A storm near extrapolation method and device based on ConvLSTM

Abstract

This invention provides a method and apparatus for storm near-term extrapolation based on ConvLSTM, relating to the technical field of meteorological early warning. The method includes: acquiring historical radar observation data and corresponding remote sensing image data; obtaining predicted extrapolation data using a radar ConvLSTM model and historical radar observation data; calculating extrapolation residual data based on the predicted extrapolation data and historical radar observation data; training an initial remote sensing ConvLSTM model using the remote sensing image data corresponding to the historical radar observation data and the extrapolation residual data to obtain a target remote sensing ConvLSTM model; and after acquiring the radar observation data to be processed and the corresponding remote sensing image data, determining the storm extrapolation result using the radar ConvLSTM model and the target remote sensing ConvLSTM model, thus solving the technical problem of poor accuracy in storm extrapolation results in existing technologies.

Description

Technical Field

[0001] This invention relates to the technical field of weather warning, and in particular to a method and apparatus for storm proximity extrapolation based on ConvLSTM. Background Technology

[0002] Storms are a crucial component of severe weather events. Their occurrence is often accompanied by intense short-duration heavy rainfall, thunderstorms, strong winds, hail, lightning, tornadoes, and other severe convective weather phenomena, causing serious damage and impacts on agriculture, geology, the environment, and aviation, resulting in enormous economic losses and casualties. Storms are characterized by their small spatial scale, sudden onset, rapid development, short lifespan, but high destructive power. Storm extrapolation technology based on radar data is a vital component of severe weather early warning systems.

[0003] Convolutional Recurrent Neural Networks (ConvLSTM) is a novel method specifically designed for feature extraction and prediction of spatiotemporal data. It leverages the inherent advantages of convolutional neural networks in image feature extraction and the strengths of recurrent neural networks in processing time-series data, resulting in a comprehensive approach. Compared to optical flow methods, this method can better fit the nonlinearity of echo motion. Furthermore, when echoes enter the monitoring area (partially), optical flow methods, unable to acquire external storm information, can only drive the internal storm at the edge of the monitoring area, leading to inaccurate gaps in storm monitoring at the edges. ConvLSTM, through automatic learning from large amounts of data, can intelligently supplement the echo information in this area, thus avoiding the extrapolation shortcomings of optical flow methods.

[0004] However, in actual operations, relying solely on radar data for storm proximity extrapolation is insufficient because radar can only monitor existing storm bodies and make certain predictions about their movement. For some locally formed storm bodies, radar cannot provide effective early warnings.

[0005] No effective solutions have yet been proposed to address the above problems. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a storm proximity extrapolation method and apparatus based on ConvLSTM, so as to alleviate the technical problem of poor accuracy of storm extrapolation results in the prior art.

[0007] In a first aspect, embodiments of the present invention provide a storm near-term extrapolation method based on ConvLSTM, comprising: acquiring historical radar observation data and remote sensing image data corresponding to the historical radar observation data, wherein the historical radar observation data is radar observation data on dates when thunderstorms occur; using a radar ConvLSTM model and the historical radar observation data to obtain predicted extrapolated data of the historical radar observation data, and calculating extrapolated residual data based on the predicted extrapolated data and the historical radar observation data; using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data to train an initial remote sensing ConvLSTM model to obtain a target remote sensing ConvLSTM model; after acquiring the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed, determining the storm extrapolation result using the radar ConvLSTM model and the target remote sensing ConvLSTM model.

[0008] Furthermore, using the radar ConvLSTM model and the historical radar observation data, the predicted extrapolation data of the historical radar observation data is obtained, including: identifying the target radar observation data in the historical radar observation data, wherein the target radar observation data is radar volume scan data with a preset time resolution for a preset duration before the storm approaches; inputting the target radar observation data into the radar ConvLSTM model to obtain the predicted extrapolation data of the historical radar observation data.

