Cloud cluster movement track prediction method and device, equipment, medium and program product

Abstract

The invention relates to the technical field of weather prediction, in particular to a cloud cluster movement track prediction method and device, equipment, a medium and a program product. Acquiring multi-source fusion numerical mode data, and satellite cloud picture data and radar echo data in a preset historical time period; inputting the satellite cloud picture data into a first prediction model to obtain satellite cloud picture extrapolation data, and inputting the radar echo data into a second prediction model to obtain radar echo extrapolation data; cloud cluster features are extracted from the satellite cloud picture extrapolation data and the radar echo extrapolation data; inputting the cloud cluster features and the multi-source fusion numerical mode data into a third prediction model to obtain a movement track of the cloud cluster within a preset duration starting from the current moment; through the operation, the nonlinear and complex dynamic change of the cloud cluster can be accurately and comprehensively described, so that the movement track predicted based on the cloud cluster characteristics and the multi-source fusion numerical mode data has higher resolution and accuracy.

Description

Technical Field

[0001] The present disclosure relates to the field of meteorological prediction technologies, and particularly to a method, apparatus, device, medium, and program product for predicting the moving trajectory of cloud clusters. Background Art

[0002] Predicting the moving trajectory of cloud clusters is an important technology in fields such as weather forecasting, disaster warning, and aerospace mission planning.

[0003] In related technologies, the movement trajectory of cloud clusters in the future for a period of time is mainly predicted relying on numerical weather prediction models and statistical models. However, these models have certain limitations in dealing with the non-linear and complex dynamic changes of cloud clusters, thus showing the drawback of insufficient prediction accuracy.

[0004] Therefore, a method for predicting the moving trajectory of cloud clusters with high accuracy has become an urgent problem to be solved currently. Summary of the Invention

[0005] To solve the above technical problems, the present disclosure provides a method, apparatus, device, medium, and program product for predicting the moving trajectory of cloud clusters.

[0006] In a first aspect, the present disclosure provides a method for predicting the moving trajectory of cloud clusters, including:

[0007] Obtain multi-source fusion numerical model data, satellite cloud map data, and radar echo data within a preset historical time period; input the satellite cloud map data into a pre-trained first prediction model to obtain satellite cloud map extrapolation data, input the radar echo data into a pre-trained second prediction model to obtain radar echo extrapolation data; extract cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data; input the cloud cluster features and the multi-source fusion numerical model data into a pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset duration starting from the current moment.

[0008] In some optional embodiments, the first prediction model is an adversarial network model composed of a bidirectional autoencoder and a three-dimensional self-attention spatio-temporal long short-term memory network. Inputting the satellite cloud map data into the first prediction model to obtain the satellite cloud map extrapolation data includes:

[0009] Input the satellite cloud map data into the first prediction model, use the bidirectional autoencoder to extract spatio-temporal features in the satellite cloud map data, and input the spatio-temporal features into the three-dimensional self-attention spatio-temporal long short-term memory network; use the three-dimensional self-attention spatio-temporal long short-term memory network to capture the dynamic evolution information of the cloud cluster according to the spatio-temporal features, and make a prediction according to the dynamic evolution information to obtain the satellite cloud map extrapolation data.

[0010] In some alternative embodiments, the second prediction model is a model composed of a variational autoencoder and a Transformer decoder. Inputting the radar echo data into the pre-trained second prediction model to obtain radar echo extrapolation data, including:

[0011] Input the radar echo data into the second prediction model. Use the variational autoencoder to generate latent space samples based on the radar echo data, and input the latent space samples into the Transformer decoder. Use the Transformer decoder to capture the spatio-temporal correlation between echo data at different heights from the latent space samples, and make predictions based on the spatio-temporal correlation to obtain radar echo extrapolation data.

[0012] In some alternative embodiments, before extracting cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data, the method includes:

[0013] Preprocess the satellite cloud map extrapolation data and the radar echo extrapolation data to obtain the preprocessed satellite cloud map extrapolation data and the preprocessed radar echo extrapolation data;

[0014] Extracting cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data includes:

[0015] Use the Zernike moment method to extract the cloud cluster features carried in the preprocessed satellite cloud map extrapolation data and the preprocessed radar echo extrapolation data.

