustrations of preferred embodiments, and the above-mentioned serial numbers of the embodiments of the present invention are only for description and do not represent the merits of the embodiments.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, which is intended to cover any modifications, equivalents, improvements, etc. within the spirit and scope of the present invention.
Claims (1)
1. A downburst prediction method based on Doppler radar data is characterized by comprising the following steps:
step one, collecting Doppler weather radar data matched with a storm event report of 10 years;
identifying convection monomers at all moments from radar data of each storm event by using a storm core identification and tracking method, extracting static characteristics of the convection monomers, and tracking the convection monomers to obtain a convection monomer sequence;
the static features include reflectivity features and velocity features;
the reflectivity characteristics comprise monomer core height, monomer maximum reflectivity height, bottom height and top height;
detecting a speed jump point on a radar radial speed image, when a speed jump value is larger than a given threshold value, taking the speed jump point as a convergence point, merging adjacent convergence points, and finding a convergence point strip-shaped region in a convection cell by continuously executing the merging operation of the convergence points at adjacent positions, wherein the convergence point strip-shaped region is called a convergence strip; traversing all the convergence points, and calculating a positive speed point n in eight neighborhoods of the convergence points + Sum of positive speeds
And maximum positive velocity
Number of negative velocity points n - Negative sum of velocities
And maximum negative velocity
The summation value of the summation point
The sum of the values
Fitting a straight line by a least square method according to the coordinates of each convergence point in the convergence zone, calculating an included angle theta between the obtained fitted straight line and a radar projection straight line, and taking the A/cos theta of each convergence point in the convergence zone as a new convergence value of the convergence point;
obtaining the speed characteristics of the convection monomer through statistics, wherein the speed characteristics comprise a middle layer convergence maximum value, a middle layer convergence minimum value, a middle layer convergence average value maximum value, a middle layer convergence average value minimum value, a convergence maximum value sum, a convergence maximum layer average value, a convergence point total number, a convergence maximum layer height, a convergence maximum layer point number, a convergence maximum value height, a convergence minimum value height, a convergence average value maximum value height and a convergence average value minimum value height;
matching the convection monomer sequences with storm events, recording the time and place of disaster occurrence in a storm event report, and if the distance between one convection monomer sequence and the disaster occurrence place is less than 10km at the disaster occurrence time, regarding that the convection monomer sequence is related to the storm event report where the disaster is located; if the distance between a plurality of convection monomer sequences and a storm disaster is less than 10km, regarding that the convection monomer sequence closest to the storm disaster is related to the report of the storm event where the storm disaster is located;
for all convection monomer sequences related to the storm event report where the disaster is located, the method is divided into the following steps according to the difference between the convection monomer sequences and the disasters in the related storm event report:
convection monomer sequences associated with downburst,
a convective monomer sequence associated with hail,
convective monomer sequences associated with heavy rains;
step four, taking the disaster occurrence time as a time reference point, splitting the convection current monomer sequence into a plurality of sample sequence fragment sets; the method comprises the following steps:
the following two time parameters are first defined: time advance: taking the time when the disaster happens as a reference point, and taking the sweep number which is shifted forwards as the time lead; the value range of the time lead is 1-3 individual sweeps; length of sequence fragment: the length of the intercepted segment in the convection monomer sequence is in the range of 3-6 individual sweeps; splitting convection monomer sequences of all disaster types into 12 sequence fragment sets by combining different time leads and sequence fragment lengths;
when the convection monomer sequence is split, the moment of occurrence of a disaster is taken as a reference point, the sweep number determined by the time lead parameter is shifted forwards, and the convection monomer sequence is divided into 1-6 segments according to the length of the sequence segments according to the difference of the lengths of the convection monomer sequence; each fragment has two time parameters of time advance and sequence fragment length, and sequence fragments with the same time parameters are stored in a set, so that a sample sequence fragment set is formed;
recording a sample sequence fragment set with a disaster as downburst as a positive sample sequence fragment set, and recording a sample sequence fragment set related to the disaster as hail and rainstorm as a negative sample sequence fragment set;
