AI tropical cyclone prediction method and system based on sparse data fusion, electronic equipment and program product

Abstract

The invention provides an AI tropical cyclone prediction method and system based on sparse data fusion, electronic equipment and a program product. The AI tropical cyclone prediction method comprises the following steps: a sparse observation fusion step: obtaining input data including original sparse observation data and IFS-HRES mode forecast data, performing data fusion on the input data by using a data assimilation model, and outputting an initial field; an AI weather prediction step: inputting the initial field into an AI weather prediction model for weather prediction, obtaining weather prediction data at different moments, and extracting tropical cyclone local prediction data at different moments from the weather prediction data at different moments; a deviation correction step: inputting the local prediction results of the tropical cyclone at different moments into an AI deviation correction model, and outputting correction results of the tropical cyclone at different moments so as to obtain a corrected tropical cyclone strength prediction result, the AI deviation correction model being a Conv2Former-LSTM hybrid model.

Description

Technical Field

[0001] The present application relates to the field of weather forecasting, and in particular to an AI tropical cyclone prediction method, system, electronic equipment and program product based on sparse data fusion. Background Art

[0002] Current AI meteorological models are generally weak in tropical cyclone intensity prediction, which is mainly due to the model's error accumulation in dealing with long time scales and insufficient forecasting of small-scale structures. Traditional tropical cyclone intensity prediction methods rely on TC tracking algorithms to track and predict the development of tropical cyclones through constraints on model meteorological variables. However, these methods are not effective in intensity prediction, especially for extreme tropical cyclone events, so it is necessary to develop a prediction method that can improve the accuracy of tropical cyclone intensity forecasts. Summary of the invention

[0003] In order to solve at least some of the above technical problems, the present application provides an AI tropical cyclone prediction method, system, electronic equipment and program product based on sparse data fusion, which aims to correct the intensity deviation of the existing model through machine learning technology and provide support for more accurate tropical cyclone forecasts.

[0004] The first aspect of the present application provides an AI tropical cyclone prediction method based on sparse data fusion, comprising the following steps: a sparse observation fusion step, obtaining input data, the input data including original sparse observation data and IFS-HRES model forecast data, using a data assimilation model to perform data fusion on the input data, and outputting an initial field; an AI weather prediction step, inputting the initial field into an AI weather prediction model for prediction, obtaining weather forecast data at different times, and extracting tropical cyclone local forecast data at different times from the weather forecast data at different times; a deviation correction step, inputting the tropical cyclone local forecast data at different times into an AI deviation correction model, outputting tropical cyclone corrected data at different times, thereby obtaining a corrected tropical cyclone intensity prediction result, wherein the AI ​​deviation correction model is a Conv2Former-LSTM hybrid model.

[0005] In the first aspect of the present application, the data assimilation model is one of the following models: an AI assimilation model, a four-dimensional variational assimilation model, an ensemble Kalman filter model, or a hybrid data assimilation model.

[0006] In the first aspect of the present application, the sparse observation fusion step also includes: preprocessing the original sparse observation data to obtain sparse grid observation data, and using the data assimilation model to perform data fusion on the sparse grid observation data and the IFS-HRES model forecast data.

[0007] In the first aspect of the present application, the sparse observation fusion step further includes: the input data also includes ocean variable data provided by a hybrid coordinate ocean model, and the ocean variable data includes sea surface temperature and ocean current velocity.

[0008] The second aspect of the present application provides an AI tropical cyclone prediction system based on sparse data fusion, comprising: a sparse observation fusion unit, the sparse observation fusion unit is used to obtain input data, the input data includes original sparse observation data and IFS-HRES model forecast data, the input data is fused using a data assimilation model, and an initial field is output; an AI weather prediction unit, the AI ​​weather prediction unit is used to input the initial field into an AI weather prediction model for prediction, obtain weather prediction data at different times, and extract tropical cyclone local prediction data at different times from the weather prediction data at different times; a deviation correction unit, the deviation correction unit is used to input the tropical cyclone local prediction data at different times into an AI deviation correction model, and output tropical cyclone correction data at different times, thereby obtaining a corrected tropical cyclone intensity prediction result, wherein the AI ​​deviation correction model is a Conv2Former-LSTM hybrid model.

[0009] The third aspect of the present application provides an electronic device, comprising: one or more processors; one or more memories; the one or more memories storing one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device executes the AI ​​tropical cyclone prediction method described in any one of claims 1 to 5.

[0010] The fourth aspect of the present application provides a computer program product, comprising a computer program / instruction, which, when executed by a processor, implements the AI ​​tropical cyclone prediction method according to any one of claims 1 to 5.

