Civil aviation low-level wind shear detection method and system with adaptive weather background
By constructing an adaptive dynamic threshold model and utilizing logistic regression trees and gradient boosting decision trees, the problems of high false alarm rates and missed detections in low-altitude wind shear detection are solved. Dynamic adjustments based on airport weather background are achieved, improving detection accuracy and robustness, and making it suitable for various airport environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE SECOND RES INST OF CIVIL AVIATION ADMINISTRATION OF CHINA
- Filing Date
- 2026-04-28
- Publication Date
- 2026-05-29
AI Technical Summary
Existing technologies for low-altitude wind shear detection suffer from high false alarm rates, high risk of missed detections, and difficulty in adapting to the wind field structure of different airports. They also cannot be dynamically adjusted according to the real-time weather background of the airport, resulting in insufficient detection accuracy.
A dynamic threshold model composed of logistic regression tree and gradient boosting decision tree is adopted. By acquiring historical meteorological data and wind shear factor, correlation analysis is performed to train the dynamic threshold model. The adaptive wind shear discrimination threshold is calculated based on real-time meteorological data and updated online to achieve adaptive wind shear detection.
It improves the accuracy and robustness of wind shear detection, reduces the false alarm rate and the probability of missed detection, adapts to the meteorological environment changes of different airports, provides higher flight safety assurance, and does not require hardware upgrades, making it suitable for large and medium-sized airports.
Smart Images
Figure CN122116700A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of civil aviation technology, specifically relating to a method and system for detecting low-altitude wind shear in civil aviation under adaptive weather conditions. Background Technology
[0002] Low-level wind shear refers to an atmospheric phenomenon occurring in the space below approximately 500 meters above the ground, where wind speed or direction changes drastically over a short distance. Based on spatial structure characteristics, it can be divided into various forms, including horizontal wind shear and vertical wind shear. The causes of low-level wind shear are complex and diverse, exhibiting characteristics such as varying scales, rapid changes, and strong suddenness. Its horizontal scale typically ranges from tens of meters to several kilometers, its duration varies from several minutes to tens of minutes, and wind speed changes can reach tens of meters per second.
[0003] Low-altitude wind shear poses a serious threat to civil aviation safety. During takeoff and landing, aircraft are extremely sensitive to changes in wind fields due to their low speed and limited altitude. When encountering strong wind shear, the aircraft's airspeed and lift change drastically, potentially leading to a sudden loss of altitude and difficulty in handling. Historical data shows that many major aviation accidents are directly related to low-altitude wind shear. Even with continuous improvements in aircraft performance, wind shear remains a significant challenge to modern aviation safety.
[0004] Currently, the detection of low-level wind shear is mainly achieved through a combination of airborne and ground-based systems. Airborne detection relies primarily on pilot visual observation and airborne wind shear warning systems (including reactive and predictive systems); ground-based detection includes equipment such as Doppler weather radar, wind profiler radar, and lidar. The International Civil Aviation Organization (ICAO) defines severe wind shear as a rapid change in wind direction or speed greater than 15 knots (approximately 7.7 meters per second), providing an important reference threshold for wind shear detection. However, wind shear exhibits high spatiotemporal variability, and its intensity and scale are closely related to specific weather conditions. Existing technologies still have significant limitations in practical applications. Each airport has different climatic characteristics; simply using the aforementioned fixed threshold for judgment often fails to match the actual wind field structure of the airport, leading to a high false alarm rate under complex weather conditions or missed detections in areas bordering weak wind shear.
[0005] Although some airports have deployed low-level wind shear warning systems, their detection performance is still limited by the aforementioned factors. Currently used fixed threshold criteria cannot be dynamically adjusted according to real-time weather background at the airport, making them ill-suited to changing meteorological environments. To improve the reliability and timeliness of detection, there is an urgent need to develop a new detection method capable of intelligently sensing changes in weather background. Specifically, adaptive detection thresholds related to the weather background should be constructed based on the climate characteristics and weather types of different airports, thereby overcoming the false alarm and missed detection problems caused by existing fixed criteria and achieving accurate identification and effective early warning of low-level wind shear. Summary of the Invention
[0006] One of the objectives of this invention is to provide a civil aviation low-altitude wind shear detection method that is adaptive to weather backgrounds. This method can dynamically adjust the detection threshold according to the real-time meteorological environment and weather type of the airport, so as to solve the problems of high false alarm rate, high risk of missed detection and difficulty in adapting to different airport wind field structures caused by using fixed thresholds in the prior art.