[0009] Further, calculating the extrapolation residual data based on the predicted extrapolation data and the historical radar observation data includes: determining the predicted reflectance value for each grid point based on the predicted extrapolation data, and determining the observed reflectance value for each grid point based on the target radar observation data; calculating the extrapolation residual data for each grid point based on the predicted reflectance value for each grid point, the observed reflectance value for each grid point, and the calculation formula for the extrapolation residual data; wherein, the calculation formula for the extrapolation residual data is: E t (E,X,Y)=R t (P1,X,Y)-R' t (P2,X,Y), where E t (E,X,Y) represents the extrapolated residual data of the grid point in row X and column Y at time t, where E and R are the values ​​of E and R. t (P1,X,Y) represents the predicted reflectance value of the grid point in row X and column Y at time t, which is P1,R'. t (P2,X,Y) represents the observed reflectance value of the grid point in row X and column Y at time t.

[0010] Furthermore, using the radar ConvLSTM model and the target remote sensing ConvLSTM model, the storm extrapolation result is determined, including: inputting the radar observation data to be processed into the radar ConvLSTM model to obtain target extrapolation data; inputting the remote sensing image data corresponding to the radar observation data to be processed into the target remote sensing ConvLSTM model to obtain target extrapolation residual data; and using the target extrapolation residual data to correct the target extrapolation data to obtain the storm extrapolation result.

[0011] Furthermore, the remote sensing image data corresponding to the historical radar observation data is geostationary satellite remote sensing data at the same time point as the historical radar observation data, wherein the remote sensing image data corresponding to the historical radar observation data contains multiple infrared channel data.

[0012] Furthermore, the key equation of the initial remote sensing ConvLSTM model is: Where * represents the convolution operator. For Hadamard products, W represents the input at time t. xi W hi W ci W xf W hf W cf W xc W hc W xo W ho and W co b represents the weight coefficients that need to be trained in each formula. i b f b c and b o i is the constant term in the corresponding formula; t For the input gate, f t For the Gate of Oblivion In cellular state, o t For output gate, This is a hidden state parameter.

[0013] Secondly, embodiments of the present invention also provide a storm near-term extrapolation device based on ConvLSTM, comprising: an acquisition unit, configured to acquire historical radar observation data and remote sensing image data corresponding to the historical radar observation data, wherein the historical radar observation data is radar observation data on dates when thunderstorms occur; a calculation unit, configured to use a radar ConvLSTM model and the historical radar observation data to obtain predicted extrapolation data of the historical radar observation data, and to calculate extrapolation residual data based on the predicted extrapolation data and the historical radar observation data; a training unit, configured to use the remote sensing image data corresponding to the historical radar observation data and the extrapolation residual data to train an initial remote sensing ConvLSTM model to obtain a target remote sensing ConvLSTM model; and an execution unit, configured to, after acquiring the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed, use the radar ConvLSTM model and the target remote sensing ConvLSTM model to determine the storm extrapolation result.

[0014] Further, the computing unit is used to: determine the target radar observation data in the historical radar observation data, wherein the target radar observation data is radar volume scan data with a preset time resolution for a preset duration before the approach of the storm; input the target radar observation data into the radar ConvLSTM model to obtain the predicted extrapolated data of the historical radar observation data.

[0015] Thirdly, embodiments of the present invention also provide an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the method described in the first aspect above, and the processor is configured to execute the program stored in the memory.

[0016] Fourthly, embodiments of the present invention also provide a computer-readable storage medium on which a computer program is stored.

[0017] In this embodiment of the invention, historical radar observation data and corresponding remote sensing image data are acquired, wherein the historical radar observation data are radar observation data on dates when thunderstorms occur; a radar ConvLSTM model and the historical radar observation data are used to obtain predicted extrapolated data of the historical radar observation data, and extrapolated residual data are calculated based on the predicted extrapolated data and the historical radar observation data; an initial remote sensing ConvLSTM model is trained using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data to obtain a target remote sensing ConvLSTM model; after acquiring the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed, the radar ConvLSTM model and the target remote sensing ConvLSTM model are used to determine the storm extrapolation result, achieving the purpose of using remote sensing image data and radar observation data for storm near-term extrapolation, thereby solving the technical problem of poor accuracy of storm extrapolation results in the prior art, and thus achieving the technical effect of improving the accuracy of storm extrapolation results.

