Typhoon center identification method, system, equipment and medium
By constructing training samples and optimizing the convolutional neural network model, the problem of typhoon center identification bias in complex terrain areas was solved, achieving high-precision and automated typhoon center identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 航天天目(重庆)卫星科技有限公司
- Filing Date
- 2025-12-23
- Publication Date
- 2026-05-15
AI Technical Summary
Existing typhoon center identification methods are easily affected by terrain and pattern errors when typhoons are close to land or complex terrain areas, resulting in identification deviations or center location jumps, and lacking adaptive capabilities.
Training samples were constructed, and the initial convolutional neural network model was trained using historical multi-layer meteorological element field data of multiple historical typhoons. The target convolutional neural network model was optimized by a preset loss function, which can accurately identify the typhoon center under complex terrain conditions.
It improves the accuracy of typhoon center identification, reduces identification deviations caused by terrain influence and model errors, and achieves high-precision, automated identification.
Smart Images
Figure CN122045804A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of meteorological forecasting and artificial intelligence technology, and in particular to a method, system, device and medium for identifying the center of a typhoon. Background Technology
[0002] Typhoon track and intensity are core elements of tropical cyclone forecasting, and accurate identification of the typhoon center is a crucial prerequisite for accurate forecasting. Currently, commonly used operational methods for typhoon center identification mainly include diagnostic algorithms based on the lowest sea-level pressure point, the maximum relative vorticity point, or the maximum tangential wind speed point.
[0003] However, while these existing methods perform well in open sea areas, they often exhibit identification biases or even cause center positioning jumps or loss when typhoons approach land or complex terrain areas due to terrain influences and model errors. Summary of the Invention
[0004] To overcome the problem that existing typhoon center identification methods are easily affected by terrain and model errors, resulting in identification deviations, this application provides a typhoon center identification method, system, device, and medium.
[0005] Firstly, in order to solve the aforementioned technical problems, this application provides a method for identifying the center of a typhoon, including: A training sample was constructed, which included historical multi-layer meteorological element field data of multiple historical typhoons. Based on a preset loss function, the initial convolutional neural network model is trained using training samples to obtain the target convolutional neural network model. Acquire multi-layer meteorological element field data of the target typhoon's location; The target multi-layer meteorological element field data is input into the target convolutional neural network model for identification, and the target center parameters of the target typhoon are obtained. The target center parameters include the center latitude and longitude coordinates and the center pressure value.
[0006] Furthermore, based on a preset loss function, the initial convolutional neural network model is trained using training samples to obtain the target convolutional neural network model, including: Using an initial convolutional neural network model, feature extraction was performed on historical multi-layer meteorological element field data to obtain central features; Global feature classification is performed on the central features to obtain the predicted center parameters; Based on a preset loss function, the initial convolutional neural network model is optimized using multiple prediction center parameters to obtain the target convolutional neural network model.
[0007] Furthermore, feature extraction was performed on historical multi-layer meteorological element field data to obtain central features, including: Multi-level convolution is performed on historical multi-level meteorological element field data to obtain initial central feature data; The initial central feature data is activated to obtain nonlinear feature data; Batch normalization is performed on the nonlinear feature data to obtain the central feature.
[0008] Furthermore, global feature classification is performed on the central features to obtain the predicted center parameters, including: Global average pooling is applied to the central features to obtain the global feature vector; The global feature vector is convolved to obtain the prediction center parameters.
[0009] Furthermore, based on a preset loss function, the initial convolutional neural network model is optimized using multiple prediction center parameters to obtain the target convolutional neural network model, including: The predicted center parameters and the corresponding true center parameters are substituted into the preset loss function to calculate the loss value. Based on multiple loss values and preset requirements, the network parameters in the initial convolutional neural network model are adjusted to obtain the target convolutional neural network model.
[0010] Furthermore, the formula for the loss function is as follows: in, Indicates the loss value. This represents the first weighting coefficient. This represents the spherical distance between the predicted center parameters and the corresponding true center parameters. This represents the second weighting coefficient. This indicates that the forecast center parameters include the forecast center pressure value. This indicates that the true center parameter includes the true center pressure value. This represents the latitude and longitude coordinates of the prediction center corresponding to the prediction center parameters. The true center latitude and longitude coordinates represent the true center parameters. This represents the Earth's radius.