[0007] The second objective of this invention is to provide a civil aviation low-altitude wind shear detection system that adapts to weather backgrounds, for performing the above-mentioned detection method.
[0008] A third objective of this invention is to provide a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the above-described adaptive weather background civil aviation low-altitude wind shear detection method.
[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention discloses a method for detecting low-altitude wind shear in civil aviation under adaptive weather conditions, which includes the following steps: Historical meteorological data, historical wind shear factor data, and pilot report real labels are obtained. Correlation analysis is performed on the historical meteorological data and historical wind shear factor to obtain a feature set. Using the feature set as the model input and the true labels of the pilot reports as the model output labels, a dynamic threshold model composed of a logistic regression tree and a gradient boosting decision tree (GBDT) is trained. Initiate the detection cycle, collect real-time radial wind speed data and real-time meteorological data, and calculate the wind shear factor for the current cycle based on the real-time radial wind speed data; The real-time meteorological data and the calculated current period wind shear factor are input into the trained dynamic threshold model. The model automatically calculates and outputs an adaptive wind shear discrimination threshold suitable for the current weather background based on the combination characteristics of the current meteorological elements and the wind shear factor. Based on the adaptive wind shear discrimination threshold, a threshold decision is made on the wind shear factor of the current cycle, and an alarm or no alarm result is output. Add new data from the current period to the training set, update the dynamic threshold model online, and wait for the next detection period.
[0010] In some embodiments of the present invention, the meteorological data includes four types of meteorological element data, namely temperature, humidity, air pressure and rainfall.
[0011] In some embodiments of the present invention, the correlation analysis specifically involves performing normality tests on historical sample data of four meteorological elements (temperature, humidity, air pressure, and rainfall) and historical sample data of wind shear factor. When the historical sample data of a certain type of meteorological element to be analyzed and the historical sample data of wind shear factor both meet the normal distribution condition, the Pearson correlation coefficient is used to calculate the correlation between the meteorological element and the historical wind shear factor. When the historical sample data of a certain type of meteorological element to be analyzed does not meet the normal distribution condition, or the historical sample data of the historical wind shear factor does not meet the normal distribution condition, the Spearman rank correlation coefficient is used to analyze the correlation between the meteorological element and the historical wind shear factor. While calculating the correlation coefficient, also calculate the coefficient of determination R. 2 The statistical significance level (P-value) is used to assess the strength of the correlation between various meteorological elements and historical wind shear factors, as well as the statistical reliability. Based on the analysis results, meteorological elements that are strongly correlated with historical wind shear factors and whose statistical significance meets the preset requirements were selected as key input features for the subsequent construction of a dynamic threshold model.
[0012] In some embodiments of the present invention, the process of training the dynamic threshold model specifically includes: Training data preparation: Strongly correlated meteorological data, wind shear factor and pilot report real labels are extracted from historical data at the same time to form a training dataset, in which the pilot report real labels are used as annotation labels for model training. Base learner creation: Create a logistic regression tree model as a gradient boosting base learner to initially fit the nonlinear relationship between meteorological features and the probability of wind shear occurrence; Integrated iterative training: Based on the logistic regression tree-based learner, the gradient boosting decision tree algorithm is used for iterative training. The prediction residuals of the preceding learners are fitted through multiple rounds of iteration. Finally, the weighted sum of multiple weak learners is obtained to obtain the dynamic threshold model.
[0013] In some embodiments of the present invention, the configuration parameters of the logistic regression tree model include: a maximum number of splits of 80, a minimum number of samples per leaf node of 5, and each tree using all input features; The configuration parameters for the integrated iterative training include: using least squares boosting (LSBoost) as the boosting method, 200 iterations, a learning rate of 0.1, real-time meteorological data as the model input, and an adaptive wind shear discrimination threshold for the current meteorological background as the output.
[0014] In some embodiments of the present invention, the threshold decision is specifically as follows: the wind shear factor calculated in the current period is compared with the adaptive wind shear discrimination threshold. If the wind shear factor exceeds the adaptive wind shear discrimination threshold, the current indicator state is determined to be abnormal and an alarm result is output; otherwise, the current indicator state is determined to be normal and no alarm result is output.
[0015] In some embodiments of the present invention, the online update of the dynamic threshold model specifically includes: adding the real-time radial wind speed data collected in the current period, real-time meteorological data, the wind shear factor calculated in the current period, and the decision result label obtained from the threshold decision to the historical training set; and re-executing the feature screening and model training steps based on the updated historical training set to complete the incremental update of the dynamic threshold model.
