Airport visibility prediction method, storage medium and related device

Abstract

The invention relates to the technical field of airport meteorological element prediction, and discloses an airport visibility prediction method, a storage medium and a related device. The method comprises the following steps: acquiring numerical weather forecast data and weather live data of an airport which take effect in the airport and nearby areas; respectively extracting data associated with visibility from the numerical weather forecast data and the weather live data as numerical weather factor forecast data and weather factor live data; performing two-dimensional extraction on grid type data in the numerical meteorological element forecast data and the meteorological element live data to obtain a first spatial correlation feature, and performing one-dimensional extraction on single-point time sequence meteorological data in the numerical meteorological element forecast data and the meteorological element live data to obtain a first time evolution feature; and inputting the first space correlation feature and the first time evolution feature into a pre-trained visibility prediction model to obtain a visibility short-term prediction result and an approaching prediction result. The method can improve the accuracy and efficiency of visibility prediction.

Description

Technical Field

[0001] This invention relates to the field of airport meteorological element prediction technology, specifically to a method and storage medium for predicting airport visibility. Background Technology

[0002] Low visibility weather is a weather phenomenon that occurs in the near-surface layer. Severe visibility impairment threatens the safety and efficiency of flight operations. Therefore, the need to establish quantitative and objective airport visibility prediction models as technical support is becoming increasingly prominent.

[0003] Domestic and international researchers have conducted extensive studies and proposed various solutions, mainly including statistical forecasting methods and numerical forecasting methods. Statistical forecasting methods use the relationship between independent and dependent variables to establish statistical models, such as linear regression and logistic regression, which have a certain predictive ability. However, due to the nonlinear nature of the formation and dissipation of low visibility weather, this forecasting method has limitations. Numerical models can simulate the microphysical and thermodynamic processes of the formation and dissipation of low visibility weather, and methods for forecasting fog using this approach have gradually developed, but are still not ideal.

[0004] In practical operations, airport visibility forecasting is mostly based on real-time and numerical weather predictions of weather conditions and meteorological element trends, combined with historical experience, to make qualitative forecasts of visibility. This method suffers from inconsistent accuracy due to variations in forecasters' skill levels and experience. Existing decision tree-based forecasting models do not fully consider the structural differences in meteorological data and employ a uniform feature extraction method, resulting in insufficient feature extraction accuracy and slow model convergence. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the purpose of this invention is to provide a method for predicting airport visibility, which can improve the accuracy of feature extraction and the convergence speed of the model, thereby improving the efficiency and accuracy of airport visibility prediction.

[0006] To solve the above problems, the technical solution adopted by the present invention is as follows: A method for predicting airport visibility, comprising: Collect numerical weather forecast data and real-time weather data of the airport and its surrounding areas; Data related to visibility were extracted from numerical weather prediction data and actual meteorological data, respectively, and used as numerical weather element forecast data and actual meteorological element data. Two-dimensional extraction is performed on gridded data in numerical meteorological element forecast data and meteorological element real-time data to obtain the first spatial correlation feature, and one-dimensional extraction is performed on single-point time-series meteorological data in numerical meteorological element forecast data and meteorological element real-time data to obtain the first temporal evolution feature. The first spatial correlation feature and the first temporal evolution feature are input into the pre-trained visibility prediction model to obtain short-term and near-term visibility prediction results.

[0007] Compared with existing technologies, the beneficial effects of this invention are as follows: This airport visibility prediction method obtains the first spatial correlation feature by performing two-dimensional extraction on the grid-type data in the collected data, and obtains the first temporal evolution feature by performing one-dimensional extraction on the time-series meteorological data in the data, so as to realize the differential feature extraction of meteorological data, improve the accuracy of feature extraction for data with different structures, thereby reducing the invalid search process in the model inference process, significantly improving the convergence speed of the model, and enabling more accurate visibility prediction results to be output at a faster speed, providing more efficient and reliable support for airport management decisions.