[0011] Furthermore, based on multiple loss values and preset requirements, the network parameters in the initial convolutional neural network model are adjusted to obtain the target convolutional neural network model, including: If multiple loss values are all less than or equal to the preset value, then the initial convolutional neural network model is determined as the target convolutional neural network model; Otherwise, adjust the network parameters in the initial convolutional neural network model to obtain an intermediate convolutional neural network model; The training samples are processed using a transitional convolutional neural network model to obtain several new loss values; When multiple new loss values are all less than or equal to the preset value, the corresponding transitional convolutional neural network model is determined as the target convolutional neural network model.
[0012] Secondly, this application also provides a typhoon center identification system, including: The sample construction module is used to construct training samples, which include historical multi-layer meteorological element field data of multiple historical typhoons. The model training module is used to train the initial convolutional neural network model using training samples based on a preset loss function, so as to obtain the target convolutional neural network model. The data acquisition module is used to acquire multi-layer meteorological element field data of the target typhoon's location. The typhoon center identification module is used to input multi-layer meteorological element field data of the target into the target convolutional neural network model for identification, and obtain the target center parameters of the target typhoon, including the center latitude and longitude coordinates and the center pressure value.
[0013] Thirdly, this application also provides a computing device, including a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the typhoon center identification method described above.
[0014] Fourthly, this application also provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the steps of a typhoon center identification method.
[0015] The beneficial effects of this application are as follows: Based on a preset loss function, the initial convolutional neural network model is trained using training samples containing historical multi-layer meteorological element field data of multiple historical typhoons to obtain a target convolutional neural network model. The target multi-layer meteorological element field data of the area where the target typhoon is located is then input into the target convolutional neural network model for identification, yielding the target center parameters of the target typhoon. Thus, by using multi-channel (multi-layer) historical typhoon data for model training, a target convolutional neural network model that meets the prediction accuracy requirements can be obtained. Using this model for typhoon center identification can reduce identification bias caused by terrain influence and model errors, thereby improving the accuracy of typhoon center identification. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a typhoon center identification method as an exemplary embodiment of this application; Figure 2 This is a network structure diagram of the target convolutional neural network model in an exemplary embodiment of this application; Figure 3 This is a schematic diagram illustrating the structure of a typhoon center identification system as an exemplary embodiment of this application. Detailed Implementation
[0017] The following embodiments are further explanations and supplements to this application and do not constitute any limitation on this application.
[0018] Typhoon track and intensity are core elements in operational tropical cyclone forecasting. Currently, commonly used operational typhoon center identification methods mainly include diagnostic algorithms based on the lowest sea-level pressure point, the maximum relative vorticity point, or the maximum tangential wind speed point. These methods perform well in open sea areas, but when typhoons approach land or complex terrain areas, identification errors often occur due to topographical influences and model errors, sometimes even leading to changes or loss of center location. Furthermore, because different numerical weather prediction models, such as the WRF (Weather Research and Forecasting Model), ECMWF (European Centre for Medium-Range Weather Forecasts), and GFS (Global Forecast System), differ in resolution, physical processes, and output variables, traditional diagnostic methods require manual adjustments to different model parameters, lacking a unified adaptive capability.
[0019] To address the aforementioned issues, embodiments of this application provide a method, system, device, and medium for identifying the center of a typhoon. These embodiments will be described in detail below.
[0020] The typhoon center identification method provided in this application can be specifically executed by a server. It should be noted that the server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. No limitation is imposed here.
[0021] The typhoon center identification method of this application extracts historical multi-layer meteorological element field data of multiple historical typhoons from different numerical weather prediction models to construct training samples. Based on a preset loss function, the initial convolutional neural network model is trained using the training samples. The model then extracts and learns features from the multi-layer meteorological field output by the numerical weather prediction model to obtain a target convolutional neural network model. This target convolutional neural network model can then automatically identify the center location (center latitude and longitude coordinates) and central pressure value of various typhoons. This method can automatically adapt to different model characteristics and accurately identify typhoon centers even under complex terrain conditions, improving the reliability and automation level of model forecast results. It achieves high-precision, automated identification and can be widely applied in operational forecasting, disaster monitoring, and intelligent meteorological analysis systems.
