An airport radiation fog refined prediction method and device

Abstract

The application belongs to the technical field of radiation fog prediction, and provides an airport radiation fog fine prediction method and device, which comprises the following steps: S1, collecting open source data and various historical meteorological data before and after the airport fog, and performing data cleaning, format unification and standardization processing; S2, constructing a deep learning model suitable for time series data, and automatically extracting features and identifying meteorological patterns of sample data; S3, constructing a preliminary prediction model based on the extracted features, and adjusting model parameters and structures through iteration to optimize prediction performance; S4, combining numerical weather prediction data, and further optimizing the fog prediction model by using time series analysis method; S5, verifying the model by using reserved sample data; the application realizes fine, high-time-efficiency prediction of the generation, continuation and dissipation process of airport radiation fog by combining multi-source data fusion, deep learning and dynamic calibration, and provides reliable support for airport operation decision.

Description

Technical Field

[0001] This invention belongs to the field of radiation fog forecasting technology, specifically a method and device for refined forecasting of airport radiation fog. Background Technology

[0002] Radiation fog is a meteorological phenomenon that occurs on clear, calm nights when the near-surface air temperature drops below the dew point due to radiative cooling of the Earth's surface. It is characterized by sudden formation, high concentration, strong locality, and rapid dissipation, posing a serious threat to airport operations and being one of the leading weather phenomena causing widespread flight delays, diversions, and cancellations. Although the global aviation industry has developed and applied a range of technologies and procedures to address radiation fog, many challenges remain regarding the accuracy and coordination of forecasts.

[0003] Specifically, current numerical weather prediction and AI models have errors in simulating the critical conditions for the formation and dissipation of radiation fog (such as the wind speed threshold of a light breeze and the vertical distribution of humidity), resulting in a high rate of false alarms and missed alarms in the forecasts.

[0004] To address the problems raised in the background art, those skilled in the art have proposed a method and apparatus for refined forecasting of airport radiation fog. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a refined forecasting method and apparatus for airport radiation fog, which solves the problems of high false alarm and false negative rates in the existing numerical forecasting and AI models when simulating the critical conditions for radiation fog generation and dissipation (such as wind speed threshold and vertical humidity distribution).

[0006] A method for refined forecasting of airport radiation fog includes the following steps:

[0007] S1. Collect open-source data and various historical meteorological data before and after airport fog, and perform data cleaning, format unification and standardization processing;

[0008] S2. Construct a deep learning model suitable for time series data to automatically extract features and identify meteorological patterns from sample data;

[0009] S3. Construct a preliminary prediction model based on the extracted features, and optimize the prediction performance by iteratively adjusting the model parameters and structure.

[0010] S4. Combine numerical weather prediction (NWP) data and use time series analysis methods to further optimize the fog prediction model;

[0011] S5. Use reserved sample data to validate the model, compare the model prediction results with the actual observation data, optimize the model, and introduce the statistical threshold of airport radiation fog elements for result calibration.

[0012] Preferably, the specific operation steps of step S1 include:

[0013] S1.1 Collect open-source data and various historical meteorological data before and after airport fog, including but not limited to temperature, humidity, wind speed, wind direction, air pressure, cloud cover and runway visual range (RVR).

[0014] S1.2 Apply data cleaning algorithms to remove outliers, missing values, and duplicate data to ensure data quality;

[0015] S1.3 Convert data from different sources and in different formats into a unified data format to facilitate subsequent processing;

[0016] S1.4. Use Z-score standardization or Min-Max standardization to standardize the data in order to eliminate the influence of different units on model training.

[0017] The Z-score standardization formula is as follows: ,in This is the original data. The mean of the data. The standard deviation of the data;

[0018] The Min-Max standardization formula is: ,in The original data is given, where min and max represent the minimum and maximum values ​​of the data, respectively. This is the standardized data.

[0019] Preferably, step S2 includes the following specific steps:

[0020] S2.1. Feature extraction of time-series meteorological data based on the physical mechanism of airport fog formation;

[0021] S2.2. The model is trained using Deep Neural Networks (DNNs), as shown in the formula: ,in The kernel size is [size]. These are the weighting coefficients;

[0022] S2.3 Extract short-term and long-term time-dependent features using multi-scale convolutional kernels (such as 3×1, 5×1, 7×1);

[0023] S2.4 The regression task uses mean squared error (MSE) loss: The classification task uses cross-entropy loss: ;in, This represents the number of categories.

[0024] Preferably, the specific operation steps of step S3 are as follows:

[0025] During model training, an attention weight layer is introduced to adjust the weights in real time based on the predictive contribution of features in historical data. The attention weight layer includes attention weight calculation and feature fusion.

[0026] The attention weight is calculated as follows: ;in, For the first The hidden representation of each feature, For a multilayer perceptron, the weight allocation is optimized through gradient descent;

[0027] The feature fusion is as follows: ;in, For the first The original value of a feature or the feature extracted by a deep fully connected neural network.