[0018] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 A flowchart of a storm proximity extrapolation method based on ConvLSTM provided in this embodiment of the invention;

[0022] Figure 2 This is a structural diagram of the initial remote sensing ConvLSTM model provided in an embodiment of the present invention;

[0023] Figure 3 A schematic diagram of a storm proximity extrapolation device based on ConvLSTM provided in an embodiment of the present invention;

[0024] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1:

[0027] According to an embodiment of the present invention, an embodiment of a storm proximity extrapolation method based on ConvLSTM is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] Figure 1 This is a flowchart of a storm proximity extrapolation method based on ConvLSTM according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0029] Step S102: Obtain historical radar observation data and remote sensing image data corresponding to the historical radar observation data, wherein the historical radar observation data is radar observation data on the dates when thunderstorms occurred.

[0030] Step S104: Using the radar ConvLSTM model and the historical radar observation data, the predicted extrapolation data of the historical radar observation data is obtained, and the extrapolation residual data is calculated based on the predicted extrapolation data and the historical radar observation data.

[0031] It should be noted that the time period corresponding to the predicted extrapolated data is the same as the time period of the historical radar observation data corresponding to the predicted extrapolated data.

[0032] ConvLSTM (Convolutional LSTM): It extends the fully connected LSTM (FC-LSTM) to have convolutional structures in both input-to-state and state-to-state transitions.

[0033] Long Short-Term Memory (LSTM) is a type of time-recurrent neural network specifically designed to address the long-term dependency problem inherent in general RNNs (Recurrent Neural Networks).

[0034] Step S106: Using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data, train the initial remote sensing ConvLSTM model to obtain the target remote sensing ConvLSTM model.

[0035] Step S108: After obtaining the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed, the storm extrapolation result is determined using the radar ConvLSTM model and the target remote sensing ConvLSTM model.

[0036] In this embodiment of the invention, historical radar observation data and corresponding remote sensing image data are acquired, wherein the historical radar observation data are radar observation data on dates when thunderstorms occur; a radar ConvLSTM model and the historical radar observation data are used to obtain predicted extrapolated data of the historical radar observation data, and extrapolated residual data are calculated based on the predicted extrapolated data and the historical radar observation data; an initial remote sensing ConvLSTM model is trained using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data to obtain a target remote sensing ConvLSTM model; after acquiring the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed, the radar ConvLSTM model and the target remote sensing ConvLSTM model are used to determine the storm extrapolation result, achieving the purpose of using remote sensing image data and radar observation data for storm near-term extrapolation, thereby solving the technical problem of poor accuracy of storm extrapolation results in the prior art, and thus achieving the technical effect of improving the accuracy of storm extrapolation results.

[0037] In this embodiment of the invention, step S104 includes the following steps:

[0038] The target radar observation data in the historical radar observation data is determined, wherein the target radar observation data is radar volume scan data with a preset time resolution for a preset duration before the approach of the storm.

[0039] The target radar observation data is input into the radar ConvLSTM model to obtain the predicted extrapolated data of the historical radar observation data.

[0040] Based on the predicted extrapolation data, the predicted reflectance value of each grid point is determined, and based on the target radar observation data, the observed reflectance value of each grid point is determined.

[0041] Based on the calculation formulas of the predicted reflectance value of each grid point, the observed reflectance value of each grid point, and the extrapolated residual data, the extrapolated residual data of each grid point is calculated.

[0042] In this embodiment of the invention, firstly, radar volume scan data with a time resolution of 6 minutes and an approaching storm of 1 hour are obtained from historical radar observation data. This data is used as input data, and the radar ConvLSTM model is used to calculate the extrapolation result of the radar storm in the next 3 hours (i.e., the predicted extrapolation data of historical radar observation data).