[0022] Please see Figure 1 , Figure 1 A typhoon center identification method is illustrated in an exemplary embodiment of this application, such as... Figure 1 As shown, this application provides a method for identifying the center of a typhoon, including: S11, construct training samples, which include historical multi-layer meteorological element field data of multiple historical typhoons; S12, based on the preset loss function, train the initial convolutional neural network model using training samples to obtain the target convolutional neural network model; S13, acquire multi-layer meteorological element field data of the target typhoon's location; S14. Input the multi-layer meteorological element field data of the target into the target convolutional neural network model for identification, and obtain the target center parameters of the target typhoon. The target center parameters include the center latitude and longitude coordinates and the center pressure value.
[0023] The typhoon center identification method provided in this application, based on a preset loss function, trains an initial convolutional neural network model using training samples including historical multi-layer meteorological element field data from multiple historical typhoons to obtain a target convolutional neural network model. The target multi-layer meteorological element field data of the area where the target typhoon is located is then input into the target convolutional neural network model for identification, yielding the target typhoon's target center parameters. Thus, by using multi-channel (multi-layer) historical typhoon data for model training, a target convolutional neural network model that meets the prediction accuracy requirements can be obtained. Using this model for typhoon center identification can reduce identification bias caused by terrain influence and model errors, thereby improving the accuracy of typhoon center identification. The historical multi-layer meteorological element field data refers to multi-channel meteorological field data output from different numerical weather prediction models.
[0024] In an exemplary embodiment provided in this application, the specific steps for constructing training samples are as follows: Obtain multi-layer meteorological fields of historical typhoons at various corresponding time points from different numerical weather prediction models (such as WRF model, ECMWF model, GFS model); Each multi-level meteorological field is formatted into a multi-channel historical multi-level meteorological element field data according to a unified spatial grid, and an initial sample is constructed. The historical multi-level meteorological element field data includes, but is not limited to, sea level pressure (SLP), 10 low-level wind field components (u component, v component), topographic height field, 500hPa geopotential height field, grid longitude and grid latitude. Based on historical typhoon path data, the true center latitude and longitude coordinates and true center pressure values of each corresponding historical typhoon are labeled in the initial samples to form training samples. This allows the true center latitude and longitude coordinates and true center pressure values to be used in the loss function to optimize the initial convolutional neural network model.
[0025] Optionally, based on a preset loss function, the initial convolutional neural network model is trained using training samples to obtain the target convolutional neural network model, including: Using an initial convolutional neural network model, feature extraction was performed on historical multi-layer meteorological element field data to obtain central features; Global feature classification is performed on the central features to obtain the predicted center parameters; Based on a preset loss function, the initial convolutional neural network model is optimized using multiple prediction center parameters to obtain the target convolutional neural network model.
[0026] In the embodiment provided in this application, an initial convolutional neural network model is used to sequentially extract features and classify global features from historical multi-layer meteorological element data to obtain prediction center parameters. Based on a preset loss function, the initial convolutional neural network model is optimized using multiple prediction center parameters to obtain a target convolutional neural network model. In this way, by combining the loss function with historical multi-layer meteorological element data for model optimization, the optimized target convolutional neural network model is suitable for typhoon center prediction in complex terrain scenarios, reducing the identification bias caused by terrain influence and model errors, thereby improving the accuracy of typhoon center identification.
[0027] In an exemplary embodiment provided in this application, before training an initial convolutional neural network model using training samples, preprocessing operations such as normalization and cropping are performed on the training samples to reduce interference data in the training samples, thereby improving the error introduced during subsequent training and increasing training efficiency.
