Dynamic Monitoring and Identification Method of Geostationary Satellite Thunderstorm Cloud Clusters Based on Long Short-Term Memory Network

Abstract

The present invention discloses a method for dynamically monitoring and identifying thunderstorm cloud clusters of geostationary satellites based on a long short-term memory network. A dynamic monitoring and identification model of thunderstorm cloud clusters is constructed based on a long short-term memory network model, and training is completed using a dynamic monitoring machine learning data set. The dynamic monitoring machine learning data set includes several groups of dynamic monitoring machine learning arrays, and each group of dynamic monitoring machine learning arrays includes a set of historical mesoscale convective cloud cluster characteristic parameters that are spatio-temporally matched and lightning activity occurrence label data. The input of the trained dynamic monitoring and identification model of thunderstorm cloud clusters is the physical characteristic parameters corresponding to the previous m consecutive moments of all mesoscale convective cloud clusters identified in the target area at the current moment, and the output is the thunderstorm cloud cluster identification result at the current moment. Therefore, through the dynamic monitoring of the characteristic parameters of mesoscale convective cloud clusters, the present invention can achieve high-precision identification of thunderstorm cloud clusters in coastal and offshore areas.

Description

Technical Field

[0001] The present invention belongs to the technology of real-time monitoring and recognition of cloud cluster characteristics, and particularly relates to a method for dynamically monitoring and recognizing thunderstorm cloud clusters of geostationary satellites based on long short-term memory networks. Background Art

[0002] Mesoscale convective systems are often accompanied by disastrous weather such as strong winds, hailstorms, and convective precipitation, which often cause huge direct and indirect losses to transportation, power grids, agriculture, forestry, and fisheries in China. Improving the fine monitoring and prediction capabilities of mesoscale convective systems is of great significance for comprehensively enhancing China's natural disaster prevention level and reducing economic and property losses caused by extreme weather disasters.

[0003] The spatial resolution of geostationary meteorological satellites reaches 0.5 - 4 km, and the full-disk observation time resolution reaches 15 minutes, with the monitoring capabilities of all-weather and all-time monitoring. Combining with ground-based lightning observations can provide strong assistance for the real-time monitoring and dynamic prediction of mesoscale convective systems in China and its adjacent seas. However, in the past, the dynamic monitoring methods of mesoscale convective systems based on satellites mainly obtained the average brightness temperature information of the infrared channels of mesoscale convective systems by using traditional averaging methods. However, due to the unevenness of pixel information within mesoscale convective cloud clusters themselves, they are extremely vulnerable to the influence of extremely large or extremely small values, resulting in difficulty in reflecting the development degree and disaster-causing nature of mesoscale convective systems. If a correction factor reflecting the height related to the development within the cloud is introduced to highlight the role of cloud systems with higher development in mesoscale convective systems and weaken the role of cloud systems with lower development in mesoscale convective systems, it will help to further enhance the ability to monitor and predict the state of mesoscale convective cloud clusters in real time. In addition, in the dynamic monitoring of mesoscale convective cloud clusters, the macroscopic and microscopic parameters of cloud clusters such as the morphology, area, and development degree of convective cores during the occurrence and development process of mesoscale convective systems themselves have not been considered in the past. Physically speaking, these parameters have important indicative significance for the occurrence and development of mesoscale convective systems. Summary of the Invention

[0004] For the above reasons, the present invention proposes a method for dynamically monitoring and recognizing thunderstorm cloud clusters of geostationary satellites based on long short-term memory networks. A dynamic monitoring and recognition model of thunderstorm cloud clusters is constructed based on the long short-term memory network model, and the constructed dynamic monitoring and recognition model of thunderstorm cloud clusters is trained using a dynamic monitoring machine learning data set, so that the trained dynamic monitoring and recognition model of thunderstorm cloud clusters can dynamically monitor the evolution of characteristic parameters of mesoscale convective cloud clusters well, and further achieve more accurate dynamic monitoring of mesoscale convective systems in China's offshore and far-sea areas.

[0005] To achieve the above object, the present invention will provide the following technical solutions:

[0006] A method for dynamically monitoring and identifying thunderstorm cloud clusters of geostationary satellites based on long short-term memory networks, comprising the following steps:

[0007] Construct a dynamic monitoring and identification model of thunderstorm cloud clusters based on the long short-term memory network model, and complete the training using a dynamic monitoring machine learning data set;

[0008] The dynamic monitoring machine learning data set includes several groups of dynamic monitoring machine learning arrays, and each group of dynamic monitoring machine learning arrays includes a set of historical mesoscale convective cloud cluster characteristic parameters that are spatio-temporally matched and lightning activity occurrence label data;

[0009] The set of historical mesoscale convective cloud cluster characteristic parameters is formed by statistically calculating the physical characteristic parameters of the mesoscale convective cloud cluster at m consecutive moments before any historical forecast moment Tm+1;

[0010] The lightning activity occurrence label data is obtained from the lightning activity at the location of the corresponding mesoscale convective cloud cluster at the historical forecast moment Tm+1 observed by the lightning positioning system;

[0011] The input of the trained dynamic monitoring and identification model of thunderstorm cloud clusters is the physical characteristic parameters corresponding to the m consecutive moments before the current moment t of all the mesoscale convective cloud clusters identified in the target area, and the output is the thunderstorm cloud cluster identification result at the current moment t m+1 before the m consecutive moments, and the output is the thunderstorm cloud cluster identification result at the current moment t m+1 of the thunderstorm cloud cluster.

[0012] Preferably, the physical characteristic parameters of the mesoscale convective cloud cluster at any moment include cloud cluster macroscopic parameters, cloud cluster microscopic parameter A, and improved cloud cluster microscopic parameter B; where:

[0013] The cloud cluster macroscopic parameters include the central longitude and latitude of the mesoscale convective cloud cluster, the major axis length, the minor axis length, and the rotation angle of the circumscribed ellipse;

[0014] The cloud cluster microscopic parameter A and the improved cloud cluster microscopic parameter B are calculated based on the brightness temperature data detected by the infrared imager carried on the geostationary satellite in the target area; where:

[0015] For any mesoscale convective cloud cluster j, the elements included in the cloud cluster microscopic parameter A correspond to: the mean value of the brightness temperature in the 10.4-micron channel , the lowest value and the highest value , the brightness temperature difference between the 6.9-micron and 7.3-micron channels the mean value of TB1 , the lowest value Mi and the highest value Max , the brightness temperature difference between the 7.3-micron and 10.4-micron channels the mean value of TB2 the lowest value Mi and the maximum value Max , the brightness temperature difference between the 12.4-μm and 10.4-μm channels The mean value of TB3 , the minimum value Mi and the maximum value Max ;

[0016] Any element in the improved cloud microphysical parameter B is obtained by optimizing the mean element in the cloud microphysical parameter A using the infrared channel brightness temperature correction factor;

[0017] For any mesoscale convective cloud cluster j, the elements included in the cloud microphysical parameter B correspond to: the mean value of the improved 10.4-μm channel brightness temperature ; the mean value of the improved brightness temperature difference between the 6.9-μm and 7.3-μm channels ; the mean value of the improved brightness temperature difference between the 7.3-μm and 10.4-μm channels ; the mean value of the improved brightness temperature difference between the 12.4-μm and 10.4-μm channels .