[0028] Preferably, the specific operation steps of step S4 are as follows:

[0029] S4.1. Transform the gridded data of NWP (such as temperature, humidity, wind speed, cloud cover, etc.) into time series features and integrate them with ground observation data; specifically including:

[0030] S4.01 and NWP data are typically gridded outputs (e.g., latitude and longitude grids), while ground observation station data is point data, requiring spatial matching, including:

[0031] Interpolate NWP grid data to station coordinates: ;in, Distance weights The value is the value of the four nearest grid points;

[0032] Synchronize the hourly forecast data of NWP with the ground observation data by timestamp and handle time lag (e.g., NWP forecast 6 hours in advance).

[0033] S4.02 and NWP data contain information at different time scales (such as short-term fluctuations and long-term trends), which need to be decomposed and modeled separately, including:

[0034] The NWP time series is decomposed into IMF (Intrinsic Mode Function) and residuals: ;in, Representing different frequency components, This is a trend item;

[0035] S4.2 Input NWP characteristics and ground observation data into a time series model (such as Prophet) to capture dynamic evolution patterns; the formula of the time series model includes: ;in, For trend items, For seasonal items, Due to the holiday effect, The regression coefficients for the NWP trend;

[0036] S4.3 To address the spatiotemporal inconsistency between NWP and ground observation data and improve model robustness, a weighted loss function is added, the formula of which includes: ;in, For the MSE loss of ground observations, The loss is used for NWP feature reconstruction (e.g., autoencoder reconstruction error). , These are the weighting coefficients;

[0037] S4.4 To ensure the model adapts to the update cycle of NWP data and sudden changes in fog conditions, a Bayesian hyperparameter optimization is introduced for the update cycle of NWP data. The formula includes: ;in, This is the latest NWP data; a mutation detection technique is introduced to address fog abrupt changes in the NWP data. This technique uses the CUSUM (cumulative sum) algorithm to monitor abrupt changes in prediction error, and the formula includes: Among them, when > Threshold, triggering model retraining.

[0038] Preferably, the statistical thresholds mentioned in step S5 include, but are not limited to, humidity thresholds, wind speed thresholds, and temperature vertical distribution characteristics, which are used for secondary identification and judgment of the model output results.

[0039] Preferably, step S5, which involves introducing statistical thresholds for airport radiation fog elements to calibrate the results, includes:

[0040] S5.1 Threshold Calibration Stage: Dynamic threshold adjustment based on Bayesian probability, by treating the threshold as a random variable and updating its posterior distribution by combining historical data and real-time observations. The formula includes:

[0041] ;

[0042] in, For a moment The threshold; For a moment Observational data; prior distribution Determined by historical threshold statistics; likelihood function It reflects the degree of matching between the current data and the threshold;

[0043] S5.2 Fuzzy Logic Calibration: Fuzzy sets are introduced to describe the uncertain relationship between elements and fog conditions. The comprehensive fog probability is calculated through membership functions, as shown in the following formula:

[0044] ;

[0045] in, This is the membership function (such as the trapezoidal function) for each element.

[0046] Preferably, the method further includes the steps of outputting hourly refined forecast values ​​of runway visual range (RVR) at each runway end of the airport, and predicting the estimated time to reach Category I, Category II and Category III operational standards.

[0047] An airport radiation fog fine forecasting device, applicable to the aforementioned airport radiation fog fine forecasting method, includes:

[0048] Data collection and preprocessing module: Collects open source data and various historical meteorological data before and after airport fogging, and performs data cleaning, format unification and standardization processing;

[0049] Deep learning feature extraction and pattern recognition module: Constructs deep learning models suitable for time series data, and performs automated feature extraction and meteorological pattern recognition on sample data;

[0050] Preliminary model building and training module: Based on the extracted features, a preliminary prediction model is built, and the prediction performance is optimized by iteratively adjusting the model parameters and structure;

[0051] Time series modeling and prediction optimization module: Combining numerical weather prediction data, time series analysis methods are used to further optimize the fog prediction model;

[0052] Model Validation and Optimization Module: The model is validated using reserved sample data, the model prediction results are compared with the actual observation data, the model is optimized, and the statistical threshold of airport radiation fog elements is introduced for result calibration.

[0053] An airport radiation fog refinement forecasting program, implemented based on the above method, is executed on a computer device to achieve refined airport radiation fog forecasting.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. This invention achieves high-precision time-series modeling of the entire process of radiation fog generation, persistence and dissipation by deeply fusing multi-source meteorological data (historical observations, ground conditions, and numerical weather predictions) and combining them with a deep fully connected neural network. At the same time, it introduces an attention mechanism to dynamically optimize feature weights, enabling the model to adaptively focus on key meteorological factors, which greatly improves the accuracy and spatiotemporal consistency of forecasts and effectively overcomes the errors of traditional methods in critical condition simulation.

[0056] 2. This invention combines data-driven forecasting with meteorological and physical laws by introducing a dynamic threshold calibration based on Bayesian updates and a fuzzy logic judgment mechanism, thereby achieving secondary identification and physical consistency correction of the model output results. This method significantly suppresses abnormal forecasts caused by model overfitting or data noise, improves the discrimination ability under complex weather conditions, and thus greatly reduces the false alarms and missed alarms in radiation fog forecasts.

