A method and system for identifying severe convective wind classification

Abstract

The present invention belongs to the field of meteorological identification technology, and discloses a method and system for graded identification of severe convective gales. The method includes: collecting three-dimensional radar network jigsaw puzzles, lightning locators, and minute-level ground automatic observation station data, and performing corresponding processing; constructing a data set based on a "face point" identification scheme; using the SCWINet method to construct a model structure, loss function, and immune evolutionary algorithm for identification; and performing comprehensive testing and individual case testing. The present invention uses a deep learning method (SCWINet) for graded identification of severe convective gales of levels 8, 9, 10, and ≥11 using radar network data such as three-dimensional reflectivity and vertical liquid water content. The method constructs a loss function (TS) that is a weighted sum of a differentiable TS and a root mean square error (MSE). diff ) is used to solve the serious imbalance problem of samples in strong convective winds.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological identification, and in particular relates to a classification identification method and system for severe convective gale. Background Art

[0002] Severe convective gale refers to the strong wind weather with instantaneous wind force reaching or exceeding level 8 (17.2m / s) on the ground and accompanied by thunderstorms. It has seriously threatened the safety of life and property of the people and social and economic development. At present, the forecasting ability of global operational numerical forecast models and mesoscale numerical models for severe convective gale is very limited, and nowcasting and warning are the most important means for business to deal with severe convective gale. However, although the automation and spatiotemporal refinement of the national observation network are increasing, it is still difficult to meet the forecast and warning needs of severe convective gale. Secondly, affected by factors such as the friction of the underlying surface and the changes in the temperature and pressure field, the intensity and suddenness of severe convective gale are easily underestimated in operational forecasts, especially when the convective system enters the vicinity of the ocean and lakes from the rough and gentle surface, the wind force is prone to increase sharply and the wind direction suddenly changes. The existing strong convective wind recognition technology is mainly based on the typical radar echo features summarized by meteorological experts. It can be mainly divided into traditional recognition algorithms such as fuzzy logic and support vector machines and deep learning algorithms such as fully connected neural networks, convolutional neural networks, and recurrent convolutional neural networks. The above methods have achieved certain results in identifying strong convective winds. However, so far, research on the hierarchical recognition of strong convective winds is extremely rare, and the serious imbalance of samples is the main factor leading to this problem.

[0003] Wind speed samples decrease sharply with the increase of wind speed. There are very few strong convective winds above level 8, and there is a serious sample imbalance problem. Extremely strong convective winds above level 10 are relatively rare, which is an extreme sample imbalance problem. Due to the extremely small number of samples, it is difficult to fully characterize the complex nonlinear characteristics that cause severe convective weather, which seriously restricts various existing objective identification methods. Even though deep learning methods have developed rapidly in the field of meteorology and have achieved better application results than other objective methods, it is still difficult to solve the serious sample imbalance problem of strong convective winds, so related research and applications are extremely rare.

[0004] However, so far, research on the graded identification of strong convective winds, especially extreme winds (above 25.8 m / s, level 10), is extremely rare. The main reasons for this problem are: First, it is difficult for traditional objective identification methods to capture the complex nonlinear mechanism that causes strong convective winds, and their identification effect is often poor (Zhou Kanghui et al., 2021; Yang Xuan et al., 2022). Second, affected by multiple factors such as the underlying surface and observation stations, the weather system that causes strong convective winds has a large deviation from the location of the recording station (Fang Chong et al., 2014), which easily leads to poor applicability of the data set required by the identification algorithm. Third, there are serious and even extreme sample imbalance problems in strong convective winds. For example, winds above level 10 are very rare in a strong convective process, which seriously restricts the applicability of most methods such as traditional methods and emerging artificial intelligence algorithms.

[0005] In order to effectively alleviate the above-mentioned difficulties in the classification and identification of severe convective gale, the present invention proposes a universal classification and identification method for severe convective gale, which provides scientific support for improving the refined forecasting and early warning capabilities of severe convective gale, and realizes the classification and identification of severe convective gale of levels 8, 9, 10, and ≥11 through loss function optimization based on immune evolutionary algorithm and convolutional neural network. Summary of the invention

[0006] In view of the problems existing in the prior art, the present invention provides a method and system for classifying and identifying severe convective strong winds.