[0011] The AI ​​tropical cyclone prediction method provided in this application significantly reduces the error of tropical cyclone intensity prediction, improves the prediction accuracy, and achieves faster calculation speed and higher model efficiency through technical means such as sparse observation data fusion, introduction of ocean variables, and error correction using the Conv2Former-LSTM hybrid model. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Flowchart of an AI tropical cyclone prediction method according to an embodiment of the present application;

[0013] Figure 2 is a schematic diagram of an AI tropical cyclone prediction system according to an embodiment of the present application;

[0014] Figure 3 It is a structural schematic diagram of an electronic device according to an embodiment of the present application;

[0015] Figure 4 This is a schematic diagram of the Conv2former model architecture;

[0016] Figure 5 This is an intensity correction effect diagram of the AI ​​tropical cyclone prediction method according to the present application;

[0017] Figure 6 Another intensity correction effect diagram of the AI ​​tropical cyclone prediction method according to the present application;

[0018] Figure 7 This is another intensity correction effect diagram of the AI ​​tropical cyclone prediction method according to the present application. DETAILED DESCRIPTION

[0019] The present application is further described below in conjunction with specific embodiments and drawings. It is to be understood that the illustrative embodiments of the present disclosure are only intended to explain the present application, rather than to limit the present application. In addition, for ease of description, only some, but not all, structures or processes related to the present application are shown in the drawings.

[0020] The following specific embodiments illustrate the implementation of the present application, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Although the description of the present application will be introduced in conjunction with the preferred embodiment, this does not mean that the features of this invention are limited to this implementation. On the contrary, the purpose of introducing the invention in conjunction with the implementation is to cover other options or modifications that may extend based on the claims of the present application. In order to provide an in-depth understanding of the present application, the following description will include many specific details. The present application can also be implemented without using these details. In addition, in order to avoid confusion or blurring the focus of the present application, some specific details will be omitted in the description.

[0021] Unless the context dictates otherwise, the terms "comprising," "having," and "including" are synonymous. The phrase "A / B" means "A or B." The phrase "A and / or B" means "(A and B) or (A or B)."

[0022] It should be understood that although the terms "first", "second", "one", "another", etc. may be used herein to describe various components, units, data, or things, these components, units, data, or things should not be limited by these terms. These terms are used only to distinguish one feature from another. For example, without departing from the scope of the exemplary embodiment, a first feature may be referred to as a second feature, and similarly a second feature may be referred to as a first feature. In addition, one may be referred to as another, and similarly another may be referred to as one.

[0023] It should be noted that in this specification, similar reference numerals and letters represent the same or similar items in the drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0024] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0025] Figure 1 FIG. 1 is a flow chart of an AI tropical cyclone prediction method according to an embodiment of the present application. Figure 1 As shown, the present application provides an AI tropical cyclone prediction method based on sparse data fusion, which includes the following steps.

[0026] Step S1, sparse observation fusion step, obtains input data, the input data includes original sparse observation data and IFS-HRES model forecast data, uses a data assimilation model to perform data fusion on the input data, and outputs an initial field.

[0027] Raw sparse observation data refers to observation data that is relatively discrete and non-dense in terms of temporal and spatial distribution, with large intervals between adjacent observation values ​​and limited overall coverage completeness and detail.

[0028] The sources of original sparse observation data mainly include the following aspects.

[0029] First, data from ground-based meteorological observation stations. This is one of the most common sources of observation data, including observations of meteorological elements such as temperature, air pressure, humidity, wind direction, wind speed, precipitation, and sunshine. Ground-based meteorological observation stations are distributed in different geographical locations, but are sparsely distributed in some remote areas, oceans, deserts, and other places due to factors such as construction costs and terrain.

[0030] The second aspect is high-altitude meteorological observation data. The information on high-altitude meteorological elements, such as wind direction, wind speed, air pressure, temperature and humidity, is obtained mainly through sounding balloons, aircraft detection, weather radar and other means. Among them, there are relatively few observation sites for sounding balloons, and the observation time interval is long, so the data sparsity is more obvious.

[0031] The third aspect is satellite remote sensing data. Satellite remote sensing technology can continuously observe the meteorological conditions of the earth's surface and atmosphere over a large area, such as cloud cover, surface temperature, water vapor content, etc. However, the spatial resolution and temporal resolution of satellite observation data are limited, and in some areas, they may be affected by weather conditions, satellite orbits and other factors, resulting in missing or incomplete data.