[0016] In some alternative embodiments, using the Zernike moment method to extract the cloud cluster features carried in the satellite cloud map extrapolation data and the radar echo extrapolation data includes:

[0017] Based on an external selection, determine the first region of interest corresponding to the satellite cloud map extrapolation data and the second region of interest corresponding to the radar echo extrapolation data; calculate the first Zernike moment features corresponding to the first region of interest, and calculate the second Zernike moment features corresponding to the second region of interest; determine at least some of the first Zernike moment features and at least some of the second Zernike moment features as cloud cluster features.

[0018] In some alternative embodiments, the third prediction model is a model composed of a graph convolutional layer and a long short-term memory network. Input the cloud cluster features and multi-source fusion numerical model data into the pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset time duration starting from the current moment, including:

[0019] Input the cloud cluster features and multi-source fusion numerical model data into the third prediction model. Use the graph convolutional layer to capture the spatial dependencies of different cloud clusters based on the cloud cluster features and multi-source fusion numerical model data, and input the spatial dependencies into the long short-term memory network. Use the long short-term memory network to capture the temporal dependencies of the cloud clusters, and perform prediction through the spatial dependencies and temporal dependencies to obtain the movement trajectory.

[0020] In a second aspect, the present disclosure provides a prediction device for the movement trajectory of a cloud cluster, including:

[0021] An acquisition module for acquiring multi-source fusion numerical model data, satellite cloud map data, and radar echo data within a preset historical time period; a first prediction module for inputting the satellite cloud map data into a pre-trained first prediction model to obtain satellite cloud map extrapolation data, and inputting the radar echo data into a pre-trained second prediction model to obtain radar echo extrapolation data; an extraction module for extracting cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data; a second prediction module for inputting the cloud cluster features and multi-source fusion numerical model data into a pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset duration starting from the current moment.

[0022] In a third aspect, the present disclosure provides a computer device, including:

[0023] A memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the prediction method for the movement trajectory of the cloud cluster corresponding to the first aspect and any of its embodiments.

[0024] In a fourth aspect, the present disclosure provides a computer-readable storage medium, on which computer instructions are stored. The computer instructions are used to cause a computer to execute the prediction method for the movement trajectory of the cloud cluster corresponding to the first aspect and any of its embodiments.

[0025] In a fifth aspect, the present disclosure provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the prediction method for the movement trajectory of the cloud cluster corresponding to the first aspect and any of its embodiments are implemented.

[0026] The technical solutions provided by the embodiments of the present disclosure have the following advantages compared with the prior art:

[0027] The cloud cluster movement trajectory prediction method provided by this embodiment obtains multi-source fusion numerical model data, satellite cloud map data, and radar echo data within a preset historical time period; inputs the satellite cloud map data into a pre-trained first prediction model to obtain satellite cloud map extrapolation data, and inputs the radar echo data into a pre-trained second prediction model to obtain radar echo extrapolation data; extracts cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data; inputs the cloud cluster features and the multi-source fusion numerical model data into a pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset time period starting from the current moment; in this solution, first, by introducing data from diverse sources, it provides comprehensive data support for the prediction of the cloud cluster movement trajectory; second, by separately predicting the satellite cloud map extrapolation data and the radar echo extrapolation data, it improves the accuracy and fine-grainedness of the extrapolation data; finally, extracting cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data can accurately and comprehensively describe the non-linear and complex dynamic changes of the cloud cluster, so that the movement trajectory predicted based on the cloud cluster features and the multi-source fusion numerical model data has higher resolution and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings herein are incorporated into and constitute a part of this specification, showing embodiments consistent with the present disclosure and, together with the specification, are used to explain the principles of the present disclosure.

[0029] To more clearly illustrate the technical solutions in the embodiments of the present disclosure or in the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0030] Figure 1 It is a flowchart of the cloud cluster movement trajectory prediction method provided by the embodiment of the present disclosure;

[0031] Figure 2 It is a structural connection diagram of the cloud cluster movement trajectory prediction device provided by the embodiment of the present disclosure;

[0032] Figure 3 It is a structural connection diagram of the computer device provided by the embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] In order to be able to more clearly understand the above-mentioned objects, features, and advantages of the present disclosure, the following will further describe the solution of the present disclosure. It should be noted that, without conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other.

[0034] In the following description, numerous specific details are set forth to provide a thorough understanding of the present disclosure, but the present disclosure may be practiced in other ways different from those described herein. Obviously, the embodiments in the specification are only a part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present disclosure.