[0018] Preferably, the infrared channel brightness temperature correction factor is calculated by the following formula:

[0019] ;

[0020] In the formula: represents the 10.4-μm channel brightness temperature at pixel i, represents the infrared channel brightness temperature correction factor at pixel i;

[0021] In the improved cloud microphysical parameter B, any element is calculated by the following formula:

[0022] ;

[0023] In the formula: represents the th element in the cloud microphysical parameter B, ; represents the mean value of the 10.4-μm channel brightness temperature ; represents the mean value of the brightness temperature difference between the 6.9-μm and 7.3-μm channels ; represents the mean value of the brightness temperature difference between the 7.3-μm and 10.4-μm channels ; represents the mean value of the brightness temperature difference between the 12.4-μm and 10.4-μm channels ; N represents the total number of pixels within any mesoscale convective cloud cluster j.

[0024] Preferably, for any mesoscale convective cloud cluster j, based on the contour points of the mesoscale convective cloud cluster j and using the least squares method to minimize the distance between the circumscribed ellipse of the mesoscale convective cloud cluster j and the contour points, the major axis length, minor axis length, and rotation angle of the circumscribed ellipse of any mesoscale convective cloud cluster j can be obtained;

[0025] For any mesoscale convective cloud cluster j, based on the brightness temperature of the 10.4 - micron channel of all pixel points within the mesoscale convective cloud cluster j, the mean value of the brightness temperature of the 10.4 - micron channel within the mesoscale convective cloud cluster can be calculated , the minimum value and the maximum value ;

[0026] For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference TB1 between the 6.9 - micron and 7.3 - micron channels of all pixels within the mesoscale convective cloud cluster j, and then statistically obtain the mean value of the brightness temperature difference TB1 between the 6.9 - micron and 7.3 - micron channels within the mesoscale convective cloud cluster j , the minimum and the maximum value ; ;

[0027] For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference TB2 between the 7.3 - micron and 10.4 - micron channels of all pixels within the mesoscale convective cloud cluster j, and then statistically obtain the mean value of the brightness temperature difference TB2 between the 7.3 - micron and 10.4 - micron channels within the mesoscale convective cloud cluster j , the minimum and the maximum value ; ;

[0028] For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference TB3 between the 12.4 - micron and 10.4 - micron channels of all pixels within the mesoscale convective cloud cluster j, and then statistically obtain the mean value of the brightness temperature difference TB3 between the 12.4 - micron and 10.4 - micron channels within the mesoscale convective cloud cluster j TB3 , the minimum and the maximum value .

[0029] Preferably, the dynamic monitoring machine - learning data set is constructed through the following steps:

[0030] Step 1, Mesoscale convective cloud cluster identification:

[0031] Obtain the brightness temperature data of the target area detected by the infrared imager carried on the geostationary satellite during the historical period, and identify the continuous low-infrared brightness temperature area at the initial moment of the detected historical period as a mesoscale convective cloud cluster;

[0032] Step 2. Statistically analyze the characteristic parameters of the mesoscale convective cloud clusters:

[0033] According to the physical connotations of the mesoscale convective cloud clusters identified in Step 1, statistically analyze the physical characteristic parameters of the corresponding mesoscale convective cloud clusters;

[0034] Step 3. Track the trajectories of the mesoscale convective cloud clusters:

[0035] Use the mesoscale convective system tracking algorithm to determine the movement trajectories of the mesoscale convective cloud clusters identified in Step 1, so as to dynamically monitor the mesoscale convective cloud clusters identified in Step 1;

[0036] Step 4. Construct a machine learning dataset for dynamic monitoring:

[0037] Combining the physical characteristic parameters of any mesoscale convective cloud cluster statistically analyzed in Step 2 and the movement trajectories of the corresponding mesoscale convective cloud clusters tracked in Step 3, and using the lightning activity information observed by the lightning location system, a machine learning dataset for dynamic monitoring can be constructed.

[0038] Preferably, the mesoscale convective cloud cluster is identified based on the brightness temperature data of the 10.4-μm channel detected by the infrared imager in the target area, and specifically includes the following steps:

[0039] Step 1.1. In the target area, obtain the brightness temperature of the corresponding 10.4-μm channel observed by the infrared imager at any pixel i position ;

[0040] Step 1.2. By comparing the brightness temperature of the 10.4-μm channel with the preset brightness temperature threshold Tb0, determine whether this corresponding pixel i belongs to the mesoscale convective cloud cluster pixel: when the judgment result shows that ≥Tb0, the pixel i is identified as a mesoscale convective cloud cluster pixel, otherwise it is a non-mesoscale convective cloud cluster pixel; i is a positive integer;

[0041] Step 1.3. Traverse the surrounding pixels of the mesoscale convective cloud cluster pixel i. If there are mesoscale convective cloud cluster pixels around the mesoscale convective cloud cluster pixel i, continue to traverse until all the adjacent pixels around the mesoscale convective cloud cluster pixel i are non-mesoscale convective cloud cluster pixels, then the identification of a complete mesoscale convective cloud cluster ends, and the identified mesoscale convective cloud cluster is numbered, denoted as mesoscale convective cloud cluster j, where j is a positive integer.

[0042] Preferably, the mesoscale convective cloud cluster trajectory tracking in step 3 specifically includes:

[0043] Step 3.1: According to the central position information of the mesoscale convective cloud cluster MCC at time Tk, Tk preset a search radius R, and search for the mesoscale convective cloud cluster within the search radius R at time Tk+1. The number of mesoscale convective cloud clusters whose central positions fall within the search radius R is denoted as N, and each mesoscale convective cloud cluster at time Tk+1 obtained by the search is denoted as MCC 1 Tk+1 , MCC 2 Tk+1 , ……, MCC N Tk+1 , where N is a positive integer;

[0044] Step 3.2: If N = 1, then record the central position of the only mesoscale convective cloud cluster MCC 1 Tk+1 as the position of the mesoscale convective cloud cluster MCC Tk at time Tk+1; if N ≥ 2, then among the mesoscale convective cloud clusters obtained by the search at time Tk+1, record the central position of the mesoscale convective cloud cluster closest to the central position of the mesoscale convective cloud cluster MCC at time Tk Tk as the predicted position of the mesoscale convective cloud cluster MCC at time Tk Tk at time Tk+1;

[0045] Step 3.3: For subsequent times Tk+2, Tk+3, ……, Tk+n, perform the corresponding mesoscale convective cloud cluster trajectory search according to steps 3.1 and 3.2 until no mesoscale convective cloud cluster is found within the search radius R, then the trajectory search of this mesoscale convective cloud cluster MCC Tk ends.