[0057] 3. This invention not only outputs the probability of fog formation, but also provides hourly detailed forecasts of the visual range of each runway end at the airport, and predicts the estimated time when Category I, II, and III operational standards will be met, directly transforming weather forecasts into decision-making information usable for aviation operations. This provides forward-looking and quantitative technical support for airport low-visibility operation management, flight scheduling, and runway resource allocation, significantly improving airport operational safety and efficiency under complex weather conditions. Attached Figure Description

[0058] Figure 1 This is a flowchart of the airport radiation fog refinement forecasting method of the present invention. Detailed Implementation

[0059] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0060] Example 1: This invention provides a method for refined forecasting of airport radiation fog, such as... Figure 1 As shown, it includes the following steps:

[0061] S1. Collect open-source data and various historical meteorological data before and after airport fog, and perform data cleaning, format unification and standardization processing;

[0062] S2. Construct a deep learning model suitable for time series data to automatically extract features and identify meteorological patterns from sample data;

[0063] S3. Construct a preliminary prediction model based on the extracted features, and optimize the prediction performance by iteratively adjusting the model parameters and structure.

[0064] S4. Combine numerical weather prediction (NWP) data and use time series analysis methods to further optimize the fog prediction model;

[0065] S5. Use reserved sample data to validate the model, compare the model prediction results with the actual observation data, optimize the model, and introduce the statistical threshold of airport radiation fog elements for result calibration.

[0066] As shown above, this method achieves accurate forecasting of airport radiation fog through a systematic data processing and model building process. Specific operations include: First, in step S1, various historical meteorological data and open-source data before and after airport fog formation are collected. Data cleaning, format unification, and standardization improve data quality and consistency, laying a reliable data foundation for subsequent modeling. Second, in step S2, a deep learning model suitable for time series is constructed to achieve automated feature extraction and pattern recognition of meteorological data, effectively capturing key temporal features of fog formation and dissipation. Next, in step S3, a preliminary prediction model is built based on the extracted features. An attention mechanism is introduced to dynamically adjust feature weights, enhancing the model's ability to predict important weather patterns. The method enhances the sensitivity of meteorological elements to improve forecast accuracy. Then, in step S4, numerical weather prediction (NWP) data and ground observation data are fused, and time series analysis methods (such as the Prophet model) and weighted loss functions are used to address the spatiotemporal inconsistency of data, enhancing the model's adaptability and robustness under complex meteorological conditions. Finally, in step S5, reserved samples are used to validate and optimize the model, and the results are calibrated by combining statistical thresholds for airport radiation fog elements (such as humidity and wind speed thresholds), achieving secondary analysis and correction of the forecast results and significantly reducing the rates of false alarms and missed alarms. This method, through the combination of multi-source data fusion, deep learning, and dynamic calibration, achieves refined and timely forecasts of the formation, persistence, and dissipation processes of airport radiation fog, providing reliable support for airport operational decisions.

[0067] Specifically, the specific operation steps of step S1 include:

[0068] S1.1 Collect open-source data and various historical meteorological data before and after airport fog, including but not limited to temperature, humidity, wind speed, wind direction, air pressure, cloud cover and runway visual range (RVR).

[0069] S1.2 Apply data cleaning algorithms to remove outliers, missing values, and duplicate data to ensure data quality;

[0070] S1.3 Convert data from different sources and in different formats into a unified data format to facilitate subsequent processing;

[0071] S1.4. Use Z-score standardization or Min-Max standardization to standardize the data in order to eliminate the influence of different units on model training.

[0072] The Z-score standardization formula is as follows: ,in This is the original data. The mean of the data. The standard deviation of the data;

[0073] The Min-Max standardization formula is: ,in The original data is given, where min and max represent the minimum and maximum values ​​of the data, respectively. This is the standardized data.

[0074] As shown above, in step S1, the system first collects open-source data and various historical meteorological data before and after airport fog, covering key elements such as temperature, humidity, wind speed, wind direction, air pressure, cloud cover, and runway visual range, providing a multi-dimensional and high-coverage data foundation for model construction. Subsequently, data cleaning algorithms are applied to remove outliers, complete missing data, and remove duplicate records, significantly improving the consistency and reliability of the dataset. Then, raw data from different sources and formats are converted into a unified structured format to ensure the efficiency and compatibility of subsequent processing. Finally, the data is normalized using Z-score standardization or Min-Max standardization methods to eliminate training bias caused by different units of measurement of various meteorological variables, enabling the model to learn various features more evenly. This provides high-quality, scale-uniform input data for subsequent deep learning models, effectively improving model convergence speed and forecast stability.

[0075] Specifically, step S2 includes the following steps:

[0076] S2.1. Feature extraction of time-series meteorological data based on the physical mechanism of airport fog formation;

[0077] S2.2. The model is trained using Deep Neural Networks (DNNs), as shown in the formula: ,in The kernel size is [size]. These are the weighting coefficients;

[0078] S2.3 Extract short-term and long-term time-dependent features using multi-scale convolutional kernels (such as 3×1, 5×1, 7×1);

[0079] S2.4 The regression task uses mean squared error (MSE) loss: The classification task uses cross-entropy loss: ;in, This represents the number of categories.