step five, extracting a feature set of the sample sequence fragment set: calculating a time difference information characteristic sequence of the static characteristics of the sample sequence segment, wherein the time difference information characteristic sequence refers to the change rate of the static characteristics of each convection monomer in the sample sequence segment set in a certain time period, and for each static characteristic f of the convection monomer, at the moment t, the time difference information of the convection monomer
Wherein, the first and the second end of the pipe are connected with each other,
for static features f at t + d t The value of the time of day is,
for static features f at t-d t A value of a time of day; d t The value range of (A) is 3-9min; if the convection current monomer is positioned at two ends of the sample sequence fragment, filling the missing value with 0; the characteristics of the sample sequence segment are the maximum value and the minimum value of all static characteristics and time difference information characteristics on the segment, namely, one characteristic vector is used for describing the sample sequence segment; after sequence fragment characteristics of all positive and negative sample sequence fragment sets are obtained, the characteristic set of the sample sequence fragment set can be extracted; the characteristic set of the sample sequence fragment set has the same time parameter with the corresponding sequence fragment set;
taking each feature of the positive and negative sample feature sets with the same time parameter as a group of input, respectively carrying out statistical significance test, and taking all features with significance level alpha less than 0.05 as the significant feature sets of the positive and negative sample feature sets under the time parameter;
seventhly, performing dimensionality reduction and decoupling on the significance feature sets of all positive and negative sample feature sets with the same time parameter through principal component analysis, and reserving 85% of effective features as the effective features of the positive and negative sample feature sets under the time parameter;
step eight, training a classifier model according to the effective characteristics of the positive and negative sample characteristic sets with the same time parameters to obtain a group of classifier models; and (4) combining and applying a classifier model set, and detecting and forecasting 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 CN109164450A (en) | 2019-01-08 |
| CN109164450B true 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) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
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 |
|---|
| 华北地区夏季一次致雹强风暴的分析;李茜 等;《内蒙古气象》;20120130;正文全文 * |
| 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;熊小伏 等;《电力自动化设备》;20180430;正文全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109164450A (en) | 2019-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109164450B (en) | Downburst prediction method based on Doppler radar data | |
| CN110728411B (en) | High-low altitude area combined rainfall prediction method based on convolutional neural network | |
| CN116644373B (en) | Automobile flow data analysis management system based on artificial intelligence | |
| CN102682601B (en) | Expressway traffic incident detection method based on optimized support vector machine (SVM) | |
| CN111462488B (en) | Intersection safety risk assessment method based on deep convolutional neural network and intersection behavior characteristic model | |
| CN101403676B (en) | Insulator hydrophobicity rank amalgamation judging method based on D-S evidence theory | |
| CN101975940B (en) | Segmentation combination-based adaptive constant false alarm rate target detection method for SAR image | |
| CN107238821A (en) | The airfield runway foreign matter detecting method and device of a kind of feature based spectrum signature | |
| CN106980113A (en) | Article detection device and object detecting method | |
| CN103559508B (en) | A kind of based on continuous Adaboost video vehicle detection method | |
| CN105809193B (en) | A kind of recognition methods of the illegal vehicle in use based on kmeans algorithm | |
| CN102147866A (en) | Target identification method based on training Adaboost and support vector machine | |
| CN103455820A (en) | Method and system for detecting and tracking vehicle based on machine vision technology | |
| CN104237890B (en) | The heavy rain identification that one is caused by " train effect " and forecasting procedure | |
| CN111582380A (en) | Ship track density clustering method and device based on space-time characteristics | |
| CN111640304B (en) | Automatic quantitative extraction method for traffic jam propagation characteristics of continuous flow traffic facility | |
| CN105046259A (en) | Coronal mass ejection (CME) detection method based on multi-feature fusion | |
| CN110059713A (en) | Precipitation type identification method based on precipitation particle multi-feature parameters | |
| CN108154173B (en) | Device and method for measuring oil-water interface of crude oil storage tank | |
| CN106910334B (en) | Method and device for predicting road section conditions based on big data | |
| CN111736156B (en) | Upwind area identification method and device based on weather radar | |
| CN113447931A (en) | Short-time strong precipitation identification method based on Doppler radar data | |
| CN110515081B (en) | Intelligent identification and early warning method for bright band of radar echo zero-degree layer | |
| CN100507555C (en) | Fractal elimination method for mass data in ultrasonic intelligent detection | |
| CN113296072B (en) | Method and system for automatically identifying thunderstorm strong wind based on YOLOv3 model |
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 |