[0035] According to an embodiment of the present disclosure, an embodiment of a method for predicting the moving trajectory of a cloud cluster is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from that here.

[0036] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0037] The prediction of the moving trajectory of a cloud cluster is an important technology in the fields of weather forecasting, disaster warning, and aerospace mission planning.

[0038] In related technologies, the prediction of the moving trajectory of a cloud cluster in a future period mainly relies on numerical weather prediction models and statistical models. Among them, the numerical weather prediction model is a tool for predicting weather changes based on physical principles by numerically solving the atmospheric motion equations. The statistical model is a model for predicting future weather changes based on historical data through statistical methods. The above two models cannot comprehensively and accurately describe the non-linear and complex dynamic changes of cloud clusters, especially showing the disadvantage of insufficient prediction accuracy under the requirements of high spatio-temporal resolution and short-term prediction.

[0039] To this end, the cloud cluster movement trajectory prediction method, device, equipment, medium and program product provided by the embodiments of the present disclosure first provide comprehensive data support for the prediction of the cloud cluster movement trajectory by introducing data from diverse sources; secondly, improve the accuracy and fine-grainedness of extrapolated data by separately predicting satellite cloud map extrapolated data and radar echo extrapolated data; finally, extract cloud cluster features from the satellite cloud map extrapolated data and radar echo extrapolated data, which can accurately and comprehensively describe the non-linear and complex dynamic changes of the cloud cluster, so that the movement trajectory predicted based on the cloud cluster features and multi-source fusion numerical model data has higher resolution and accuracy.

[0040] In this embodiment, a method for predicting the movement trajectory of a cloud cluster is provided, which can be used in a device for predicting the movement trajectory of a cloud cluster. The device can be set in devices such as a PC or a meteorological monitoring platform. Figure 1 It is a flowchart of the method for predicting the movement trajectory of a cloud cluster according to the embodiments of the present disclosure, as Figure 1 shown. The process includes the following steps:

[0041] S101, obtain multi-source fusion numerical model data, satellite cloud map data and radar echo data within a preset historical time period.

[0042] Among them, the multi-source fusion numerical model data refers to meteorological data predicted by using a numerical weather prediction model based on multi-source fusion data for a preset duration starting from the current moment. The multi-source fusion data includes satellite remote sensing data, radar observation data and ground observation data. Among them, the satellite remote sensing data includes high-resolution atmospheric, oceanic, and surface data, etc., the radar observation data includes data such as precipitation and wind speed, and the ground observation data includes data such as temperature, humidity, and air pressure.

[0043] The preset historical time period is the time period between a historical moment and the current moment. The satellite cloud map data refers to obtaining cloud cluster data from satellite images, including but not limited to the position, shape, thickness, etc. of the cloud cluster. The radar echo data refers to the data reflected by the radar when encountering cloud clusters at different heights. Therefore, the radar echo data in this embodiment is the radar echo data corresponding to different heights at each moment within the preset historical time period. The radar echo data includes but not limited to information such as the position, intensity, and speed of the cloud cluster.

[0044] Specifically, obtain the multi-source fusion numerical model data output by the numerical weather prediction model, obtain the corresponding satellite cloud map data within the preset historical time period from a satellite remote sensing platform or a satellite data website, and obtain the corresponding radar echo data within the preset historical time period from a meteorological data platform or website.

[0045] S102. Input the satellite cloud image data into the pre-trained first prediction model to obtain the extrapolated satellite cloud image data, and input the radar echo data into the pre-trained second prediction model to obtain the extrapolated radar echo data.

[0046] Among them, the extrapolated satellite cloud image data is the predicted satellite cloud image data within a preset time duration starting from the current moment, and the extrapolated radar echo data is the predicted radar echo data within a preset time duration starting from the current moment. The first prediction model is a deep learning model or an adversarial network model trained with historical satellite cloud images as training samples. The second prediction model is a deep learning network model obtained with historical extrapolated radar echo data as training samples, including but not limited to ConvLSTM model, CAST-LSTM model, Unet model, SAConvLSTM model, etc.