[0016] The judgment was confirmed in conjunction with subsequent pilot reports.
[0017] A second aspect of this invention discloses a civil aviation low-altitude wind shear detection system adapted to weather conditions, which is used to perform the aforementioned civil aviation low-altitude wind shear detection method adapted to weather conditions. The detection system includes: The feature preprocessing module is used to acquire historical meteorological data, historical wind shear factors, and pilot report real labels, and to perform correlation analysis on the historical meteorological data and historical wind shear factors to obtain a feature set. The model training module is used to train a dynamic threshold model consisting of a logistic regression tree and a gradient boosting decision tree (GBDT) based on the feature set and the true labels reported by the pilots. The real-time data processing module is used to start the detection cycle, collect real-time radial wind speed data and real-time meteorological data, and calculate the wind shear factor for the current cycle based on the real-time radial wind speed data. The threshold generation module is used to input the current period wind shear factor and real-time meteorological data to complete the training of the dynamic threshold model and generate an adaptive wind shear discrimination threshold. The anomaly judgment module is used to make a threshold judgment on the current indicator status based on the adaptive wind shear discrimination threshold, and output an alarm or no alarm result. The incremental update module is used to add new data from the current period to the training set and update the dynamic threshold model online.
[0018] In some embodiments of the present invention, the feature preprocessing module is specifically used to perform the correlation analysis using Pearson correlation coefficient or Spearman correlation coefficient to screen and obtain a feature set that meets the correlation requirements; The model training module is specifically used to jointly train the gradient boosting decision tree (GBDT) and the logistic regression tree to obtain a dynamic threshold model that can output an adaptive wind shear discrimination threshold based on the input features.
[0019] The third aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the above-described civil aviation low-altitude wind shear detection method adapted to weather background.
[0020] Compared with the prior art, the present invention has the following beneficial effects: This invention addresses key technical issues in airport wind shear detection by proposing a systematic solution that offers benefits in terms of detection accuracy, implementation cost, and engineering practicality, forming a wind shear detection technology system that combines high performance, high compatibility, and self-evolution capabilities.
[0021] In terms of detection accuracy and environmental adaptability, this invention overcomes the limitations of traditional fixed-threshold detection methods that rely on a single threshold. By constructing a dynamic threshold model that has a clear correlation with airport temperature, humidity, air pressure, and rainfall, the wind shear detection system possesses the ability to automatically adjust according to changes in weather background. Under complex meteorological conditions such as thunderstorms, strong winds, and temperature inversions, the system can automatically match the corresponding discrimination threshold based on real-time meteorological characteristics, thereby effectively filtering out false alarms caused by meteorological clutter and simultaneously achieving accurate capture of weak-energy shear signals. Furthermore, this invention designs a closed-loop self-optimization mechanism, using each detection result for online learning and model updates, enabling the detection performance to progressively improve with data accumulation and adaptively track the seasonal changes and long-term evolution patterns of the airport's local climate. Compared to traditional methods, this invention improves the accuracy and robustness of wind shear identification, providing higher safety assurance for aircraft during approach and landing phases.
[0022] In terms of implementation cost and deployment flexibility, this invention does not rely on upgrading airborne equipment or adding external dedicated detection hardware. Instead, it improves detection performance through algorithm-level optimization based on existing automatic weather station data and lidar wind measurement data at airports. This method is compatible: firstly, it is not limited to specific wind shear inversion algorithms and can adapt to various existing wind shear detection algorithms. As long as a standardized wind shear factor is output, all can be integrated into the dynamic threshold decision process of this invention. Secondly, it has no special requirements for the lidar scanning mode; it is applicable to PPI scanning, RHI scanning, and glide slope scanning modes. This technical solution does not require large-scale hardware replacement or modification, avoiding corresponding investments. It is suitable for large hub airports and small to medium-sized regional airports, and can improve safety detection capabilities based on existing equipment.
[0023] In terms of practicality and engineering application value, the dynamic threshold model designed in this invention is interpretable and transferable. By training the model using historical meteorological and wind shear data, tailored detection parameters can be generated for different airports with varying geographical environments and climate characteristics, thus meeting the requirements for engineering applications. Simultaneously, the online self-optimization mechanism allows the system to continuously improve after deployment, eliminating the need for frequent manual parameter adjustments and reducing maintenance requirements. This method can be embedded into existing terminal area meteorological early warning platforms without altering existing workflows, helping to improve the on-site decision-making efficiency of meteorologists and air traffic controllers, and providing support for civil aviation operational safety. Attached Figure Description
[0024] Appendix Figure 1 This is a flowchart of the civil aviation low-altitude wind shear detection method with adaptive weather background according to the present invention.