[0046] Preferably, the construction of the dynamic monitoring machine learning data set in step 4 specifically includes:

[0047] Step 4.1: Using all the mesoscale convective cloud cluster trajectories tracked in step 3, for any mesoscale convective cloud cluster trajectory case, according to the lightning occurrence position and lightning occurrence time observed by the lightning location system, count the lightning activities of the mesoscale convective cloud cluster at different times. Lightning occurrence is marked as "1", and no lightning occurrence is marked as "0";

[0048] Step 4.2: Using the physical characteristic parameters of the mesoscale convective cloud clusters obtained in Step 2, for any mesoscale convective cloud cluster trajectory case that lasts for more than m + 1 time steps, take the physical characteristic parameters of the mesoscale convective cloud clusters at consecutive times T1 to Tm and the lightning activity information at time step Tm+1 as a set of machine learning data;

[0049] Step 4.3: Take the physical characteristic parameters of the mesoscale convective cloud clusters at consecutive times T2 to Tm+1 and the lightning activity information at time step Tm+2 as a set of machine learning data, and so on until the mesoscale convective cloud cluster trajectory terminates, finally constructing a dynamic monitoring machine learning data set.

[0050] Preferably, the training of the thunderstorm cloud cluster dynamic monitoring and recognition model specifically includes the following steps:

[0051] Step 5.1: Divide the dynamic monitoring machine learning data set constructed in Step 4 into three parts: a training set, a test set, and a validation set according to 6:2:2;

[0052] Step 5.2: Using the training set and the test set, adopt a long short-term memory network model, design different inputs of mesoscale convective cloud cluster characteristic parameters, and use the binary classification BCEwithLogitsLoss loss function to test the optimal combination of input characteristic parameters of the thunderstorm cloud cluster dynamic monitoring and recognition model;

[0053] Step 5.3: Using the optimal combination of input characteristic parameters obtained in Step 5.2, adopt a long short-term memory network model with the training set and the test set, design multiple groups of network hyperparameters, and use the binary classification BCEwithLogitsLoss loss function to optimize the network hyperparameters of the thunderstorm cloud cluster dynamic monitoring and recognition model; for the validation set, use detection rate and false alarm rate indicators to evaluate the accuracy of the output of the thunderstorm cloud cluster dynamic monitoring and recognition model.

[0054] Preferably, the formula of the BCEwithLogitsLoss loss function in Step 5.2 and Step 5.3 is as follows:

[0055]

[0056]

[0057] Among them, is the input historical mesoscale convective cloud cluster characteristic parameter group, is the sample label, is the predicted value; represents the Sigmoid activation function;

[0058] In Step 5.2 and 5.3, the calculation formula for the output result of the thunderstorm cloud cluster dynamic monitoring and recognition model is as follows:

[0059]

[0060] Among them, out is the model output, and net represents the dynamic monitoring and recognition model of thunderstorm cloud clusters. It is the Sigmoid activation function.

[0061] According to the above technical solution, compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. The method for dynamically monitoring and recognizing thunderstorm cloud clusters of geostationary satellites based on long short-term memory networks provided by the present invention, on the basis of constructing a dynamic monitoring and recognition model of thunderstorm cloud clusters based on long short-term memory networks, uses the characteristic parameters of mesoscale convective cloud clusters as the model input. Compared with directly using the multi-channel observations of geostationary satellites only as the model input in the past, it has more physical interpretability.

[0063] 2. By testing the input of multiple groups of characteristic parameters of mesoscale convective cloud clusters, the present invention proves that introducing the macroscopic parameter A and the improved microscopic parameter B of the cloud cluster as the model input at the same time can achieve a higher-precision recognition result of thunderstorm cloud clusters, which has a beneficial effect on the dynamic monitoring of thunderstorms in the offshore and open ocean. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic flow chart of the method for dynamically monitoring and recognizing thunderstorm cloud clusters of geostationary satellites based on long short-term memory networks provided by an embodiment of the present invention.

[0065] Figure 2 It is a schematic trajectory diagram of a mesoscale convective cloud cluster (white area) recognized by an example of the present invention. The coloring is the brightness temperature of the 10.4-micron channel, the black star represents the center of the mesoscale convective cloud cluster in the example, and the white numbers represent the numbers of the mesoscale convective cloud clusters. In the figure: (a) represents the position of the mesoscale convective cloud cluster at 10:50 (International Time) on July 3, 2018, (b) represents the position of the mesoscale convective cloud cluster at 11:10 on July 3, 2018, (c) represents the position of the mesoscale convective cloud cluster at 11:30 on July 3, 2018, (d) represents the position of the mesoscale convective cloud cluster at 11:50 on July 3, 2018, (e) represents the position of the mesoscale convective cloud cluster at 12:10 on July 3, 2018, (f) represents the position of the mesoscale convective cloud cluster at 12:30 on July 3, 2018, (g) represents the position of the mesoscale convective cloud cluster at 12:50 on July 3, 2018.

[0066] Figure 3Cost function convergence curves of the training set of the thunderstorm cloud cluster dynamic monitoring and recognition model of the long short-term memory network (a), detection rate (b), and false alarm rate (c) of the validation set under the input of the characteristic parameter group of five groups of historical mesoscale convective cloud clusters in the examples of the present invention.

[0067] Figure 4 Cost function convergence curves of the training set and test set of the thunderstorm cloud cluster dynamic monitoring and recognition model of the long short-term memory network in the examples of the present invention, as well as the detection rate and false alarm rate of the validation set. Detailed implementation manners

[0068] The following further describes the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. 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.

[0069] The present invention utilizes the Himawari-8 geostationary satellite infrared imaging instrument to construct a characteristic parameter group of historical mesoscale convective cloud clusters. On this basis, a thunderstorm cloud cluster dynamic monitoring and recognition model is constructed based on the long short-term memory network, combined with the lightning activity occurrence label data of spatio-temporal matching, to dynamically monitor the characteristics of historical and future mesoscale convective cloud clusters of the geostationary satellite, achieve the purpose of identifying thunderstorm cloud clusters, and thus make a relatively accurate prediction of the occurrence of thunderstorm events. The following will further describe the embodiments of the present invention in detail in conjunction with the accompanying drawings.

[0070] The present invention discloses a method for dynamically monitoring and recognizing geostationary satellite thunderstorm cloud clusters based on the long short-term memory network, as Figure 1 shown, which constructs a thunderstorm cloud cluster dynamic monitoring and recognition model based on the long short-term memory network model and completes the training using a dynamic monitoring machine learning data set;

[0071] The dynamic monitoring machine learning data set includes several groups of dynamic monitoring machine learning arrays, and each group of dynamic monitoring machine learning arrays includes a spatio-temporally matched historical mesoscale convective cloud cluster characteristic parameter group and lightning activity occurrence label data.

[0072] The historical mesoscale convective cloud cluster characteristic parameter group is constituted by statistically analyzing the physical characteristic parameters of the mesoscale convective cloud cluster at m consecutive moments before any historical forecast moment Tm+1; m is a constant, and its value can be 2. The lightning activity occurrence label data is obtained through the lightning activity at the location of the corresponding mesoscale convective cloud cluster at the historical forecast moment Tm+1 observed by the lightning location system. It can be seen that the mesoscale convective cloud cluster used to construct the dynamic monitoring machine learning data set lasts at least m+1 time steps before it disappears.

