CN112255611B - Intelligent downburst identification method based on radar detection data - Google Patents
Intelligent downburst identification method based on radar detection data Download PDFInfo
- Publication number
- CN112255611B CN112255611B CN202011171264.5A CN202011171264A CN112255611B CN 112255611 B CN112255611 B CN 112255611B CN 202011171264 A CN202011171264 A CN 202011171264A CN 112255611 B CN112255611 B CN 112255611B
- Authority
- CN
- China
- Prior art keywords
- downburst
- radar
- model
- radar detection
- 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
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- 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)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The embodiment of the invention discloses an intelligent downburst identification method based on radar detection data in the technical field of atmospheric science, which comprises the following steps: s1, establishing radar detection data D (t) changing along with radar detection time t, and defining D (t-1), D (t-2) and D (t-n) to be radar detection data of 1-n time periods before t moment respectively; s2, establishing a radar echo image Pa (D (t)) and a radar radial velocity image Pb (D (t)) which change along with the radar detection data D (t). The invention avoids falling into local extreme value and seeks optimal parameter by obtaining effective information as much as possible from limited data set; the multiple models can be used for model integration, and accuracy and stability of model identification are improved.
Description
Technical Field
The embodiment of the invention relates to the technical field of atmospheric science, in particular to a downburst intelligent identification method based on radar detection data.
Background
Downburst forms in strong convection storms, which form strong divergent straight strong winds when they reach the ground. When the wind power generation device is used, the sharp increase of wind speed and the strong wind shear of a local area are caused, so that the taking off and landing of an aviation airplane, the navigation of a ship, the crash of the aviation airplane and the overturning of the ship are seriously caused, the serious property loss is caused, and even the life is threatened. On radar intensity echoes, bow-shaped echoes are a significant indicator of downburst. If there is a strong rear inflow rush at the center of the bow echo, a strong downburst wind tends to occur. On a radar radial velocity diagram, downburst appears as radial velocity divergence, namely, echoes in a shape of a 'bulls eye' are presented, and the range and the intensity of the downburst can be roughly judged through the distance between extreme values of a pair of radial velocity centers.
The existing identification of downburst mainly depends on subjective analysis of the characteristics by weather service personnel to make identification and early warning measures. But accurate identification is very difficult because of the small time and spatial dimensions of downburst occurrence.
Based on the method, the invention designs an intelligent downburst identification method based on radar detection data to solve the problems.
Disclosure of Invention
The embodiment of the invention provides an intelligent downburst identification method based on radar detection data, which aims to solve the technical problems in the background art.
The embodiment of the invention provides an intelligent downburst identification method based on radar detection data. In one possible solution, the method comprises the following steps:
s1, establishing radar detection data D (t) changing along with radar detection time t, and defining D (t-1), D (t-2) and D (t-n) to be radar detection data of 1-n time periods before t moment respectively;
s2, establishing a radar echo image Pa (D (t)) and a radar radial velocity image Pb (D (t)) which are changed along with radar detection data D (t), and recording a radar echo time sequence image formed by Pa (D (t)), pa (D (t-1)), pa (D (t-2)) and Pa (D (t-n)) as PA (D, t); recording one radar radial velocity time series image composed of Pb (D (t)), pb (D (t-1)), pb (D (t-2)) and Pb (D (t-n)) as PB (D, t);
s3, establishing a data pair between a radar echo time sequence image PA (D, T), a radar radial velocity time sequence image PB (D, T) and a ground meteorological station observation wind speed information label Lab (T) of whether the downburst occurs or not, recording the occurrence of the downburst as T (T) =01, and recording the occurrence of the downburst as T (T) =10;
s4, converting the observation wind speed information of the ground meteorological station into observation time consistent with radar detection time;
s5, collecting radar detection data and ground meteorological data in the time period of the past whole year, processing and calculating to obtain all data pairs in the time period, and further obtaining a case data set DS for intelligent identification of downburst;
s6, reducing the probability that downburst flows do not exist in the model training result;
s7, constructing a DBIM network model by fusing downburst flow identification of a radar echo time-space sequence with a deep neural network as a basis and a radar echo time sequence image and a radial velocity field time sequence image as input;
s8, finding out the optimal network hyper-parameter at the similar time, and carrying out grouping inspection optimization on the DBIM network model;
wherein, the optimization of the DBIM network model in S8 includes the following steps:
a. randomly splitting an original data set DS into K parts;
b. selecting any one as a verification set, taking the rest as training sets, training the DBIM network model to obtain a model with network parameters, testing on the verification set, and storing an evaluation index E of the model;
c. repeating the step b for K times to ensure that all subsets have one chance as a verification set;
d. and taking the mean value of each group of evaluation indexes as the accurate evaluation of the model and as the comprehensive evaluation index of the DBIM network model under the current grouping verification.