[0028] Please see Figure 2 , Figure 2 As shown in an exemplary embodiment of this application, the network structure diagram of the target convolutional neural network model is as follows: Figure 2 As shown, the network structure of the target convolutional neural network model includes an input layer, a hidden layer, and an output layer. The hidden layer includes multiple convolutional layers and pooling layers.
[0029] Specifically, the input layer receives historical multi-layer meteorological element field data from multiple historical typhoons included in the training samples; the hidden layer extracts features from the resampled historical multi-layer meteorological element field data to obtain the central features; and the output layer performs global feature classification on the central features, selecting the coordinate point with the highest probability to obtain the predicted center parameters. Resampling refers to re-representing the original training sample data sequence with a new sampling rate (or a new number of data points), which can solve the problem of imbalanced datasets.
[0030] Optionally, feature extraction is performed on historical multi-layer meteorological element field data to obtain central features, including: Multi-level convolution is performed on historical multi-level meteorological element field data to obtain initial central feature data; The initial central feature data is activated to obtain nonlinear feature data; Batch normalization is performed on the nonlinear feature data to obtain the central feature.
[0031] In the embodiment provided in this application, historical multi-layer meteorological element field data is sequentially subjected to multi-layer convolution, activation processing, and batch normalization. This process effectively extracts the central features of historical typhoons that conform to the format standards, reducing interference information in the central features. This not only lowers the difficulty of model optimization based on the predicted center parameters corresponding to the central features but also improves the accuracy of model optimization, thereby enhancing the accuracy of typhoon center identification based on the model. The multi-layer convolution is implemented using multi-layer convolution kernels, and the activation processing is implemented using ReLU activation.
[0032] Optionally, global feature classification is performed on the center features to obtain the predicted center parameters, including: Global average pooling is applied to the central features to obtain the global feature vector; The global feature vector is convolved to obtain the prediction center parameters.
[0033] In the embodiment provided in this application, the central features are sequentially subjected to global average pooling and convolution processing to ignore unnecessary spatial details in the central features. This extracts ordered, global, and essential predicted center parameters from the chaotic central features, directly regressing longitude, latitude, and central pressure. This improves the matching degree between the predicted center parameters and the corresponding historical typhoon's true center parameters, thereby enhancing the accuracy of model training and ultimately improving the accuracy of typhoon center identification based on the model. The convolution processing is implemented using 1x1 convolution.
[0034] Optionally, based on a preset loss function, the initial convolutional neural network model is optimized using multiple prediction center parameters to obtain the target convolutional neural network model, including: The predicted center parameters and the corresponding true center parameters are substituted into the preset loss function to calculate the loss value. Based on multiple loss values and preset requirements, the network parameters in the initial convolutional neural network model are adjusted to obtain the target convolutional neural network model.
[0035] In the embodiment provided in this application, the predicted center parameters and the corresponding true center parameters are substituted into a preset loss function for calculation to obtain a loss value. Based on multiple loss values and preset requirements, the network parameters in the initial convolutional neural network model are adjusted so that the final target convolutional neural network model can meet the prediction accuracy requirements, thereby improving the accuracy of typhoon center identification based on the model.
[0036] Alternatively, the loss function can be formulated as follows: in, Indicates the loss value. This represents the first weighting coefficient. This represents the spherical distance (in km) between the predicted center parameters and the corresponding true center parameters. This represents the second weighting coefficient, which can be used to set weights. This indicates that the forecast center parameters include the forecast center pressure value. This indicates that the true center parameter includes the true center pressure value. This represents the latitude and longitude coordinates of the prediction center corresponding to the prediction center parameters. The true center latitude and longitude coordinates represent the true center parameters. This represents the Earth's radius.
[0037] In the embodiment provided in this application, the initial convolutional neural network model is optimized and converged based on the loss function combined with the Earth's spherical distance and air pressure error. This can "minimize the geographical distance error of the typhoon center," making the accuracy indicators of the model's output (such as average position error) more physically meaningful. This makes it suitable for complex terrain scenarios and allows for efficient embedding into operational forecasting processes, thereby improving the model's prediction accuracy and, consequently, the accuracy of typhoon center identification based on the model. .