[0032] Fourthly, ocean buoy observation data is used to monitor meteorological elements and marine environmental parameters on the ocean. However, the number of ocean buoys is limited and their distribution range is wide. Data acquisition has certain temporal and spatial limitations, especially in vast ocean areas, where data sparsity is a prominent problem.

[0033] IFS-HRES (Integrated Forecast System High Resolution) model forecast data is weather forecast data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). The IFS-HRES model forecast data selected in this application has the following characteristics.

[0034] In terms of temporal and spatial resolution, the horizontal spatial resolution is about 0.25°×0.25° longitude and latitude, which can provide detailed geographic spatial information. The data is divided and collected at intervals of 0.25°, which can more accurately depict the changes in meteorological elements. The time step is 6 hours, and the weather conditions in the next 10 days can be forecast. There is a set of meteorological element forecast values ​​every 6 hours, which can better capture short-term change trends and meet short-term to medium-term forecast needs.

[0035] The IFS-HRES model forecast data includes the following variables: z, t, u, v of the isobaric surface variables 200hPa, 500hPa, 700hPa, and 850hPa, as well as surface variables t2m, u10, v10, MSL, SST, etc. Among them, 200hPa, 500hPa, 700hPa, and 850hPa in the isobaric surface variables represent different pressure height levels, z represents the potential height, t represents the temperature, u and v represent the zonal wind speed and the meridional wind speed, respectively. Among the surface variables, t2m represents the temperature at a height of 2 meters, u10 and v10 represent the zonal wind speed and the meridional wind speed, respectively, at a height of 10 meters, MSL represents the mean sea level pressure, and SST represents the sea surface temperature.

[0036] Data assimilation model is a technical means to combine observation data with numerical forecast models, which can improve the accuracy and reliability of weather forecasts. In the field of weather forecasting, data assimilation models mainly include the following types.

[0037] AI assimilation model: AI assimilation model is a new method of data assimilation using artificial intelligence technology (especially deep learning). Its core function is to combine multi-source observation data (such as satellites, ground stations, radars, etc.) with numerical weather forecast models to generate more accurate initial conditions. Compared with traditional data assimilation methods, AI assimilation models rely on an end-to-end data-driven approach. By learning the complex nonlinear relationship between observation data and model states, it can automatically extract features and complete the entire process from input data to assimilated output. In addition, AI models can efficiently process massive amounts of data, especially in areas where observations are sparse or unevenly distributed (such as oceans, polar regions, and central Africa), showing excellent capabilities, providing the possibility of improving the balanced performance of global weather forecasts.

[0038] Specifically, the AI ​​assimilation model can be a deep learning-based data assimilation framework (DL-based DAframework). Typical deep learning-based data assimilation framework components include: encoder-decoder structure and attention mechanism.

[0039] Encoder-decoder structure: The encoder is used to reduce the dimensionality of the input data (including observation data and model data), extract key features and map them to a latent space. For example, when processing multi-source meteorological data, the encoder can fuse and compress data of different resolutions and types (such as temperature, wind speed, etc.) into a unified feature representation. The decoder is responsible for remapping the features of the latent space back to the original data space and generating the assimilated system state estimate.

[0040] Attention mechanism: The attention mechanism enables the model to focus on the key parts of the data during the data assimilation process. For example, when assimilating meteorological data, for a developing tropical cyclone system, the attention mechanism can guide the model to focus on observational data related to the storm (such as wind speed and air pressure observations near the center of the storm) instead of treating all data equally, thereby improving the efficiency and accuracy of assimilation.

[0041] Four-dimensional variational assimilation (4D-Var) model: The three-dimensional variational assimilation (3D-Var) model finds an optimal analysis field by minimizing the difference between the observed data and the model background field, so that the analysis field is as close to the background field as possible while satisfying the observation constraints. The 4D-Var model is based on the 3D-Var model and takes the time dimension into account. That is, within an assimilation window, the observation data at multiple times are used to adjust the initial field and trajectory of the model, thereby making better use of the spatiotemporal information of the observation data. It can more accurately describe the spatiotemporal evolution of meteorological elements and improve the ability to predict the development and changes of weather systems.

[0042] Ensemble Kalman Filter (EnKF) Model: The Kalman Filter (KF) model is a filtering method based on linear systems. It continuously corrects the state estimation of the model according to new observation data through two steps: prediction and update. The KF model can effectively deal with noise and uncertainty in linear systems, but its application is limited to systems with strong nonlinear characteristics such as meteorological systems. In order to overcome this problem, the Ensemble Kalman Filter model based on the Monte Carlo method generates multiple ensemble forecast members and uses the differences between the ensemble members to estimate the error covariance of the model, thereby better dealing with nonlinear and non-Gaussian problems. The EnKF model can effectively assimilate various observation data, including conventional observation data and satellite remote sensing data, and shows good performance when processing sparse observation data. It has been widely used in meteorological numerical forecasting and data assimilation systems.