S9, introducing a category weight to a loss function of the DBIM network model, and giving a penalty term w under the condition that the actual violent stream exists but the model is identified as the condition that no violent stream exists so as to improve the sensitivity to T (T) =01 in training, wherein the improved loss function is as follows:
where yi is an indicative function of downburst weather, ti is the output of the CNN model corresponding to downburst weather, and represents the probability that the region is identified as downburst weather;
s10, substituting a data set DS into a DBIM network model to train a neural network to obtain a model for intelligently identifying downburst;
and S11, according to the model intelligently identified by the downburst, taking the real-time radar detection data as the input of the model, and after model calculation, obtaining a result DBIM _ Output (D (t 0)) =01 which indicates that the downburst is identified or real-time radar detection data DBIM _ Output (D (t 0)) =10 which indicates that the downburst is not identified, wherein the result DBIM _ Output (D (t 0)) =01 is Output by the model identified by classification.
The embodiment of the invention provides an intelligent downburst identification method based on radar detection data. In a possible solution, the condition that the downburst T (T) =01 in S3 needs to be satisfied at the same time includes:
a. the instantaneous wind speed reaches or exceeds WS1;
b. in the past IM1 minute, the wind speed changes beyond WS2;
c. identifying a storm core or a strong echo center from radar detection data of each time to have a rapid descending phenomenon within the latest 1 hour of observing the wind speed by the weather station, wherein the descending speed reaches or exceeds WS3;
wherein, WS1 belongs to [17.2,100] and the unit is meter/second; WS2 belongs to [11.7,100] and has the unit of meter/second; IM1 belongs to [5,120] and has the unit of minutes; WS3 ∈ [20,200], in units of kilometers per hour.
The embodiment of the invention provides an intelligent downburst identification method based on radar detection data. In one possible solution, the algorithm for converting in S4 is an interpolation algorithm.
The embodiment of the invention provides an intelligent downburst identification method based on radar detection data. In a possible scheme, the building of the downburst identification DBIM network model in S7 includes the following steps:
a. decomposing the data set into a radar echo time sequence image and a radial velocity field time sequence image, wherein each group of images comprises 8 elevation planes and 6 adjacent moments, and the size of each image is 100 multiplied by 100 pixels;
b. processing a radar echo image, designing a 4-channel sparse structure to generate dense data, wherein each channel comprises 1-2 convolution layers, the sizes of convolution kernels are 1 multiplied by 1, 3 multiplied by 3 and 5 multiplied by 5, and the sizes of input images and output images are kept consistent by properly filling 4 channels by extracting echo intensity space information under different time scales;
c. and connecting the output of each channel in the channel dimension to obtain a four-dimensional vector, and outputting the four-dimensional vector to a subsequent layer.
Based on the scheme, the intelligent downburst identification model based on the convolutional neural network has the advantages that weather radar data is used as input, and a function mapping relation between a radar echo image and a radial velocity field image and the condition whether downburst occurs or not is searched through self-learning of the deep neural network. By applying a series of optimization techniques, the problem that the recognition result is biased to the probable event in the sample due to too few cases of downburst in the data set is solved. Meanwhile, the packet inspection is adopted, and the advantages are that: on one hand, effective information as much as possible can be obtained from a limited data set, local extreme values are avoided, and optimal parameters are sought; on the other hand, a plurality of models can be used for model integration, and accuracy and stability of model identification are improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of an identification method of the present invention;
FIGS. 2-3 are diagrams of neural network structures for intelligent identification of downburst in accordance with the present invention;
FIG. 4 is a table diagram of a partial neural network structure parameter for intelligent identification of downburst according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it is to be understood that the terms "central," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "axial," "radial," "circumferential," and the like are used in the indicated orientations and positional relationships, based on the orientation or positional relationship illustrated in the drawings, for convenience in describing the invention and to simplify the description, and are not intended to indicate or imply that the referenced device or element must have a particular orientation, be constructed and operated in a particular orientation, and are not to be construed as limiting the invention.