[0007] The present invention is achieved by providing a method for classifying and identifying severe convective gale winds, the method comprising:

[0008] S1: Collect the data of 3D radar network mosaic, lightning locator, and minute-level ground automatic observation station, and process them accordingly;

[0009] S2: Dataset construction based on the “face point” recognition scheme;

[0010] S3: Use SCWINet method to build model structure, loss function and immune evolutionary algorithm for recognition;

[0011] S4: Perform comprehensive tests and individual case tests.

[0012] Furthermore, the corresponding processing in S1 specifically includes:

[0013] First, the data quality is controlled. Based on the above-mentioned three-dimensional radar network data, the vertical and horizontal characteristics of the convective system are extracted and used as dependent variables for identifying strong convective winds. After quality control, the data set is divided into training, validation, and test sets.

[0014] Furthermore, S2 specifically includes: using a 5×5 grid area to identify whether strong convective wind occurs at the center point.

[0015] Furthermore, the number of input features of the model in the model structure is 16, and 5×5 represents the spatial area size of a single input feature. After that, it undergoes two convolutions to expand it to 32 and 16 dimensions, and then enters 2 fully connected layers with a dimension of 16, and finally outputs the severe convective wind speed.

[0016] Furthermore, the loss function proposes a differentiable TS (TS diff ), as shown in formula (1), which can make the model pay more attention to strong convective winds during training, thereby alleviating the serious imbalance problem of wind speed intensity samples. However, its loss function and its multi-order derivatives are complex, and the gradient is very unstable. MSE has better convergence and stability, and the recognition result tends to be biased towards low wind speed values. Therefore, MSE and TS are combined. diff The weighted sum is used to construct MTLoss, and the MSE and TS are brought into play by appropriately adjusting the weight coefficient. diff The advantages of both and minimize their disadvantages:

[0017]

[0018] MTLoss=MSELoss+α×TSLoss (3)

[0019]

[0020] Where thld k is the wind speed threshold used by MTLoss; k represents the kth wind speed threshold; α is a weight parameter to be determined. By adjusting α, the recognition BIAS of GaleINet in the test set is ≤1 and TS is maximized.

[0021] Furthermore, in the immune evolutionary algorithm:

[0022] Loss uses the non-differentiable TS as the loss function and cannot directly constrain the BIAS of the model. Therefore, when the Loss is low, the BIAS of the model will be too high or too low. To solve this problem, BIAS constraints are imposed on the validation set during the training of SCWINet, as shown in formula (4). Based on this, the specific implementation process of IEA training SCWINet is as follows:

[0023] F(x)=Loss (5)

[0024] max(BIAS k )≤1 (6)

[0025]

[0026] Where F(x) is the fitness function of IEA optimization; x is the parameter to be optimized, which corresponds to the α parameter in formula (4). When F(x) reaches the minimum, the TS of SCWINet is the highest, and its parameters are the optimal parameters.

[0027] Furthermore, the comprehensive test calculates the TS and BIAS of the training set, validation set, and test set respectively;

[0028] The case test selected a severe convective wind event to compare and test the applicability of SCWINet.

[0029] Another object of the present invention is to provide a system for classifying and identifying strong convective and strong winds for implementing the method for classifying and identifying strong convective and strong winds, the system comprising:

[0030] The data collection and processing module is used to collect data from the three-dimensional radar network puzzle, lightning locator, and minute-level ground automatic observation station, and perform corresponding processing;

[0031] The dataset construction module is connected with the data collection and processing module to construct the dataset based on the “face point” recognition scheme;

[0032] Construct a recognition module, connect it with the dataset construction module, and use the SCWINet method to build the model structure, loss function, and immune evolutionary algorithm for recognition;

[0033] The inspection module is connected with the construction identification module to perform comprehensive inspection and individual case inspection.