[0047] Specifically, input the satellite cloud image data within a preset historical time period into the pre-trained first prediction model, and use the first prediction model to predict the satellite cloud image data within a preset time duration starting from the current moment to obtain the extrapolated satellite cloud image data. Similarly, input the radar echo data within a preset historical time period into the pre-trained second prediction model, and use the second prediction model to predict the radar echo data within a preset time duration starting from the current moment to obtain the extrapolated radar echo data. The preset time duration starting from the current moment, that is, a certain future time period, can be set by those skilled in the art according to actual needs and is not limited here. It can be the next 3 months, the next 6 months, etc.

[0048] In some alternative embodiments, the first prediction model is an adversarial network model composed of a bidirectional autoencoder and a three-dimensional self-attention spatio-temporal long short-term memory network. Inputting the satellite cloud image data into the first prediction model to obtain the extrapolated satellite cloud image data includes: inputting the satellite cloud image data into the first prediction model, using the bidirectional autoencoder to extract the spatio-temporal features in the satellite cloud image data, and inputting the spatio-temporal features into the three-dimensional self-attention spatio-temporal long short-term memory network; using the three-dimensional self-attention spatio-temporal long short-term memory network to capture the dynamic evolution information of the cloud clusters according to the spatio-temporal features and make predictions according to the dynamic evolution information to obtain the extrapolated satellite cloud image data.

[0049] Exemplarily, in this embodiment, the first prediction model is a spatio-temporal generative adversarial network (ST-GRAN) model composed of a bidirectional autoencoder and a three-dimensional self-attention spatio-temporal long short-term memory network (3D self-attention spatio-temporal LSTM). After inputting the satellite cloud image data into the ST-GRAN model, first, the bidirectional autoencoder is used to perform encoding and decoding operations on the input satellite cloud image data, so as to extract the spatio-temporal features carried in the satellite cloud image data through the encoding and decoding operations (the spatio-temporal features characterize the variation law of cloud clusters in time and space), and the extracted spatio-temporal features are input into the 3D self-attention spatio-temporal LSTM. Then, the 3D self-attention spatio-temporal LSTM is used to capture the dynamic evolution information of cloud clusters in different time and space dimensions from the spatio-temporal features, and based on the captured dynamic evolution information, the change of cloud clusters in the next period of time is predicted to obtain the extrapolated data of the satellite cloud image.

[0050] In this embodiment, the ST-GRAN model is used to predict the satellite cloud image data, which can not only capture the local detail changes in the satellite cloud image data, but also capture the global dynamic trend of the satellite cloud image data, and has strong non-linear modeling ability, thus ensuring that the extrapolated data of the satellite cloud image obtained in this way has higher fine-grainedness and accuracy.

[0051] In some other alternative embodiments, the second prediction model is a model composed of a variational autoencoder and a Transformer decoder. Inputting the radar echo data into the pre-trained second prediction model to obtain the extrapolated data of the radar echo, including: inputting the radar echo data into the second prediction model, using the variational autoencoder to generate latent space samples according to the radar echo data, and inputting the latent space samples into the Transformer decoder; using the Transformer decoder to capture the spatio-temporal correlation between the echo data at different heights from the latent space samples, and making a prediction according to the spatio-temporal correlation to obtain the extrapolated data of the radar echo.

[0052] Exemplarily, in this embodiment, the second prediction model is a VAE-Transformer model composed of a variational autoencoder (VAE) and a Transformer decoder. Inputting the radar echo data in a preset historical time period into the VAE-Transformer model, through the reparameterization trick, the mean and variance output by the VAE are used to generate latent space samples, and the generated latent space samples contain the main features in the radar echo data. After generating the latent space samples, the latent space samples are input into the Transformer decoder, and the multi-head self-attention mechanism and the feed-forward neural network of the Transformer decoder are used to capture the spatio-temporal correlation between the echo data at different heights from the latent space samples, so as to predict the radar echo data in the next period of time according to the spatio-temporal correlation to obtain the extrapolated data of the radar echo.

[0053] In this embodiment, the VAE-Transformer model is used for radar echo data, which can not only accurately capture the intensity and structural evolution of cloud clusters, but also significantly improve the calculation efficiency and the spatial resolution of radar echo extrapolation data.

[0054] S103: Extract cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data.