[0008] In the aforementioned airport visibility prediction method, the step of inputting spatial correlation features and temporal evolution features into a pre-trained visibility prediction model to obtain short-term and near-term visibility prediction results is described in the following steps: Collect global meteorological reanalysis data and historical meteorological data of airports from the past few years; Data related to visibility in and around the airport are extracted from global meteorological reanalysis data as meteorological element reanalysis data, and data related to visibility are extracted from historical meteorological data of the airport as historical meteorological element data of the airport. Two-dimensional extraction is performed on the grid-type data in the meteorological element reanalysis data and the historical real-time data of airport meteorological elements to obtain the second spatial correlation feature, and one-dimensional extraction is performed on the single-point time-series meteorological data in the meteorological element reanalysis data and the historical real-time data of airport meteorological elements to obtain the second temporal evolution feature. The second spatial correlation features and the second temporal evolution features are transformed into the initial model training data for prediction and used as the input of the decision tree model. The historical visibility of the airport in the historical real-time data of airport meteorological elements is used as the output of the decision tree model to train the decision tree model and obtain the visibility prediction model.

[0009] The aforementioned airport visibility prediction method, wherein the step of converting the second spatial correlation features and the second temporal evolution features into initial prediction model training data includes: Convert the data of the second spatial correlation feature and the second temporal evolution feature into a unified format; Outliers in the second spatial correlation feature and the second temporal evolution feature are removed using anomaly analysis. Missing values ​​in the second spatial correlation feature and the second temporal evolution feature are filled in using the interpolation algorithm.

[0010] The aforementioned airport visibility prediction method uses a gradient boosting decision tree model. During training, the parameters of the decision tree model's iterative framework and weak learner regression tree are optimized through grid search.

[0011] The aforementioned airport visibility prediction method determines the size of the two-dimensional convolutional kernel for extracting the first and second spatial correlation features based on the data resolution. The higher the resolution, the larger the size of the two-dimensional convolutional kernel. The step size of the one-dimensional convolutional kernel for extracting the first and second temporal evolution features is determined based on the data time interval. The shorter the time interval, the shorter the step size.

[0012] In the aforementioned airport visibility prediction method, the decision tree model calculates the contribution of each second spatial correlation feature and second temporal evolution feature to visibility prediction after each iteration during training, and adjusts the weight of each feature based on the contribution.

[0013] A storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the aforementioned airport visibility prediction method.

[0014] An electronic device is characterized by comprising a processor and a memory, wherein the processor is communicatively connected to the memory via a data bus, and when the processor calls and executes a computer program in the memory, it implements the aforementioned airport visibility prediction method.

[0015] An airport visibility prediction system includes: an acquisition module for collecting numerical weather forecast data and actual meteorological data of the airport and its surrounding area; a data processing module for extracting visibility-related data from the numerical weather forecast data and the actual meteorological data, respectively, as numerical weather element prediction data and actual meteorological element data; a differential feature extraction module for performing two-dimensional extraction on grid-type data in the numerical weather element forecast data and the actual meteorological element data to obtain a first spatial correlation feature, and performing one-dimensional extraction on single-point time-series meteorological data in the numerical weather element forecast data and the actual meteorological element data to obtain a first temporal evolution feature; and a visibility prediction module for inputting the first spatial correlation feature and the first temporal evolution feature into a pre-trained visibility prediction model to obtain short-term visibility prediction results and near-term visibility prediction results.

[0016] The aforementioned airport visibility prediction system includes a differential feature extraction module comprising a kernel matching module and an extraction module. The kernel matching module determines the size of the two-dimensional convolutional kernel for extracting the first spatial correlation feature based on the data resolution; the higher the resolution, the larger the size of the two-dimensional convolutional kernel. It also determines the stride of the one-dimensional convolutional kernel for extracting the first temporal evolution feature based on the data time interval; the shorter the time interval, the shorter the stride of the one-dimensional convolutional kernel. The extraction module extracts the first spatial correlation feature using the corresponding two-dimensional convolutional kernel and extracts the first temporal evolution feature using the corresponding one-dimensional convolutional kernel.

