CN117237677B - Precipitation prediction correction method for overall similarity of strong precipitation space based on deep learning - Google Patents
Precipitation prediction correction method for overall similarity of strong precipitation space based on deep learning Download PDFInfo
- Publication number
- CN117237677B CN117237677B CN202311518546.1A CN202311518546A CN117237677B CN 117237677 B CN117237677 B CN 117237677B CN 202311518546 A CN202311518546 A CN 202311518546A CN 117237677 B CN117237677 B CN 117237677B
- Authority
- CN
- China
- Prior art keywords
- precipitation
- model
- field
- gan
- steps
- 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
- 238000001556 precipitation Methods 0.000 title claims abstract description 209
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000012937 correction Methods 0.000 title claims abstract description 45
- 238000013135 deep learning Methods 0.000 title claims abstract description 28
- 238000012360 testing method Methods 0.000 claims abstract description 7
- 238000005457 optimization Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 30
- 238000012549 training Methods 0.000 claims description 13
- 238000013136 deep learning model Methods 0.000 claims description 11
- 238000001914 filtration Methods 0.000 claims description 8
- 238000009499 grossing Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention discloses a precipitation prediction correction method of the overall similarity of a strong precipitation space based on deep learning, which comprises the following steps: (1) identifying precipitation attributes using YOLOv 5; (2) establishing a GAN-based rainfall forecast correction model; (3) Establishing a strong precipitation correction model O-GAN based on GAN and fused with precipitation space characteristics; (4) Substituting the numerical mode forecast data of the test period into the model O-GAN to generate a rainfall forecast after post-treatment; the correction method effectively improves the correction skill of the traditional optimization-only point-by-point error model; the end-to-end output from the precipitation picture to the precipitation rain cluster space attribute is realized, and the objective identification efficiency is improved; the problem of prediction fuzzification possibly occurring in the traditional point-by-point correction model is avoided, meanwhile, the characteristic of strong precipitation can be effectively captured, and the accuracy of precipitation prediction is improved.
Description
Technical Field
The invention relates to the technical field of weather forecast, in particular to a precipitation forecast correction method of strong precipitation space overall similarity based on deep learning.
Background
The method is limited by factors such as chaotic characteristics of the atmosphere, errors in initial field data assimilation, imperfect physical process parameterization and the like, and the original forecast output by each meteorological center numerical mode generally has systematic deviations of different degrees. Especially, the model forecasting capability is obviously insufficient for forecasting variables which are strong in nonlinearity such as strong precipitation, ambiguous in microphysics and affected by complex multiscales. Due to the extreme unbalance between the strong and weak precipitation samples, a certain correction difficulty is caused to a data-driven deep learning algorithm, and the problem of weak prediction of a strong precipitation event often occurs. In addition, because the weather forecast field has complex multi-scale space-time change characteristics, especially precipitation has the characteristics of biased distribution, non-negative data distribution, sparse local spatial distribution, temporal discontinuity and the like, the problem of spatial deviation of forecast variables should be considered and solved in the construction of the deep learning model.
The distance error of 'point-to-point' forecast is often adopted as a target loss function in the structure of the traditional deep learning mode precipitation forecast post-processing algorithm, the traditional target loss function is directly applied to the precipitation forecast regulation, the problem of 'double punishment' is usually limited, and the whole forecast value is close to the average value of a forecast object, so that the forecast distortion and fuzzification are caused. Therefore, a new method for combining the spatial structure attribute of precipitation with the deep learning technology is explored, the deep learning strong precipitation prediction model is optimized, and the method has important application value and guiding significance for improving the strong precipitation prediction level.
Disclosure of Invention
The invention aims to: the invention aims to provide a precipitation prediction correction method for the overall similarity of a strong precipitation space based on deep learning, which aims to solve the problems of insufficient strong precipitation prediction capacity and large deviation of raindrop position and shape prediction.
The technical scheme is as follows: the invention relates to a precipitation prediction correction method of the overall similarity of a strong precipitation space based on deep learning, which comprises the following steps:
(1) Identifying the precipitation attribute by utilizing YOLOv 5;
(2) Establishing a GAN-based rainfall forecast correction model;
(3) Establishing a strong precipitation correction model O-GAN based on GAN and fused with precipitation space characteristics;
(4) Substituting the numerical mode forecast data of the test period into the model O-GAN to generate a rainfall forecast after post-treatment.
Further, the step (1) includes the following steps:
(11) Preprocessing the actually measured precipitation image comprises the following steps: intercepting precipitation rain clusters, and filtering secondary precipitation rain clusters;
(12) Objective recognition of precipitation space attributes is carried out on an observed precipitation field by using a MODE method, and the extracted precipitation attributes are used as truth labels of a deep learning recognition model; wherein the precipitation spatial properties include: precipitation rain cluster, rain cluster area, rain cluster position, average rain cluster surface strength, rain cluster length ratio and rain cluster slope;
(13) And (5) establishing a mapping model of the precipitation field and the spatial attribute of the corresponding precipitation object by utilizing the YOLOv 5.