[0043] Hybrid data assimilation model: In order to give full play to the advantages of different assimilation models and improve the effect of data assimilation, some hybrid data assimilation models have been proposed. For example, 3D-Var is combined with EnKF, and the efficient computing power of 3D-Var and the processing ability of EnKF for nonlinear problems are used to improve the assimilation accuracy while ensuring the computing efficiency.

[0044] In one embodiment of the present invention, the sparse observation fusion step further includes preprocessing the original sparse observation data to obtain sparse grid observation data, and using a data assimilation model to perform data fusion on the sparse grid observation data and the IFS-HRES model forecast data.

[0045] The preprocessing of the original sparse observation data to obtain sparse grid observation data mainly includes the following steps: data cleaning, data interpolation, coordinate transformation and grid division.

[0046] Data cleaning includes checking and handling missing values, and removing outliers.

[0047] Check and process missing values: Since the original sparse observation data has a large missing ratio, consider filling it based on spatiotemporal correlation. For example, use the temperature data of adjacent stations at the same time for interpolation filling, or use the temperature change pattern of the same station at different times for filling.

[0048] Removing outliers: Outliers may be caused by instrument failure, human error, or extreme meteorological events. Outliers can be identified by statistical methods, such as using the 3-times standard deviation method. Assuming that the data of a certain meteorological element follows a normal distribution, data points that exceed the mean ±3 times the standard deviation can be considered outliers. For identified outliers, you can choose to delete or correct them. If the outlier is caused by an instrument failure, and the data point has reasonable before and after values ​​in the time series, the average of the before and after values ​​can be used for correction.

[0049] Data interpolation includes spatial interpolation and temporal interpolation.

[0050] Spatial interpolation: Since the observation sites are sparsely distributed, spatial interpolation is required to construct a sparse grid. Commonly used spatial interpolation methods include Inverse Distance Weighted Interpolation (IDW) and Kriging Interpolation (Kriging Interpolation).

[0051] The basic principle of inverse distance weighted interpolation (IDW) is to assume that the attribute value of the interpolation point is affected by the surrounding known observation points, and the degree of influence is inversely proportional to the distance. For example, for the temperature interpolation of a grid point, the observation stations closer to the grid point contribute more to its temperature value, while those farther away contribute less.

[0052] Kriging interpolation is an interpolation method based on spatial autocorrelation. It assumes that meteorological elements have a certain degree of autocorrelation in space, and describes this autocorrelation by constructing a semivariogram. For example, when performing Kriging interpolation on precipitation data, the spatial autocorrelation of precipitation data must first be analyzed, and a suitable semivariogram model (such as a spherical model, an exponential model, etc.) must be determined. Then, the estimated value of the interpolation point is calculated based on the data of the known observation points and the semivariogram. Compared with IDW, Kriging interpolation can better take into account the spatial structure and autocorrelation of data, and has better application effects in some areas with complex terrain and complex changes in meteorological elements.

[0053] Time interpolation: If the data in the time series is discontinuous, time interpolation is required. For time series with equal time intervals, linear interpolation can be used.

[0054] Coordinate transformation and meshing include two parts: coordinate transformation and meshing.

[0055] Coordinate conversion: According to actual needs, the geographic coordinates of the observation data, such as longitude and latitude, are converted into plane rectangular coordinates, which is more convenient for grid division and subsequent calculations. For example, map projection methods such as Mercator projection and Lambert projection are used to convert the longitude and latitude coordinates on the earth's surface into plane coordinates. When performing projection, it is necessary to consider the deformation properties and applicable scope of the projection. For example, the Mercator projection has less deformation in low-latitude areas and is suitable for use in fields such as navigation; the Lambert projection maintains the area unchanged in mid-latitude areas and is suitable for meteorological regional analysis.

[0056] Grid division: Grid the data according to the size of the study area and the required resolution. For example, to study the meteorological conditions of a region, determine the grid resolution to be 0.25°×0.25° longitude and latitude, and then divide the entire area into multiple grid cells according to this resolution. For each grid cell, the corresponding meteorological element value can be assigned according to the interpolation result to obtain sparse grid observation data. When dividing the grid, boundary processing also needs to be considered. For example, for grid cells close to the boundary of the study area, special interpolation methods or boundary conditions may be required to determine their meteorological element values.