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0018] Figure 1 This is a flowchart of an airport visibility prediction method according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic block diagram of an electronic device according to an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the airport visibility prediction system according to an embodiment of the present invention. Detailed Implementation

[0021] The embodiments of the present invention are described in detail below, with reference to... Figure 1 The present invention provides a method for predicting airport visibility, comprising the following steps: Collect numerical weather forecast data and real-time weather data of the airport and its surrounding areas; Data related to visibility were extracted from numerical weather prediction data and actual meteorological data, respectively, and used as numerical weather element forecast data and actual meteorological element data. Two-dimensional extraction is performed on gridded data in numerical meteorological element forecast data and meteorological element real-time data to obtain the first spatial correlation feature, and one-dimensional extraction is performed on single-point time-series meteorological data in numerical meteorological element forecast data and meteorological element real-time data to obtain the first temporal evolution feature. The first spatial correlation feature and the first temporal evolution feature are input into the pre-trained visibility prediction model to obtain short-term and near-term visibility prediction results.

[0022] This method employs a two-dimensional convolution kernel to extract two-dimensional features from gridded data (such as 3×3 gridded relative humidity and pressure field data) in numerical meteorological forecast and real-time meteorological data. This extraction captures the first spatial correlation features of the gridded data, yielding spatial characteristics such as pressure gradient and humidity distribution uniformity in the airport area. For single-point time-series meteorological data (such as hourly updated expected and wind speed data) in both numerical meteorological forecast and real-time meteorological data, a one-dimensional convolution kernel is used for one-dimensional extraction. This captures the evolution of these meteorological elements over continuous time, extracting temporal features such as the rate of temperature change and wind speed fluctuations. This method employs differentiated feature extraction techniques—two-dimensional and one-dimensional—for gridded and single-point time-series meteorological data in numerical meteorological element forecast data and meteorological element real-time data, respectively. This approach ensures better adaptation of extracted features to the data structure, resulting in more accurate features and reducing ineffective search processes during model inference. Consequently, it significantly improves the model's convergence speed, enabling the pre-trained visibility prediction model to output more accurate visibility predictions more quickly, providing more efficient and reliable support for airport management decisions. Using this method, a visibility prediction model trained with meteorological element forecast data and historical meteorological data from Zhuhai Jinwan Airport from 2018 to 2023 achieved an 18% improvement in accuracy compared to a prediction scheme using a uniform feature extraction method. The prediction accuracy for low visibility (≤1000m) reached 89%, and the number of model convergence iterations was reduced by 25%.

[0023] Understandably, visibility prediction models can employ decision tree models, backpropagation (BP) neural networks, long short-term memory (LSTM) networks, etc. The training data for the visibility prediction model is composed of spatial correlation features obtained from differentiated feature extraction of recent years' global meteorological reanalysis data and historical airport meteorological data, as well as temporal evolution features of single-point time-series data. This data serves as the model's input, further improving the model's training efficiency and accuracy. Specifically, in this embodiment, the visibility prediction model employs a decision tree model, and the model's training process includes the following steps: Collect global meteorological reanalysis data and historical meteorological data of airports from the past few years, such as the last 10 or 20 years. Data related to visibility in and around the airport are extracted from global meteorological reanalysis data as meteorological element reanalysis data, and data related to visibility are extracted from historical meteorological data of the airport as historical meteorological element data of the airport. Two-dimensional extraction is performed on the grid-type data in the meteorological element reanalysis data and the historical real-time data of airport meteorological elements to obtain the second spatial correlation feature, and one-dimensional extraction is performed on the single-point time-series meteorological data in the meteorological element reanalysis data and the historical real-time data of airport meteorological elements to obtain the second temporal evolution feature. The second spatial correlation features and the second temporal evolution features are transformed into the initial model training data for prediction and used as the input of the decision tree model. The historical visibility of the airport in the historical real-time data of airport meteorological elements is used as the output of the decision tree model to train the decision tree model and obtain the visibility prediction model.

[0024] Understandably, in this embodiment, to further improve the convergence speed and prediction accuracy of the model, the decision tree model adopts a gradient boosting decision tree. During training, the weak learners in the gradient boosting decision tree are first initialized, and a constant value that minimizes the loss function is estimated. In each iteration, the negative gradient of the loss function is calculated as a residual estimate, and a new regression tree is trained based on the residual. The best-fit value is calculated for the leaf region of the regression tree, and the weak learners are updated. This iteration is repeated until the preset maximum number of iterations is reached, resulting in the final model. During the iteration process, the model parameters are selected through grid search, choosing the parameters corresponding to the maximum accuracy value. In this embodiment, the maximum number of iterations is set to 80, the learning rate is 0.05, the maximum depth is 3, the minimum number of samples required for internal node re-division is 60, and the remaining parameters use default values. After the model is trained, it is validated using accuracy, risk score, and Heidrick skill score. If the validation is successful, the model is actually applied to visibility prediction; if the validation fails, the model is retrained.