[0073] The input of the trained thunderstorm cloud cluster dynamic monitoring and recognition model is all the mesoscale convective cloud clusters identified in the target area at the current moment t m+1The physical characteristic parameters corresponding to the first m consecutive moments are output as the recognition result of the thunderstorm cloud cluster at the current moment t. m+1 The recognition result of the thunderstorm cloud cluster.

[0074] As Figure 1 shown, the specific steps of the dynamic monitoring machine learning data set of the present invention are as follows:

[0075] Step 1. Mesoscale convective cloud cluster recognition:

[0076] Identify the continuous low infrared brightness temperature area within the detection range of the geostationary satellite infrared imager during the specified historical period as the mesoscale convective cloud cluster. In this step, the recognition of the low infrared brightness temperature area is based on the brightness temperature detected by the 10.4-micron channel in the geostationary satellite infrared imager of the sunflower. The specific method is as follows:

[0077] Step 1.1. Within the detection range of the geostationary satellite infrared imager of the sunflower, the infrared imager observes the corresponding brightness temperature of the 10.4-micron channel at any pixel i position during the specified historical period ;

[0078] Step 1.2. By comparing the brightness temperature of the 10.4-micron channel with the preset brightness temperature threshold Tb0 (in this embodiment, the preset brightness temperature threshold can be set to Tb0 = 241K), to determine whether this corresponding pixel i belongs to the mesoscale convective cloud cluster pixel: when ≥Tb0, the pixel i is recognized as a mesoscale convective cloud cluster pixel, otherwise it is a non-mesoscale convective cloud cluster pixel; i is a positive integer;

[0079] Step 1.3. Traverse the surrounding pixels of the mesoscale convective cloud cluster pixel i. If there are mesoscale convective cloud cluster pixels around the mesoscale convective cloud cluster pixel i, continue to traverse until all the adjacent pixels around the mesoscale convective cloud cluster pixel i are non-mesoscale convective cloud cluster pixels, then the recognition of a complete mesoscale convective cloud cluster ends, and the recognized mesoscale convective cloud cluster is numbered, denoted as the mesoscale convective cloud cluster j, where j is a positive integer. It can be seen that the obtained mesoscale convective cloud cluster j is composed of continuous mesoscale convective cloud cluster pixels i.

[0080] Step 2. Statistically analyze the physical characteristic parameters of the mesoscale convective cloud cluster:

[0081] According to the physical connotations of the mesoscale convective cloud clusters identified in Step 1, statistically analyze the physical characteristic parameters of the corresponding mesoscale convective cloud clusters; and introduce an infrared channel brightness temperature correction factor to improve the physical characteristic parameters of the mesoscale convective cloud clusters;

[0082] In the present invention, the physical characteristic parameters of mesoscale convective cloud clusters include cloud cluster microscopic parameters, improved cloud cluster microscopic parameters, and cloud cluster macroscopic parameters, where: the cloud cluster macroscopic parameters include the central longitude and latitude of the mesoscale convective cloud cluster, the major axis length, minor axis length, and rotation angle of the circumscribed ellipse;

[0083] For any mesoscale convective cloud cluster j, the elements included in the cloud cluster microscopic parameter A correspond to: the mean value of the brightness temperature in the 10.4-μm channel , the minimum value and the maximum value , the brightness temperature difference between the 6.9-μm and 7.3-μm channels the mean value of TB1 , the minimum value Mi and the maximum value Max , the brightness temperature difference between the 7.3-μm and 10.4-μm channels the mean value of TB2 , the minimum value Mi and the maximum value Max , the brightness temperature difference between the 12.4-μm and 10.4-μm channels the mean value of TB3 , the minimum value Mi and the maximum value Max ;

[0084] For any mesoscale convective cloud cluster j, the elements included in the cloud cluster microscopic parameter B correspond to: the mean value of the improved brightness temperature in the 10.4-μm channel ; the mean value of the improved brightness temperature difference between the 6.9-μm and 7.3-μm channels ; the mean value of the improved brightness temperature difference between the 7.3-μm and 10.4-μm channels ; the mean value of the improved brightness temperature difference between the 12.4-μm and 10.4-μm channels .

[0085] In a specific implementation, for any mesoscale convective cloud cluster j, by calculating the average values of the longitude values and latitude values of all the pixel points within the mesoscale convective cloud cluster j, the central longitude and latitude information of the mesoscale convective cloud cluster can be obtained;

[0086] For any mesoscale convective cloud cluster j, based on the contour points of the mesoscale convective cloud cluster j and using the least squares method to minimize the distance between the circumscribed ellipse of the mesoscale convective cloud cluster j and the contour points, the major axis length, minor axis length, and rotation angle of the circumscribed ellipse of any mesoscale convective cloud cluster j can be obtained;

[0087] For any mesoscale convective cloud cluster j, based on the brightness temperature in the 10.4-μm channel of all the pixel points within the mesoscale convective cloud cluster j, the mean value of the brightness temperature in the 10.4-μm channel within the mesoscale convective cloud cluster can be calculated and the minimum value and the maximum value , which is specifically calculated by the following formula:

[0088] ;

[0089] ;

[0090] ;

[0091] wherein, represents the brightness temperature of the 10.4 - micron channel at pixel position i, represents all the pixel numbers within any mesoscale convective cloud cluster j.

[0092] For any mesoscale convective cloud cluster j, the mean value of the improved 10.4 - micron channel brightness temperature is calculated according to the following formula:

[0093] ;

[0094] wherein, represents the brightness temperature of the 10.4 - micron channel at pixel position i, represents all the pixel numbers within any mesoscale convective cloud cluster j, represents the correction coefficient at pixel position i, which is calculated by the following formula:

[0095] .

[0096] In addition, in practical applications, if the brightness temperature is relatively high, for example when, it is not appropriate to use the above formula to calculate the correction coefficient , and at this time, the value of the correction coefficient is directly 0.

[0097] Based on the above - mentioned calculation formula of the correction coefficient , if the brightness temperature = 210K, = 1, if the brightness temperature is lower than 210K, can be greater than 1.

[0098] For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference TB1 between the 6.9 - micron and 7.3 - micron channels of all pixels within the mesoscale convective cloud cluster j, and then statistically obtain the mean value of the brightness temperature difference TB1 between the 6.9 - micron and 7.3 - micron channels within the mesoscale convective cloud cluster j, the minimum and the maximum value ; Specifically, it is calculated by the following formula:

[0099] ;

[0100] Mi ;

[0101] ;

[0102] In the formula: represents the brightness temperature difference between the 6.9 - micron and 7.3 - micron channels at pixel position i, represents all the pixel numbers within any mesoscale convective cloud cluster j.

[0103] For any mesoscale convective cloud cluster j, the mean value of the improved brightness temperature difference between the 6.9 - micron and 7.3 - micron channels , is calculated according to the following formula:

[0104] = ;

[0105] ;

[0106] Among them, represents the brightness temperature difference between the 6.9 - micron and 7.3 - micron channels at pixel position i, represents the correction coefficient at position i, represents the brightness temperature of the 10.4 - micron channel at pixel position i, represents all the pixel numbers within any mesoscale convective cloud cluster j.

