Plateau airport refined weather forecast system based on deep learning

Abstract

The invention discloses a plateau airport refined weather forecast system based on deep learning. The system comprises a data acquisition module, a data processing module, a model training module, a prediction output module and an evaluation and optimization module. According to the system, an improved Transform deep learning model is adopted, a high-resolution numerical forecasting product of the European mid-term weather forecasting center and airport high-frequency automatic observation data are fused, and single-point and hour-level accurate meteorological element forecasting is achieved. On the basis of a 0.4-degree global numerical forecasting product of the ECMWF, the system obtains key meteorological factors such as temperature, rainfall and wind field. For the complex terrain and extreme weather conditions of the plateau airport, the model structure is optimized, and the multi-source data fusion and transfer learning technology is introduced, so that the forecasting precision and timeliness are remarkably improved. The method effectively solves the problems that a traditional numerical forecasting mode is large in forecasting error and poor in timeliness in a plateau area, and provides intelligent guarantee for aviation safety and operation efficiency of a plateau airport.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and meteorological science and technology, and specifically relates to a refined weather forecast system for plateau airports based on deep learning, which is suitable for refined weather forecasting of plateau airports with complex terrain and extreme weather conditions. Background Art

[0002] With the rapid development of meteorological data processing technology and artificial intelligence technology, deep learning models have shown great application potential in the field of weather forecasting. Due to their special geographical location, plateau airports (such as Lanzhou Airport) are often faced with extreme weather threats such as thunderstorms, strong winds, and low visibility. These weather phenomena seriously affect aviation safety and operational efficiency.

[0003] The current mainstream weather forecasting methods mainly rely on numerical weather forecast models. However, these traditional methods show obvious limitations in plateau areas: on the one hand, due to the influence of complex terrain, the model does not accurately capture local meteorological conditions, resulting in large forecast errors; on the other hand, in application scenarios such as single-point prediction and short-term forecast, its timeliness and accuracy are difficult to meet the actual needs of plateau airports.

[0004] Although artificial intelligence technology provides new ideas for solving the above problems, its application in the field of weather forecasting, especially in complex terrain and plateau areas, is still in the exploratory stage and has not yet formed a mature solution. Therefore, developing a deep learning weather forecasting system that can integrate numerical forecast products and real-time observation data has important practical significance and application value for improving the weather support capabilities of plateau airports. Summary of the invention

[0005] Therefore, in order to solve the above-mentioned shortcomings, the present invention provides a plateau airport refined weather forecast system based on deep learning. Through the improved Transformer deep learning model, it fully integrates the ECMWF numerical forecast data and the airport automatic observation data to solve the problems of prediction accuracy and timeliness under complex meteorological conditions in plateau areas, and provide real-time and refined meteorological support for airport operations.

[0006] The present invention is implemented by constructing a plateau airport refined weather forecast system based on deep learning, including: Data acquisition module, data processing module, model training module, prediction output module, evaluation and optimization module; The data acquisition module is used to obtain ECMWF numerical weather forecast data and airport automatic observation data; The data processing module is used to clean, interpolate, normalize and align the acquired data in time and space to construct standardized input data; The model training module is based on an improved Transformer model, which achieves accurate prediction of hourly meteorological elements at a single point in a plateau airport (forecast time is 48 hours) through an optimized self-attention mechanism and position encoding method. The prediction output module is used to generate single-point weather forecast results and display them through a visual interface; The prediction output module is also used to regularly evaluate system performance and further optimize model structure and parameters through error analysis; The evaluation and optimization module is used to regularly evaluate model performance and automatically optimize model structure and parameters through error analysis.

[0007] Furthermore; the data acquisition module includes an interface module for automatically acquiring historical data; the collected data sources include ECMWF numerical weather forecast products and airport automatic observation data; the ECMWF numerical weather forecast product data include: temperature, humidity, wind speed, wind direction, and air pressure; the airport automatic observation data include: visibility, cloud height, and weather phenomena.

[0008] Further, the specific processing steps of the data processing module are: First, data quality control is performed to eliminate outliers and erroneous data; missing data is interpolated and completed using the spatiotemporal interpolation method; then data normalization is performed to convert data of different dimensions into standardized data within a unified range; finally, data spatiotemporal alignment is performed to ensure the consistency of different data sources in time and space.

[0009] Furthermore, the model training module includes: an improved Transformer model module, a multimodal data fusion module, and an encoder-decoder architecture module; The improved Transformer model module improves the feature extraction capability under complex meteorological conditions through an optimized self-attention mechanism; The multimodal data fusion module can realize the joint modeling of ECMWF numerical forecast data and airport observation data; The encoder-decoder architecture module is used to learn the time series characteristics of meteorological elements and output single-point hourly prediction results (forecast time is 48 hours).