[0043] Next, the predicted reflectance image is spatiotemporally matched with the observed reflectance image frame by frame, and the residuals are extrapolated. The formula for calculating the extrapolated residual data is: E t (E,X,Y)=R t (P1,X,Y)-R' t (P2,X,Y), where E t (E,X,Y) represents the extrapolated residual data of the grid point in row X and column Y at time t, where E and R are the values ​​of E and R. t (P1,X,Y) represents the predicted reflectance value of the grid point in row X and column Y at time t, which is P1,R'. t (P2,X,Y) represents the observed reflectance value of the grid point in row X and column Y at time t.

[0044] The time interval is usually 6 minutes. If the current time is T, then t = T+6, T+12, ..., T+180, which calculates the radar extrapolation residual over 3 hours.

[0045] The following is a detailed explanation of step S106.

[0046] The radar input data used to calculate the radar extrapolation residuals are geostationary satellite remote sensing data at the same time point (i.e., remote sensing image data corresponding to the historical radar observation data), including multiple infrared channel multi-time series historical data D(IR,X,Y), where IR is infrared channel data and IR contains n infrared channels.

[0047] Labeled data: Extrapolated residual data E t .

[0048] Construct a dataset consisting of input data and labeled data.

[0049] The key equations for ConvLSTM are shown below:

[0050]

[0051]

[0052]

[0053]

[0054]

[0055] Where * represents the convolution operator. For Hadamard products, W represents the input at time t. xi W hi W ci W xf W hf W cf W xc W hc W xo W ho and W co b represents the weight coefficients that need to be trained in each formula. i b f b c and b o i is the constant term in the corresponding formula; t For the input gate, f t Forget Gate For cell state, o t It is an output gate. This refers to the hidden state parameter.

[0056] like Figure 2 As shown, the initial remote sensing ConvLSTM model structure is as follows:

[0057] Input layer: The input layer allows input of multi-time series and multi-channel satellite remote sensing data.

[0058] Fusion Layer: Employs multi-channel linear fusion of input values ​​from each channel, resulting in the fusion value C of the multi-infrared channels for the i-th pixel. i The calculation formula is shown below. Compared to the ConvLSTM model, c ij and b i These are the newly added model parameters.

[0059] The calculation formula is as follows:

[0060] n is the number of infrared channels, ir ij c is the value of the j-th pixel in the j-th infrared channel. ij b is the weighting coefficient. i is the regression coefficient.

[0061] ConvLSTM1: The first layer of ConvLSTM.

[0062] ConvLSTM2: The second layer of ConvLSTM.

[0063] Output layer: Multi-time extrapolation residuals.

[0064] Finally, 60% of the dataset was used as the training set, 20% as the validation set, and 20% as the test set to train the initial remote sensing ConvLSTM model, thus obtaining the target remote sensing ConvLSTM model.

[0065] In this embodiment of the invention, step S108 includes the following steps:

[0066] The radar observation data to be processed is input into the radar ConvLSTM model to obtain target extrapolation data;

[0067] Input the remote sensing image data corresponding to the radar observation data to be processed into the target remote sensing ConvLSTM model to obtain the target extrapolation residual data;

[0068] The target extrapolation data is corrected using the target extrapolation residual data to obtain the storm extrapolation result.

[0069] In this embodiment of the invention, remote sensing data from 3 hours prior to the storm (i.e., remote sensing image data corresponding to the radar observation data to be processed) is used as input data. The target satellite remote sensing ConvLSTM model is used to calculate the input data, and the target extrapolation residual data is output. The specific steps are as follows:

[0070] Acquire multi-channel infrared data from geostationary meteorological satellites up to 3 hours before the storm.

[0071] Calculate the target extrapolation residual data E′ using the target remote sensing ConvLSTM model t (E,X,Y)(t=T+6,T+12,...,T+180).