[0107] For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference between the 7.3 - micron and 10.4 - micron channels of all pixels within the mesoscale convective cloud cluster j TB2, and then statistically obtain the mean value of the brightness temperature difference TB2, the minimum and the maximum value ; Specifically, it is calculated by the following formula:

[0108] ;

[0109] ;

[0110] ;

[0111] For any mesoscale convective cloud cluster j, the mean value of the improved brightness temperature difference between the 7.3 - micron and 10.4 - micron channels , is calculated according to the following formula:

[0112] = ;

[0113] ;

[0114] Among them, represents the brightness temperature difference between the 7.3-μm and 10.4-μm channels at pixel position i, represents the correction coefficient at pixel position i, represents the brightness temperature of the 10.4-μm channel at pixel position i, represents the total number of all pixels within any mesoscale convective cloud cluster j.

[0115] For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference between the 12.4-μm and 10.4-μm channels for all pixels within this mesoscale convective cloud cluster j TB3, and then statistically obtain the mean value of the brightness temperature difference TB3 between the 12.4-μm and 10.4-μm channels within this mesoscale convective cloud cluster j , the minimum and the maximum value ; specifically calculated by the following formula:

[0116] ;

[0117] ;

[0118] ;

[0119] For any mesoscale convective cloud cluster j, the mean value of the improved brightness temperature difference between the 12.4-μm and 10.4-μm channels is calculated according to the following formula:

[0120] = ;

[0121] ;

[0122] Among them, represents the brightness temperature difference between the 12.4-μm and 10.4-μm channels at pixel position i, represents the correction coefficient at pixel position i, represents the brightness temperature of the 10.4-μm channel at pixel position i, represents the total number of all pixels within any mesoscale convective cloud cluster j.

[0123] Step 3. Mesoscale convective cloud cluster trajectory tracking:

[0124] Using the mesoscale convective system tracking algorithm, determine the movement trajectory of the mesoscale convective cloud cluster identified in Step 1. The specific method is as follows:

[0125] Step 3.1: According to the central position information of the mesoscale convective cloud cluster MCC at time Tk Ti preset a search radius R, and search for the mesoscale convective cloud cluster within the range of this search radius R at time Tk+1. The number of mesoscale convective cloud clusters whose central positions fall within the search radius R is denoted as N. Each mesoscale convective cloud cluster obtained at time Tk+1 is denoted as MCC 1 Tk+1 , MCC 2 Tk+1 , ……, MCC N Tk+1 , where N is a positive integer;

[0126] Step 3.2: If N = 1, record the central position of the only mesoscale convective cloud cluster MCC 1 Tk+1 as the position of the mesoscale convective cloud cluster MCC Tk at time Tk+1; if N ≥ 2, among the mesoscale convective cloud clusters obtained by searching at time Tk+1, record the central position of the mesoscale convective cloud cluster that is closest to the central position of the mesoscale convective cloud cluster MCC at time Tk Tk as the predicted position of the mesoscale convective cloud cluster MCC at time Tk Tk at time Tk+1;

[0127] Step 3.3: For subsequent times Tk+2, Tk+3, ……, Tk+n, perform the corresponding mesoscale convective cloud cluster trajectory search according to Steps 3.1 and 3.2 until no mesoscale convective cloud cluster is found within the search radius R, then the trajectory search of this mesoscale convective cloud cluster MCC Tk ends.

[0128] Figure 2 is a schematic diagram of the application example of the mesoscale convective cloud cluster tracking method provided by the example of the present invention. (White area) Trajectory schematic diagram, the filled color is the brightness temperature of the 10.4-micron channel, the black star represents the center of the example mesoscale convective cloud cluster, and the white numbers represent the numbers of the mesoscale convective cloud clusters. Figure 2 It can well reflect the birth, development, and extinction processes of the mesoscale convective cloud cluster.

[0129] Step Four: Construction of the dynamic monitoring machine learning data set:

[0130] Combined with the physical characteristic parameters of any mesoscale convective cloud cluster counted in Step 2 and the movement trajectory of the corresponding mesoscale convective cloud cluster tracked in Step 3, using the lightning activity information observed by the lightning location system, a dynamic monitoring machine learning dataset is constructed. Specifically, using all the mesoscale cloud cluster trajectories tracked in Step 3, count the number of lightning occurrences within the mesoscale convective cloud cluster at any moment, and label thunderstorm and non-thunderstorm events; then combined with the physical characteristic parameters of the mesoscale convective cloud cluster counted in Step 2, take the physical characteristic parameters of the mesoscale convective cloud cluster at consecutive moments from T1 to Tm, and the lightning activity information of the mesoscale convective system at the (m + 1)-th moment as a set of input and output data; and so on, to construct a dynamic monitoring machine learning dataset. The specific implementation method includes the following steps:

[0131] Step 4.1. Using all the mesoscale convective cloud cluster trajectories tracked in Step 3, for any mesoscale convective cloud cluster trajectory case (MCC Tk ,MCC Tk+1 ,MCC Tk+2 ,……MCC Tk+n ), according to the lightning occurrence locations observed by the lightning location system, count the number of lightning strikes within the pixel of this mesoscale convective cloud cluster and within 5 minutes before and after, that is, the number of lightning strikes of the mesoscale convective cloud cluster at the corresponding moment (L Tk ,L Tk+1 ,……,L Tk+n ). If the number of lightning strikes is greater than 0, label it as "1", which is a thunderstorm event; if the number of lightning strikes is equal to 0, label it as "0", which is a non-thunderstorm event;

[0132] Step 4.2. Using the physical characteristic parameters of the mesoscale convective cloud cluster counted in Step 2, for any mesoscale convective system case (MCC Tk ,MCC Tk+1 ,MCC Tk+2 ,……MCC Tk+n ) that lasts for more than m + 1 (in the present invention, the value of m can be set to m = 2) time steps, take the physical characteristic parameters of the mesoscale convective cloud cluster at consecutive moments from T1 to Tm and the lightning activity information at the (m + 1)-th time step as a set of machine learning data;

[0133] Step 4.3. Take the physical characteristic parameters of the mesoscale convective cloud cluster at consecutive time steps from T2 to Tm + 1 and the lightning activity information at the (m + 2)-th moment as a set of machine learning data, and so on, until the mesoscale convective cloud cluster trajectory terminates, and finally construct a dynamic monitoring machine learning dataset.

[0134] It can be seen that the dynamically monitored machine learning dataset of the mesoscale convective cloud cluster characteristics parameters constructed includes the characteristics parameter combination of the mesoscale convective cloud cluster and the lightning activity information matched with the characteristics parameter combination of the mesoscale convective cloud cluster.