[0057] In one embodiment of the present application, the AI ​​assimilation model is preferentially used as the data assimilation model to fuse the sparse grid observation data and the IFS-HRES model forecast data, thereby outputting the initial field.

[0058] In one embodiment of the present application, the sparse observation fusion step further includes: the input data includes ocean variable data provided by a hybrid coordinate ocean model, and the ocean variable data includes sea surface temperature and ocean current velocity.

[0059] The Hybrid Coordinate Ocean Model (HYCOM) is an advanced ocean numerical model that can provide ocean physical field data and ocean dynamic process related data. Ocean physical field data include temperature field data, salinity field data and ocean current data. By integrating ocean variable data such as sea surface temperature and ocean current speed into the input data, the model's ability to predict tropical cyclone intensity can be further enhanced, significantly improving the overall prediction accuracy of tropical cyclone behavior, especially in the response to extreme weather events.

[0060] Step S2, AI weather prediction step, inputs the initial field into the AI ​​weather prediction model for prediction, obtains weather prediction data at different times, and extracts tropical cyclone local prediction data at different times from the weather prediction data at different times.

[0061] The initial field preferably includes weather data at at least two time points, for example, weather data at time t and weather data at time t-6, where time t-6 is 6 hours before time t.

[0062] In one embodiment of the present application, Fudan University's Fuxi model is used as an AI weather forecast model. Based on the weather data provided by the initial field, the Fuxi model can output a series of weather forecast data at different times with the same data structure, for example, weather forecast data at t+6, t+12, ..., t+240. Based on the characteristics of tropical cyclones themselves, such as low central pressure and high wind speed, the location of the tropical cyclone can be identified in the weather forecast data at different times above, thereby extracting local forecast data for tropical cyclones at different times, for example, local forecast data for tropical cyclones is data within ±5° of longitude and latitude centered on the tropical cyclone.

[0063] Step S3, the deviation correction step, inputs the tropical cyclone local prediction data at different times into the AI ​​deviation correction model, outputs the tropical cyclone correction data at different times, and thus obtains the corrected tropical cyclone intensity prediction result, wherein the AI ​​deviation correction model is a Conv2Former-LSTM hybrid model.

[0064] The Conv2former model is a new architecture that combines the advantages of convolutional neural networks (ConvNet) and visual transformers. The Conv2former model simplifies the self-attention mechanism. By comparing the design principles of convolutional neural networks and visual transformers, the Conv2former model uses convolution modulation operations to simplify the self-attention mechanism. This operation can better utilize the relatively large convolution kernels nested in the convolution layer, thereby more effectively encoding spatial features. Moreover, the Conv2former model retains the effectiveness and local receptive field advantages of ConvNet, can excellently extract local features of input data, and retain key meteorological information. At the same time, it introduces the global interaction characteristics of the Transformer, which can handle long-distance dependencies in the input data, so that the model can better capture the complex relationship of tropical cyclones during their development.

[0065] Figure 4 This is a schematic diagram of the Conv2former model architecture. Figure 4In the figure, “Patch Embed” means patch embedding block, “Conv Block” means convolution block, “FC” means fully connected layer, “Lat” means latitude, “Lon” means longitude, “MSL” means mean sea level pressure, “WS10” means level 10 wind, “Conv” means convolution, “DConv” means depthwise separable convolution, and “Hadamard Prod” means Hadamard product.

[0066] like Figure 4 As shown in the figure, the Conv2former model adopts a pyramid architecture with four stages, each with different feature map resolutions. The input of the Conv2former model is the local prediction data of tropical cyclones at different times, which can be considered as images with C×H×W dimensions, where C is the number of channels, H is the height of the image, and W is the width of the image. Each stage includes multiple convolution blocks. Between two adjacent stages, a patch embedding block is used to reduce the resolution. The patch embedding block is usually a 2×2 convolution with a stride of 2. Figure 4 The “×M2” in the figure indicates that there are two more stages with the same structure.

[0067] The convolution kernel size of the Conv2former model can be selected according to actual needs, usually between 5×5 and 21×21, with 11×11 being the preferred size.

[0068] LSTM (Long-Short Term Memory), or long short-term memory network, is a special recurrent neural network (RNN) designed to solve the gradient vanishing and exploding problems of traditional RNN when processing long sequences. Its core components include: Input Gate, which is used to control the impact of current input on memory cells; Forget Gate, which is used to decide how much past information to retain; and Output Gate, which is used to decide the output of the current memory cell.