[0080] As shown above, firstly, features of time-series meteorological data are extracted based on the physical mechanism of airport fog formation to ensure that the feature selection closely matches the actual conditions for radiation fog formation. Secondly, a deep fully connected neural network (DNN) is used for model training, which automatically learns the complex relationships between meteorological elements through nonlinear transformation, thereby improving the model's expressive power. Furthermore, short-term and long-term time-dependent features are extracted using multi-scale convolutional kernels (3×1, 5×1, 7×1), effectively capturing the multi-level temporal patterns in fog evolution and enhancing prediction stability. Finally, S2.4 employs mean squared error (MSE) loss for regression tasks and cross-entropy loss for classification tasks, ensuring that the model training objective accurately matches specific forecasting needs, significantly improving the accuracy of fog formation probability prediction, and laying a reliable foundation for subsequent refined forecasting.

[0081] Specifically, the specific operation steps of step S3 are as follows:

[0082] During model training, an attention weight layer is introduced to adjust the weights in real time based on the predictive contribution of features in historical data. The attention weight layer includes attention weight calculation and feature fusion.

[0083] The attention weight is calculated as follows: ;in, For the first The hidden representation of each feature, For a multilayer perceptron, the weight allocation is optimized through gradient descent;

[0084] The feature fusion is as follows: ;in, For the first The original value of a feature or the feature extracted by a deep fully connected neural network.

[0085] As shown above, in step S3, by introducing an attention weight layer, weights are dynamically allocated during model training based on the predictive contribution of each feature in historical data. Specifically, this includes: first, calculating attention weights by using a multilayer perceptron (MLP) to perform a nonlinear transformation on the hidden representation of each feature, and optimizing the weight allocation through gradient descent, enabling the model to automatically focus on key meteorological elements related to fog conditions; then, performing feature fusion by weighting and fusing the weighted features with their original values ​​or deep features extracted by a deep fully connected neural network to form a comprehensive meteorological representation. This method enables the model to have adaptive feature selection capabilities, suppressing noise interference and highlighting key influencing factors, thereby significantly improving the discriminativeness and stability of feature expression, and providing more robust and interpretable feature inputs for subsequent accurate predictions.

[0086] Specifically, the specific operation steps of step S4 are as follows:

[0087] S4.1. Transform the gridded data of NWP (such as temperature, humidity, wind speed, cloud cover, etc.) into time series features and integrate them with ground observation data; specifically including:

[0088] S4.01 and NWP data are typically gridded outputs (e.g., latitude and longitude grids), while ground observation station data is point data, requiring spatial matching, including:

[0089] Interpolate NWP grid data to station coordinates: ;in, Distance weights The value is the value of the four nearest grid points;

[0090] Synchronize the hourly forecast data of NWP with the ground observation data by timestamp and handle time lag (e.g., NWP forecast 6 hours in advance).

[0091] S4.02 and NWP data contain information at different time scales (such as short-term fluctuations and long-term trends), which need to be decomposed and modeled separately, including:

[0092] The NWP time series is decomposed into IMF (Intrinsic Mode Function) and residuals: ;in, Representing different frequency components, This is a trend item;

[0093] S4.2 Input NWP characteristics and ground observation data into a time series model (such as Prophet) to capture dynamic evolution patterns; the formula of the time series model includes: ;in, For trend items, For seasonal items, Due to the holiday effect, The regression coefficients for the NWP trend;

[0094] S4.3 To address the spatiotemporal inconsistency between NWP and ground observation data and improve model robustness, a weighted loss function is added, the formula of which includes: ;in, For the MSE loss of ground observations, The loss is used for NWP feature reconstruction (e.g., autoencoder reconstruction error). , These are the weighting coefficients;

[0095] S4.4 To ensure the model adapts to the update cycle of NWP data and sudden changes in fog conditions, a Bayesian hyperparameter optimization is introduced for the update cycle of NWP data. The formula includes: ;in, This is the latest NWP data; a mutation detection technique is introduced to address fog abrupt changes in the NWP data. This technique uses the CUSUM (cumulative sum) algorithm to monitor abrupt changes in prediction error, and the formula includes: Among them, when > Threshold, triggering model retraining.

[0096] As shown above, in step S4, through a multi-stage fusion and optimization mechanism, the numerical weather prediction (NWP) grid data and ground observation data are deeply fused: First, through spatial interpolation and time synchronization, the NWP gridded data is matched to the coordinates of the observation stations and time lag is eliminated, achieving spatiotemporal alignment of the data; Second, the NWP sequence is decomposed at multiple scales to extract its different frequency components and trend terms, so as to separate short-term fluctuations and long-term evolution characteristics; Subsequently, the fused multi-source data is input into a time series model (such as Prophet) to model the dynamic change law of meteorological elements, and a weighted loss function is introduced to balance the ground observation error and the NWP feature reconstruction loss, effectively mitigating the impact of data spatiotemporal inconsistency on the model; In addition, for the NWP update cycle and fog change characteristics, Bayesian hyperparameter optimization is used to adaptively adjust the model parameters, and the CUSUM algorithm is used to monitor the change of prediction error in real time, triggering dynamic retraining of the model. This series of operations enables the model to efficiently fuse multi-source heterogeneous meteorological data, dynamically model complex weather processes, and respond quickly to sudden fog events, thereby significantly improving the spatiotemporal consistency, stability, and adaptability to weather changes in forecasts.

[0097] Specifically, the statistical thresholds mentioned in step S5 include, but are not limited to, humidity thresholds, wind speed thresholds, and temperature vertical distribution characteristics, which are used for secondary identification and judgment of the model output results.