Further, the step (12) includes the steps of:
(121) Identifying a study object of interest by spatial smoothing and precipitation threshold control; spatially convolving an original precipitation field with a radiusSmoothing, namely convolution processing is carried out, and the formula is as follows:
;
wherein, for the original data field +.>For convolved fields +.>Is a filtering function; />And->Is the grid point coordinate;
(122) Threshold control is carried out on the convolution field to obtain a mask fieldI.e. the intensity of precipitation in the convolution field is greater than or equal to the threshold valueIs defined as follows:
;
assigning all the grid points in the continuous area of M (x, y) =1 to the corresponding grid points in the original precipitation field to obtain a reconstructed fieldThe formula is as follows:
;
(123) Calculating spatial attributes of the identified precipitation object, namely the centroid position, the area size and the aspect ratio; the shaft angle is used as a truth value label corresponding to original precipitation observation; wherein the aspect ratio comprises: the ratio of the short axis to the long axis, the aspect ratio of the circular object is 1.0, otherwise <1; the axis angle is the degree by which the spindle rotates counterclockwise from the x-axis.
Further, the step (13) specifically comprises the following steps: error between the rainfall object attribute true value obtained by MODE calculation and the predicted value obtained by YOLO-V5 recognition; using Euclidean distance as a loss function based on a YOLO-V5 precipitation attribute identification model; training the model by an optimization method based on a deep learning model to obtain a deep learning model with high accuracy for identifying a precipitation object and attributes thereof from a precipitation field.
Further, the step (2) includes the following steps:
(21) Establishing a multi-element weather forecast factor and a corresponding precipitation actual measurement data set;
(22) The meteorological element data preprocessing comprises the steps of taking logarithm of precipitation related variables, carrying out maximum and minimum normalization and carrying out missing value processing;
(23) Establishing a model for numerical mode release between a plurality of weather predictors and precipitation variables by using a U-Net model, wherein the corresponding loss function isThe formula is as follows:
;
wherein,is a rainfall forecast field obtained by a U-Net model, < >>Is a corresponding precipitation observation field.
(24) Taking the established U-Net as a generator in a numerical mode release model based on GAN, and selecting a convolution network as a discriminator to obtain corresponding rainfall forecast and corresponding loss functionThe formula is as follows:
;
wherein,is->Weight constant coefficient of (2); />The loss function representing GAN is given by:
。
further, the step (3) includes the following steps:
(31) Respectively extracting the precipitation prediction of the GAN generator and the precipitation attribute corresponding to the actually measured precipitation in the step (2) by using the training stable YOLO-v5 precipitation attribute identification model in the step (1), and establishing a loss function based on a precipitation objectThe formula is as follows:
;
wherein,the training YOLO-v5 model identifies and predicts the rainfall object attribute of the rainfall field; />Representing precipitation attributes corresponding to the measured precipitation;
(32) And constructing a mixed loss function of the point-to-point error, the precipitation data distribution error and the overall similarity of the strong precipitation spatial characteristics, and constructing a deep learning prediction model integrating the precipitation spatial characteristics.
Further, the formula of the mixing loss function in the step (32) is as follows:
;
wherein,and->Respectively->And->Is a weight constant coefficient of (a).
The invention relates to a precipitation prediction correction system based on deep learning and with strong precipitation space overall similarity, which comprises the following components:
precipitation attribute identification module: for identifying precipitation attributes using YOLOv 5;
and a GAN correction module: the method is used for establishing a GAN-based rainfall forecast correction model;
an O-GAN correction module: the method comprises the steps of establishing a strong precipitation correction model O-GAN based on GAN and fusing precipitation space characteristics;
precipitation forecasting module: and substituting the numerical mode forecast data of the test period into the model O-GAN to generate a rainfall forecast after post-treatment.
The device comprises a memory, a processor and a program stored on the memory and capable of running on the processor, wherein the processor realizes the steps in any one of the precipitation prediction correction methods based on the overall similarity of the deep learning strong precipitation space when executing the program.
The storage medium of the present invention stores a computer program designed to implement, when running, the steps in any one of the precipitation prediction correction methods based on the deep learning strong precipitation space overall similarity.
The beneficial effects are that: compared with the prior art, the invention has the following remarkable advantages:
1. according to the method, the actual requirements of the rainfall spatial properties, such as the rainfall center, the rainfall cluster shape and the like, in the rainfall forecast are fully considered, the spatial properties of the rainfall are used as one of targets for training the deep learning model, and the correction skill of the traditional only-optimized point-by-point error model is effectively improved.