[0072] Using radar observation data from the hour immediately preceding the storm (i.e., the radar observation data to be processed) as input, the radar ConvLSTM model is used to extrapolate the storm's development and changes over a 3-hour period. The specific steps are as follows:

[0073] Acquire radar volume scan data with a time resolution of 6 minutes, one hour before the storm. t (P,X,Y)(t=T-60,T-54,...,T-6), after data quality control and preprocessing, are copied sequentially to the data directory of the radar ConvLSTM model according to the time series of the base data;

[0074] R t As input, target extrapolation data is calculated using a radar ConvLSTM model to obtain 30 sets of target extrapolation data R′ for the next 3 hours. t (P,X,Y)(t=T+6,T+12,...,T+180).

[0075] Finally, the target extrapolated data is corrected using the target extrapolated residual data, where the target extrapolated residual data is E′. t (E,X,Y), the target extrapolated data is R′ t (P,X,Y)(t=T+6,T+12,...,T+180), where T is the current time, the weather radar extrapolation results are corrected grid by grid according to the following formula to obtain the storm extrapolation result R″. t Using (P,X,Y) can improve the accuracy of extrapolation results. The specific process is as follows:

[0076] First, time alignment is performed. The extrapolated results of 30 radar observation data (i.e., target extrapolated data) and the extrapolated residual results of 30 data (i.e., target extrapolated residual data) are matched one-to-one according to their predicted time. The rationality of the data files is checked. If the predicted time of the two data files differs by more than 2 minutes, the two data that cannot be aligned are removed.

[0077] Storm extrapolation result R″ t The formula for calculating (P,X,Y) is as follows:

[0078] R″ t =R′ t +E′ t (t=T+6, T+12,..., T+180).

[0079] In this embodiment of the invention, an improved satellite remote sensing ConvLSTM model is trained using multiple infrared channel data from geostationary meteorological satellites and radar extrapolation residual data (the residual information between the radar extrapolation result and the actual observation data corresponding to the extrapolation time). This establishes a fitting relationship between satellite remote sensing infrared data and storm formation and dissipation information, thereby obtaining storm formation and dissipation evolution information based on satellite observations. Combined with radar extrapolation results, a more accurate storm extrapolation forecast is obtained.

[0080] Meanwhile, based on the ConvLSTM model, the input data types are expanded and a data fusion layer is added to realize multi-channel infrared data input. This can fully utilize the characteristics of the temperature and humidity field at high altitudes of the cloud top in different infrared bands, capture more accurate temperature and humidity field feature information and input it into other layers of the model, thereby improving the inversion accuracy.

[0081] Example 2:

[0082] This invention also provides a storm proximity extrapolation device based on ConvLSTM. This ConvLSTM-based storm proximity extrapolation device is used to execute the storm proximity extrapolation method based on ConvLSTM provided in the above-described embodiments of this invention. The following is a detailed description of the storm proximity extrapolation device based on ConvLSTM provided in this invention.

[0083] like Figure 3 As shown, Figure 3 This is a schematic diagram of the storm proximity extrapolation device based on ConvLSTM described above. The ConvLSTM-based storm proximity extrapolation device includes:

[0084] The acquisition unit 10 is used to acquire historical radar observation data and remote sensing image data corresponding to the historical radar observation data, wherein the historical radar observation data is radar observation data on the dates when thunderstorms occurred.

[0085] The calculation unit 20 is used to obtain the predicted extrapolation data of the historical radar observation data using the radar ConvLSTM model and the historical radar observation data, and to calculate the extrapolation residual data based on the predicted extrapolation data and the historical radar observation data.

[0086] Training unit 30 is used to train the initial remote sensing ConvLSTM model using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data, so as to obtain the target remote sensing ConvLSTM model.

[0087] The execution unit 40 is used to determine the storm extrapolation result by using the radar ConvLSTM model and the target remote sensing ConvLSTM model after acquiring the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed.