In the present invention, unless otherwise specifically stated or limited, the terms "mounted," "connected," "fixed," and the like are to be construed broadly and may, for example, be fixedly connected, detachably connected, or integrally formed; the connection can be mechanical connection, electrical connection or communication connection; either directly or indirectly through intervening media, either internally or in any other suitable relationship, unless expressly stated otherwise. The specific meanings of the above terms in the present invention can be understood according to specific situations by those of ordinary skill in the art.
The technical solution of the present invention will be described in detail below with specific examples. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments.
Fig. 1-4 illustrate an intelligent downburst identification method based on radar detection data according to the present invention; the method comprises the following steps:
s1, establishing radar detection data D (t) changing along with radar detection time t, and defining D (t-1), D (t-2) and D (t-n) to be radar detection data of 1-n time periods before t moment respectively;
s2, establishing a radar echo image Pa (D (t)) and a radar radial velocity image Pb (D (t)) which change along with radar detection data D (t), and recording a radar echo time sequence image formed by Pa (D (t)), pa (D (t-1)), pa (D (t-2)) and Pa (D (t-n)) as PA (D, t); recording one radar radial velocity time series image composed of Pb (D (t)), pb (D (t-1)), pb (D (t-2)) and Pb (D (t-n)) as PB (D, t);
s3, establishing a data pair between a radar echo time sequence image PA (D, T), a radar radial velocity time sequence image PB (D, T) and a ground meteorological station observation wind speed information label Lab (T) of whether the downburst occurs or not, recording the occurrence of the downburst as T (T) =01, and recording the occurrence of the downburst as T (T) =10;
s4, converting the observation wind speed information of the ground meteorological station into observation time consistent with radar detection time;
s5, collecting radar detection data and ground meteorological data in the time period of the past whole year, processing and calculating to obtain all data pairs in the time period, and further obtaining a case data set DS for intelligent identification of downburst;
s6, reducing the probability that downbeat storm flow does not exist in the model training result;
s7, constructing a DBIM network model by fusing downburst flow identification of a radar echo time-space sequence with a deep neural network as a basis and a radar echo time sequence image and a radial velocity field time sequence image as input;
s8, finding out the optimal network hyper-parameter at the similar time, and carrying out grouping inspection optimization on the DBIM network model;
s9, introducing category weight to a loss function of the DBIM network model, and giving a penalty term w under the condition that the downburst actually exists but the model is identified as not existing to improve the sensitivity to T (T) =01 in training, wherein the improved loss function is as follows:
where yi is an indicative function of downburst weather, ti is the output of the CNN model corresponding to downburst weather, and represents the probability that the region is identified as downburst weather;
s10, substituting a data set DS into a DBIM network model to train a neural network to obtain a model for intelligently identifying downburst;
s11, according to the model intelligently identified by the downburst, real-time radar detection data is used as the input of the model, and after model calculation, a result DBIM _ Output (D (t 0)) =01 of the model Output of classification identification can be obtained to indicate that the downburst is identified or the real-time radar detection data DBIM _ Output (D (t 0)) =10 indicates that the downburst is not identified.