2. According to the invention, by utilizing the recognition model YOLO-v5 in computer vision and combining the rainfall and rain cluster space attribute characteristics extracted by utilizing the meteorological objective recognition model MODE, the end-to-end output from the rainfall image to the rainfall and rain cluster space attribute is realized, the objective recognition efficiency is improved, and the rainfall and rain cluster space attribute is effectively fused in a deep learning model frame.
3. According to the method, the generated network GAN is further introduced into the rainfall forecast correction task, the problem of forecast fuzzification possibly occurring in the traditional point-by-point correction model is avoided, meanwhile, the strong rainfall characteristic can be effectively captured, and the rainfall forecast accuracy is improved.
Drawings
FIG. 1 is a general flow chart of the present invention;
FIG. 2 is a schematic diagram of a model for identifying the spatial attribute of precipitation based on YOLO-v 5;
FIG. 3 is a schematic diagram of a rainfall forecast correction model based on U-Net in the invention;
FIG. 4 is a schematic diagram of a correction model for a GAN-based precipitation forecast according to the present invention;
FIG. 5 is a schematic diagram of a correction model for rainfall forecast based on O-GAN.
Description of the embodiments
The technical scheme of the invention is further described below with reference to the accompanying drawings.
As shown in fig. 1, the embodiment of the invention provides a precipitation prediction correction method of the overall similarity of a strong precipitation space based on deep learning, which comprises the following steps:
(1) Identifying the precipitation attribute by utilizing YOLOv 5; as shown in fig. 2, the method comprises the following steps:
(11) Preprocessing the actually measured precipitation image comprises the following steps: intercepting precipitation rain clusters, and filtering secondary precipitation rain clusters;
(12) Objective recognition of precipitation space attributes is carried out on an observed precipitation field by using a MODE method, and the extracted precipitation attributes are used as truth labels of a deep learning recognition model; wherein the precipitation spatial properties include: precipitation rain cluster, rain cluster area, rain cluster position, average rain cluster surface strength, rain cluster length ratio and rain cluster slope; the method comprises the following steps:
(121) Identifying a study object of interest by spatial smoothing and precipitation threshold control; spatially convolving an original precipitation field with a radiusSmoothing, namely convolution processing is carried out, and the formula is as follows:
;
wherein, for the original data field +.>For convolved fields +.>Is a filtering function; />And->Is the grid point coordinate;
(122) Threshold control is carried out on the convolution field to obtain a mask fieldI.e. the intensity of precipitation in the convolution field is greater than or equal to the threshold valueIs defined as follows:
;
assigning all the grid points in the continuous area of M (x, y) =1 to the corresponding grid points in the original precipitation field to obtain a reconstructed fieldThe formula is as follows:
;
(123) Calculating spatial attributes of the identified precipitation object, namely the centroid position, the area size and the aspect ratio; the shaft angle is used as a truth value label corresponding to original precipitation observation; wherein the aspect ratio comprises: the ratio of the short axis to the long axis, the aspect ratio of the circular object is 1.0, otherwise <1; the axis angle is the degree by which the spindle rotates counterclockwise from the x-axis.
(13) And (5) establishing a mapping model of the precipitation field and the spatial attribute of the corresponding precipitation object by utilizing the YOLOv 5. The method comprises the following steps: error between the rainfall object attribute true value obtained by MODE calculation and the predicted value obtained by YOLO-V5 recognition; using Euclidean distance as a loss function based on a YOLO-V5 precipitation attribute identification model; training the model, such as Adam, by an optimization method based on a deep learning model, and obtaining the deep learning model for identifying the precipitation object and the attribute thereof from the precipitation field with high accuracy.
(2) Establishing a GAN-based rainfall forecast correction model; the method comprises the following steps:
(21) Establishing a multi-element weather forecast factor and a corresponding precipitation actual measurement data set;
(22) The meteorological element data preprocessing comprises the steps of taking logarithm of precipitation related variables, carrying out maximum and minimum normalization and carrying out missing value processing;
(23) A numerical model release model between a plurality of weather predictors and precipitation variables is established by using a U-Net model, and as shown in figure 3, the U-Net mainly comprises four parts: the method comprises a convolution layer, a pooling layer, an up-sampling layer and a jump connection layer, wherein the whole network structure is like a letter U, the left half part is a down-sampling process, namely an encoding process, the spatial resolution is reduced along with the depth, the right half part is an up-sampling process, namely a decoding process, and the spatial resolution is gradually improved. The convolution layers extract data space features, and under the cooperation of other parts of the model, different convolution layers extract feature information of space fields with different scales, so that multi-scale features of a weather system can be considered. The skip connection layer retains and transfers the high resolution information extracted in the down-sampling process to the corresponding up-sampling layer, while avoiding the problem of pixel distortion during up-sampling. U-Net typically uses a loss function based on point-by-point errorThe optimization model is given by the following formula:
;
wherein,is a rainfall forecast field obtained by a U-Net model, < >>Is a corresponding precipitation observation field.