[0034] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the severe convective gale classification identification method.

[0035] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the severe convective gale classification identification method.

[0036] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0037] First, the present invention proposes a deep learning algorithm SCWINet for hierarchical identification of severe convective winds, which is based on a convolutional neural network and uses a differentiable TS (TS diff ) and the weighted sum of the root mean square error are used to construct the loss function of the convolutional neural network, thereby solving the problem of significantly low wind speed recognition caused by severe imbalance of strong convective wind samples. The main conclusions are as follows:

[0038] (1) SCWINet has good recognition ability for strong convective winds of different levels under different weather types, and its recognition effect for winds above 17.2 m / s is significantly better than fuzzy logic.

[0039] (2) The loss function based on the weighted sum of differentiable TS and MSE can effectively deal with the serious imbalance of samples of strong convective winds, especially extreme strong convective winds, thereby significantly enhancing the recognition ability of strong convective winds. However, using only MSE will cause the recognized wind speed to be significantly lower than 17.2 m / s due to the sample imbalance problem. TS diff This will lead to unstable gradients and make it difficult to train the neural network.

[0040] (3) There are inevitably some false alarms in the monitoring and identification of strong convective winds. The main reason is that there are still some inaccuracies in the construction of the strong wind data set. For example, the elimination of non-strong convective wind samples also eliminates information that is conducive to the elimination of empty reports to a certain extent. Although the strong convective wind samples in this paper have serious and even extreme imbalance problems, the proportion is still significantly higher in the real atmosphere. Secondly, the identification characteristics of strong convective winds, especially extremely strong convective winds, are still inaccurate. The existing radar three-dimensional characteristics and their polarization indicators for strong convective winds are not only highly nonlinear, but also have unclear relationships. They are also affected by the underlying surface on the automatic station observations, the changes in atmospheric environmental characteristics on the dual-polarization radar observations, the lack of radar quality control, and the temporal and spatial drift of strong convective winds.

[0041] (4) The non-differentiable TS is differentiable and applied to the loss function of the neural network, which is a deep learning optimization strategy based on meteorological expert knowledge. It can be seen that the application of artificial intelligence technology in the field of meteorology still needs to rely on the guidance of subjective forecasting experience, mathematical and physical mechanisms, and other meteorological expert knowledge.

[0042] Second, the present invention proposes a deep learning method (SCWINet) for classifying severe convective winds of level 8, 9, 10, and ≥11 using radar network data such as three-dimensional reflectivity and vertical liquid water content. The method constructs a loss function (TS) that is a weighted sum of differentiable TS and root mean square error (MSE). diff ) to solve the serious imbalance problem of strong convective wind samples. The main results are as follows: SCWINet has good recognition ability for strong convective winds of different levels under different weather types, and its recognition effect for strong convective winds above 17.2m / s is significantly better than the existing strong convective wind recognition algorithm based on storm recognition and fuzzy logic (fuzzy logic method) of the provincial station. In addition, subjective forecasting experience, mathematical and physical mechanisms and other meteorological expert knowledge have important guiding significance for the effective application of artificial intelligence technology in the meteorological field.

[0043] Third, the technical solution of the present invention fills the technical gap in the industry at home and abroad: the gap in the technology for identifying graded severe convective gales of levels 8-12.

[0044] The technical solution of the present invention solves a technical problem that people have been eager to solve but have never been able to solve: it solves the technical problem existing in the identification of graded severe convective gales of levels 8-12.

[0045] Fourthly, the severe convective wind classification identification method provided by the present invention has brought about significant technical progress, which is specifically manifested in the following aspects:

[0046] 1. Integration and utilization of multi-source data: The method integrates multiple data sources such as three-dimensional radar network puzzle, lightning locator and minute-level ground automatic observation station. This integration of multi-source data can provide more comprehensive and rich information, thereby improving the accuracy and precision of strong wind identification. The comprehensive utilization of multi-source data provides a new technical means for the classification and identification of severe convective strong winds under complex meteorological conditions.