[0055] Specifically, a feature extraction method is used to extract cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data. The feature extraction method includes, but is not limited to, feature extraction methods based on image processing and feature extraction methods based on machine learning, etc. Feature extraction methods based on image processing such as Zernike moments (i.e., Zernike moments) and Hu moments (i.e., Hu moments). Feature extraction methods based on machine learning such as principal component analysis and autoencoders. Cloud cluster features include, but are not limited to, the geometric shape, coverage area, moving speed and direction of cloud clusters, etc. The cloud cluster features extracted in this embodiment can comprehensively depict the spatial shape and moving trend of cloud clusters, providing high-quality basic data for the subsequent prediction of the cloud cluster movement trajectory.

[0056] In some alternative embodiments, before extracting cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data, preprocessing is performed on the satellite cloud map extrapolation data and the radar echo extrapolation data to obtain preprocessed satellite cloud map extrapolation data and preprocessed radar echo extrapolation data. The preprocessing methods include, but are not limited to, denoising, normalization, and scale adjustment, etc. In this embodiment, through the preprocessing of the satellite cloud map extrapolation data and the radar echo extrapolation data, the data quality and the consistency of the data format are ensured. After obtaining the preprocessed satellite cloud map extrapolation data and the preprocessed radar echo extrapolation data, Zernike moments are used to extract the cloud cluster features carried in the preprocessed satellite cloud map extrapolation data and the preprocessed radar echo extrapolation data.

[0057] In some other alternative embodiments, the extraction process of cloud cluster features is further optimized, specifically including: determining a first region of interest corresponding to the satellite cloud map extrapolation data and a second region of interest corresponding to the radar echo extrapolation data based on external selection; calculating a first Zernike moment feature corresponding to the first region of interest and a second Zernike moment feature corresponding to the second region of interest; determining at least some of the first Zernike moment features and at least some of the second Zernike moment features as cloud cluster features.

[0058] First, determine a first region of interest corresponding to the satellite cloud image extrapolation data and a second region of interest corresponding to the radar echo extrapolation data according to the user's selection. The first region of interest may be part or all of the satellite cloud image extrapolation data, the second region of interest may be part or all of the radar echo extrapolation data, and the number of the first region of interest and the second region of interest may be one or more.

[0059] Secondly, convert the satellite cloud image extrapolation data corresponding to the first region of interest and the radar echo extrapolation data corresponding to the second region of interest into data in polar coordinates. According to the Zernike polynomials, calculate a first Zernike moment feature corresponding to the first region of interest and a second Zernike moment feature corresponding to the second region of interest. The order and degree of the Zernike polynomials can be selected by the user.

[0060] Thirdly, all the first Zernike moment features can be organized into a first feature vector, and all the second Zernike moment features can be organized into a second feature vector. Alternatively, in order to reduce the computational complexity, some Zernike moment features can also be selected from the first Zernike moment features according to the order, repetition number, numerical size or other statistical characteristics of the Zernike moments to organize into a first feature vector, and some Zernike moment features can be selected from the second Zernike moment features to organize into a second feature vector. After obtaining the first feature vector and the second feature vector, through dimensionality reduction processing, reduce the vector dimension to further improve the computational efficiency. It should be noted that all the eigenvalue related to the cloud cluster movement trajectory is included in the first feature vector and the second feature vector.

[0061] Finally, to improve the accuracy of cloud cluster features, eigenvalues can be screened from the first eigenvector and the second eigenvector, the screened eigenvalues are fused, and the fused eigenvector is preprocessed, and the preprocessed eigenvector is determined as the cloud cluster feature. Through the preprocessing operation on the fused eigenvector, the fused eigenvector is processed into an input data format suitable for the third prediction model. In addition, it should be added that the above method of screening eigenvalues from the first eigenvector and the second eigenvector includes screening eigenvalues from the first eigenvector and the second eigenvector according to the importance of each eigenvalue in the first eigenvector and the second eigenvector. And / or, screening eigenvalues from the first eigenvector and the second eigenvector according to the degree of association between each eigenvalue in the first eigenvector and the second eigenvector and the task objectives (i.e., the moving direction and speed of the cloud cluster). It should be noted that the importance of the eigenvalues can be evaluated by machine learning algorithms (such as random forest, gradient boosting tree, etc.). The degree of association between the eigenvalues and the task objectives can be evaluated by methods such as Pearson correlation coefficient and Spearman rank correlation coefficient. By screening eigenvalues from the first eigenvector and the second eigenvector, not only can the purpose of simplifying the features be achieved, but also the finally extracted cloud cluster features can be made more accurate.