[0088] In this embodiment of the invention, historical radar observation data and corresponding remote sensing image data are acquired, wherein the historical radar observation data are radar observation data on dates when thunderstorms occur; a radar ConvLSTM model and the historical radar observation data are used to obtain predicted extrapolated data of the historical radar observation data, and extrapolated residual data are calculated based on the predicted extrapolated data and the historical radar observation data; an initial remote sensing ConvLSTM model is trained using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data to obtain a target remote sensing ConvLSTM model; after acquiring the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed, the radar ConvLSTM model and the target remote sensing ConvLSTM model are used to determine the storm extrapolation result, achieving the purpose of using remote sensing image data and radar observation data for storm near-term extrapolation, thereby solving the technical problem of poor accuracy of storm extrapolation results in the prior art, and thus achieving the technical effect of improving the accuracy of storm extrapolation results.

[0089] Example 3:

[0090] This invention also provides an electronic device, including a memory and a processor. The memory is used to store a program that supports the processor in executing the method described in Embodiment 1 above, and the processor is configured to execute the program stored in the memory.

[0091] See Figure 4 The present invention also provides an electronic device 100, including: a processor 50, a memory 51, a bus 52 and a communication interface 53, wherein the processor 50, the communication interface 53 and the memory 51 are connected through the bus 52; the processor 50 is used to execute executable modules, such as computer programs, stored in the memory 51.

[0092] The memory 51 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 53 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0093] Bus 52 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0094] The memory 51 is used to store programs. After receiving an execution instruction, the processor 50 executes the programs. The method executed by the device for defining the flow process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 50 or implemented by the processor 50.

[0095] Processor 50 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 50 or by instructions in software form. Processor 50 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 51. The processor 50 reads the information in memory 51 and, in conjunction with its hardware, completes the steps of the above method.

[0096] Example 4:

[0097] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method described in Embodiment 1 above.

[0098] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0099] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0100] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0101] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0102] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0103] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A storm nowcasting method based on ConvLSTM, characterized in that, include: Acquire historical radar observation data and corresponding remote sensing image data, wherein the historical radar observation data are radar observation data on dates when thunderstorms occurred; Using the radar ConvLSTM model and the historical radar observation data, the predicted extrapolated data of the historical radar observation data is obtained, and the extrapolated residual data is calculated based on the predicted extrapolated data and the historical radar observation data. Using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data, the initial remote sensing ConvLSTM model is trained to obtain the target remote sensing ConvLSTM model. The structure of the initial remote sensing ConvLSTM model is as follows: Input layer: input multi-temporal, multi-channel remote sensing image data; Fusion layer: multivariate linear fusion of the input values ​​of each channel, and the fused value of the i-th pixel multi-infrared channel. The calculation formula is shown below. Relative to the ConvLSTM model, and These are the newly added model parameters; the calculation formula is: n is the number of infrared channels. Let be the value of the j-th pixel in the j-th infrared channel. These are the weighting coefficients. Regression coefficients; ConvLSTM1: First ConvLSTM layer; ConvLSTM2: Second ConvLSTM layer; Output layer: Outputs multi-temporal extrapolation residuals; After acquiring the radar observation data to be processed and the corresponding remote sensing image data, the storm extrapolation result is determined using the radar ConvLSTM model and the target remote sensing ConvLSTM model. This includes: inputting the radar observation data to be processed into the radar ConvLSTM model to obtain target extrapolation data; inputting the remote sensing image data corresponding to the radar observation data to be processed into the target remote sensing ConvLSTM model to obtain target extrapolation residual data; and correcting the target extrapolation data using the target extrapolation residual data to obtain the storm extrapolation result.

2. The method according to claim 1, characterized in that, Using the radar ConvLSTM model and the historical radar observation data, the predicted extrapolated data of the historical radar observation data are obtained, including: The target radar observation data in the historical radar observation data is determined, wherein the target radar observation data is radar volume scan data with a preset time resolution for a preset duration before the approach of the storm. The target radar observation data is input into the radar ConvLSTM model to obtain the predicted extrapolated data of the historical radar observation data.