[0047] 2. Point-oriented recognition scheme: Based on the "point-oriented" recognition scheme, the data set is constructed. This method can more accurately identify and locate the location and intensity of strong winds. Compared with the previous overall area or path recognition methods, it greatly improves the recognition efficiency and accuracy, and helps to achieve more accurate strong wind warnings and predictions.

[0048] 3. Deep learning model and algorithm: The SCWINet method is used to build the model structure, and the loss function and immune evolutionary algorithm are combined for identification. The deep learning model has significant advantages in dealing with complex nonlinear problems, while the immune evolutionary algorithm can optimize the model parameters and improve the generalization ability and robustness of the model. The application of this technology has significantly improved the accuracy and automation of high wind classification.

[0049] 4. Comprehensive and detailed testing: The identification method is verified and evaluated through comprehensive testing and case testing. Comprehensive testing can evaluate the overall performance of the model, while case testing can deeply analyze the performance of the model in different scenarios, thereby discovering potential problems and making targeted improvements. This comprehensive testing method ensures the reliability and stability of the method.

[0050] 5. Quality control and data division: During the data processing stage, the collected data was strictly quality controlled to ensure the accuracy and reliability of the data. At the same time, the data set was divided into training, validation and test sets, which facilitated the effective training and performance evaluation of the model.

[0051] The severe convective gale classification identification method provided by the present invention significantly improves the accuracy and automation of gale identification through technical innovations such as multi-source data integration, point-oriented identification scheme, application of deep learning models and algorithms, and comprehensive and detailed inspection, providing strong technical support for meteorological warning and forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a method for classifying and identifying severe convective gale winds provided by an embodiment of the present invention;

[0053] Figure 2 is a frequency distribution diagram of severe convective gale at different wind speed levels provided by an embodiment of the present invention;

[0054] Figure 3 is a structural diagram of the SCWINet model provided by an embodiment of the present invention;

[0055] Figure 4 It is a structural diagram of a severe convective gale classification identification system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0057] Based on the provided severe convective wind classification identification method, the following are two specific embodiments:

[0058] Example 1: Severe convective wind forecast in urban areas

[0059] In this embodiment, the severe convective gale classification identification method is applied to a specific urban area, with the purpose of identifying and predicting the occurrence of severe convective gale events in advance so as to take corresponding preventive measures.

[0060] 1. Data collection and processing: Collect the 3D radar network mosaic, lightning locator data, and minute-level ground automatic observation station data in the urban area. Through professional data processing and quality control, ensure the accuracy and reliability of the data. Based on these data, extract the vertical and horizontal characteristic parameters of the severe convective system as the basis for subsequent identification work.

[0061] 2. Model construction and training: Based on the "face point" recognition scheme, the recognition model is constructed using the SCWINet method. By setting a suitable loss function and using the immune evolutionary algorithm, the model parameters are optimized to ensure that the model can accurately identify severe convective wind events of different levels. The data set is divided into training set, validation set and test set for model training, validation and testing.

[0062] 3. Comprehensive test and case analysis: After the model training and testing are completed, the model is comprehensively tested, including the evaluation of key indicators such as accuracy and recall. At the same time, representative cases are selected for detailed analysis to verify the effect and accuracy of the model in practical applications.

[0063] Example 2: Monitoring of severe convective winds in agricultural areas

[0064] In this embodiment, the method is used to monitor a large agricultural area with the goal of identifying severe convective winds in advance and reducing crop losses.

[0065] 1. Data preparation and preprocessing: Collect relevant 3D radar network, lightning location and ground automatic observation station data for a wide range of agricultural areas. Through data quality control, ensure the data is accurate and reliable, and extract key convective system features.

[0066] 2. Model establishment and optimization: The SCWINet method is used to establish a strong convective wind identification model specific to agricultural areas. The model structure and parameters are adjusted to adapt to the specific meteorological conditions and geographical characteristics of agricultural areas. The immune evolutionary algorithm is used to optimize the model to improve the generalization ability and prediction accuracy of the model.