As shown in FIG. 4, (24) the established U-Net is used as a generator in the model for numerical mode release based on GAN, and a convolution network is used as a discriminator to obtain corresponding rainfall forecast and corresponding loss functionThe formula is as follows:
;
wherein,is->Weight constant coefficient of (2); />The loss function representing GAN is given by:
。
(3) Establishing a strong precipitation correction model O-GAN based on GAN and fused with precipitation space characteristics; as shown in fig. 5, the method comprises the following steps:
(31) Respectively extracting the precipitation prediction of the GAN generator and the precipitation attribute corresponding to the actually measured precipitation in the step (2) by using the training stable YOLO-v5 precipitation attribute identification model in the step (1), and establishing a loss function based on a precipitation objectThe formula is as follows:
;
wherein,the training YOLO-v5 model identifies and predicts the rainfall object attribute of the rainfall field; />Representing precipitation attributes corresponding to the measured precipitation;
(32) And constructing a mixed loss function of the point-to-point error, the precipitation data distribution error and the overall similarity of the strong precipitation spatial characteristics, and constructing a deep learning prediction model integrating the precipitation spatial characteristics.
The formula of the mixing loss function is as follows:
;
wherein,and->Respectively->And->Is a weight constant coefficient of (a).
(4) Substituting the numerical mode forecast data of the test period into the model O-GAN to generate a rainfall forecast after post-treatment.
The embodiment of the invention also provides a precipitation prediction correction system based on the deep learning and with the overall similarity of the strong precipitation space, which comprises the following steps:
precipitation attribute identification module: for identifying precipitation attributes using YOLOv 5;
and a GAN correction module: the method is used for establishing a GAN-based rainfall forecast correction model;
an O-GAN correction module: the method comprises the steps of establishing a strong precipitation correction model O-GAN based on GAN and fusing precipitation space characteristics;
precipitation forecasting module: and substituting the numerical mode forecast data of the test period into the model O-GAN to generate a rainfall forecast after post-treatment.
The embodiment of the invention also provides equipment, which comprises a memory, a processor and a program stored on the memory and capable of running on the processor, wherein the processor realizes the steps in the precipitation forecast correction method based on the deep learning strong precipitation space overall similarity when executing the program.
The embodiment of the invention also provides a storage medium, which stores a computer program designed to realize the steps in the precipitation prediction correction method based on the overall similarity of the deep learning strong precipitation space when running.
Claims (4)
1. The precipitation prediction correction method for the strong precipitation space overall similarity based on deep learning is characterized by comprising the following steps of:
(1) Identifying the precipitation attribute by utilizing YOLOv 5; the method comprises the following steps:
(11) Preprocessing the actually measured precipitation image comprises the following steps: intercepting precipitation rain clusters, and filtering secondary precipitation rain clusters;
(12) Objective recognition of precipitation space attributes is carried out on an observed precipitation field by using a MODE method, and the extracted precipitation attributes are used as truth labels of a deep learning recognition model; wherein the precipitation spatial properties include: precipitation rain cluster, rain cluster area, rain cluster position, average rain cluster surface strength, rain cluster length ratio and rain cluster slope; the method comprises the following steps:
(121) Identifying a study object of interest by spatial smoothing and precipitation threshold control; spatially convolving an original precipitation field with a radiusSmoothing, namely convolution processing is carried out, and the formula is as follows:
;
wherein, for the original data field +.>For convolved fields +.>Is a filtering function; />And->Is the grid point coordinate;
(122) Threshold control is carried out on the convolution field to obtain a mask fieldI.e. the intensity of precipitation in the convolution field is greater than or equal to the threshold value +.>Is defined as follows:
;
assigning all the grid points in the continuous area of M (x, y) =1 to the corresponding grid points in the original precipitation field to obtain a reconstructed fieldThe formula is as follows:
;
(123) Calculating spatial attributes of the identified precipitation object, namely the centroid position, the area size and the aspect ratio; the shaft angle is used as a truth value label corresponding to original precipitation observation; wherein the aspect ratio comprises: the ratio of the short axis to the long axis, the aspect ratio of the circular object is 1.0, otherwise <1; the shaft angle is the degree of counterclockwise rotation of the spindle from the x-axis;