[0098] As shown above, in step S5, based on historical statistics and physical mechanism analysis, the system introduces statistical thresholds for key meteorological elements such as humidity threshold, wind speed threshold, and vertical temperature distribution characteristics to perform secondary identification and judgment on the preliminary prediction results output by the model. First, by combining historical fog cases and real-time observation data, the range of each threshold is dynamically set and calibrated. For example, when the near-surface relative humidity is consistently higher than the set threshold, the wind speed is lower than the static stability condition threshold, and a significant temperature inversion stratification occurs, the probability of radiation fog formation is determined to be significantly increased. Subsequently, these threshold conditions are used to perform logical verification and physical consistency correction on the model prediction results, filtering out abnormal prediction outputs that do not conform to meteorological mechanisms. This method, by integrating data-driven prediction and physical threshold constraints, effectively compensates for the shortcomings of pure model methods in simulating critical conditions, significantly improves the physical rationality and reliability of the forecast results, greatly reduces false alarms and missed alarms caused by model errors, and provides a more reliable decision-making basis for the classification forecast of airport runway operation standards.

[0099] Specifically, step S5, which involves introducing statistical thresholds for airport radiation fog elements to calibrate the results, includes:

[0100] S5.1 Threshold Calibration Stage: Dynamic threshold adjustment based on Bayesian probability, by treating the threshold as a random variable and updating its posterior distribution by combining historical data and real-time observations. The formula includes:

[0101] ;

[0102] in, For a moment The threshold; For a moment Observational data; prior distribution Determined by historical threshold statistics; likelihood function It reflects the degree of matching between the current data and the threshold;

[0103] S5.2 Fuzzy Logic Calibration: Fuzzy sets are introduced to describe the uncertain relationship between elements and fog conditions. The comprehensive fog probability is calculated through membership functions, as shown in the following formula:

[0104] ;

[0105] in, This is the membership function (such as the trapezoidal function) for each element.

[0106] As shown above, in step S5, a secondary judgment method based on dynamic thresholds and physical mechanisms is used to calibrate the model prediction results. First, based on historical radiation fog cases and real-time observation data, key parameters such as humidity threshold, wind speed threshold, and vertical temperature distribution characteristics are dynamically calibrated and adaptively updated. Then, the preliminary forecast output by the model is compared and logically verified item by item with these threshold conditions. For example, only when the forecast scenario simultaneously meets the conditions of "relative humidity higher than the dynamic threshold," "wind speed lower than the static stability threshold," and "significant temperature inversion stratification," does the system finally confirm the high risk of radiation fog formation and correct or remove abnormal outputs that do not conform to physical laws. This method, by introducing prior meteorological knowledge as a constraint, embeds a physical consistency verification mechanism into the data-driven prediction framework, thereby significantly enhancing the reliability and interpretability of the forecast results. It effectively suppresses "false alarms" and "missed alarms" caused by the model's over-reliance on statistical laws, providing airport operation decisions with high-quality forecast products that combine intelligence and physical reliability.

[0107] Comparative Example 1: The existing technical solution is: an airport radiation fog forecasting method, the steps of which include:

[0108] Step 1, Data Sources: Primarily relies on numerical weather prediction (NWP) grid data, with limited use of ground observation data;

[0109] Step 2, Data Processing: Simply interpolate NWP data to airport location without performing multi-source data fusion and standardization.

[0110] Step 3, Prediction Model: Traditional machine learning methods such as multiple linear regression, logistic regression, or decision trees are used, but time-series deep learning models are not used;

[0111] Step 4, Feature Extraction: A few meteorological factors (such as temperature, humidity, and wind) were manually selected, without achieving automated feature learning and multi-scale temporal modeling;

[0112] Step 5, Calibration Method: The results are judged using fixed thresholds or empirical formulas, lacking dynamic adjustment and integration with physical mechanisms;

[0113] Step 6, Output format: Output "fog / no fog" classification results, or a rough visibility level, without providing hourly runway visual range (RVR) forecasts and operational standard time predictions.

[0114] Based on the technical content of Embodiment 1 and the performance comparison with Comparative Example 1, the following table is obtained:

[0115] Performance indicators Example 1 (This Invention) Comparative Example (Prior Art) Improvement range / explanation Fog forecast accuracy (%) 92-95 75-80 An increase of approximately 12-20 percentage points False alarm rate (%) 5-8 15-20 Reduced by approximately 7-15 percentage points Missed report rate (%) 3-6 10-15 Reduced by approximately 5-12 percentage points RVR forecast mean absolute error (m) 50-100 200-300 Error reduced by approximately 60-70% Forecast lead time (hours) 24-48 3-6 Extend the effective period of the forecast System response time (seconds) <5 <2 Slightly lower real-time performance but richer output

[0116] As shown in the table above, the performance comparison data clearly demonstrates the significant progress of this invention compared to existing technologies using quantitative indicators: In terms of forecast accuracy, the fog forecast accuracy has been greatly improved from the original 75%-80% to 92%-95%, while the false alarm rate and missed alarm rate have been reduced by about 7-15 and 5-12 percentage points respectively, proving that its prediction results are more reliable; In terms of forecast refinement, the average absolute error of runway visual range (RVR) has been greatly reduced from 200-300 meters to 50-100 meters, a reduction of about 60%-70%, achieving a precise characterization of visibility conditions; In addition, the effective forecast lead time has been extended from 3-6 hours to 24-48 hours, and the system response time can still be kept within a few seconds, fully meeting the urgent needs of airports for high-efficiency and refined operational decision support.

