CN-121997022-A - Low-altitude weather forecast vertical encryption method and device based on deep learning, electronic equipment and storage medium

CN121997022ACN 121997022 ACN121997022 ACN 121997022ACN-121997022-A

Abstract

The application relates to the technical field of meteorological data processing, in particular to a low-altitude meteorological prediction vertical encryption method and device based on deep learning, electronic equipment and storage medium, wherein the method comprises the steps of obtaining three-dimensional meteorological prediction field data with first vertical resolution output by a global or regional numerical weather prediction mode; preprocessing the three-dimensional weather forecast field data to obtain preprocessed three-dimensional weather forecast field data. The application establishes and realizes direct and efficient nonlinear mapping from a low-vertical-resolution three-dimensional weather forecast field to a high-vertical-resolution three-dimensional weather element forecast field by means of a pre-trained deep learning model, thereby rapidly processing a low-resolution numerical forecast product input in real time and generating a refined three-dimensional weather field with obviously increased vertical layer number, and further providing higher-precision key data support for low-altitude flight safety decision.

Inventors

LIU SHIMIN
CHEN YUNGANG
YE XIANCAI
LIN CHAO

Assignees

北京弘象科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. A low-altitude weather forecast vertical encryption method based on deep learning, comprising: acquiring three-dimensional weather forecast field data with a first vertical resolution output by a global or regional numerical weather forecast mode; Preprocessing the three-dimensional weather forecast field data to obtain preprocessed three-dimensional weather forecast field data; inputting the preprocessed three-dimensional weather forecast field data into a pre-trained deep learning vertical encryption model, and outputting a three-dimensional weather element forecast field with a second vertical resolution; The second vertical resolution is higher than the first vertical resolution, and the deep learning vertical encryption model is used for establishing a nonlinear mapping relation from the first vertical resolution to the second vertical resolution.
2. The method of claim 1, wherein the deep-learning vertical encryption model is an encoder-decoder based neural network architecture; the encoder is composed of a plurality of three-dimensional convolution layers and is used for carrying out feature extraction and downsampling on the input three-dimensional weather forecast field data; the decoder is comprised of a plurality of three-dimensional deconvolution layers for upsampling the features extracted by the encoder and reconstructing them into the three-dimensional weather element prediction field having the second vertical resolution.
3. Method according to claim 2, characterized in that a jump connection is provided between the encoder and the corresponding level of the decoder for passing the low-level spatial detail features extracted by the encoder to the decoder and fusing with the high-level semantic features of the corresponding level of the decoder.
4. The method of claim 2, wherein an attention mechanism module is integrated in the decoder for enabling the model to enhance attention weights to critical vertical layers including boundary layers, wind shear layers, or inverse temperature layers when reconstructing the three-dimensional meteorological element prediction field.
5. The method according to any one of claims 1 to 4, wherein the training method of the deep learning vertical encryption model comprises: Acquiring a training data set, wherein the training data set comprises historical forecast data samples output by a numerical weather forecast mode and high-resolution analysis data labels corresponding to the historical forecast data samples in time and space; Respectively preprocessing the history forecast data sample and the high-resolution analysis data label to obtain a preprocessed history forecast data sample and a preprocessed high-resolution analysis data label; and training the deep learning vertical encryption model by using the preprocessed historical forecast data sample as input and the preprocessed high-resolution analysis data label as a supervision target through optimizing a loss function.
6. The method of claim 5, wherein the loss function is a joint loss function comprising at least a mean square error loss for a wind vector field and an L1 loss for a scalar field of wind speeds.
7. The method of claim 1, wherein the step of inputting the preprocessed three-dimensional weather forecast field data into a pre-trained deep-learning vertical encryption model, outputting a three-dimensional weather element forecast field having a second vertical resolution, further comprises: Calculating vertical wind shear strength at a designated course key point based on the three-dimensional weather element forecast field having the second vertical resolution; and generating flight risk warning information in response to the vertical wind shear strength exceeding a preset warning threshold.
8. A low-altitude weather forecast vertical encryption device based on deep learning, comprising: The data acquisition module is used for acquiring three-dimensional weather forecast field data with first vertical resolution output by a global or regional numerical weather forecast mode; the preprocessing module is used for preprocessing the three-dimensional weather forecast field data to obtain preprocessed three-dimensional weather forecast field data; the vertical encryption module is used for inputting the preprocessed three-dimensional weather forecast field data into a pre-trained deep learning vertical encryption model and outputting a three-dimensional weather element forecast field with a second vertical resolution; The second vertical resolution is higher than the first vertical resolution, and the deep learning vertical encryption model is used for establishing a nonlinear mapping relation from the first vertical resolution to the second vertical resolution.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the method according to any of claims 1 to 7.