[0067] 3. Effect evaluation and application: Through comprehensive evaluation of the model, analyze its performance in the prediction of severe convective winds in agricultural areas. Select severe convective wind events in specific time periods and regions as individual cases, and deeply analyze the prediction effect of the model to guide disaster prevention work in agricultural production.

[0068] These two examples demonstrate the application of the severe convective wind classification and identification method in urban and agricultural areas. Through precise data processing, model construction and optimization, the prediction accuracy of severe convective winds is effectively improved, providing a scientific basis for taking timely preventive measures.

[0069] like Figure 1 As shown, an embodiment of the present invention provides a method for classifying and identifying severe convective gale winds, the method comprising:

[0070] S1: Collect the data of 3D radar network mosaic, lightning locator, and minute-level ground automatic observation station, and process them accordingly;

[0071] S2: Dataset construction based on the “face point” recognition scheme;

[0072] S3: Use SCWINet method to build model structure, loss function and immune evolutionary algorithm for recognition;

[0073] S4: Perform comprehensive tests and individual case tests. Specific embodiment:

[0075] 1 Materials and methods

[0076] 1.1 Information

[0077] The 2022-2023 Zhejiang Province three-dimensional radar network puzzle, lightning locator, and minute-level ground automatic observation station data were selected. First, the data quality control was performed, see Table 1 for details. Many studies have shown that the vertical and horizontal characteristics of the convective system are closely related to the occurrence of strong convective winds, and this is used to identify whether strong convective winds occur. To this end, this paper extracted the vertical and horizontal characteristics of the convective system based on the above-mentioned three-dimensional radar network data, and used them as the dependent variable for identifying strong convective winds, see Table 2 for details. After quality control in Table 1, the data set is divided into training, validation, and test sets, and the detailed information is shown in Table 3. In addition, the frequency distribution of the total convective wind speed of the three data sets is as follows Figure 2 As shown. Figure 2 It can be seen that the sample frequency decreases rapidly with the increase of wind speed level, and there is an extremely unbalanced distribution problem in the samples of extremely strong convective winds above level 10.

[0078] Table 1 Data quality control methods

[0079]

[0080] Table 2 Dependent variables used to identify severe convective winds

[0081]

[0082] Table 3 Training, validation, and test set information

[0083]

[0084] 1.2 Dataset Construction Based on “Face Point” Recognition Scheme

[0085] There are two main data set construction schemes for severe convective wind identification: “point-to-point” and “face-to-point”. The specific schemes are shown in Table 4. Since the triggering of severe convective winds is closely related to the small-scale spatiotemporal characteristics of the convective system, this paper chooses the “face-to-point” identification scheme, which uses a 5×5 grid area to identify whether a strong convective wind occurs at the center point.

[0086] Table 4 Data construction scheme

[0087]

[0088] 1.3SCWINet

[0089] 1.3.1 Model structure

[0090] The network structure of SCWINet is as follows Figure 3As shown in the figure, the number of input features of the model is 16, 5×5 represents the spatial area size of a single input feature, and then it undergoes two convolutions to expand it to 32 and 16 dimensions, and then enters 2 fully connected layers with a dimension of 16, and finally outputs the severe convective wind speed.

[0091] 1.3.2 Loss Function

[0092] TS is an important evaluation index in the field of meteorology. It can effectively deal with the problem of severe sample imbalance, but it is not differentiable and cannot be directly used in the gradient descent method. Therefore, a differentiable TS (TSdiff) is proposed based on the non-differentiable TS, as shown in formula (1). It can make the model pay more attention to strong convective winds during training, thereby alleviating the problem of severe sample imbalance of wind speed intensity. However, its loss function and its multi-order derivatives are complex and the gradient is very unstable. MSE has better convergence and stability, and the recognition result tends to be biased towards low wind speed values. Therefore, MSE and TSdiff are weighted and summed to construct MTLoss. By appropriately adjusting the weight coefficient, the advantages of MSE and TSdiff can be brought into play and the defects of both can be minimized:

[0093]