[0135] The training of the thunderstorm cloud cluster dynamic monitoring and recognition model specifically includes the following steps:

[0136] Using the dynamically monitored machine learning dataset of the mesoscale convective cloud cluster characteristics parameters constructed in step four, train the thunderstorm cloud cluster dynamic monitoring and recognition model constructed based on the long short-term memory network. In this embodiment, the dynamically monitored machine learning dataset is divided into a training set, a test set, and a validation set according to a certain ratio; further, according to the training set and the test set, design different mesoscale convective cloud cluster characteristics parameter combinations and multiple groups of hyperparameter combinations, and train the training dataset and the test dataset based on the thunderstorm cloud cluster dynamic monitoring and recognition model. Through multiple iterations, construct the optimal model for thunderstorm cloud cluster recognition; finally, input the validation set to verify the accuracy of thunderstorm cloud cluster recognition. The specific implementation method includes the following steps:

[0137] Step 5.1, Dataset division:

[0138] Divide the dynamically monitored machine learning dataset constructed in step four into three parts: a training set, a test set, and a validation set according to 6:2:2;

[0139] Step 5.2, Design of model input characteristic parameters:

[0140] Design a set of long short-term memory network model hyperparameters: the learning rate is 0.001, and the time series length is 3; in the network hyperparameters, the hidden layer is set to 50 layers, the batch size is 100, the number of network layers is set to 3 layers, and the number of iterations is 300 times.

[0141] Design the characteristics parameter combination of the mesoscale convective cloud cluster as the model input:

[0142] The first category is the cloud cluster microparameter combination (Input1): the average, minimum, and maximum brightness temperatures of the 10.4-micron channel, the average, minimum, and maximum values of the brightness temperature difference between the 6.9-micron and 7.3-micron channels, the average, minimum, and maximum values of the brightness temperature difference between the 7.3- and 10.4-micron channels, and the average, minimum, and maximum values of the brightness temperature difference between the 12.4- and 10.4-micron channels.

[0143] The second category is the improved combination of cloud microparameters (Input2): the average value of the improved 10.4-μm channel, the minimum and maximum brightness temperatures of the 10.4-μm channel, the average value of the improved brightness temperature difference between the 6.9-μm and 7.3-μm channels, the minimum and maximum values of the brightness temperature difference between the 6.9-μm and 7.3-μm channels, the average value of the improved brightness temperature difference between the 7.3-μm and 10.4-μm channels, the minimum and maximum values of the brightness temperature difference between the 7.3-μm and 10.4-μm channels, the average value of the improved brightness temperature difference between the 12.4-μm and 10.4-μm channels, and the minimum and maximum values of the brightness temperature difference between the 12.4-μm and 10.4-μm channels.

[0144] The third category is the combination of cloud macroparameters (Input3): the major axis length, minor axis length, and rotation angle of the mesoscale convective cloud cluster.

[0145] The fourth category is the combination of cloud microparameters and cloud macroparameters (Input4): the average, minimum, and maximum brightness temperatures of the 10.4-μm channel, the average, minimum, and maximum values of the brightness temperature difference between the 6.9-μm and 7.3-μm channels, the average, minimum, and maximum values of the brightness temperature difference between the 7.3-μm and 10.4-μm channels, the average, minimum, and maximum values of the brightness temperature difference between the 12.4-μm and 10.4-μm channels; the major axis length, minor axis length, and rotation angle of the mesoscale convective cloud cluster.

[0146] The fifth category is the improved combination of cloud microparameters and cloud macroparameters (Input5): the average value of the improved 10.4-μm channel, the minimum and maximum brightness temperatures of the 10.4-μm channel, the average value of the improved brightness temperature difference between the 6.9-μm and 7.3-μm channels, the minimum and maximum values of the brightness temperature difference between the 6.9-μm and 7.3-μm channels, the average value of the improved brightness temperature difference between the 7.3-μm and 10.4-μm channels, the minimum and maximum values of the brightness temperature difference between the 7.3-μm and 10.4-μm channels, the average value of the improved brightness temperature difference between the 12.4-μm and 10.4-μm channels, and the minimum and maximum values of the brightness temperature difference between the 12.4-μm and 10.4-μm channels; the major axis length, minor axis length, and rotation angle of the mesoscale convective cloud cluster.

[0147] Taking Inupt1, Input2, Input3, Input4, and Input5 in the training set and test set in 5.1 as the model inputs respectively, using the long short-term memory network model, minimizing the error between the predicted and observed values with the binary classification BCEwithLogitsLoss loss function, and through 300 iterations, taking the model with the minimum value of the test set loss function as the optimal model, denoted as Model1, Model2, Model3, Model4, and Model5 respectively.

[0148] The formula of the BCEwithLogitsLoss loss function is as follows:

[0149]

[0150]

[0151] Among them, is the input historical mesoscale convective cloud cluster characteristic parameter group, is the true value (i.e., the above-mentioned lightning activity information, which is the sample label), is the predicted value. represents the Sigmoid activation function. Finally, use Inupt1, Input2, Input3, Input4, and Input5 in the validation set as the model inputs, and use Model1, Model2, Model3, Model4, and Model5 to calculate the detection rate and false alarm rate indicators respectively.

[0152] The calculation formula for the output result of the thunderstorm cloud cluster dynamic monitoring and recognition model is as follows:

[0153]

[0154] Among them, out is the model output, and net represents the optimal thunderstorm cloud cluster dynamic monitoring and recognition model, is the Sigmoid activation function.

[0155] Figure 3 This is the cost function convergence curve of the training set of the thunderstorm cloud cluster dynamic monitoring and recognition model under the input of two groups of historical mesoscale convective cloud cluster characteristic parameter groups in the examples of the present invention [refer to Figure 3 in (a)], and the detection rate of the validation set [refer to Figure 3 in (b)] and the false alarm rate [refer to Figure 3 in (c)]. It can be seen that: in the case of adopting the improved cloud microparameters (Input2), compared with before the improvement (Input1), the loss value of the model cost function decreases, the detection rate increases by about 3 percentage points, and the false alarm rate is about 20.6%; in addition, when introducing the combination of the improved cloud microparameters and macroparameters (Input5) at the same time, compared with the case of using the unimproved cloud microparameters and macroparameters (Input4), the detection rate increases by about 7 percentage points, and the false alarm rate is about 17%. Therefore, considering the above quantitative evaluation indicators comprehensively, using the combination of the improved cloud microparameters and macroparameters as the input of the thunderstorm cloud cluster dynamic monitoring and recognition model can improve the accuracy of thunderstorm cloud cluster recognition to a certain extent.

[0156] Step 5.3 Model hyperparameter design and optimal model accuracy verification:

[0157] Use the training set and the test set, and use the improved mesoscale convective cloud microphysical characteristic parameters and macroparameter combination (Input5) obtained in Step 5.2 as the model input.

[0158] Design the hyperparameters of the long short-term memory network model: the learning rate is 0.001, and the time series length is 3; design multiple sets of hyperparameters, the hidden layer: 40, 50, 60, 80, 100 layers, batch size: 100, 300, 500, 700, 800, the number of network layers: 2, 3, 4, 6, 8, 10 layers. Use the binary classification BCEwithLogitsLoss loss function to minimize the error between the prediction and the observed value, and the number of iterations is 600 times; through 600 iterations, obtain the LOSS distribution curves under different hyperparameter combinations, and take the hyperparameter combination with the smallest value of the test set loss function as the optimal hyperparameter combination.