[0010] Further, the self-attention mechanism optimization includes: A dynamically weighted multi-head attention mechanism is used to enhance the coupling relationship between local meteorological features and remote spatiotemporal features. A combination of periodic position coding and local position coding is introduced to better adapt to the temporal periodicity and spatial variability of meteorological data.

[0011] Furthermore, the prediction output module is used to output basic elements and key meteorological elements for airport operations. The basic elements include: temperature, humidity, wind speed, wind direction, and air pressure. The key meteorological elements for airport operations include: visibility, cloud base height, and weather phenomena.

[0012] Furthermore, the evaluation and optimization module adopts the following methods: evaluating the model performance through indicators such as root mean square error (RMSE) and mean absolute error (MAE); comparing with traditional numerical weather prediction methods, and based on the error analysis results, optimizing the model structure and parameter configuration; introducing transfer learning and sample augmentation techniques to improve the model's prediction ability for extreme weather conditions.

[0013] The present invention has the following advantages: The present invention discloses a refined meteorological forecasting system for plateau airports based on deep learning. The system uses an improved Transformer deep learning model, integrates high-resolution numerical forecast products of the European Centre for Medium-Range Weather Forecasts (ECMWF) and high-frequency automatic observation data of airports, and realizes accurate meteorological element prediction at the single-point and hourly levels. The system is based on ECMWF's 0.4-degree (about 40 kilometers) global numerical forecast products to obtain key meteorological elements such as temperature, precipitation, and wind fields. Aiming at the complex terrain and extreme weather conditions of plateau airports, by optimizing the model structure, introducing multi-source data fusion and transfer learning technologies, the accuracy and timeliness of the forecast have been significantly improved. The present invention effectively solves the problems of large forecast errors and poor timeliness of traditional numerical forecast models in plateau areas, provides intelligent guarantee for the aviation safety and operation efficiency of plateau airports, and has wide popularization and application value. The present invention effectively improves the accuracy of meteorological forecasting, especially the prediction ability for short-term and single-point weather changes, by integrating multi-source data and deep learning technologies. The system has real-time data processing capabilities, can quickly respond and output prediction results. It has been specifically optimized for the complex terrain and extreme weather conditions of plateau airports, and the model performs excellently in complex meteorological scenarios. Using deep learning algorithms to realize intelligent analysis and prediction of meteorological data, reducing the dependence on traditional manual experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is the overall system architecture diagram; Figure 2 Structural diagram of the improved Transformer model; Figure 3 Data processing and fusion flowchart. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The following will be combined with the attached Figures 1-3The present invention is described in detail, and the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] The present invention provides a high-altitude airport refined weather forecast system based on deep learning, such as Figures 1-3 As shown, the present invention includes the following technical links: Embodiment 1: The weather forecast system comprises: Data acquisition module, data processing module, model training module, prediction output module, evaluation and optimization module; The data acquisition module is used to obtain ECMWF numerical weather forecast data and airport automatic observation data; The data processing module is used to clean, interpolate, normalize and align the collected data in time and space to construct standardized input data; The model training module is based on an improved Transformer model, which achieves accurate prediction of hourly meteorological elements at a single point in a plateau airport (forecast time is 48 hours) through an optimized self-attention mechanism and position encoding method. The prediction output module is used to generate single-point weather forecast results and display them through a visual interface; The prediction output module is also used to regularly evaluate system performance and further optimize model structure and parameters through error analysis; The evaluation and optimization module is used to regularly evaluate model performance and automatically optimize model structure and parameters through error analysis.

[0017] When this embodiment is implemented, the data acquisition module includes an interface module for automatically acquiring historical data. The collected data sources include ECMWF numerical weather forecast products and airport automatic observation data. The ECMWF numerical weather forecast product data include: temperature, humidity, wind speed, wind direction, and air pressure; the airport automatic observation data include: visibility, cloud height, and weather phenomena.

[0018] When this embodiment is implemented, the specific processing steps of the data processing module are: First, data quality control is performed to eliminate outliers and erroneous data; missing data is interpolated and completed using the spatiotemporal interpolation method; then data normalization is performed to convert data of different dimensions into standardized data within a unified range; finally, data spatiotemporal alignment is performed to ensure the consistency of different data sources in time and space.