[0025] In some embodiments, to further improve the model's convergence speed, after each iteration, the contribution to visibility prediction accuracy is calculated by the average reduction in impurity (such as the Gini index or entropy) resulting from node splitting of each second spatial correlation feature and second temporal evolution feature across all trees. The weight of each feature is then adjusted based on this contribution. Alternatively, after each iteration, the contribution of a feature can be determined by directly observing the rate of decrease in model accuracy after shuffling or removing it. By dynamically adjusting feature weights based on their contribution to visibility prediction, the weights of features strongly correlated with low visibility, such as humidity-related features, can be increased, while the weights of features with weaker impacts on visibility, such as wind speed-related features, can be decreased, further improving the model's convergence speed, prediction accuracy, and prediction efficiency.

[0026] It is understandable that numerical weather prediction (NMR) data can utilize weather forecasts from different numerical weather prediction model agencies such as ECWMF and GFS, while airport real-time weather data can be collected from real-time airport data using automatic weather observation equipment products manufactured by companies such as Vaisala or Metarit. NMR data can include data at various resolutions, while weather forecast data primarily consists of single-point time-series meteorological data. The start time for NMR data can be 2:00, 8:00, 14:00, and 20:00 in the local time zone. Generally, these data can be acquired at 6:00, 12:00, 18:00, and 24:00 in the local time zone, respectively. The time interval for NMR data can be 1 hour, 3 hours, 6 hours, etc. The latest time for real-time weather data can be 1 minute, 2 minutes, 5 minutes, or 10 minutes prior to the current time, and the time interval for real-time weather data can be 10 seconds, 20 seconds, 30 seconds, or 60 seconds, etc. Global meteorological data in analysis can utilize different in-analysis datasets such as ERA5 and FNL, while historical airport weather data can be obtained from the airport's monthly observation log.

[0027] In some embodiments, to further improve the accuracy of feature extraction and enhance the convergence speed and prediction accuracy of the model, during the extraction of the first and second spatial correlation features, the size of the corresponding two-dimensional convolutional kernel is determined based on the resolution of the extracted data. Higher resolution results in a larger two-dimensional convolutional kernel size. The system stores a first mapping relationship between resolution and two-dimensional convolutional kernel size. Based on the resolution of the input data, the system queries the size of the corresponding two-dimensional convolutional kernel according to the first mapping relationship and uses the corresponding two-dimensional convolutional kernel to extract the spatial correlation features of the data. For example, for data with a resolution of 0.25° to 0.5°, such as ECWMF numerical weather prediction, a 3×3 two-dimensional convolutional kernel is used to extract features; for high-resolution data of 0.1°, such as localized numerical model data, a 5×5 two-dimensional convolutional kernel is used to avoid missing spatial features.

[0028] Similarly, in some embodiments, to improve the extraction of first and second temporal evolution features, the stride of the corresponding one-dimensional convolutional kernel is determined based on the time interval of the extracted data. Shorter time intervals and higher frequencies require a lower stride for the one-dimensional convolutional kernel, while longer time intervals and lower frequencies require a longer stride, thus avoiding temporal noise interference. The system stores a second mapping relationship between time intervals and one-dimensional convolutional kernel strides. Based on the time interval of the input single-point time-series data, the system queries the corresponding one-dimensional convolutional kernel stride according to the second mapping relationship and uses the one-dimensional convolutional kernel with the corresponding stride to extract the temporal evolution features of the data. For example, for high-frequency real-time data with time intervals of 10 seconds or 1 minute, such as AWS real-time monitoring data, a one-dimensional convolutional kernel with a stride of 3 is used to extract features; for low-frequency historical data with time intervals of 1 hour or 3 hours, such as monthly observation logs, a one-dimensional convolutional kernel with a stride of 5 is used to extract features.