Through the above, it can be easily found that, in the process of identifying the downburst by using the intelligent downburst identification method based on radar detection data, one piece of radar detection data is defined as D (t), wherein t represents the radar detection time. And D (t-1), D (t-2) and D (t-n) are respectively defined to represent radar detection data of 1 time period, 2 time periods and n time periods before the time t, wherein n is a positive integer of [1,100], and a radar echo image at the time t is defined to be Pa (D (t)), and Pa (D (t)), pa (D (t-1)), pa (D (t-2)), \8230;, and Pa (D (t-n)) form a radar echo time sequence image, which is marked as PA (D, t). The radar radial velocity image at time t is Pb (D (t)), and one radar radial velocity time-series image is constituted by Pb (D (t)), pb (D (t-1)), pb (D (t-2)), \ 8230;, and Pb (D (t-n)), and is denoted as Pb (D, t). The PA (D, t) and the PB (D, t) are used for recording some characteristics of particles such as water vapor in the atmosphere in time and space, and both the radar echo image and the radar radial velocity image can be obtained by a meteorology-disclosed algorithm or professional meteorology software; and establishing a data pair between the radar echo time sequence image PA (D, t) and the radar radial velocity time sequence image PB (D, t) and the label Lab (t) whether the downburst occurs or not. Wherein Lab (t) is wind speed information observed through a ground meteorological station; because the time period of radar detection is different from the time period observed by the ground meteorological station, the wind speed information observed by the ground meteorological station is converted into the observation time consistent with the radar detection time by adopting an interpolation algorithm; processing and calculating radar detection data and ground meteorological station data for a long period of time (such as the last year and the whole year) to generate all 'data pairs' in the period of time, thereby obtaining a use case data set for intelligent identification of downburst, and recording the use case data set as DS; since the occurrence of downburst belongs to a small probability event, the number of use cases of T (T) =01 in the DS is much smaller than the number of use cases of T (T) =10. In order to prevent the following calculation step from biasing the result of model training to a high probability event, that is, no downburst occurs, thereby affecting the accuracy of recognition, the processing method of the step is as follows: generating a batch of new use cases by small-amplitude translation, rotation, deformation, noise increase and other modes for radar echo images and radial velocity images with possible downburst characteristics, and adding the new use cases to a data set DS, so that the proportion of the downburst identified in the DS is increased; constructing a downburst identification network model which is based on a deep neural network, takes a radar echo time sequence image and a radial velocity field time sequence image as input, integrates various characteristics of a radar echo time-space sequence, and is marked as DBIM, wherein the network structure of the model is shown in figure 2; since the occurrence of downburst has a certain seasonal characteristic, data sets based on a large number of historical meteorological data are organized chronologically. In order to improve the generalization performance of the model, a better network hyper-parameter is found in a similar training time, and an optimized DBIM network model is obtained; in order to help the DBIM network model to have sensitivity to T (T) =01 in the training process, the step introduces category weight into a loss function of the DBIM network model, and gives a larger penalty term to the situation that the downburst exists actually but the model identifies that the downburst does not exist, wherein w is a judgment weight term, namely the penalty term. The larger the value is, the more radar images can be judged by the model to be downburst, so that the higher false alarm rate is caused, and the recognition success rate is correspondingly improved; substituting the data set DS into a DBIM network model to train a neural network, and finally obtaining a model for intelligently identifying downburst; then, real-time radar detection data is used as an input of the model, and after model calculation, a result DBIM _ Output (D (t 0)) =01 or DBIM _ Output (D (t 0)) =10 of classification recognition can be obtained. Wherein, DBIM _ Output represents the result of model Output, and D (t 0) represents real-time radar detection data. Output result 01 indicates that the occurrence of downburst was recognized, and output result 10 indicates that the occurrence of downburst was not recognized.
The invention provides an intelligent downburst identification model based on a convolutional neural network. The model takes weather radar data as input, and seeks a functional mapping relation between a radar echo image and a radial velocity field image and whether downburst occurs or not through self-learning of a deep neural network. By applying a series of optimization techniques, the problem that the recognition result is biased to the probable event in the sample due to too few cases of downburst in the data set is solved. Meanwhile, the packet inspection is adopted, and the advantages are that: on one hand, effective information as much as possible can be obtained from a limited data set, local extreme values are avoided, and optimal parameters are sought; on the other hand, a plurality of models can be used for model integration, and accuracy and stability of model identification are improved.