[0094] MTLoss=MSELoss+α×TSLoss (3)

[0095]

[0096] In the formula, thldk is the wind speed threshold used by MTLoss; k represents the kth wind speed threshold; α is the weight parameter to be determined. By adjusting α, the recognition BIAS of GaleINet in the test set is ≤1, and TS reaches the maximum

[0097] 1.3.3 Immune Evolutionary Algorithm

[0098] Loss uses the non-differentiable TS as the loss function and cannot directly constrain the BIAS of the model. Therefore, when Loss is low, the BIAS of the model will be too high or too low. In order to solve this problem, BIAS constraints are imposed on the validation set during the training process of SCWINet, as shown in formula (4). Based on this, the specific implementation process of IEA training SCWINet is as follows.

[0099] F(x)=Loss (5)

[0100] max(BIAS k )≤1 (6)

[0101]

[0102] Where F(x) is the fitness function optimized by IEA; x is the parameter to be optimized, which corresponds to the α parameter in equation (4). When F(x) reaches the minimum, the TS of SCWINet is the highest, and its parameters are the optimal parameters. The principle of IEA is detailed in the literature (Qu et al., 2013, Wang et al., 2000), which will not be repeated in this paper.

[0103] 2Test results

[0104] 2.1 Comprehensive inspection

[0105] The TS and BIAS of the training set, validation set, and test set were calculated respectively, as shown in Table 1. As can be seen from Table 1, all three datasets achieved good training results.

[0106] Table 4 Comparative test of strong convective wind

[0107]

[0108] 2.2 Case Test

[0109] The severe convective wind events at 17:12 on July 22, 2023 and 19:36 on May 26, 2023 were selected to compare and test the applicability of SCWINet.

[0110] 3.2.1 Scattered severe convective weather at 17:12 on July 22, 2023

[0111] The severe convective gale of level 8-11 mainly occurred in the southwest of Quzhou and the east of Jiande, with the strongest wind speed reaching 26.2m / s. The maximum reflectivity factor of the live radar combination was 67.93dBz. At 17:06, SCWINet accurately identified that the region would experience severe convective gale of level 8-11 6 minutes in advance, and its location was basically consistent with the actual situation. There were some false alarms for Shaoxing and southern Jinhua, but the overall false alarms were not significant. However, the fuzzy logic had serious false alarms in the strong wind identification near 17:12. The TS, POD, FAR, and BIAS of SCWINet were 0.099, 1, 0.901, and 10.110, respectively, which were significantly better than the fuzzy logic's 0.003, 1.000, 0.991, and 42.440.

[0112]

[0113] 3.2.2 Severe dispersed convection at 19:36 on May 26, 2023

[0114] The 8-12 strong convective gale mainly occurred in the eastern part of Quzhou, with the strongest wind speed reaching 34.5m / s. The maximum reflectivity factor of the live radar combination was 64.23dBz. At 19:24, SCWINet accurately identified in advance that there would be 9-11 strong convective gale in the area, and its location was basically consistent with the actual situation. However, the fuzzy logic completely missed the strong wind identification at 19:36, and its location was obviously biased to the southwest. TS, POD, FAR, and BIAS of SCWINet were 0.130, 1.000, 0.870, and 6.250, respectively, which were significantly better than 0, 1.000, 0, and 7.500 of fuzzy logic.

[0115]

[0116] like Figure 4 As shown, an embodiment of the present invention provides a system for classifying and identifying strong convective and strong winds for implementing the method for classifying and identifying strong convective and strong winds, the system comprising:

[0117] The data collection and processing module is used to collect data from the three-dimensional radar network puzzle, lightning locator, and minute-level ground automatic observation station, and perform corresponding processing;

[0118] The dataset construction module is connected with the data collection and processing module to construct the dataset based on the “face point” recognition scheme;

[0119] Construct a recognition module, connect it with the dataset construction module, and use the SCWINet method to build the model structure, loss function, and immune evolutionary algorithm for recognition;

[0120] The inspection module is connected with the construction identification module to perform comprehensive inspection and individual case inspection.