3. The method according to claim 2, characterized in that, Extrapolated residual data are calculated based on the predicted extrapolated data and the historical radar observation data, including: Based on the predicted extrapolation data, the predicted reflectance value of each grid point is determined, and based on the target radar observation data, the observed reflectance value of each grid point is determined. Based on the calculation formulas of the predicted reflectance value of each grid point, the observed reflectance value of each grid point, and the extrapolated residual data, the extrapolated residual data of each grid point is calculated. The formula for calculating the extrapolated residual data is as follows: ,in, Used for characterization Time of the first Line number The extrapolated residual data of the column grid points are , Used for characterization Time of the first Line number The predicted reflectance value of the grid points in the column is Used for characterization Time of the first Line number The observed reflectance values ​​of the grid points in the column are .

4. The method according to claim 1, characterized in that, The remote sensing image data corresponding to the historical radar observation data is geostationary satellite remote sensing data at the same time point as the historical radar observation data, wherein the remote sensing image data corresponding to the historical radar observation data contains multiple infrared channel data.

5. The method according to claim 1, characterized in that, The key equations of the initial remote sensing ConvLSTM model are: , in, For convolution operators, For Hadamard products, This represents the input at time t. , , , , , , , , , These are the weight coefficients that need to be trained in each formula. , , and This is the constant term in the corresponding formula; For input gate, For the Gate of Oblivion In cellular state, For output gate, This is a hidden state parameter.

6. A storm proximity extrapolation device based on ConvLSTM, characterized in that, include: The acquisition unit is used to acquire historical radar observation data and remote sensing image data corresponding to the historical radar observation data, wherein the historical radar observation data is radar observation data on the dates when thunderstorms occurred. The calculation unit is used to obtain the predicted extrapolation data of the historical radar observation data using the radar ConvLSTM model and the historical radar observation data, and to calculate the extrapolation residual data based on the predicted extrapolation data and the historical radar observation data. The training unit is used to train the initial remote sensing ConvLSTM model using the remote sensing image data corresponding to the historical radar observation data and the extrapolated residual data, to obtain the target remote sensing ConvLSTM model. The structure of the initial remote sensing ConvLSTM model is as follows: Input layer: input multi-temporal, multi-channel remote sensing image data; Fusion layer: multivariate linear fusion of the input values ​​of each channel, and the fused value of the i-th pixel multi-infrared channel. The calculation formula is shown below. Relative to the ConvLSTM model, and These are the newly added model parameters; the calculation formula is: n is the number of infrared channels. Let be the value of the j-th pixel in the j-th infrared channel. These are the weighting coefficients. Regression coefficients; ConvLSTM1: First ConvLSTM layer; ConvLSTM2: Second ConvLSTM layer; Output layer: Outputs multi-temporal extrapolation residuals; The execution unit is configured to, after acquiring the radar observation data to be processed and the remote sensing image data corresponding to the radar observation data to be processed, determine the storm extrapolation result using the radar ConvLSTM model and the target remote sensing ConvLSTM model, including: inputting the radar observation data to be processed into the radar ConvLSTM model to obtain target extrapolation data; inputting the remote sensing image data corresponding to the radar observation data to be processed into the target remote sensing ConvLSTM model to obtain target extrapolation residual data; and correcting the target extrapolation data using the target extrapolation residual data to obtain the storm extrapolation result.

7. The apparatus according to claim 6, characterized in that, The computing unit is used for: The target radar observation data in the historical radar observation data is determined, wherein the target radar observation data is radar volume scan data with a preset time resolution for a preset duration before the approach of the storm. The target radar observation data is input into the radar ConvLSTM model to obtain the predicted extrapolated data of the historical radar observation data.

8. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a program that enables the processor to execute the method of any one of claims 1 to 5, and the processor being configured to execute the program stored in the memory.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When a computer program is run by a processor, it performs the steps of the method described in any one of claims 1 to 5.

Images