Optionally, the condition that the downburst T (T) =01 in S3 needs to be satisfied at the same time includes:
a. the instantaneous wind speed reaches or exceeds WS1;
b. in the past IM1 minute, the wind speed changes beyond WS2;
c. within the last 1 hour of the wind speed observed by the meteorological station, the phenomenon that the storm core or the strong echo center is rapidly descended is identified from radar detection data of each time, and the descent speed reaches or exceeds WS3. It is worth noting that in this example, WS1 ∈ [17.2,100], in meters/second; WS2 belongs to [11.7,100] and has the unit of meter/second; IM1 belongs to [5,120] and has the unit of minutes; WS3 belongs to [20,200], and the unit is kilometer/hour; when the 3 conditions are all satisfied, marking that downburst flow occurs at the time, and recording as T (T) =01; otherwise, the samples are marked as no downburst and are marked as T (T) =10.
In addition, the algorithm for conversion in S4 is an interpolation algorithm.
More specifically, the building of the downburst identification DBIM network model in S7 includes the following steps:
a. decomposing the data set into a radar echo time sequence image and a radial velocity field time sequence image, wherein each group of images comprises 8 elevation planes and 6 adjacent moments, and the size of each image is 100 multiplied by 100 pixels;
b. processing a radar echo image, designing a 4-channel sparse structure to generate dense data, wherein each channel comprises 1-2 convolution layers, the sizes of convolution kernels are preferably 1 × 1, 3 × 3 and 5 × 5, and the sizes of input images and output images are kept consistent by extracting echo intensity space information under different time scales and properly filling the 4 channels;
c. connecting the output of each channel in channel dimension to obtain a four-dimensional vector, and outputting the four-dimensional vector to a subsequent layer; the structure of these layers is shown in fig. 4, conv denotes convolution layers, and the parenthesized numerals denote the window size or the number of outputs of convolution kernels. The input layer is a four-dimensional vector, the functions of other layers are consistent with the meanings of English names, wherein, a Drapout discarding layer, max Pooling is a maximum Pooling layer, full Connection is a Full-Connection layer, and a Flatten layer is used for converting the multi-dimensional vector of the previous layer into a one-dimensional vector, and the layers are also some main basic units in the neural network; and processing the time sequence image of the radial velocity field by using the same network model until the time sequence image of the radial velocity field and the time sequence image of the radial velocity field are respectively subjected to final convolution Conv (3 multiplied by 64) and Dropout, then entering a Flatten layer and a full connection layer, and finally outputting the data in a 2-classification one-hot coding data form.
Further, the optimization of the DBIM network model in S8 includes the following steps:
a. randomly splitting an original data set DS into K parts;
b. selecting any one as a verification set, taking the rest as training sets, training the DBIM network model to obtain a model with network parameters, testing on the verification set, and storing an evaluation index E of the model;
c. repeating the step b for K times to ensure that all subsets have one chance as a verification set;
d. taking the mean value of each group of evaluation indexes as the accurate evaluation of the model and as the comprehensive evaluation index of the DBIM network model under the current grouping verification; and comprehensive evaluation index of DBIM network model
In the present invention, unless otherwise explicitly specified or limited, the first feature "on" or "under" the second feature may be directly contacting the first feature and the second feature or indirectly contacting the first feature and the second feature through an intermediate.
Also, a first feature "on," "above," and "over" a second feature may be directly on or obliquely above the second feature, or may simply mean that the first feature is at a higher level than the second feature. A first feature being "under," "below," and "beneath" a second feature may be directly under or obliquely under the first feature, or may simply mean that the first feature is at a lower level than the second feature.
In the description herein, reference to the description of the term "one embodiment," "some embodiments," "an example," "a specific example" or "some examples," or the like, means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.