[0159] The formula of the BCEwithLogitsLoss loss function is as follows:

[0160]

[0161]

[0162] Among them, is the characteristic parameter group of the mesoscale convective cloud cluster in the input history, is the true value, is the predicted value. represents the Sigmoid activation function.

[0163] Using the optimal long short-term memory network model, for the validation set, quantitative evaluation is carried out using the detection rate and false alarm rate indicators.

[0164] The calculation formula for the output result of the thunderstorm cloud cluster dynamic monitoring and recognition model is as follows:

[0165]

[0166] Among them, out is the model output, net represents the optimal thunderstorm cloud cluster dynamic monitoring and recognition model, is the Sigmoid activation function.

[0167] Figure 4 This is the cost function convergence curve of the training set and test set of the thunderstorm cloud cluster dynamic monitoring and recognition model of the example of the present invention, and the detection rate and false alarm rate of the validation set. In this example, the optimal hyperparameter combination of the thunderstorm cloud cluster dynamic monitoring and recognition model is that the number of network layers is 4, the number of hidden layers is 50, and the batch size is 100. The detection rate of thunderstorms in the validation set is 81.1%, and the false alarm rate is 17.3%.

[0168] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for dynamically monitoring and identifying thunderstorm cloud clusters of geostationary satellites based on long short-term memory networks, characterized in that, It includes the following steps: Construct a dynamic monitoring and recognition model of thunderstorm cloud clusters based on the long short-term memory network model, and complete the training using the dynamic monitoring machine learning dataset; The dynamic monitoring machine learning dataset includes several groups of dynamic monitoring machine learning arrays, and each group of dynamic monitoring machine learning arrays includes a spatio-temporally matched historical mesoscale convective cloud cluster feature parameter group and lightning activity occurrence label data; The historical mesoscale convective cloud cluster feature parameter group is constructed by statistically analyzing the physical feature parameters of the mesoscale convective cloud cluster at m consecutive moments before any historical forecast moment Tm+1; The lightning activity occurrence label data is obtained from the lightning activity at the location of the corresponding mesoscale convective cloud cluster at the historical forecast moment Tm+1 observed by the lightning positioning system; The input of the trained dynamic monitoring and recognition model for thunderstorm cloud clusters is the physical characteristic parameters corresponding to the first m consecutive moments before the current moment t of all mesoscale convective cloud clusters identified in the target area m+1 and the output is the recognition result of thunderstorm cloud clusters at the current moment t m+1 ; The physical feature parameters of the mesoscale convective cloud cluster at any moment include cloud cluster macroscopic parameters, cloud cluster microscopic parameter A, and improved cloud cluster microscopic parameter B; where: The cloud cluster macroscopic parameters include the central longitude and latitude of the mesoscale convective cloud cluster, the major axis length, minor axis length, and rotation angle of the circumscribed ellipse; The cloud cluster microscopic parameter A and the improved cloud cluster microscopic parameter B are calculated based on the brightness temperature data detected by the infrared imager carried on the geostationary satellite in the target area; where: For any mesoscale convective cloud cluster j, the elements included in the cloud cluster microparameter A correspond to: the mean value of the brightness temperature in the 10.4-μm channel , the minimum value and the maximum value , the brightness temperature difference between the 6.9-μm and 7.3-μm channels the mean value of TB1 , the minimum value Mi and the maximum value Max , the brightness temperature difference between the 7.3-μm and 10.4-μm channels the mean value of TB2 , the minimum value Mi and the maximum value Max , the brightness temperature difference between the 12.4-μm and 10.4-μm channels the mean value of TB3 , the minimum value Mi and the maximum value Max ; Any element in the improved cloud cluster microscopic parameter B is obtained by optimizing the mean element in the cloud cluster microscopic parameter A using the infrared channel brightness temperature correction factor; For any mesoscale convective cloud cluster j, the elements included in the cloud cluster microparameter B correspond to: the mean of the improved brightness temperature in the 10.4-μm channel ; the mean of the improved brightness temperature difference between the 6.9-μm and 7.3-μm channels ; the mean of the improved brightness temperature difference between the 7.3-μm and 10.4-μm channels ; the mean of the improved brightness temperature difference between the 12.4-μm and 10.4-μm channels .

2. The dynamic monitoring and recognition method of geostationary satellite thunderstorm cloud clusters based on the long short-term memory network according to claim 1, characterized in that The infrared channel brightness temperature correction factor is calculated by the following formula: ； Wherein: represents the brightness temperature of the 10.4-μm channel at pixel i; represents the correction factor of the brightness temperature of the infrared channel at pixel i; In the improved cloud microparameter B, any element is calculated by the following formula: ； In the formula: represents the th element in the cloud cluster microparameter B, ; represents the mean of the brightness temperature of the 10.4-μm channel ; represents the mean of the brightness temperature difference between the 6.9-μm and 7.3-μm channels ; represents the mean of the brightness temperature difference between the 7.3-μm and 10.4-μm channels ; represents the mean of the brightness temperature difference between the 12.4-μm and 10.4-μm channels ; N represents all the pixel numbers within any mesoscale convective cloud cluster j.

3. The method for dynamically monitoring and identifying geostationary satellite thunderstorm cloud clusters based on a long short-term memory network according to claim 2, wherein For any mesoscale convective cloud cluster j, based on the contour points of the mesoscale convective cloud cluster j, using the least squares method to minimize the distance between the circumscribed ellipse of the mesoscale convective cloud cluster j and the contour points, the major axis length, minor axis length, and rotation angle of the circumscribed ellipse of any mesoscale convective cloud cluster j can be obtained; For any mesoscale convective cloud cluster j, based on the brightness temperatures of the 10.4-μm channel of all pixel points within the mesoscale convective cloud cluster j, the mean value of the brightness temperatures of the 10.4-μm channel within the mesoscale convective cloud cluster can be calculated. , the minimum value and the maximum value ; For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference between the 6.9-μm and 7.3-μm channels for all pixels within the mesoscale convective cloud cluster j TB1, and then statistically obtain the mean value of the brightness temperature difference TB1, the minimum and the maximum value within the mesoscale convective cloud cluster j; For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference between the 7.3-μm and 10.4-μm channels for all pixels within the mesoscale convective cloud cluster j TB2, and then statistically obtain the mean value of the brightness temperature difference TB2, the minimum and the maximum value within the mesoscale convective cloud cluster j; For any mesoscale convective cloud cluster j, first calculate the brightness temperature difference between the 12.4-μm and 10.4-μm channels for all pixels within the mesoscale convective cloud cluster j TB3, and then statistically obtain the mean value of the brightness temperature difference TB3 between the 12.4-μm and 10.4-μm channels, the minimum and the maximum values within the mesoscale convective cloud cluster j 4. The method for dynamically monitoring and identifying geostationary satellite thunderstorm cloud clusters based on a long short-term memory network according to claim 1, characterized in that, The dynamic monitoring machine learning dataset is constructed through the following steps: Step 1. Mesoscale convective cloud cluster recognition: Obtain the brightness temperature data in the target area detected by the infrared imager carried on the geostationary satellite during the historical period, and identify the continuous low infrared brightness temperature area at the initial moment of the detected historical period as the mesoscale convective cloud cluster; Step 2. Statistical feature parameters of mesoscale convective cloud clusters: According to the physical connotations of the mesoscale convective cloud clusters identified in Step 1, statistically analyze the physical feature parameters of the corresponding mesoscale convective cloud clusters; Step 3. Mesoscale convective cloud cluster trajectory tracking: Use the mesoscale convective system tracking algorithm to determine the movement trajectories of the mesoscale convective cloud clusters identified in Step 1, so as to dynamically monitor the mesoscale convective cloud clusters identified in Step 1; Step 4. Construction of dynamic monitoring machine learning dataset: Combining the physical feature parameters of any mesoscale convective cloud cluster statistically analyzed in Step 2 and the movement trajectories of the corresponding mesoscale convective cloud clusters tracked in Step 3, and using the lightning activity information observed by the lightning positioning system, the dynamic monitoring machine learning dataset can be constructed.