[0019] When this embodiment is implemented; the model training module includes: an improved Transformer model module, a multi-modal data fusion module, and an encoder-decoder architecture module; The improved Transformer model module enhances the feature extraction ability under complex meteorological conditions through an optimized self-attention mechanism; The multi-modal data fusion module can realize the joint modeling of ECMWF numerical forecast data and airport observation data; The encoder-decoder architecture module is used to learn the time series features of meteorological elements and output single-point hourly prediction results (forecast lead time of 48 hours).

[0020] When this embodiment is implemented; the optimization of the self-attention mechanism includes: A dynamically weighted multi-head attention mechanism, which is used to enhance the coupling relationship between local meteorological features and remote spatio-temporal features; introducing a combination of periodic positional encoding and local positional encoding to better adapt to the time periodicity and spatial variability of meteorological data.

[0021] When this embodiment is implemented; the prediction output module is used to output the following meteorological elements: basic elements such as temperature, humidity, wind speed, wind direction, and air pressure; key meteorological elements for airport operations such as visibility, cloud height, and weather phenomena.

[0022] When this embodiment is implemented; the evaluation and optimization module adopts the following methods: evaluating the model performance through indicators such as root mean square error (RMSE) and mean absolute error (MAE); comparing with traditional numerical weather prediction methods, and optimizing the model structure and parameter configuration based on the error analysis results; introducing transfer learning and sample augmentation techniques to improve the model's prediction ability for extreme weather conditions.

[0023] For the system architecture of this application, as Figure 1 shown, the system architecture of the present invention includes the following modules: Data acquisition module: Automatically obtain ECMWF numerical forecast data and airport automatic observation data to construct a high-quality meteorological data set.

[0024] Data processing module: Ensure the effectiveness and consistency of data input through steps such as data cleaning, quality control, interpolation and completion, and normalization.

[0025] Model training module: Based on historical data, use the improved Transformer model for training, and achieve accurate prediction of meteorological elements through a multi-layer encoder-decoder architecture.

[0026] Prediction module: Realize real-time prediction of single-point and hourly meteorological elements, including core indicators such as temperature, humidity, wind speed, and visibility, with a forecast lead time of 48 hours.

[0027] Evaluation and Optimization Module: The system regularly evaluates the model performance and automatically optimizes the model structure and parameters through error analysis.

[0028] Example 2, the operation process of the meteorological forecasting system: As Figure 1 shown, data is automatically collected and stored in the data storage module; the data is preprocessed and then input into the deep learning model; the model makes real-time predictions based on the input data; the prediction results are output and displayed through a visualization interface.

[0029] Example 3, the implementation effect of the meteorological forecasting system: Verified by experiments, the refined meteorological forecasting system for plateau airports based on deep learning of the present invention performs excellently in the following aspects: Prediction accuracy: Compared with traditional numerical forecasting methods, the average prediction error is reduced by more than 30%.

[0030] Response speed: Hourly prediction output is achieved to meet the high timeliness requirements.

[0031] Adaptation to extreme weather: The prediction accuracy for typical plateau extreme weather such as thunderstorms and strong winds is significantly improved.

[0032] In the present invention application, the data sources for data collection include ECMWF numerical weather forecast products (such as temperature, humidity, wind speed, wind direction, air pressure, etc.) and airport automatic observation data (such as visibility, cloud height, weather phenomena, etc.).

[0033] Data processing is carried out through methods such as data quality control, interpolation to fill in missing values, and normalization processing to ensure the accuracy, consistency, and standardization of the data.

[0034] In the present invention application, the deep learning model adopts an improved Transformer model, optimizing its self-attention mechanism and position encoding method to adapt to the spatio-temporal sequence characteristics of meteorological data. The model inputs the fused data and processes it through a multi-layer encoder-decoder architecture, outputting single-point hourly meteorological element prediction results.

[0035] In the present invention application, for model training and optimization, it is necessary to construct a high-quality training dataset covering historical data of different seasons, weather conditions, and extreme meteorological events. Adopt cross-validation and hyperparameter search techniques to optimize the model structure and parameters (such as learning rate, number of layers, number of attention heads, etc.). Introduce transfer learning and sample augmentation techniques to improve the generalization ability of the model under extreme weather conditions.

[0036] In this invention application, the development and integration of the forecasting system require the development of a meteorological forecasting system based on deep learning, integrating modules for automatic data acquisition, model inference, and result visualization. The system has the characteristics of high timeliness and low latency, and can output short-term forecasting results of single-point meteorological elements (such as temperature, wind speed, visibility, etc.) at the airport in real time.