(13) Utilizing YOLOv5 to establish a mapping model of a precipitation field and the spatial attribute of a corresponding precipitation object; the method comprises the following steps: error between the rainfall object attribute true value obtained by MODE calculation and the predicted value obtained by YOLO-V5 recognition; using Euclidean distance as a loss function based on a YOLO-V5 precipitation attribute identification model; training the model by an optimization method based on a deep learning model to obtain a deep learning model with high accuracy for identifying a precipitation object and attributes thereof from a precipitation field;
(2) Establishing a GAN-based rainfall forecast correction model; the method comprises the following steps:
(21) Establishing a multi-element weather forecast factor and a corresponding precipitation actual measurement data set;
(22) The meteorological element data preprocessing comprises the steps of taking logarithm of precipitation related variables, carrying out maximum and minimum normalization and carrying out missing value processing;
(23) Establishing a model for numerical mode release between a plurality of weather predictors and precipitation variables by using a U-Net model, wherein the corresponding loss function isThe formula is as follows:
;
wherein,is a rainfall forecast field obtained by a U-Net model, < >>Is a corresponding precipitation observation field;
(24) Taking the established U-Net as a generator in a numerical mode release model based on GAN, and selecting a convolution network as a discriminator to obtain corresponding rainfall forecast and corresponding loss functionThe formula is as follows:
;
wherein,is->Weight constant coefficient of (2); />The loss function representing GAN is given by:
;
(3) Establishing a strong precipitation correction model O-GAN based on GAN and fused with precipitation space characteristics; the method comprises the following steps:
(31) Respectively extracting the precipitation prediction of the GAN generator and the precipitation attribute corresponding to the actually measured precipitation in the step (2) by using the training stable YOLO-v5 precipitation attribute identification model in the step (1), and establishing a loss function based on a precipitation objectThe formula is as follows:
;
wherein,the training YOLO-v5 model identifies and predicts the rainfall object attribute of the rainfall field; />Representing precipitation attributes corresponding to the measured precipitation;
(32) Constructing a mixed loss function of the point-to-point error, precipitation data distribution error and overall similarity of the strong precipitation space characteristics, and constructing a deep learning prediction model fused with the precipitation space characteristics; the formula of the mixing loss function is as follows:
;
wherein,and->Respectively->And->Weight constant coefficient of (2);
(4) Substituting the numerical mode forecast data of the test period into the model O-GAN to generate a rainfall forecast after post-treatment.
2. Precipitation forecast correction system of strong precipitation space overall similarity based on degree of depth study, characterized by comprising:
precipitation attribute identification module: for identifying precipitation attributes using YOLOv 5; comprising the following steps:
(11) Preprocessing the actually measured precipitation image comprises the following steps: intercepting precipitation rain clusters, and filtering secondary precipitation rain clusters;
(12) Objective recognition of precipitation space attributes is carried out on an observed precipitation field by using a MODE method, and the extracted precipitation attributes are used as truth labels of a deep learning recognition model; wherein the precipitation spatial properties include: precipitation rain cluster, rain cluster area, rain cluster position, average rain cluster surface strength, rain cluster length ratio and rain cluster slope; the method comprises the following steps:
(121) Identifying a study object of interest by spatial smoothing and precipitation threshold control; spatially convolving an original precipitation field with a radiusSmoothing, namely convolution processing is carried out, and the formula is as follows:
;
wherein, for the original data field +.>For convolved fields +.>Is a filtering function; />And->Is the grid point coordinate;
(122) Threshold control is carried out on the convolution field to obtain a mask fieldI.e. the intensity of precipitation in the convolution field is greater than or equal to the threshold value +.>Is defined as follows:
;
assigning all the grid points in the continuous area of M (x, y) =1 to the corresponding grid points in the original precipitation field to obtain a reconstructed fieldThe formula is as follows:
;
(123) Calculating spatial attributes of the identified precipitation object, namely the centroid position, the area size and the aspect ratio; the shaft angle is used as a truth value label corresponding to original precipitation observation; wherein the aspect ratio comprises: the ratio of the short axis to the long axis, the aspect ratio of the circular object is 1.0, otherwise <1; the shaft angle is the degree of counterclockwise rotation of the spindle from the x-axis;