Description

Low-altitude weather forecast vertical encryption method and device based on deep learning, electronic equipment and storage medium Technical Field The invention relates to the technical field of meteorological data processing, in particular to a low-altitude meteorological prediction vertical encryption method and device based on deep learning, electronic equipment and a storage medium. Background Low altitude airspace (commonly referred to as the airspace below 3000 meters from the ground to altitude) is the primary operating space for unmanned aerial vehicles, general aviation aircraft, and urban air traffic aircraft. The atmosphere is directly influenced by the terrain, the ground cover and the human activities, and meteorological elements (such as wind, temperature and humidity) of the atmosphere have large gradient and severe change in the vertical direction, so that the phenomena of micro-scale turbulence, strong wind shear, temperature reversal layer and the like which endanger the flight safety are easily generated. Therefore, the acquisition of a fine three-dimensional aerial image field with high space-time resolution, particularly high vertical resolution is a key premise for realizing low-altitude flight safety planning and intelligent management and control. Currently, global or regional numerical weather forecast modes (e.g., ECMWF, GRAPES, etc.) of business operations are the primary data sources for weather forecast. However, limited by the computing resources and mode theory, such modes are generally low in resolution in the vertical direction, have only a few levels in the near-ground to low-altitude range, and are difficult to accurately delineate the fine meteorological structure evolution within the boundary layer. In order to improve the resolution, the conventional method mainly depends on increasing the number of vertical layering of the numerical mode or performing dynamic downscaling, but the calculation cost is exponentially increased, and the real-time requirement of low-altitude flight on rapid update of weather information cannot be met. In addition, some researches attempt to obtain finer wind fields by coupling a mesoscale mode with a computational fluid dynamics model, but such methods are mostly specific to specific scenes, time-consuming in computation, complex in flow, and difficult to be suitable for large-scale and business forecasting services. In recent years, the data-driven deep learning method provides a new idea for the refinement of weather forecast. The convolutional neural network and other models can effectively extract spatial features in meteorological data, and have potential in the fields of meteorological element prediction, prediction error correction and the like. However, the existing researches focus on the improvement of horizontal resolution or the prediction of single meteorological elements, and are specially aimed at the requirements of low-altitude flight safety guarantee, so that a deep learning solution for efficiently and accurately improving the vertical resolution of a meteorological field is not disclosed and reported at present. Therefore, the intelligent encryption method capable of overcoming the bottleneck of the traditional numerical mode and rapidly generating the high-vertical-resolution three-dimensional gas image field is developed, and has important application value and urgency. Disclosure of Invention In view of the above, the present invention aims to provide a method and apparatus for vertical encryption of low-altitude weather forecast based on deep learning, an electronic device, and a storage medium. In a first aspect, an embodiment of the present invention provides a method for vertical encryption of low-altitude weather forecast based on deep learning, including: acquiring three-dimensional weather forecast field data with a first vertical resolution output by a global or regional numerical weather forecast mode; preprocessing the three-dimensional weather forecast field data to obtain preprocessed three-dimensional weather forecast field data; Inputting the preprocessed three-dimensional weather forecast field data into a pre-trained deep learning vertical encryption model, and outputting a three-dimensional weather element forecast field with a second vertical resolution; The second vertical resolution is higher than the first vertical resolution, and the deep learning vertical encryption model is used for establishing a nonlinear mapping relation from the first vertical resolution to the second vertical resolution. With reference to the first aspect, the deep learning vertical encryption model is an encoder-decoder based neural network architecture; the encoder is composed of a plurality of three-dimensional convolution layers and is used for carrying out feature extraction and downsampling on input three-dimensional weather forecast field data; the decoder is formed of a plurality of three-dimensional deconvolution