[0037] In this invention application, the forecasting effect is evaluated by indicators such as root mean square error (RMSE) and mean absolute error (MAE) to evaluate the model performance, and it is compared with traditional numerical weather forecasting methods. The reliability and stability of the system are verified in experiments and practical applications.

[0038] The system is applicable to meteorological forecasting scenarios under complex terrains and extreme weather conditions at plateau airports, including but not limited to accurate prediction of extreme weather such as thunderstorms, strong winds, and low visibility.

[0039] The system has the characteristics of high timeliness and low latency, and can achieve hourly-level data prediction, meeting the rapid response requirements of plateau airport operations.

[0040] The system can be extended to meteorological forecasting application scenarios at other airports with complex terrains, and has good adaptability and popularization value.

[0041] The development and integration of the forecasting system require the development of a meteorological forecasting system based on deep learning, integrating modules for automatic data acquisition, model inference, and result visualization. The system has the characteristics of high timeliness and low latency, and can output short-term forecasting results of single-point meteorological elements (such as temperature, wind speed, visibility, etc.) at the airport in real time.

[0042] The deep learning model adopts an improved Transformer model, optimizing its self-attention mechanism and position encoding method to adapt to the spatio-temporal sequence characteristics of meteorological data. The model input fuses data, which is processed through a multi-layer encoder-decoder architecture, and outputs single-point hourly-level meteorological element prediction results.

[0043] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. 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 invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A plateau airport refined weather forecast system based on deep learning, characterized in that: include; Data acquisition module, data processing module, model training module, prediction output module, and evaluation and optimization module; The data acquisition module is used to obtain ECMWF numerical weather forecast data and airport automatic observation data; The data processing module is used to clean, interpolate, normalize and align the acquired data in time and space to construct standardized input data; The model training module is based on an improved Transformer model, which achieves accurate prediction of hourly meteorological elements at a single point at a plateau airport through an optimized self-attention mechanism and position encoding method. The prediction output module is used to generate single-point weather forecast results and display them through a visual interface; The prediction output module is also used to regularly evaluate system performance and further optimize model structure and parameters through error analysis; The evaluation and optimization module is used to regularly evaluate model performance and automatically optimize model structure and parameters through error analysis.

2. The plateau airport refined weather forecast system based on deep learning according to claim 1 is characterized in that: The data acquisition module includes an interface module for automatically acquiring historical data; the collected data sources include ECMWF numerical weather forecast products and airport automatic observation data; The ECMWF numerical weather forecast product data includes: temperature, humidity, wind speed, wind direction, and air pressure; the airport automatic observation data includes: visibility, cloud height, and weather phenomena.

3. The plateau airport refined weather forecast system based on deep learning according to claim 1 or 2, characterized in that: The specific processing steps of the data processing module are: First, data quality control is performed to eliminate outliers and erroneous data; missing data is interpolated and completed using the spatiotemporal interpolation method; then data normalization is performed to convert data of different dimensions into standardized data within a unified range; finally, data spatiotemporal alignment is performed to ensure the consistency of different data sources in time and space.

4. The plateau airport refined weather forecast system based on deep learning according to claim 1 is characterized in that: The model training module includes: an improved Transformer model module, a multimodal data fusion module, and an encoder-decoder architecture module; The improved Transformer model module improves the feature extraction capability under complex meteorological conditions through an optimized self-attention mechanism; The multimodal data fusion module can realize the joint modeling of ECMWF numerical forecast data and airport observation data; The encoder-decoder architecture module is used to learn the time series characteristics of meteorological elements and output single-point hourly prediction results.

5. The plateau airport refined weather forecast system based on deep learning according to claim 4 is characterized in that: Self-attention mechanism optimization includes: A dynamically weighted multi-head attention mechanism is used to enhance the coupling relationship between local meteorological features and remote spatiotemporal features. A combination of periodic position coding and local position coding is introduced to better adapt to the temporal periodicity and spatial variability of meteorological data.

6. The plateau airport refined weather forecast system based on deep learning according to claim 1, Its characteristics are The prediction output module is used to output basic elements and key meteorological elements for airport operations. The basic elements include: temperature, humidity, wind speed, wind direction, and air pressure; the key meteorological elements for airport operations include: visibility, cloud height, and weather phenomena.

7. The plateau airport refined weather forecast system based on deep learning according to claim 1 is characterized in that: The evaluation and optimization module adopts the following methods: evaluating the model performance through root mean square error and mean absolute error; comparing with the traditional numerical weather forecasting method, optimizing the model structure and parameter configuration based on the error analysis results; introducing transfer learning and sample enhancement technology to improve the model's prediction ability for extreme weather conditions.

Images