[0069] LSTM can capture long-term dependencies. Through the gate mechanism, LSTM can effectively memorize and utilize information over a long time span, and is suitable for tasks that require understanding of dynamic changes in time. Moreover, LSTM has a stable training process, and the gate mechanism effectively alleviates the gradient vanishing and explosion problems, ensuring stable training and prediction of the model on long sequences.

[0070] The AI ​​bias correction model of this application is a Conv2Former-LSTM hybrid model, which combines the respective advantages of the Conv2Former model and the LSTM model to correct the intensity of tropical cyclones.

[0071] The Conv2Former-LSTM hybrid model includes: an input layer that accepts spatiotemporal data (for example, local prediction data of tropical cyclones at different times in this application) as input; a Conv2Former module that extracts spatial features for each frame and generates a high-dimensional spatial feature representation; a feature sequence construction that arranges the spatial features of each frame in chronological order to form time series data; an LSTM module that performs temporal modeling on the spatial feature sequence to capture dynamic changes and long-term dependencies in the sequence; and an output layer that generates a final output based on a specific task (for example, outputting tropical cyclone correction data at different times, including latitude, longitude, average sea level pressure, etc.).

[0072] The Conv2Former module and the LSTM module can be optimized through joint training to ensure that the two modules work together to improve the overall model performance. In addition, the Conv2Former module can be trained first to extract high-quality spatial features, and then the LSTM module can be trained for temporal modeling, thereby gradually optimizing the Conv2Former-LSTM hybrid model.

[0073] The AI ​​deviation correction model can output tropical cyclone correction data at different times, for example, tropical cyclone correction data at t+6, t+12,..., t+240, so as to obtain the corrected tropical cyclone intensity forecast results.

[0074] The AI ​​tropical cyclone prediction method of the present application significantly reduces the error of tropical cyclone intensity prediction, improves the prediction accuracy, and achieves faster calculation speed and higher model efficiency through technical means such as sparse observation data fusion, introduction of ocean variables, and error correction using the Conv2Former-LSTM hybrid model.

[0075] Figure 2 FIG. 1 is a schematic diagram of an AI tropical cyclone prediction system according to an embodiment of the present application. Figure 2 As shown, the AI ​​tropical cyclone prediction system 20 includes a sparse observation fusion unit 201, an AI weather prediction unit 202 and a deviation correction unit 203.

[0076] The sparse observation fusion unit 201 is used to obtain input data, which includes original sparse observation data and IFS-HRES model forecast data, and use the data assimilation model to perform data fusion on the input data to output the initial field.

[0077] The AI ​​weather prediction unit 202 is used to input the initial field into the AI ​​weather prediction model for prediction, obtain weather prediction data at different times, and extract tropical cyclone local prediction data at different times from the weather prediction data at different times.

[0078] The deviation correction unit 203 is used to input the tropical cyclone local prediction data at different times into the AI ​​deviation correction model, and output the tropical cyclone correction data at different times, so as to obtain the corrected tropical cyclone intensity prediction result, wherein the AI ​​deviation correction model is a Conv2Former-LSTM hybrid model.

[0079] Figure 5-7 This is a diagram showing the intensity correction effect of the AI ​​tropical cyclone prediction method according to the present application. Figure 5-7 The horizontal axis is the time step, and the vertical axis is the minimum air pressure (hPa), which respectively shows the intensity correction effects of Hurricane Fabian, Hurricane Idalia, and Hurricane Calvin. Among them, the black square solid line represents the true value of the hurricane's minimum air pressure, the dotted solid line is the predicted value before correction, and the dotted line is the predicted value after correction. Since the central air pressure of a tropical cyclone is negatively correlated with its intensity, generally speaking, the lower the central air pressure of a tropical cyclone, the stronger the tropical cyclone intensity, and the higher the central air pressure, the weaker the tropical cyclone intensity, so the air pressure data of a tropical cyclone can be used to represent its intensity. Figure 5-7 As shown, the intensity of the tropical cyclone corrected by the AI ​​tropical cyclone prediction method of the present application is closer to the actual value of the intensity than before correction, which has a significant beneficial effect.

[0080] Figure 3 The figure shows a schematic diagram of the structure of an electronic device 30 according to an embodiment of the present application.

[0081] like Figure 3 As shown, in some embodiments, the electronic device 30 may include one or more processors 300, a system control logic 301 connected to at least one of the processors 300, a system memory 302 connected to the system control logic 301, a non-volatile memory (NVM) 303 connected to the system control logic 301, and a network interface 304 connected to the system control logic 301.