Claims (4)
1. An intelligent downburst identification method based on radar detection data is characterized by comprising the following steps:
s1, establishing radar detection data D (t) changing along with radar detection time t, and defining D (t-1), D (t-2) and D (t-n) to be radar detection data of 1-n time periods before t moment respectively;
s2, establishing a radar echo image Pa (D (t)) and a radar radial velocity image Pb (D (t)) which change along with radar detection data D (t), and recording a radar echo time sequence image formed by Pa (D (t)), pa (D (t-1)), pa (D (t-2)) and Pa (D (t-n)) as PA (D, t); recording as Pb (D, t) one radar radial velocity sequence image composed of Pb (D (t)), pb (D (t-1)), pb (D (t-2)), and up to Pb (D (t-n));
s3, establishing a data pair between a radar echo time sequence image PA (D, T), a radar radial velocity time sequence image PB (D, T) and a ground meteorological station observation wind speed information label Lab (T) of whether the downburst occurs or not, recording the occurrence of the downburst as T (T) =01, and recording the occurrence of the downburst as T (T) =10;
s4, converting the observation wind speed information of the ground meteorological station into observation time consistent with radar detection time;
s5, collecting radar detection data and ground meteorological data in the time period of the past whole year, processing and calculating to obtain all data pairs in the time period, and further obtaining a case data set DS for intelligent identification of downburst;
s6, reducing the probability that downburst flows do not exist in the model training result;
s7, constructing a DBIM network model by fusing downburst flow identification of a radar echo time-space sequence with a deep neural network as a basis and a radar echo time sequence image and a radial velocity field time sequence image as input;
s8, finding out the optimal network hyper-parameter at the similar time, and carrying out grouping inspection optimization on the DBIM network model;
wherein, the optimization of the DBIM network model in S8 includes the following steps:
a. randomly splitting an original data set DS into K parts;
b. selecting any one as a verification set, taking the rest as training sets, training the DBIM network model to obtain a model with network parameters, testing on the verification set, and storing an evaluation index E of the model;
c. repeating the step b for K times to ensure that all subsets have one chance as a verification set;
d. taking the mean value of each group of evaluation indexes as the accurate evaluation of the model and as the comprehensive evaluation index of the DBIM network model under the current grouping verification;
s9, introducing category weight to a loss function of the DBIM network model, and giving a penalty term w under the condition that the downburst actually exists but the model is identified as not existing to improve the sensitivity to T (T) =01 in training, wherein the improved loss function is as follows:
where yi is an indicative function of downburst weather, ti is an output of the CNN model corresponding to downburst weather, and represents a probability that the region is identified as downburst weather;
s10, substituting a data set DS into a DBIM network model to train a neural network to obtain a model for intelligently identifying downburst;
s11, according to the model intelligently identified by the downburst, real-time radar detection data is used as the input of the model, and after model calculation, a result DBIM _ Output (D (t 0)) =01 of the model Output of classification identification can be obtained to indicate that the downburst is identified or the real-time radar detection data DBIM _ Output (D (t 0)) =10 indicates that the downburst is not identified.
2. The method for intelligently identifying downburst based on radar detection data according to claim 1, wherein the condition that downburst T (T) =01 in S3 needs to be satisfied simultaneously includes:
a. the instantaneous wind speed reaches or exceeds WS1;
b. in the past IM1 minute, the wind speed changes beyond WS2;
c. identifying the phenomenon that the storm core or the strong echo center is rapidly descended from radar detection data of each time within the latest 1 hour of observing the wind speed by the meteorological station, wherein the descent speed reaches or exceeds WS3;
wherein, WS1 belongs to [17.2,100] and the unit is meter/second; WS2 belongs to [11.7,100] and has the unit of meter/second; IM1 belongs to [5,120] and has the unit of minutes; WS3 ∈ [20,200], in units of kilometers per hour.
3. The intelligent thunderstorm flow identification method based on radar detection data as claimed in claim 1, wherein the algorithm converted in S4 is an interpolation algorithm.