[0029] Understandably, the contribution of various data points from numerical weather prediction data, real-time meteorological data, global meteorological reanalysis data, and historical airport meteorological data to visibility can be calculated to determine the visibility-related data, which can then be used as input and training samples for the model. In this embodiment, the numerical weather prediction data and meteorological reanalysis data include the southwest jet stream index, warm advection criterion, temperature advection index, isobaric field index, inversion layer index, relative vorticity, and divergence. The real-time meteorological data and historical meteorological data include wind direction, wind speed, air temperature, dew point temperature, relative humidity, surface pressure, dominant visibility, and temperature-dew point difference.

[0030] In some embodiments, before obtaining visibility prediction through the visibility prediction model, it is also determined whether a low visibility scene has occurred by using the dominant visibility in the meteorological element real-time data. If the dominant visibility exceeds 10,000 meters, the moment is recorded as 10,000 meters; if the dominant visibility is less than 1,000 meters, it is recorded as foggy; otherwise, it is recorded as no fog and the original value is retained.

[0031] It is understandable that outliers in the second spatial correlation feature and the second temporal evolution feature are removed by anomaly analysis, and missing values ​​in the second spatial correlation feature and the second temporal evolution feature after outlier removal are filled by interpolation algorithm. This results in preprocessed quality control data, which forms the initial model training data for prediction, serving as training samples for the decision tree model.

[0032] In some embodiments, to further ensure the accuracy of airport visibility prediction, the system records the data predicted by the model each time and compares the recorded prediction data with the actual visibility at the corresponding time to calculate the model's prediction accuracy. When the model's prediction accuracy decreases by more than or equal to 3% for three consecutive months, a parameter re-optimization process is triggered to retrain and optimize the visibility prediction model to ensure prediction accuracy.

[0033] In some embodiments, to further improve the training convergence speed of the airport visibility prediction model, the airport uploads the gradient change of the trained gradient boosting decision tree model relative to the initial model to the server, along with the airport's inherent attributes, including geographical location, altitude, surrounding terrain, climate type, and urbanization level. The server processes the uploaded inherent attributes of different airports, converting them into static feature vectors for each airport using a pre-trained encoder. Simultaneously, based on the airport's geographical location, the server extracts daily or hourly meteorological time-series data of the surrounding area from global meteorological reanalysis data, and processes this data using a bidirectional LSTM (Long Short-Term Memory) network to obtain a meteorological context feature vector for each airport. Finally, the server concatenates the static feature vector and the meteorological context feature vector for each airport to obtain a meta-knowledge vector for each airport. When a new airport joins the system, or when an airport encounters a rare weather event (i.e., a new scenario), the server calculates a static feature vector for the new scenario based on the inherent attributes uploaded by the airport. Then, based on the daily or hourly meteorological time-series data uploaded by the airport, it calculates a meteorological scenario feature vector for the new scenario using a bidirectional LSTM network. This feature vector is then combined with the corresponding static feature vector to form a meta-knowledge vector for the new scenario. The server calculates the similarity between the meta-knowledge vector of the new scenario and the existing meta-knowledge vectors of other airports. Using the corresponding similarity as weights, it aggregates the gradient changes of existing airports to obtain the similarity gradient changes for the new scenario. Based on these similarity gradient changes, the initial model's parameters are updated to obtain a similar model for the new scenario. The server distributes this similar model to the airport experiencing the new scenario. The airport uses this similar model as the initial model for its visibility prediction model and trains it using the initial model's training data. Since the initial parameters of the similar model are mainly influenced by existing similar scenarios (i.e., knowledge from existing similar scenarios is transferred to the similar model), the similar model can adapt to the new scenario with fewer iterations, such as 5-10 iterations. By forming a federated learning system with multiple airports, knowledge can be shared among different airports, enabling faster deployment of visibility forecasting capabilities for new airports and allowing airports to quickly adapt to rare weather events.

[0034] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the above-described airport visibility prediction method.