[0038] Optionally, based on multiple loss values and preset requirements, the network parameters in the initial convolutional neural network model are adjusted to obtain the target convolutional neural network model, including: If multiple loss values are all less than or equal to the preset value, then the initial convolutional neural network model is determined as the target convolutional neural network model; Otherwise, adjust the network parameters in the initial convolutional neural network model to obtain an intermediate convolutional neural network model; The training samples are processed using a transitional convolutional neural network model to obtain several new loss values; When multiple new loss values are all less than or equal to the preset value, the corresponding transitional convolutional neural network model is determined as the target convolutional neural network model.
[0039] In the embodiment provided in this application, if multiple loss values are all less than or equal to preset values, it indicates that the current model has reached a convergence state that meets the accuracy requirements. At this point, the initial convolutional neural network model is determined as the target convolutional neural network model. Otherwise, it indicates that the current model has not yet reached a convergence state that meets the accuracy requirements. In this case, the network parameters in the initial convolutional neural network model are continuously adjusted until multiple new loss values corresponding to the adjusted transitional convolutional neural network model are all less than or equal to preset values, so that the model reaches a convergence state that meets the accuracy requirements. That is, the predicted latitude and longitude of the typhoon center and the central pressure value output by the model gradually approach the true values. At this point, the corresponding transitional convolutional neural network model is determined as the target convolutional neural network model, which can guarantee the prediction accuracy of the final target convolutional neural network model, thereby improving the accuracy of typhoon center identification based on the model. The network parameters include the convolutional kernel weights and bias terms of the convolutional layers, the scaling and translation parameters of the batch normalization layers, and the weights and bias terms of the fully connected layers. The number of parameters in the training samples can be 2 million. The initial convolutional neural network model uses the Adam optimizer (adaptive moment estimation algorithm with a learning rate of 0.001).
[0040] In one exemplary embodiment provided in this application, after the model training is completed, the target convolutional neural network model can also be applied to typhoon cases that have not participated in the training for verification, so as to evaluate the model's generalization ability and recognition accuracy.
[0041] Taking the WRF model forecast output as an example, the model output field during the path of the verification typhoon is selected. The parameters included in the multi-layer meteorological element field verification data are extracted, including sea level pressure, lower-level 10 wind field components (u, v), topographic height field, 500 hPa geopotential height field, grid longitude, and grid latitude. After inputting the multi-layer meteorological element field verification data into the target convolutional neural network model, the corresponding time-specific center latitude and longitude verification coordinates and center pressure verification values of the verification typhoon are automatically output. Then, the center latitude and longitude verification coordinates and center pressure verification values are compared and verified with the actual center latitude and longitude coordinates and actual center pressure values of the verification typhoon. Thus, after verification on multiple typhoon processes, the verification results demonstrate that the recognition accuracy of the target convolutional neural network model in this application is higher than 99%, and the recognition results are stable and reliable.
[0042] In summary, the typhoon center identification method proposed in this application has the following significant advantages: ① High identification accuracy under complex terrain: The model can learn the spatial structural features of the pressure field and wind field, and can still stably identify the center location even when the typhoon makes landfall or in complex terrain areas; ② Automatic adaptation to different modes: By including output samples from different modes in the training set, the model has good generalization ability and can adapt to the input of different regions and different numerical weather forecast models; ③ Reduced manual intervention: Compared with traditional diagnostic algorithms, no manual parameter adjustment is required, achieving fully automatic identification; ④ Strong operational integration: The model can be directly embedded into existing operational forecasting processes, improving automated analysis and real-time monitoring capabilities.
[0043] Please see Figure 3 , Figure 3 A typhoon center identification system is illustrated in an exemplary embodiment of this application, such as... Figure 3 As shown, this application provides a typhoon center identification system 300, including: The sample construction module 301 is used to construct training samples, which include historical multi-layer meteorological element field data of multiple historical typhoons. The model training module 302 is used to train the initial convolutional neural network model using training samples based on a preset loss function to obtain the target convolutional neural network model. Data acquisition module 303 is used to acquire multi-layer meteorological element field data of the target typhoon's location; The typhoon center identification module 304 is used to input the multi-layer meteorological element field data of the target into the target convolutional neural network model for identification, and obtain the target center parameters of the target typhoon, including the center latitude and longitude coordinates and the center pressure value.