5. The method for dynamically monitoring and identifying geostationary satellite thunderstorm cloud clusters based on a long short-term memory network according to claim 4, wherein, The mesoscale convective cloud cluster is identified based on the 10.4-micron channel brightness temperature data detected by the infrared imager in the target area, and specifically includes the following steps: Step 1.

1. In the target area, obtain the brightness temperature at the 10.4-μm channel observed by the infrared imager at any pixel i position ; Step 1.

2. By comparing the brightness temperature of the 10.4-μm channel with the preset brightness temperature threshold Tb0, to determine whether the corresponding pixel i belongs to the mesoscale convective cloud cluster pixel: when the judgment result shows that ≥Tb0, the pixel i is identified as a mesoscale convective cloud cluster pixel, otherwise it is a non-mesoscale convective cloud cluster pixel; i is a positive integer; Step 1.3: Traverse the surrounding pixels of the mesoscale convective cloud cluster pixel i. If there are mesoscale convective cloud cluster pixels around the mesoscale convective cloud cluster pixel i, continue the traversal until all adjacent pixels around the mesoscale convective cloud cluster pixel i are non-mesoscale convective cloud cluster pixels. Then, the identification of a complete mesoscale convective cloud cluster ends, and the identified mesoscale convective cloud cluster is numbered, denoted as mesoscale convective cloud cluster j, where j is a positive integer.

6. The method for dynamically monitoring and identifying geostationary satellite thunderstorm cloud clusters based on a long short-term memory network according to claim 4 or 5, characterized in that The mesoscale convective cloud cluster trajectory tracking in Step 3 specifically includes: Step 3.1: According to the mesoscale convective cloud cluster MCC Tk center position information at time Tk, preset a search radius R, and conduct a search for mesoscale convective cloud clusters within the search radius R at time Tk+1. The number of mesoscale convective cloud clusters whose center positions fall within the search radius R is denoted as N, and each mesoscale convective cloud cluster obtained at time Tk+1 is denoted as MCC 1 Tk+1 , MCC 2 Tk+1 , ……, MCC N Tk+1 , where N is a positive integer; Step 3.2: If N = 1, record the center position of the only mesoscale convective cloud cluster MCC 1 Tk+1 as the position of the mesoscale convective cloud cluster MCC at time Tk+1; if N ≥ 2, among the mesoscale convective cloud clusters searched at time Tk+1, record the center position of the mesoscale convective cloud cluster that is closest to the center position of the mesoscale convective cloud cluster MCC at time Tk Tk as the predicted position of the mesoscale convective cloud cluster MCC at time Tk+1; Tk Tk ​​ Step 3.

3. For the subsequent moments Tk+2, Tk+3, ……, Tk+n, conduct the trajectory search of mesoscale convective cloud clusters at corresponding moments according to Step 3.1 and Step 3.2 until no mesoscale convective cloud cluster is found within the search radius R, then the trajectory search of this mesoscale convective cloud cluster MCC Tk ends.

7. The method for dynamically monitoring and identifying geostationary satellite thunderstorm cloud clusters based on a long short-term memory network according to claim 4, wherein The construction of the dynamic monitoring machine learning dataset in Step 4 specifically includes: Step 4.1: Using all the mesoscale convective cloud cluster trajectories tracked in Step 3, for any mesoscale convective cloud cluster trajectory case, according to the lightning occurrence location and lightning occurrence time observed by the lightning location system, count the lightning activities of the mesoscale convective cloud cluster at different times. Lightning occurrence is marked as "1", and no lightning occurrence is marked as "0". Step 4.2: Using the physical characteristic parameters of the mesoscale convective cloud cluster statistically obtained in Step 2, for any mesoscale convective cloud cluster trajectory case that lasts for more than m + 1 time steps, take the physical characteristic parameters of the mesoscale convective cloud cluster at T1 to Tm consecutive times and the lightning activity information at the Tm+1 time step as a set of machine learning data. Step 4.3: Take the physical characteristic parameters of the mesoscale convective cloud cluster at T2 to Tm+1 consecutive times and the lightning activity information at the Tm+2 time step as a set of machine learning data, and so on until the mesoscale convective cloud cluster trajectory terminates, finally constructing a dynamic monitoring machine learning dataset.

8. The method for dynamically monitoring and identifying geostationary satellite thunderstorm cloud clusters based on a long short-term memory network according to claim 5, characterized in that The training of the thunderstorm cloud cluster dynamic monitoring and identification model specifically includes the following steps: Step 5.1: Divide the dynamic monitoring machine learning dataset constructed in Step 4 into three parts: a training set, a test set, and a validation set according to 6:2:

2. Step 5.2: Using the training set and the test set, adopt the long short-term memory network model, design different input feature parameters of the mesoscale convective cloud cluster, and use the binary classification BCEwithLogitsLoss loss function to test the optimal input feature parameter combination of the thunderstorm cloud cluster dynamic monitoring and identification model. Step 5.3: Using the optimal input feature parameter combination obtained in Step 5.2, adopt the long short-term memory network model, design multiple groups of network hyperparameters, and use the binary classification BCEwithLogitsLoss loss function to optimize the network hyperparameters of the thunderstorm cloud cluster dynamic monitoring and identification model; for the validation set, use the detection rate and false alarm rate indicators to evaluate the accuracy of the output of the thunderstorm cloud cluster dynamic monitoring and identification model.

9. The method for dynamically monitoring and identifying geostationary satellite thunderstorm cloud clusters based on a long short-term memory network according to claim 8, wherein, The formula of the BCEwithLogitsLoss loss function in Steps 5.2 and 5.3 is as follows: ； ； Among them, is the characteristic parameter group of the input historical mesoscale convective cloud clusters, is the sample label, is the predicted value; represents the Sigmoid activation function; In Steps 5.2 and 5.3, the calculation formula of the output result of the thunderstorm cloud cluster dynamic monitoring and identification model is as follows: ； Among them, out is the model output, and net represents the dynamic monitoring and recognition model of thunderstorm clouds. is the Sigmoid activation function.

Images