[0035] In some possible implementations, various aspects of the airport visibility prediction method provided by the present invention can also be implemented in the form of a program product, which includes program code that, when the program product is run on a device, causes the control device to perform the steps in the airport visibility prediction method according to the various exemplary embodiments of the present application described above.

[0036] By designing and programming the processor, the code corresponding to the airport visibility prediction method described in the foregoing embodiments can be embedded into the chip, enabling the chip to execute the steps of the airport visibility prediction method shown in the embodiments of the present invention during operation. How to design and program the processor is a technique well-known to those skilled in the art and will not be elaborated upon here.

[0037] Based on the same inventive concept, referring to Figure 2 The present invention also provides an electronic device for implementing the above-described airport visibility prediction method, including a memory and a controller. The memory and the controller are connected via a data bus. The controller can implement the above-described airport visibility prediction method by calling and executing a computer program in the memory.

[0038] In one possible design, the processor may include one or more processing units. The processor and memory may be implemented on the same chip or on separate chips. The processor may be a general-purpose processor, such as a central processing unit (CPU), digital signal processor, application-specific integrated circuit, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the airport visibility prediction method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0039] Memory, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory can include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, magnetic disk, optical disk, etc. Memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited to this. The memory in the embodiments of this application can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0040] Reference Figure 3 Based on the same inventive concept, embodiments of the present invention also provide an airport visibility prediction system, including an acquisition module, a data processing module, a differential feature extraction module, and a visibility prediction module. The acquisition module collects numerical weather forecast data and actual meteorological data of the airport and its surrounding area. The data processing module extracts visibility-related data from the numerical weather forecast data and actual meteorological data, respectively, as numerical weather element prediction data and actual meteorological element data. The differential feature extraction module performs two-dimensional extraction on the gridded data in the numerical weather element forecast data and actual meteorological element data to obtain a first spatial correlation feature, and performs one-dimensional extraction on the single-point time-series meteorological data in the numerical weather element forecast data and actual meteorological element data to obtain a first temporal evolution feature. The visibility prediction module inputs the first spatial correlation feature and the first temporal evolution feature into a pre-trained visibility prediction model to obtain short-term and near-term visibility prediction results.

[0041] In some embodiments, the differential feature extraction module includes a convolutional kernel matching module and an extraction module. The convolutional kernel matching module is used to determine the size of the two-dimensional convolutional kernel for extracting the first spatial correlation feature based on the data resolution; the higher the resolution, the larger the size of the two-dimensional convolutional kernel. It also determines the stride of the one-dimensional convolutional kernel for extracting the first temporal evolution feature based on the data time interval; the shorter the time interval, the shorter the stride. The extraction module is used to extract the first spatial correlation feature using the corresponding two-dimensional convolutional kernel and to extract the first temporal evolution feature using the corresponding one-dimensional convolutional kernel.

[0042] In some embodiments, the system further includes a parameter update module, which is used to monitor the prediction accuracy of the visibility prediction model. When the prediction accuracy of the model decreases by more than or equal to 3% for three consecutive months, the parameter re-optimization process is initiated to retrain and optimize the visibility prediction model in order to ensure the prediction accuracy.

[0043] It should be noted that in the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first" or "second" is mentioned, it is only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0044] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0046] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0047] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0048] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting airport visibility, characterized in that, include: Collect numerical weather forecast data and real-time weather data of the airport and its surrounding areas; Data related to visibility were extracted from numerical weather prediction data and actual meteorological data, respectively, and used as numerical weather element forecast data and actual meteorological element data. Two-dimensional extraction is performed on gridded data in numerical meteorological element forecast data and meteorological element real-time data to obtain the first spatial correlation feature, and one-dimensional extraction is performed on single-point time-series meteorological data in numerical meteorological element forecast data and meteorological element real-time data to obtain the first temporal evolution feature. The first spatial correlation feature and the first temporal evolution feature are input into the pre-trained visibility prediction model to obtain short-term and near-term visibility prediction results.