[0121] An embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the severe convective gale classification identification method.

[0122] An embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the severe convective gale classification identification method.

[0123] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0124] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for classifying and identifying severe convective winds, characterized in that: The method includes: S1: Collect the data of 3D radar network mosaic, lightning locator, and minute-level ground automatic observation station, and process them accordingly; S2: Dataset construction based on face point recognition scheme; S3: Use SCWINet method to build model structure, loss function and immune evolutionary algorithm for recognition; S4: Conduct comprehensive and individual tests; The loss function is based on the non-differentiable TS and proposes a differentiable TS diff , as shown in formula (1), it can make the model pay more attention to strong convective winds during training, thereby alleviating the serious imbalance problem of wind speed intensity samples, but its loss function and its multi-order derivatives are complex, and the gradient is very unstable. MSE has better convergence and stability, and the recognition result tends to be biased towards low wind speed values. Therefore, MSE and TS are combined. diff The weighted sum is used to construct MTLoss, and the MSE and TS are brought into play by appropriately adjusting the weight coefficient. diff The advantages of both and minimize their disadvantages: MTLoss=MSELoss+α×TSLoss (3) Where thld k is the wind speed threshold used by MTLoss; k represents the kth wind speed threshold; α is a weight parameter to be determined. By adjusting α, the recognition BIAS of GaleINet in the test set is ≤1 and TS is maximized; In the immune evolutionary algorithm: Loss uses the non-differentiable TS as the loss function and cannot directly constrain the BIAS of the model. Therefore, when the Loss is low, the BIAS of the model will be too high or too low. To solve this problem, BIAS constraints are imposed on the validation set during the training of SCWINet, as shown in formula (4). Based on this, the specific implementation process of IEA training SCWINet is as follows: F(x)=Loss (5) max(BIAS k )≤1 (6) Where F(x) is the fitness function of IEA optimization; x is the parameter to be optimized, which corresponds to the α parameter in formula (4). When F(x) reaches the minimum, the TS of SCWINet is the highest, and its parameters are the optimal parameters.

2. The method for classifying and identifying severe convective winds according to claim 1, characterized in that: The corresponding processing in S1 specifically includes: First, the data quality is controlled. Based on the above-mentioned three-dimensional radar network data, the vertical and horizontal characteristics of the convective system are extracted and used as dependent variables for identifying strong convective winds. After quality control, the data set is divided into training, validation, and test sets.

3. The method for classifying and identifying severe convective winds according to claim 1, characterized in that: The S2 specifically includes: using a 5×5 grid area to identify whether a strong convective gale occurs at the center point.

4. The method for classifying and identifying severe convective winds according to claim 1, characterized in that: The number of input features of the model in the model structure is 16, and 5×5 represents the spatial area size of a single input feature. After that, it undergoes two convolutions to expand it to 32 and 16 dimensions, and then enters 2 fully connected layers with a dimension of 16, and finally outputs the severe convective wind speed.

5. The method for classifying and identifying severe convective winds according to claim 1, characterized in that: The comprehensive test calculates TS and BIAS for the training set, validation set, and test set respectively; The case test selected a severe convective wind event to compare and test the applicability of SCWINet.

6. A system for classifying and identifying strong convective winds that implements the method for classifying and identifying strong convective winds as claimed in any one of claims 1 to 5, characterized in that: The system includes: The data collection and processing module is used to collect data from the three-dimensional radar network puzzle, lightning locator, and minute-level ground automatic observation station, and perform corresponding processing; The dataset construction module is connected with the data collection and processing module to construct the dataset based on the face point recognition scheme; Construct a recognition module, connect it with the dataset construction module, and use the SCWINet method to build the model structure, loss function, and immune evolutionary algorithm for recognition; The inspection module is connected with the construction identification module to perform comprehensive inspection and individual case inspection.

7. A computer device, characterized in that: The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the severe convective gale classification identification method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the severe convective gale classification identification method according to any one of claims 1 to 5.

Images