(13) Utilizing YOLOv5 to establish a mapping model of a precipitation field and the spatial attribute of a corresponding precipitation object; the method comprises the following steps: error between the rainfall object attribute true value obtained by MODE calculation and the predicted value obtained by YOLO-V5 recognition; using Euclidean distance as a loss function based on a YOLO-V5 precipitation attribute identification model; training the model by an optimization method based on a deep learning model to obtain a deep learning model with high accuracy for identifying a precipitation object and attributes thereof from a precipitation field;
and a GAN correction module: the method is used for establishing a GAN-based rainfall forecast correction model; comprising the following steps:
(21) Establishing a multi-element weather forecast factor and a corresponding precipitation actual measurement data set;
(22) The meteorological element data preprocessing comprises the steps of taking logarithm of precipitation related variables, carrying out maximum and minimum normalization and carrying out missing value processing;
(23) Establishing a model for numerical mode release between a plurality of weather predictors and precipitation variables by using a U-Net model, wherein the corresponding loss function isThe formula is as follows:
;
wherein,is a rainfall forecast field obtained by a U-Net model, < >>Is a corresponding precipitation observation field;
(24) Taking the established U-Net as a generator in a numerical mode release model based on GAN, and selecting a convolution network as a discriminator to obtain corresponding rainfall forecast and corresponding loss functionThe formula is as follows:
;
wherein,is->Weight constant coefficient of (2); />The loss function representing GAN is given by:
;
an O-GAN correction module: the method comprises the steps of establishing a strong precipitation correction model O-GAN based on GAN and fusing precipitation space characteristics; comprising the following steps:
(31) Respectively extracting the precipitation prediction of the GAN generator and the precipitation attribute corresponding to the actually measured precipitation in the step (2) by using the training stable YOLO-v5 precipitation attribute identification model in the step (1), and establishing a loss function based on a precipitation objectThe formula is as follows:
;
wherein,the training YOLO-v5 model identifies and predicts the rainfall object attribute of the rainfall field; />Representing precipitation attributes corresponding to the measured precipitation;
(32) Constructing a mixed loss function of the point-to-point error, precipitation data distribution error and overall similarity of the strong precipitation space characteristics, and constructing a deep learning prediction model fused with the precipitation space characteristics; the formula of the mixing loss function is as follows:
;
wherein,and->Respectively->And->Weight constant coefficient of (2);
precipitation forecasting module: and substituting the numerical mode forecast data of the test period into the model O-GAN to generate a rainfall forecast after post-treatment.
3. An apparatus comprising a memory, a processor and a program stored on the memory and executable on the processor, wherein the processor performs the steps of a method for correcting precipitation predictions based on overall similarity of strong precipitation spaces for deep learning as claimed in any one of claims 1.
4. A storage medium storing a computer program, characterized in that the computer program is designed to, when run, implement the steps in a method for correcting precipitation forecast of strong precipitation space overall similarity based on deep learning according to any of the claims 1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311518546.1A CN117237677B (en) | 2023-11-15 | 2023-11-15 | Precipitation prediction correction method for overall similarity of strong precipitation space based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311518546.1A CN117237677B (en) | 2023-11-15 | 2023-11-15 | Precipitation prediction correction method for overall similarity of strong precipitation space based on deep learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117237677A CN117237677A (en) | 2023-12-15 |
CN117237677B true CN117237677B (en) | 2024-03-26 |
Family
ID=89084751
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311518546.1A Active CN117237677B (en) | 2023-11-15 | 2023-11-15 | Precipitation prediction correction method for overall similarity of strong precipitation space based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117237677B (en) |
Citations (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108805839A (en) * | 2018-06-08 | 2018-11-13 | 西安电子科技大学 | Combined estimator image defogging method based on convolutional neural networks |
CN109839617A (en) * | 2019-03-13 | 2019-06-04 | 中国气象局气象探测中心 | A kind of non-precipitation echo removing method and system of radar network composite product |
WO2019126707A1 (en) * | 2017-12-21 | 2019-06-27 | Climacell Inc. | Improved real-time weather forecasting system |
CN110073301A (en) * | 2017-08-02 | 2019-07-30 | 强力物联网投资组合2016有限公司 | The detection method and system under data collection environment in industrial Internet of Things with large data sets |