[0117] Example 2: This example is basically the same as the previous example, except that the method also includes the step of outputting hourly refined forecast values ​​of the runway visual range (RVR) at each runway end of the airport, and predicting the estimated time to reach Category I, Category II and Category III operation standards.

[0118] Comparative Example 2: A method for early warning of low visibility operations at airports, comprising the following steps:

[0119] Step 1, Data and Modeling: It mainly relies on a single source (such as manual observation reports from airport meteorological stations or simple visibility sensor data), or directly interpolates the visibility output of numerical weather prediction; it does not integrate multi-source, multi-scale data, nor does it build a physical / statistical transformation model from meteorological elements to RVR;

[0120] Step 2, Forecast Output:

[0121] Spatially: It usually gives a "visibility" forecast value that represents the entire airport or a certain area, but it cannot distinguish the local differences between different runways and different ends of the same airport (RVR is significantly affected by local factors such as runway lights and background brightness).

[0122] In terms of time: visibility levels are given at 3 or 6-hour intervals (e.g., "<800 meters", "<400 meters"), which cannot provide a continuous and refined evolution trend for each hour;

[0123] Decision transition: Regulators and airlines need to manually estimate the time windows that may meet Category I, II, and III operations based on the forecasted visibility range and the operational standards manual. This process relies on experience and is not timely or accurate.

[0124] Key limitations: There is a "last mile" gap between forecast information and operational decisions; the provided "visibility" cannot be automatically and accurately matched with the "runway visual range (RVR)" and "operational standards" on which aircraft take-off and landing are directly based, resulting in decision-making delays and inefficient resource allocation.

[0125] Based on the technical content of Embodiment 2 and the performance comparison with Comparative Embodiment 2, the following table is obtained:

[0126] Operational performance indicators Example 2 (Invention) Comparative Example (Prior Art) Explanation of Efficiency Improvement LVP Decision Lead Time It can make accurate start / stop predictions 2-4 hours in advance. Typically, only a rough judgment can be made 0.5-1 hour in advance. This bought valuable time for flight adjustments and ground resource allocation. Average deviation between RVR forecast and actual data ≤100 meters (near the critical threshold) Typically ≥300 meters, and fluctuates greatly around the threshold. Significantly reduce false alarms or missed alarms caused by forecast bias. Time prediction error to reach operating standards Average error ≤ 30 minutes Unable to provide quantifiable timeframes, relying on experience, errors often exceed 1 hour. Improve the predictability of flight schedule management Runway capacity utilization during low visibility periods Accurate forecasting can improve accuracy by 15%-25%. Due to conservative decision-making and delayed response, utilization rate is low. Directly improves the airport's operational capacity and economic efficiency during severe weather. Decision-making collaboration efficiency Air traffic control, airlines, and airports collaborate using the same set of quantitative data, resulting in high efficiency. Based on different interpretations and experience, various departments held numerous and lengthy collaborative meetings. Reduce communication costs and improve the consistency and speed of overall operational decision-making.

[0127] As shown in the table above, this performance comparison data clearly illustrates the fundamental breakthroughs of the technical solution in Example 2 compared to existing technologies from both technical and operational perspectives. Technically, it achieves a leap from providing general "weather visibility" warnings to directly outputting the core parameter for aviation decision-making—"Runway Visual Range (RVR)"—and improves forecast accuracy to a refined level of "runway end, hourly." More importantly, it can automatically predict the specific time to reach different operational standards, completing end-to-end output from "weather forecast" to "operational decision recommendations." In terms of benefits, these technical advantages directly translate into significant operational value: the decision lead time is extended from less than 1 hour to 2-4 hours, RVR forecast deviation is reduced by more than 60%, and the time prediction error for operational standards is controlled within 30 minutes. This significantly improves runway capacity utilization under low visibility conditions and the efficiency of collaborative decision-making among various departments, fundamentally enhancing the airport's safety and operational support capabilities.

[0128] As shown above, this method, based on fog forecasting, further outputs hourly refined forecasts of runway visual range (RVR) at each runway end of the airport, and predicts the estimated time for reaching Category I, II, and III operational standards. The system dynamically generates hourly RVR evolution sequences for each runway end based on a statistical relationship model between meteorological elements and RVR, combined with real-time observation and forecast data. Simultaneously, based on the RVR and cloud base height thresholds corresponding to different operational levels, the system automatically determines and outputs the time nodes and duration windows that meet the standards for each level. This step directly transforms meteorological forecasts into decision-making information usable for aviation operations, enabling air traffic control, flight operations, and flight crews to grasp the refined evolution trends of runway operating conditions in advance. This provides quantitative and forward-looking decision support for flight takeoffs and landings, runway allocation, and operational mode transitions, effectively improving the airport's operational efficiency and safety margin under low visibility conditions.