[0062] S104. Input the cloud cluster feature and the multi-source fusion numerical model data into the pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset time period starting from the current moment.

[0063] Specifically, input the cloud cluster feature and the multi-source fusion numerical model data into the third prediction model. The third prediction model analyzes the spatial form and movement trend hidden in the cloud cluster feature, and analyzes the meteorological changes in the multi-source fusion numerical model data, and predicts the movement trajectory of the cloud cluster within a preset time period starting from the current moment, and outputs the predicted movement trajectory. The output movement trajectory can be understood as a set of cloud cluster positions at each moment within the preset time period.

[0064] In some alternative embodiments, the third prediction model is a model composed of a graph convolutional layer and a long short-term memory network. Inputting the cloud cluster feature and the multi-source fusion numerical model data into the pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset time period starting from the current moment includes: inputting the cloud cluster feature and the multi-source fusion numerical model data into the third prediction model, using the graph convolutional layer to capture the spatial dependencies between different cloud clusters according to the cloud cluster feature and the multi-source fusion numerical model data, and inputting the spatial dependencies into the long short-term memory network; using the long short-term memory network to capture the temporal dependencies of the cloud cluster, and making a prediction through the spatial dependencies and the temporal dependencies to obtain the movement trajectory.

[0065] Exemplarily, in this embodiment, the third prediction model is a graph convolutional long short-term memory network (GC-LSTM) model composed of a graph convolutional layer and a long short-term memory network. After inputting the cloud cluster features and multi-source fusion numerical model data into the pre-trained third prediction model, the graph convolutional layer is used to perform spatial feature modeling on the input data to capture the spatial dependencies between different cloud clusters from the input data (i.e., cloud cluster features and multi-source fusion numerical model data), and the spatial dependencies between different cloud clusters are input into the long short-term memory network. Then, the long short-term memory network is used to perform time series modeling to capture the time series dependencies of the cloud clusters. Finally, by combining the spatial dependencies output by the graph convolutional layer and the time series dependencies captured by the long short-term memory network, the position movement trajectory of the cloud cluster within a preset time period starting from the current moment is predicted.

[0066] The method for predicting the moving trajectory of a cloud cluster provided in this embodiment obtains multi-source fusion numerical model data, satellite cloud map data, and radar echo data within a preset historical time period; inputs the satellite cloud map data into a pre-trained first prediction model to obtain satellite cloud map extrapolation data, and inputs the radar echo data into a pre-trained second prediction model to obtain radar echo extrapolation data; extracts cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data; inputs the cloud cluster features and the multi-source fusion numerical model data into a pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset time period starting from the current moment; in this solution, first, by introducing data from diverse sources, comprehensive data support is provided for predicting the moving trajectory of the cloud cluster; second, the accuracy and fine-grainedness of the extrapolation data are improved by separately predicting the satellite cloud map extrapolation data and the radar echo extrapolation data; finally, extracting cloud cluster features from the satellite cloud map extrapolation data and the radar echo extrapolation data can accurately and comprehensively describe the non-linear and complex dynamic changes of the cloud cluster, so that the movement trajectory predicted based on the cloud cluster features and the multi-source fusion numerical model data has higher resolution and accuracy.

[0067] In this embodiment, a device for predicting the moving trajectory of a cloud cluster is also provided. This device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0068] This embodiment provides a device for predicting the moving trajectory of a cloud cluster, as Figure 2 shown, including:

[0069] An acquisition module 201, configured to acquire multi-source fusion numerical model data, satellite cloud map data, and radar echo data within a preset historical time period.

[0070] The first prediction module 202 is configured to input satellite cloud image data into a pre-trained first prediction model to obtain satellite cloud image extrapolation data, and input radar echo data into a pre-trained second prediction model to obtain radar echo extrapolation data.

[0071] The extraction module 203 is configured to extract cloud cluster features from the satellite cloud image extrapolation data and the radar echo extrapolation data.

[0072] The second prediction module 204 is configured to input the cloud cluster features and multi-source fusion numerical model data into a pre-trained third prediction model to obtain the movement trajectory of the cloud cluster within a preset time duration starting from the current moment.