[0082] In some embodiments, the processor 300 may include one or more single-core or multi-core processors. In some embodiments, the processor 300 may include any combination of a general-purpose processor and a special-purpose processor (e.g., a graphics processor, an application processor, a baseband processor, etc.). For example, the processor 300 may include any combination of a general-purpose processor and a special-purpose processor required to implement the aforementioned AI tropical cyclone prediction method.

[0083] In some embodiments, system control logic 301 may include any suitable interface controller to provide any suitable interface to at least one of processors 300 and / or any suitable device or component in communication with system control logic 301 .

[0084] In some embodiments, the system control logic 301 may include one or more memory controllers to provide an interface to the system memory 302. The system memory 302 may be used to load and store data and / or instructions. In some embodiments, the system memory 302 of the electronic device 30 may include any suitable volatile memory, such as a suitable dynamic random access memory (DRAM).

[0085] The non-volatile memory 303 may include one or more tangible, non-transitory computer-readable media for storing data and / or instructions. In some embodiments, the non-volatile memory 303 may include any suitable non-volatile memory such as flash memory and / or any suitable non-volatile storage device, such as at least one of a HDD (hard disk drive), a CD (compact disc) drive, and a DVD (digital versatile disc) drive.

[0086] The non-volatile memory 303 may include a portion of storage resources installed on a device of the electronic device 30 , or it may be accessible by the device but not necessarily a portion of the device. For example, the non-volatile memory 303 may be accessed over a network via the network interface 304 .

[0087] In particular, the system memory 302 and the non-volatile memory 303 / storage 516 may include: a temporary copy and a permanent copy of the instruction 305, respectively. The instruction 305 may include: when executed by at least one of the processors 300, the electronic device 30 may implement the following: Figure 1 In some embodiments, instructions 305 , hardware, firmware, and / or software components thereof may additionally / alternatively be placed in system control logic 301 , network interface 304 , and / or processor 300 .

[0088] The network interface 304 may include a transceiver for providing a radio interface for the electronic device 30, thereby communicating with any other suitable device (such as a front-end module, an antenna, etc.) through one or more networks. In some embodiments, the network interface 304 may be integrated with other components of the electronic device 30. For example, the network interface 304 may be integrated with at least one of the processor 300, the system memory 302, the non-volatile memory 303, and a firmware device (not shown) having instructions. When at least one of the processors 300 executes the instructions, the electronic device 30 implements the above-mentioned Figure 1 The process shown.

[0089] The network interface 304 may further include any suitable hardware and / or firmware to provide a multiple-input multiple-output radio interface. For example, the network interface 304 may be a network adapter, a wireless network adapter, a telephone modem and / or a wireless modem.

[0090] In one embodiment, at least one of the processors 300 may be packaged together with logic for one or more controllers of the system control logic 301 to form a system in package (SiP). In one embodiment, at least one of the processors 300 may be integrated on the same die with logic for one or more controllers of the system control logic 301 to form a system on chip (SoC).

[0091] The electronic device 30 may further include an input / output (I / O) device 306. The I / O device 306 may include a user interface to enable a user to interact with the electronic device 30; and a peripheral component interface design to enable peripheral components to interact with the electronic device 30.

[0092] In some embodiments, the user interface may include, but is not limited to, a display (eg, a liquid crystal display, a touch screen display, etc.), a sensor, a speaker, a microphone, a light emitting diode, and physical buttons.

[0093] In some embodiments, the peripheral component interface may include, but is not limited to, a non-volatile memory port, an audio jack, and a power interface.

[0094] In some embodiments, the sensor may include, but is not limited to, a gyroscope sensor, an acceleration sensor, a GNSS antenna, and a positioning unit. The positioning unit may also be a part of the network interface 304 or interact with the network interface 304 to communicate with components of the positioning network (e.g., GNSS satellites).

[0095] An embodiment of the present application also provides a computer program product for implementing the AI ​​tropical cyclone prediction method provided in the above embodiments.

[0096] The various embodiments of the mechanism disclosed in the present application can be implemented in hardware, software, firmware or a combination of these implementation methods. The embodiments of the present application can be implemented as a computer program module or module code executed on a programmable system, and the programmable system includes at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device and at least one output device.

[0097] A computer program module or module code can be applied to input instructions to perform the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, a processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0098] Module code can be implemented with high-level modular language or object-oriented programming language to communicate with the processing system. When necessary, module code can also be implemented with assembly language or machine language. In fact, the mechanism described in this application is not limited to the scope of any specific programming language. In either case, the language can be a compiled language or an interpreted language.