2. The airport visibility prediction method according to claim 1, characterized in that, In the step of inputting spatial correlation features and temporal evolution features into a pre-trained visibility prediction model to obtain short-term and near-term visibility prediction results, the visibility prediction model is trained through the following steps: Collect global meteorological reanalysis data and historical meteorological data of airports from the past few years; Data related to visibility in and around the airport are extracted from global meteorological reanalysis data as meteorological element reanalysis data, and data related to visibility are extracted from historical meteorological data of the airport as historical meteorological element data of the airport. Two-dimensional extraction is performed on the grid-type data in the meteorological element reanalysis data and the historical real-time data of airport meteorological elements to obtain the second spatial correlation feature, and one-dimensional extraction is performed on the single-point time-series meteorological data in the meteorological element reanalysis data and the historical real-time data of airport meteorological elements to obtain the second temporal evolution feature. The second spatial correlation features and the second temporal evolution features are transformed into the initial model training data for prediction and used as the input of the decision tree model. The historical visibility of the airport in the historical real-time data of airport meteorological elements is used as the output of the decision tree model to train the decision tree model and obtain the visibility prediction model.

3. The airport visibility prediction method according to claim 2, characterized in that, The step of converting the second spatial correlation features and the second temporal evolution features into training data for the initial prediction model includes: Convert the data of the second spatial correlation feature and the second temporal evolution feature into a unified format; Outliers in the second spatial correlation feature and the second temporal evolution feature are removed using anomaly analysis. Missing values ​​in the second spatial correlation feature and the second temporal evolution feature are filled in using the interpolation algorithm.

4. The airport visibility prediction method according to claim 2, characterized in that, The decision tree model is a gradient boosting decision tree model. During the training process, the iterative framework of the decision tree model and the parameters of the weak learner regression tree are optimized through grid search.

5. The airport visibility prediction method according to any one of claims 2 to 4, characterized in that, The size of the two-dimensional convolutional kernel for extracting the first and second spatial correlation features is determined based on the resolution of the data. The higher the resolution, the larger the size of the two-dimensional convolutional kernel. The stride of the one-dimensional convolutional kernel for extracting the first and second temporal evolution features is determined based on the time interval of the data. The shorter the time interval, the shorter the stride.

6. The airport visibility prediction method according to any one of claims 2 to 4, characterized in that, During the training process, after each iteration, the decision tree model calculates the contribution of each second spatial correlation feature and second temporal evolution feature to visibility prediction, and adjusts the weight of each feature according to the contribution.

7. A storage medium storing a computer program, characterized in that, When the computer program is invoked and executed by the processor, it implements the airport visibility prediction method according to any one of claims 1 to 6.

8. An electronic device, characterized in that, The device includes a processor and a memory, wherein the processor is communicatively connected to the memory via a data bus, and when the processor calls and executes a computer program in the memory, it implements the airport visibility prediction method according to any one of claims 1 to 6.

9. An airport visibility prediction system, characterized in that, include: The acquisition module is used to collect numerical weather forecast data and real-time weather data of the airport and its surrounding areas. The data processing module is used to extract visibility-related data from the numerical weather forecast data and the actual weather data, respectively, as numerical weather element prediction data and actual weather element data. The differential feature extraction module is used to perform two-dimensional extraction on the grid-type data in the numerical meteorological element forecast data and the meteorological element real-time data to obtain the first spatial correlation feature, and to perform one-dimensional extraction on the single-point time-series meteorological data in the numerical meteorological element forecast data and the meteorological element real-time data to obtain the first temporal evolution feature. The visibility prediction module is used to input the first spatial correlation feature and the first temporal evolution feature into a pre-trained visibility prediction model to obtain short-term visibility prediction results and near-term visibility prediction results.

10. The airport visibility prediction system according to claim 9, characterized in that, The differential feature extraction module includes a convolution kernel matching module and an extraction module. The convolution kernel matching module is used to determine the size of the two-dimensional convolution kernel for extracting the first spatial correlation feature based on the resolution of the data. The higher the resolution, the larger the size of the two-dimensional convolution kernel. The module also determines the stride of the one-dimensional convolution kernel for extracting the first temporal evolution feature based on the time interval of the data. The shorter the time interval, the shorter the stride of the one-dimensional convolution kernel. The extraction module is used to extract the first spatial correlation feature through the corresponding two-dimensional convolution kernel and to extract the first temporal evolution feature through the corresponding one-dimensional convolution kernel.

Images