[0044] The typhoon center identification system 300 provided in this application trains an initial convolutional neural network model using a model training module 302 based on a preset loss function and training samples constructed by a sample construction module 301, which include historical multi-layer meteorological element field data of multiple historical typhoons. This results in a target convolutional neural network model. The typhoon center identification module 304 then inputs the target multi-layer meteorological element field data of the target typhoon's location, acquired by the data acquisition module 303, into the target convolutional neural network model for identification, thus obtaining the target typhoon's center parameters. In this way, by using multi-channel (multi-layer) historical typhoon data for model training, a target convolutional neural network model that meets the prediction accuracy requirements can be obtained. Using this model for typhoon center identification can reduce identification bias caused by terrain influence and model errors, thereby improving the accuracy of typhoon center identification.
[0045] Optionally, the model training module 302 is specifically used for: Using an initial convolutional neural network model, feature extraction was performed on historical multi-layer meteorological element field data to obtain central features; Global feature classification is performed on the central features to obtain the predicted center parameters; Based on a preset loss function, the initial convolutional neural network model is optimized using multiple prediction center parameters to obtain the target convolutional neural network model.
[0046] Optionally, the model training module 302 is specifically used for: Multi-level convolution is performed on historical multi-level meteorological element field data to obtain initial central feature data; The initial central feature data is activated to obtain nonlinear feature data; Batch normalization is performed on the nonlinear feature data to obtain the central feature.
[0047] Optionally, the model training module 302 is specifically used for: Global average pooling is applied to the central features to obtain the global feature vector; The global feature vector is convolved to obtain the prediction center parameters.
[0048] Optionally, the model training module 302 is specifically used for: The predicted center parameters and the corresponding true center parameters are substituted into the preset loss function to calculate the loss value. Based on multiple loss values and preset requirements, the network parameters in the initial convolutional neural network model are adjusted to obtain the target convolutional neural network model.
[0049] Optionally, in model training module 302, the loss function is formulated as follows: in, Indicates the loss value. This represents the first weighting coefficient. This represents the spherical distance between the predicted center parameters and the corresponding true center parameters. This represents the second weighting coefficient. This indicates that the forecast center parameters include the forecast center pressure value. This indicates that the true center parameter includes the true center pressure value. This represents the latitude and longitude coordinates of the prediction center corresponding to the prediction center parameters. The true center latitude and longitude coordinates represent the true center parameters. This represents the Earth's radius.
[0050] Optionally, the model training module 302 is specifically used for: If multiple loss values are all less than or equal to the preset value, then the initial convolutional neural network model is determined as the target convolutional neural network model; Otherwise, adjust the network parameters in the initial convolutional neural network model to obtain an intermediate convolutional neural network model; The training samples are processed using a transitional convolutional neural network model to obtain several new loss values; When multiple new loss values are all less than or equal to the preset value, the corresponding transitional convolutional neural network model is determined as the target convolutional neural network model.
[0051] It should be noted that the typhoon center identification system and the typhoon center identification method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the typhoon center identification system provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.
[0052] A computing device according to an embodiment of this application includes a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, it implements some or all of the steps of the typhoon center identification method described above.
[0053] The computing device can be a computer, and the corresponding program is computer software. The parameters and steps in the computing device described above can be referred to the parameters and steps in the embodiment of the typhoon center identification method above, and will not be repeated here.
[0054] This application embodiment provides a computer-readable storage medium storing instructions that, when executed, perform the steps of the aforementioned typhoon center identification method.
[0055] The computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.
[0056] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of this disclosure. The aforementioned computer-readable storage medium can be a non-transitory computer-readable storage medium, including: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, and other media capable of storing program code; it can also be a transient computer-readable storage medium.
[0057] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0058] Those skilled in the art will recognize that this application can be implemented as a system, method, or computer program product. Therefore, this disclosure can be implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "module" or "system." Furthermore, in some embodiments, this application can also be implemented as a computer program product contained in one or more computer-readable media, which contains computer-readable program code. Computer-readable storage media can be, for example, but not limited to—electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof.