[0073] In some alternative embodiments, the first prediction model is an adversarial network model composed of a bidirectional autoencoder and a three-dimensional self-attention spatio-temporal long short-term memory network. The first prediction module 202 includes:

[0074] An extraction sub-module, configured to input the satellite cloud image data into the first prediction model, extract spatio-temporal features from the satellite cloud image data by using the bidirectional autoencoder, and input the spatio-temporal features into the three-dimensional self-attention spatio-temporal long short-term memory network; a first capture sub-module, configured to use the three-dimensional self-attention spatio-temporal long short-term memory network to capture the dynamic evolution information of the cloud cluster according to the spatio-temporal features, and make a prediction according to the dynamic evolution information to obtain the satellite cloud image extrapolation data.

[0075] In some alternative embodiments, the second prediction model is a model composed of a variational autoencoder and a Transformer decoder. The first prediction module 202 includes:

[0076] A generation sub-module, configured to input the radar echo data into the second prediction model, generate a latent space sample according to the radar echo data by using the variational autoencoder, and input the latent space sample into the Transformer decoder; a second capture sub-module, configured to use the Transformer decoder to capture the spatio-temporal correlation between echo data at different heights from the latent space sample, and make a prediction according to the spatio-temporal correlation to obtain the radar echo extrapolation data.

[0077] In some alternative embodiments, the apparatus further includes:

[0078] A preprocessing module, configured to preprocess the satellite cloud image extrapolation data and the radar echo extrapolation data before extracting cloud cluster features from the satellite cloud image extrapolation data and the radar echo extrapolation data, to obtain preprocessed satellite cloud image extrapolation data and preprocessed radar echo extrapolation data;

[0079] The extraction module 203 includes:

[0080] An extraction sub-module is used to extract the cloud cluster features carried in the extrapolated data of the preprocessed satellite cloud image and the extrapolated data of the preprocessed radar echo by using the Zernike moment method.

[0081] In some alternative embodiments, the extraction sub-module includes:

[0082] A selection unit is used to determine a first region of interest corresponding to the extrapolated data of the satellite cloud image and a second region of interest corresponding to the extrapolated data of the radar echo based on an external selection; a calculation unit is used to calculate a first Zernike moment feature corresponding to the first region of interest and a second Zernike moment feature corresponding to the second region of interest; a determination unit is used to determine at least part of the first Zernike moment features and at least part of the second Zernike moment features as cloud cluster features.

[0083] In some alternative embodiments, the third prediction model is a model composed of a graph convolutional layer and a long short-term memory network. The second prediction module 204 includes:

[0084] A third capture sub-module is used to input the cloud cluster features and the multi-source fusion numerical model data into the third prediction model, and use the graph convolutional layer to capture the spatial dependencies of different cloud clusters according to the cloud cluster features and the multi-source fusion numerical model data, and input the spatial dependencies into the long short-term memory network; a fourth capture sub-module is used to use the long short-term memory network to capture the temporal dependencies of the cloud clusters, and perform predictions through the spatial dependencies and the temporal dependencies to obtain the movement trajectories.

[0085] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be repeated here.

[0086] The prediction device for the cloud cluster movement trajectory in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0087] The embodiments of the present disclosure also provide a computer device having the above-mentioned Figure 2 prediction device for the cloud cluster movement trajectory shown.

[0088] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of a computer device provided by an alternative embodiment of the present disclosure. As shown in Figure 3As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if needed, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (such as an array of servers, a set of blade servers, or a multi-processor system). Figure 3 In this, a processor 10 is taken as an example.

[0089] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above programmable logic device can be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0090] Among them, the memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.

[0091] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0092] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0093] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0094] Embodiments of the present disclosure also provide a computer-readable storage medium. The methods according to the embodiments of the present disclosure can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the methods described herein can be processed by such software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0095] In addition to the above computer devices and computer-readable storage media, embodiments of the present application can also be computer program products, which include computer program instructions that cause a processor to execute the steps of the sound source localization method provided in any embodiment of the present application when the computer program instructions are run by the processor.

[0096] The computer program product can be written in any combination of one or more programming languages to write program code for performing the operations of the embodiments of the present application. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, executed as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0097] The above are only specific implementation manners of the present disclosure, enabling those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting cloud movement trajectory, characterized in that: include: Acquire multi-source fusion numerical model data, satellite cloud image data and radar echo data within a preset historical time period; Inputting the satellite cloud image data into a pre-trained first prediction model to obtain satellite cloud image extrapolated data, and inputting the radar echo data into a pre-trained second prediction model to obtain radar echo extrapolated data; Extracting cloud cluster features from the satellite cloud image extrapolation data and the radar echo extrapolation data; The cloud features and the multi-source fusion numerical pattern data are input into a pre-trained third prediction model to obtain the movement trajectory of the cloud within a preset time period starting from the current moment.