[0099] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, instructions may be distributed over a network or through other computer-readable media. Therefore, a machine-readable medium may include any mechanism for storing or transmitting information in a machine (e.g., computer) readable form, including, but not limited to, a floppy disk, an optical disk, an optical disk, a magneto-optical disk, a read-only memory (ROM), a random access memory (RAM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic card or an optical card, a flash memory, or a tangible machine-readable memory for transmitting information (e.g., a carrier wave, an infrared signal, a digital signal, etc.) using the Internet in an electrical, optical, acoustic, or other form of propagation signal. Accordingly, machine-readable media include any type of machine-readable media suitable for storing or transmitting electronic instructions or information in a form readable by a machine (eg, a computer).

[0100] References to "one embodiment" or "an embodiment" in the specification mean that the specific features, structures, or characteristics described in conjunction with the embodiment are included in at least one exemplary implementation or technology disclosed according to the embodiment of the present application. The appearance of the phrase "in one embodiment" in various places in the specification does not necessarily all refer to the same embodiment.

[0101] The disclosure of the embodiment of the present application also relates to an operating device for executing the text. The device can be specially constructed for the required purpose or it can include a general-purpose computer selectively activated or reconfigured by a computer program stored in the computer. Such a computer program can be stored in a computer-readable medium, such as, but not limited to any type of disk, including a floppy disk, an optical disk, a CD-ROM, a magneto-optical disk, a read-only memory (ROM), a random access memory (RAM), an EPROM, an EEPROM, a magnetic or optical card, an application-specific integrated circuit (ASIC) or any type of medium suitable for storing electronic instructions, and each can be coupled to a computer system bus. In addition, the computer mentioned in the specification may include a single processor or may be an architecture involving multiple processors for increased computing power.

[0102] The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be imagined by any technician familiar with the technical field within the technical scope disclosed in the present application should be included in the protection scope of the present application; in the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An AI tropical cyclone prediction method based on sparse data fusion, characterized in that: The steps include: Sparse observation fusion step, obtaining input data, the input data including original sparse observation data and IFS-HRES model forecast data, using a data assimilation model to perform data fusion on the input data, and outputting an initial field; AI weather prediction step, inputting the initial field into the AI ​​weather prediction model for prediction, obtaining weather prediction data at different times, and extracting tropical cyclone local prediction data at different times from the weather prediction data at different times; The deviation correction step inputs the tropical cyclone local prediction data at different times into the AI ​​deviation correction model, and outputs the tropical cyclone correction data at different times, so as to obtain the corrected tropical cyclone intensity prediction result, wherein the AI ​​deviation correction model is a Conv2Former-LSTM hybrid model.

2. The AI ​​tropical cyclone prediction method according to claim 1, characterized in that: The data assimilation model is one of the following models: an AI assimilation model, a four-dimensional variational assimilation model, an ensemble Kalman filter model, or a hybrid data assimilation model.

3. The AI ​​tropical cyclone prediction method according to claim 1, wherein: The sparse observation fusion step also includes: preprocessing the original sparse observation data to obtain sparse grid observation data, and using the data assimilation model to perform data fusion on the sparse grid observation data and the IFS-HRES model forecast data.

4. The AI ​​tropical cyclone prediction method according to claim 1, wherein: The sparse observation fusion step also includes: the input data also includes ocean variable data provided by a hybrid coordinate ocean model, and the ocean variable data includes sea surface temperature and ocean current velocity.

5. An AI tropical cyclone prediction system based on sparse data fusion, characterized in that: include: A sparse observation fusion unit, wherein the sparse observation fusion unit is used to obtain input data, wherein the input data includes original sparse observation data and IFS-HRES model forecast data, and perform data fusion on the input data using a data assimilation model to output an initial field; An AI weather prediction unit, the AI ​​weather prediction unit is used to input the initial field into an AI weather prediction model for prediction, obtain weather prediction data at different times, and extract tropical cyclone local prediction data at different times from the weather prediction data at different times; A deviation correction unit, wherein the deviation correction unit is used to input the tropical cyclone local prediction data at different times into an AI deviation correction model, and output the tropical cyclone correction data at different times, so as to obtain a corrected tropical cyclone intensity prediction result, wherein the AI ​​deviation correction model is a Conv2Former-LSTM hybrid model.

6. An electronic device, characterized in that: include: one or more processors; One or more memories; the one or more memories store one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device executes the AI ​​tropical cyclone prediction method according to any one of claims 1 to 4.

7. A computer program product, characterized in that It comprises a computer program / instruction, which, when executed by a processor, implements the AI ​​tropical cyclone prediction method according to any one of claims 1 to 4.

Images