CN111624680A (en) * | 2020-05-13 | 2020-09-04 | 苏翔 | Precipitation forecast displacement correction algorithm based on object diagnosis |
CN112819830A (en) * | 2021-01-24 | 2021-05-18 | 南京林业大学 | Individual tree crown segmentation method based on deep learning and airborne laser point cloud |
CN113139329A (en) * | 2021-03-04 | 2021-07-20 | 河海大学 | Xinanjiang model parameter calibration method based on hydrological similarity and artificial neural network |
CN113407897A (en) * | 2021-08-18 | 2021-09-17 | 长江水利委员会长江科学院 | Design method of distributed water circulation model based on multi-source mutual-aid water supply mode |
CN114185114A (en) * | 2022-02-16 | 2022-03-15 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Correction method and device for rainfall forecast falling area, electronic equipment and medium |
CN114217319A (en) * | 2022-02-22 | 2022-03-22 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Method and device for correcting weather radar minute rainfall forecast value |
CN114881381A (en) * | 2022-07-11 | 2022-08-09 | 南京信息工程大学 | Urban ponding water level prediction method and system based on improved convolutional neural network |
CN114898206A (en) * | 2022-03-31 | 2022-08-12 | 中国电力科学研究院有限公司 | Short-time heavy rainfall forecasting method, computer equipment and storage medium |
CN115292667A (en) * | 2022-06-15 | 2022-11-04 | 中山大学 | Space probability analysis method and system based on rainfall forecast and remote correlation corresponding relation |
CN115294560A (en) * | 2022-08-10 | 2022-11-04 | 青岛文达通科技股份有限公司 | Vehicle tracking method and system based on attribute matching and motion trail prediction |
WO2022250686A1 (en) * | 2021-05-28 | 2022-12-01 | Google Llc | Optimizing a probability of precipitation forecast |
CN115423999A (en) * | 2022-08-02 | 2022-12-02 | 上海软杰智能设备有限公司 | YOLO-V5-based vehicle attribute identification method |
KR20220164198A (en) * | 2021-06-04 | 2022-12-13 | 한국과학기술정보연구원 | Method for predicting precipitation based on deep learning and apparatus implementing the same method |
CN115600710A (en) * | 2021-07-07 | 2023-01-13 | 重庆城市综合交通枢纽(集团)有限公司(Cn) | Method for predicting regional demand of manned vehicle, electronic device and storage medium |
CN115600483A (en) * | 2022-07-04 | 2023-01-13 | 临沂大学(Cn) | Rainfall inversion method based on deep forest |
CN115792915A (en) * | 2022-11-09 | 2023-03-14 | 河南省气象台 | 0-6 h rainfall forecasting method integrating numerical mode and radar monitoring |
WO2023052570A1 (en) * | 2021-09-29 | 2023-04-06 | Swiss Reinsurance Company Ltd. | Aerial and/or satellite imagery-based, optical sensory system and method for quantitative measurements and recognition of property damage after an occurred natural catastrophe event |
WO2023103587A1 (en) * | 2021-12-09 | 2023-06-15 | 南京邮电大学 | Imminent precipitation forecast method and apparatus |
CN116628462A (en) * | 2023-07-19 | 2023-08-22 | 中南大学 | Urban three-dimensional space land function attribute identification and space-time change monitoring analysis method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20230260045A1 (en) * | 2022-02-16 | 2023-08-17 | ExlService Holdings, Inc. | Reducing network traffic associated with generating event predictions based on cognitive image analysis systems and methods |
-
2023
- 2023-11-15 CN CN202311518546.1A patent/CN117237677B/en active Active
Patent Citations (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110073301A (en) * | 2017-08-02 | 2019-07-30 | 强力物联网投资组合2016有限公司 | The detection method and system under data collection environment in industrial Internet of Things with large data sets |
WO2019126707A1 (en) * | 2017-12-21 | 2019-06-27 | Climacell Inc. | Improved real-time weather forecasting system |
CN108805839A (en) * | 2018-06-08 | 2018-11-13 | 西安电子科技大学 | Combined estimator image defogging method based on convolutional neural networks |
CN109839617A (en) * | 2019-03-13 | 2019-06-04 | 中国气象局气象探测中心 | A kind of non-precipitation echo removing method and system of radar network composite product |
CN111624680A (en) * | 2020-05-13 | 2020-09-04 | 苏翔 | Precipitation forecast displacement correction algorithm based on object diagnosis |
CN112819830A (en) * | 2021-01-24 | 2021-05-18 | 南京林业大学 | Individual tree crown segmentation method based on deep learning and airborne laser point cloud |
CN113139329A (en) * | 2021-03-04 | 2021-07-20 | 河海大学 | Xinanjiang model parameter calibration method based on hydrological similarity and artificial neural network |
WO2022250686A1 (en) * | 2021-05-28 | 2022-12-01 | Google Llc | Optimizing a probability of precipitation forecast |
KR20220164198A (en) * | 2021-06-04 | 2022-12-13 | 한국과학기술정보연구원 | Method for predicting precipitation based on deep learning and apparatus implementing the same method |
CN115600710A (en) * | 2021-07-07 | 2023-01-13 | 重庆城市综合交通枢纽(集团)有限公司(Cn) | Method for predicting regional demand of manned vehicle, electronic device and storage medium |
CN113407897A (en) * | 2021-08-18 | 2021-09-17 | 长江水利委员会长江科学院 | Design method of distributed water circulation model based on multi-source mutual-aid water supply mode |
WO2023052570A1 (en) * | 2021-09-29 | 2023-04-06 | Swiss Reinsurance Company Ltd. | Aerial and/or satellite imagery-based, optical sensory system and method for quantitative measurements and recognition of property damage after an occurred natural catastrophe event |