[0059] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0060] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.
Claims
1. A method for identifying the center of a typhoon, characterized in that, include: Construct training samples, which include historical multi-layer meteorological element field data of multiple historical typhoons; Based on a preset loss function, the initial convolutional neural network model is trained using the training samples to obtain the target convolutional neural network model. Acquire multi-layer meteorological element field data of the target typhoon's location; The target multi-layer meteorological element field data is input into the target convolutional neural network model for identification to obtain the target center parameters of the target typhoon, which include the center latitude and longitude coordinates and the center pressure value.
2. The method according to claim 1, characterized in that, The process of training an initial convolutional neural network model using the training samples based on a preset loss function to obtain a target convolutional neural network model includes: Using an initial convolutional neural network model, feature extraction is performed on the historical multi-layer meteorological element field data to obtain the central features; The central features are subjected to global feature classification processing to obtain the predicted center parameters; Based on a preset loss function, the initial convolutional neural network model is optimized using multiple prediction center parameters to obtain the target convolutional neural network model.
3. The method according to claim 2, characterized in that, The process of extracting features from the historical multi-layer meteorological element field data to obtain central features includes: Multi-layer convolution is performed on the historical multi-layer meteorological element field data to obtain initial central feature data; The initial central feature data is activated to obtain nonlinear feature data; The nonlinear feature data is batch normalized to obtain the central feature.
4. The method according to claim 2, characterized in that, The step of performing global feature classification on the central features to obtain the predicted center parameters includes: The central feature is subjected to global average pooling to obtain a global feature vector; The global feature vector is convolved to obtain the prediction center parameters.
5. The method according to claim 2, characterized in that, The optimization of the initial convolutional neural network model based on a preset loss function and using multiple prediction center parameters to obtain the target convolutional neural network model includes: The predicted center parameters and the corresponding true center parameters are substituted into a preset loss function for calculation to obtain the loss value; Based on multiple loss values and preset requirements, the network parameters in the initial convolutional neural network model are adjusted to obtain the target convolutional neural network model.
6. The method according to claim 5, characterized in that, The formula for the loss function is as follows: in, Indicates the loss value. This represents the first weighting coefficient. This represents the spherical distance between the predicted center parameters and the corresponding true center parameters. This represents the second weighting coefficient. This indicates that the forecast center parameters include the forecast center pressure value. This indicates that the true center parameter includes the true center pressure value. This represents the latitude and longitude coordinates of the prediction center corresponding to the prediction center parameters. The true center latitude and longitude coordinates represent the true center parameters. This represents the Earth's radius.
7. The method according to claim 5, characterized in that, The step of adjusting the network parameters in the initial convolutional neural network model based on multiple loss values and preset requirements to obtain the target convolutional neural network model includes: If multiple loss values are all less than or equal to a preset value, then the initial convolutional neural network model is determined as the target convolutional neural network model; Otherwise, adjust the network parameters in the initial convolutional neural network model to obtain an intermediate convolutional neural network model; The training samples are processed using the transitional convolutional neural network model to obtain multiple new loss values; When multiple new loss values are all less than or equal to the preset value, the corresponding transitional convolutional neural network model is determined as the target convolutional neural network model.
8. A typhoon center identification system, characterized in that, include: The sample construction module is used to construct training samples, which include historical multi-layer meteorological element field data of multiple historical typhoons. The model training module is used to train the initial convolutional neural network model using the training samples based on a preset loss function, so as to obtain the target convolutional neural network model. The data acquisition module is used to acquire multi-layer meteorological element field data of the target typhoon's location. The typhoon center identification module is used to input the multi-layer meteorological element field data of the target into the target convolutional neural network model for identification, and obtain the target center parameters of the target typhoon, which include the center latitude and longitude coordinates and the center pressure value.
9. A computing device, comprising a memory, a processor, and a program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the steps of a typhoon center identification method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the steps of a typhoon center identification method as described in any one of claims 1 to 7.