[0129] Working Principle: First, the system constructs a high-quality meteorological dataset through multi-source data fusion and standardized preprocessing. Then, it automatically extracts key meteorological features and evolution patterns from the time series using a deep fully connected neural network. Next, an attention mechanism is introduced to dynamically optimize feature weights, and numerical weather prediction (NWP) grid data and ground observation data are integrated. Through spatiotemporal alignment, multi-scale decomposition, and adaptive loss function design, the model's ability to model complex weather processes is enhanced. In the prediction output stage, the system further combines dynamic thresholds based on historical statistics and physical mechanisms for result calibration, achieving refined discrimination of fog formation, persistence, and dissipation processes. Finally, meteorological element forecasts are transformed into hourly runway visual range (RVR) values ​​and forecasts of achievement times for different operational levels, forming a fully automated forecasting system from data acquisition and intelligent analysis to operational decision support. This solution, by integrating deep learning, multi-source data collaboration, and physical constraint calibration, significantly improves the spatiotemporal accuracy and reliability of radiation fog forecasts, effectively supporting operational safety and efficiency optimization at airports under low visibility conditions.

[0130] An airport radiation fog fine forecasting device, applicable to the aforementioned airport radiation fog fine forecasting method, includes:

[0131] Data collection and preprocessing module: Collects open source data and various historical meteorological data before and after airport fogging, and performs data cleaning, format unification and standardization processing;

[0132] Deep learning feature extraction and pattern recognition module: Constructs deep learning models suitable for time series data, and performs automated feature extraction and meteorological pattern recognition on sample data;

[0133] Preliminary model building and training module: Based on the extracted features, a preliminary prediction model is built, and the prediction performance is optimized by iteratively adjusting the model parameters and structure;

[0134] Time series modeling and prediction optimization module: Combining numerical weather prediction data, time series analysis methods are used to further optimize the fog prediction model;

[0135] Model Validation and Optimization Module: The model is validated using reserved sample data, the model prediction results are compared with the actual observation data, the model is optimized, and the statistical threshold of airport radiation fog elements is introduced for result calibration.

[0136] As shown above, the device systematically integrates functional modules such as data preprocessing, feature extraction, model building, multi-source fusion, and verification calibration, ensuring the continuity and integrity of the forecasting process. Its hybrid modeling approach, combining deep learning and time series analysis, effectively captures the multi-scale characteristics and dynamic patterns of radiation fog formation and evolution. By introducing a physical threshold calibration mechanism, the device embeds meteorological prior knowledge into data-driven forecasts, significantly improving the physical rationality and interpretability of the forecast results. Ultimately, the device can output refined runway visibility forecast products that can directly support operational decisions, enabling airports to grasp the impact trends of low visibility weather in advance and accurately, thereby effectively reducing operational risks, optimizing the allocation of support resources, and improving flight on-time performance, providing reliable technical support for the safe and efficient operation of airports.

[0137] This application provides an electronic device applicable to the aforementioned AI-based autonomous operation and maintenance system for satellite communication ground stations, comprising:

[0138] Memory is used to protect computer programs and data;

[0139] A processor is used to run system programs.

[0140] This application provides a computer storage medium applicable to the aforementioned AI-based autonomous operation and maintenance system for satellite communication ground stations, and performs hierarchical confidentiality management of the system and data in accordance with confidentiality management requirements.

[0141] Those skilled in the art will understand that embodiments of this application can be provided as a system or a computer program product. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] This application is described with reference to flowchart illustrations and / or block diagrams of devices (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, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0143] 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.

[0144] 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.

[0145] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0146] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0147] Computer-readable media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0148] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, article, or apparatus that includes that element.

[0149] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for refined forecasting of airport radiation fog, characterized in that, Includes the following steps: S1. Collect open-source data and various historical meteorological data before and after airport fog, and perform data cleaning, format unification and standardization processing; S2. Construct a deep learning model suitable for time series data to automatically extract features and identify meteorological patterns from sample data; S3. Construct a preliminary prediction model based on the extracted features, and optimize the prediction performance by iteratively adjusting the model parameters and structure. S4. Combine numerical weather prediction data and use time series analysis methods to further optimize the fog prediction model; S5. Use reserved sample data to validate the model, compare the model prediction results with the actual observation data, optimize the model, and introduce the statistical threshold of airport radiation fog elements for result calibration.

2. The airport radiation fog fine forecasting method as described in claim 1, characterized in that: The specific operation steps of step S1 include: S1.1 Collect open-source data and various historical meteorological data before and after airport fog, including but not limited to temperature, humidity, wind speed, wind direction, air pressure, cloud cover and runway visual range; S1.2 Apply data cleaning algorithms to remove outliers, missing values, and duplicate data to ensure data quality; S1.3 Convert data from different sources and in different formats into a unified data format to facilitate subsequent processing; S1.

4. Use Z-score standardization or Min-Max standardization to standardize the data in order to eliminate the influence of different units on model training. The Z-score standardization formula is as follows: ,in This is the original data. The mean of the data. The standard deviation of the data; The Min-Max standardization formula is: ,in The original data is given, where min and max represent the minimum and maximum values ​​of the data, respectively. This is the standardized data.

3. The method for refined forecasting of airport radiation fog as described in claim 1, characterized in that: The specific steps of step S2 include: S2.