WO2023103587A1 (en) * | 2021-12-09 | 2023-06-15 | 南京邮电大学 | Imminent precipitation forecast method and apparatus |
CN114185114A (en) * | 2022-02-16 | 2022-03-15 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Correction method and device for rainfall forecast falling area, electronic equipment and medium |
CN114217319A (en) * | 2022-02-22 | 2022-03-22 | 中国气象局公共气象服务中心(国家预警信息发布中心) | Method and device for correcting weather radar minute rainfall forecast value |
CN114898206A (en) * | 2022-03-31 | 2022-08-12 | 中国电力科学研究院有限公司 | Short-time heavy rainfall forecasting method, computer equipment and storage medium |
CN115292667A (en) * | 2022-06-15 | 2022-11-04 | 中山大学 | Space probability analysis method and system based on rainfall forecast and remote correlation corresponding relation |
CN115600483A (en) * | 2022-07-04 | 2023-01-13 | 临沂大学(Cn) | Rainfall inversion method based on deep forest |
CN114881381A (en) * | 2022-07-11 | 2022-08-09 | 南京信息工程大学 | Urban ponding water level prediction method and system based on improved convolutional neural network |
CN115423999A (en) * | 2022-08-02 | 2022-12-02 | 上海软杰智能设备有限公司 | YOLO-V5-based vehicle attribute identification method |
CN115294560A (en) * | 2022-08-10 | 2022-11-04 | 青岛文达通科技股份有限公司 | Vehicle tracking method and system based on attribute matching and motion trail prediction |
CN115792915A (en) * | 2022-11-09 | 2023-03-14 | 河南省气象台 | 0-6 h rainfall forecasting method integrating numerical mode and radar monitoring |
CN116628462A (en) * | 2023-07-19 | 2023-08-22 | 中南大学 | Urban three-dimensional space land function attribute identification and space-time change monitoring analysis method |
Non-Patent Citations (3)
Title |
---|
"CLGAN: A GAN-based video prediction model for precipitation nowcasting";Yan Ji等;《Geoscientific Model Development》;第16卷(第10期);第2737页摘要和第2740-2742页第3节,图1 * |
"Deep-learning-based post-processing for probabilistic precipitation forecasting";Yan Ji等;《Front. Earth Sci》;20220920;第10卷;第1-13页 * |
"Skilful precipitation nowcasting using deep generative models of radar";Suman Ravuri等;《Nature》;20211231;第672-677页 * |
Also Published As
Publication number | Publication date |
---|---|
CN117237677A (en) | 2023-12-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114120102A (en) | Boundary-optimized remote sensing image semantic segmentation method, device, equipment and medium | |
CN109977997B (en) | Image target detection and segmentation method based on convolutional neural network rapid robustness | |
CN112395975A (en) | Remote sensing image target detection method based on rotating area generation network | |
CN113449594B (en) | Multilayer network combined remote sensing image ground semantic segmentation and area calculation method | |
CN113592822B (en) | Insulator defect positioning method for electric power inspection image | |
CN115546630A (en) | Construction site extraction method and system based on remote sensing image characteristic target detection | |
CN110838112A (en) | Insulator defect detection method based on Hough transform and YOLOv3 network | |
CN114694038A (en) | High-resolution remote sensing image classification method and system based on deep learning | |
CN111860255A (en) | Training and using method, device, equipment and medium of driving detection model | |
US10685443B2 (en) | Cloud detection using images | |
CN115861619A (en) | Airborne LiDAR (light detection and ranging) urban point cloud semantic segmentation method and system of recursive residual double-attention kernel point convolution network | |
CN110874591A (en) | Image positioning method, device, equipment and storage medium | |
CN112330701A (en) | Tissue pathology image cell nucleus segmentation method and system based on polar coordinate representation | |
CN115436920A (en) | Laser radar calibration method and related equipment | |
CN117237677B (en) | Precipitation prediction correction method for overall similarity of strong precipitation space based on deep learning | |
CN113034511A (en) | Rural building identification algorithm based on high-resolution remote sensing image and deep learning | |
CN112232102A (en) | Building target identification method and system based on deep neural network and multitask learning | |
CN115937546A (en) | Image matching method, three-dimensional image reconstruction method, image matching device, three-dimensional image reconstruction device, electronic apparatus, and medium | |
CN112164087B (en) | Super-pixel segmentation method and device based on edge constraint and segmentation boundary search | |
CN114998570A (en) | Method and device for determining object detection frame, storage medium and electronic device | |
CN113780096A (en) | Vegetation land feature extraction method based on semi-supervised deep learning | |
CN113256581B (en) | Automatic defect sample labeling method and system based on visual attention modeling fusion | |
CN117807518B (en) | Automatic identification method, system and equipment for slot line or shear line on conventional weather diagram | |
CN116030347B (en) | High-resolution remote sensing image building extraction method based on attention network | |
CN112686254B (en) | Typhoon center positioning method based on infrared satellite cloud picture |
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 |