4. The intelligent downburst identification method based on radar detection data as claimed in claim 1, wherein the building of the downburst identification DBIM network model in S7 comprises the following steps:
a. decomposing the data set into a radar echo time sequence image and a radial velocity field time sequence image, wherein each group comprises 8 elevation planes and 6 adjacent time images, and the size of the images is 100 multiplied by 100 pixels;
b. processing radar echo images, designing a 4-channel sparse structure to generate dense data, wherein each channel comprises 1-2 convolution layers, the sizes of convolution kernels are 1 × 1, 3 × 3 and 5 × 5, and the sizes of input images and output images are kept consistent through moderate filling of 4 channels by extracting echo intensity space information under different time scales;
c. and connecting the output of each channel in the channel dimension to obtain a four-dimensional vector, and outputting the four-dimensional vector to a subsequent layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011171264.5A CN112255611B (en) | 2020-10-28 | 2020-10-28 | Intelligent downburst identification method based on radar detection data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011171264.5A CN112255611B (en) | 2020-10-28 | 2020-10-28 | Intelligent downburst identification method based on radar detection data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112255611A CN112255611A (en) | 2021-01-22 |
| CN112255611B true CN112255611B (en) | 2023-04-11 |
Family
ID=74262203
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011171264.5A Active CN112255611B (en) | 2020-10-28 | 2020-10-28 | Intelligent downburst identification method based on radar detection data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112255611B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN115169431A (en) * | 2022-05-16 | 2022-10-11 | 湖南师范大学 | Thunderstorm identification method and system |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5796611A (en) * | 1994-10-04 | 1998-08-18 | Nippon Telegraph And Telephone Corporation | Weather forecast apparatus and method based on recognition of echo patterns of radar images |
| CN108254750B (en) * | 2017-11-24 | 2021-07-30 | 南京信息工程大学 | Down-blast intelligent identification early warning method based on radar data |
| CN109765559B (en) * | 2019-02-02 | 2019-11-05 | 中国水利水电科学研究院 | A kind of mountain area disastrous rainstorm recognition methods based on Doppler radar and deep learning |
| CN111796343B (en) * | 2020-06-15 | 2022-07-01 | 兰州中心气象台(兰州干旱生态环境监测预测中心) | Strong convection weather short-time approaching recognition method based on artificial intelligence algorithm |
-
2020
- 2020-10-28 CN CN202011171264.5A patent/CN112255611B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112255611A (en) | 2021-01-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112255611B (en) | Intelligent downburst identification method based on radar detection data | |
| CN111523699A (en) | Overhead line fault probability prediction method based on comprehensive state health degree | |
| CN114994801B (en) | Lightning monitoring and early warning method and device | |
| CN113391239B (en) | Mutual inductor anomaly monitoring method and system based on edge calculation | |
| CN112069955B (en) | Typhoon intensity remote sensing inversion method based on deep learning | |
| CN110796360A (en) | Fixed traffic detection source multi-scale data fusion method | |
| CN112070103B (en) | Method for inverting atmospheric visibility through microwave link network gridding self-adaptive variable scale | |
| CN115598738A (en) | Hail early warning method and device | |
| CN115857059A (en) | Ice accumulation forecast management method, system, readable storage medium and equipment | |
| CN114386654A (en) | Multi-scale numerical weather forecasting mode fusion weather forecasting method and device | |
| CN112948352B (en) | Method for constructing atmospheric optical turbulence space-time characteristics and probabilistic database | |
| CN113606099A (en) | Method and system for monitoring icing of blades of wind turbine generator | |
| CN113176420A (en) | Wind speed forecasting and correcting system for power grid tower point | |
| CN111427101A (en) | Thunderstorm strong wind grading early warning method, system, storage medium and computer program | |
| CN115080629B (en) | Meteorological element forecasting system and method based on remote sensing technology | |
| CN116012618A (en) | Weather identification method, system, equipment and medium based on radar echo diagram | |
| CN113807404B (en) | Intelligent monitoring and emergency system of intelligent traffic network for multi-source information fusion | |
| CN114722909A (en) | Solar flare time sequence classification method based on low-dimensional convolutional neural network | |
| CN113936166A (en) | Hail echo identification method and system based on Doppler weather radar data | |
| CN113609406A (en) | High-altitude operation trend information sharing method, system and equipment based on geocoding | |
| Xuemin et al. | Research on short-term offshore wind power prediction based on sodar wind measurement | |
| Qiao | A new method for producing temperature profiles based on ERA5 and RAOB | |
| CN117991412B (en) | Extreme precipitation prediction method and system based on multi-mode data | |
| CN117131783B (en) | Multi-mode learning-based power transmission line risk prediction model, method and system | |
| CN117392572B (en) | Transmission tower bird nest detection method based on unmanned aerial vehicle inspection |
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 |