1. Feature extraction of time-series meteorological data based on the physical mechanism of airport fog formation; S2.

2. The model is trained using a deep fully connected neural network, as shown in the formula: ,in The kernel size is [size]. These are the weighting coefficients; S2.3 Extract short-term and long-term time-dependent features using multi-scale convolutional kernels; S2.4, The regression task uses mean squared error loss: The classification task uses cross-entropy loss: ;in, This represents the number of categories.

4. The method for refined forecasting of airport radiation fog as described in claim 1, characterized in that: The specific operation steps of step S3 are as follows: During model training, an attention weight layer is introduced to adjust the weights in real time based on the predictive contribution of features in historical data. The attention weight layer includes attention weight calculation and feature fusion. The attention weight is calculated as follows: ;in, For the first The hidden representation of each feature, For a multilayer perceptron, the weight allocation is optimized through gradient descent; The feature fusion is as follows: ;in, For the first The original value of a feature or the feature extracted by a deep fully connected neural network.

5. The method for refined forecasting of airport radiation fog as described in claim 1, characterized in that: The specific steps of step S4 are as follows: S4.

1. Transform the gridded data from numerical weather prediction into time series features and integrate it with ground observation data; specifically including: S4.

01. Numerical weather forecast data is typically output in a gridded format, while ground observation station data is point data, requiring spatial matching, including: Interpolate numerical weather prediction grid data to observation station coordinates: ;in, Distance weights The value is the value of the four nearest grid points; The hourly forecast data of numerical weather prediction is synchronized with the ground observation data according to the timestamp, and the time lag is processed. S4.

02. Numerical weather prediction data contains information at different time scales and needs to be decomposed and modeled separately, including: Decompose numerical weather prediction time series into IMF and residuals: ;in, Representing different frequency components, This is a trend item; S4.2 Input numerical weather prediction characteristics and ground observation data into a time series model to capture dynamic evolution patterns; the formulas of the time series model include: ;in, For trend items, For seasonal items, Due to the holiday effect, The regression coefficients represent the trends in numerical weather prediction. S4.3 To address the spatiotemporal inconsistency between numerical weather prediction and ground observation data and improve model robustness, a weighted loss function is added, the formula of which includes: ;in, This represents the mean square error loss from ground observations. For numerical weather prediction feature reconstruction loss, , These are the weighting coefficients; S4.4 To ensure the model adapts to the update cycle of numerical weather prediction data and sudden changes in fog conditions, a Bayesian hyperparameter optimization is introduced for the update cycle of numerical weather prediction data. The formula includes: ;in, This is based on the latest numerical weather prediction data. A mutation detection technique is introduced to address abrupt changes in fog conditions within the numerical weather prediction data. This technique uses the CUSUM algorithm to monitor abrupt changes in prediction errors, and the formula includes: Among them, when > Threshold, triggering model retraining.

6. The method for refined forecasting of airport radiation fog as described in claim 1, characterized in that: The statistical thresholds mentioned in step S5 include, but are not limited to, humidity thresholds, wind speed thresholds, and temperature vertical distribution characteristics, which are used for secondary identification and judgment of the model output results.

7. The method for refined forecasting of airport radiation fog as described in claim 6, characterized in that: The step S5, which involves calibrating the results by introducing statistical thresholds for airport radiation fog elements, includes: S5.1 Threshold Calibration Stage: Dynamic threshold adjustment based on Bayesian probability, by treating the threshold as a random variable and updating its posterior distribution by combining historical data and real-time observations. The formula includes: ； in, For a moment The threshold; For a moment Observational data; prior distribution Determined by historical threshold statistics; likelihood function It reflects the degree of matching between the current data and the threshold; S5.2 Fuzzy Logic Calibration: Fuzzy sets are introduced to describe the uncertain relationship between elements and fog conditions. The comprehensive fog probability is calculated through membership functions, as shown in the following formula: ; in, is the membership function for each element.

8. The method for refined forecasting of airport radiation fog as described in claim 1, characterized in that: It also includes the steps of outputting hourly refined forecast values ​​of the visual range of each runway end of the airport, and predicting the estimated time to reach Category I, Category II and Category III operational standards.

9. A refined forecasting device for airport radiation fog, characterized in that: The method for refined forecasting of airport radiation fog as described in any one of claims 1-8 includes: Data collection and preprocessing module: Collects open source data and various historical meteorological data before and after airport fogging, and performs data cleaning, format unification and standardization processing; Deep learning feature extraction and pattern recognition module: Constructs deep learning models suitable for time series data, and performs automated feature extraction and meteorological pattern recognition on sample data; Preliminary model building and training module: Based on the extracted features, a preliminary prediction model is built, and the prediction performance is optimized by iteratively adjusting the model parameters and structure; Time series modeling and prediction optimization module: Combining numerical weather prediction data, time series analysis methods are used to further optimize the fog prediction model; Model Validation and Optimization Module: The model is validated using reserved sample data, the model prediction results are compared with the actual observation data, the model is optimized, and the statistical threshold of airport radiation fog elements is introduced for result calibration.

10. A refined forecasting program for airport radiation fog, characterized in that, The program is implemented based on the method of any one of claims 1 to 8 and is used to execute on a computer device to achieve refined forecasting of airport radiation fog.

Images