2. The method according to claim 1, characterized in that The first prediction model is an adversarial network model composed of a bidirectional autoencoder and a three-dimensional self-attention spatiotemporal long short-term memory network. The satellite cloud image data is input into the first prediction model to obtain satellite cloud image extrapolation data, including: Input the satellite cloud image data into the first prediction model, use the bidirectional autoencoder to extract the spatiotemporal features in the satellite cloud image data, and input the spatiotemporal features into the three-dimensional self-attention spatiotemporal long short-term memory network; The three-dimensional self-attention spatiotemporal long short-term memory network is used to capture the dynamic evolution information of the cloud cluster according to the spatiotemporal characteristics, and prediction is performed based on the dynamic evolution information to obtain the satellite cloud image extrapolation data.

3. The method according to claim 1, characterized in that The second prediction model is a model composed of a variational autoencoder and a Transformer decoder, and the radar echo data is input into the pre-trained second prediction model to obtain radar echo extrapolation data, including: Inputting the radar echo data into the second prediction model, generating latent space samples according to the radar echo data using the variational autoencoder, and inputting the latent space samples into the Transformer decoder; The Transformer decoder is used to capture the spatiotemporal correlation between echo data at different heights from the latent space samples, and prediction is performed based on the spatiotemporal correlation to obtain the radar echo extrapolation data.

4. The method according to claim 1, characterized in that: Before extracting cloud features from the satellite cloud image extrapolation data and the radar echo extrapolation data, the method includes: Preprocessing the satellite cloud image extrapolation data and the radar echo extrapolation data to obtain preprocessed satellite cloud image extrapolation data and preprocessed radar echo extrapolation data; The extracting cloud features from the satellite cloud image extrapolation data and the radar echo extrapolation data comprises: The cloud features carried in the preprocessed satellite cloud image extrapolation data and the preprocessed radar echo extrapolation data are extracted using the Zernike moment method.

5. The method according to claim 4, characterized in that The method of extracting cloud features carried in the satellite cloud image extrapolation data and the radar echo extrapolation data by using the Zernike moment method includes: Determine, based on external selection, a first region of interest corresponding to the satellite cloud image extrapolation data and a second region of interest corresponding to the radar echo extrapolation data; Calculating a first Zernike moment feature corresponding to the first region of interest, and calculating a second Zernike moment feature corresponding to the second region of interest; At least a portion of the first Zernike moment characteristic and at least a portion of the second Zernike moment characteristic are determined as the cloud characteristic.

6. The method according to claim 1, characterized in that The third prediction model is a model composed of a graph convolution layer and a long short-term memory network. The cloud features and the multi-source fusion numerical pattern data are input into the pre-trained third prediction model to obtain the movement trajectory of the cloud within a preset time starting from the current moment, including: Inputting the cloud cluster features and the multi-source fusion numerical pattern data into the third prediction model, using the graph convolution layer to capture the spatial dependencies of different cloud clusters according to the cloud cluster features and the multi-source fusion numerical pattern data, and inputting the spatial dependencies into the long short-term memory network; The long short-term memory network is used to capture the dependency of clouds in time series, and prediction is performed through the spatial dependency and the time series dependency to obtain the motion trajectory.

7. A cloud movement trajectory prediction device, characterized in that: include: An acquisition module is used to acquire multi-source fusion numerical model data, satellite cloud image data and radar echo data within a preset historical time period; A first prediction module is used to input the satellite cloud image data into a pre-trained first prediction model to obtain satellite cloud image extrapolated data, and input the radar echo data into a pre-trained second prediction model to obtain radar echo extrapolated data; An extraction module, used for extracting cloud features from the satellite cloud image extrapolation data and the radar echo extrapolation data; The second prediction module is used to input the cloud characteristics and the multi-source fusion numerical model data into a pre-trained third prediction model to obtain the movement trajectory of the cloud within a preset time period starting from the current moment.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the cloud movement trajectory prediction method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the cloud movement trajectory prediction method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for predicting the cloud movement trajectory according to any one of claims 1 to 6 are implemented.

Images