[0025] Appendix Figure 2 This is a graph showing the normality test results of Embodiment 1 of the present invention; wherein Figure 2 In the table, A to E represent the normality test results for temperature, humidity, rainfall, air pressure, and wind shear factor, respectively. Pearson correlation coefficient was used for data that conformed to a normal distribution, while Spearman rank correlation coefficient was used for data that did not conform to a normal distribution.
[0026] Appendix Figure 3 This is a graph showing the significance test results of Embodiment 1 of the present invention; wherein Figure 3 Figure A shows the correlation analysis between wind shear factor and meteorological elements. Figure 3 Plot B shows the statistical significance (p-value) results of the correlation between various meteorological elements and wind shear factors; Figure 3 In the figure, C represents the coefficient of determination (R²) of each meteorological element on the variation of wind shear factor. 2 Result image.
[0027] Appendix Figure 4 This is a model training diagram of Embodiment 1 of the present invention; wherein Figure 4 A in the diagram represents the adaptive change of the threshold. Figure 4 Map B shows the changes in meteorological elements.
[0028] Appendix Figure 5 A schematic diagram of the radial distance library for the fitting window used in radial shear calculation.
[0029] Appendix Figure 6 A schematic diagram of the tangential distance library for the fitting window calculated for tangential shear. Detailed Implementation
[0030] The adaptive weather-background low-altitude wind shear detection method for civil aviation proposed in this invention deeply explores the intrinsic correlation between multiple meteorological elements such as temperature, humidity, and air pressure and wind shear, constructing an adaptive discrimination threshold that can be dynamically adjusted according to the weather background. This method can automatically adjust the detection sensitivity based on real-time meteorological conditions, effectively reducing the probability of false alarms and missed detections under complex and changeable weather conditions, significantly improving the accuracy and robustness of wind shear warnings, and providing more precise and reliable technical support for civil aviation flight safety. The following provides a detailed description of the adaptive weather-background low-altitude wind shear detection method for civil aviation of this invention.
[0031] The adaptive weather background civil aviation low-altitude wind shear detection method of the present invention has the following process: Figure 1 As shown, the specific steps include: Historical meteorological data, historical wind shear factor data, and pilot report real labels are obtained. Correlation analysis is performed on the historical meteorological data and historical wind shear factor to obtain a feature set. Using the feature set as the model input and the true labels of the pilot reports as the model output labels, a dynamic threshold model composed of a logistic regression tree and a gradient boosting decision tree (GBDT) is trained. Initiate the detection cycle, collect real-time radial wind speed data and real-time meteorological data, and calculate the wind shear factor for the current cycle based on the real-time radial wind speed data; The real-time meteorological data and the calculated current period wind shear factor are input into the trained dynamic threshold model. The model automatically calculates and outputs an adaptive wind shear discrimination threshold suitable for the current weather background based on the combination characteristics of the current meteorological elements and the wind shear factor. Based on the adaptive wind shear discrimination threshold, a threshold decision is made on the wind shear factor of the current cycle, and an alarm or no alarm result is output. Add new data from the current period to the training set, update the dynamic threshold model online, and wait for the next detection period.
[0032] In some embodiments of the present invention, the meteorological data includes four types of meteorological element data, namely temperature, humidity, air pressure and rainfall.
[0033] In some embodiments of the present invention, the correlation analysis specifically involves performing normality tests on historical sample data of four meteorological elements (temperature, humidity, air pressure, and rainfall) and historical sample data of wind shear factor. When the historical sample data of a certain type of meteorological element to be analyzed and the historical sample data of the historical wind shear factor both meet the normal distribution condition, the Pearson correlation coefficient is used to calculate the correlation between the meteorological element and the historical wind shear factor. When the historical sample data of a certain type of meteorological element to be analyzed does not meet the normal distribution condition, or the historical sample data of the historical wind shear factor does not meet the normal distribution condition, the Spearman rank correlation coefficient is used to analyze the correlation between the meteorological element and the historical wind shear factor. While calculating the correlation coefficient, also calculate the coefficient of determination R. 2 The statistical significance level (P-value) is used to assess the strength of the correlation between various meteorological elements and historical wind shear factors, as well as the statistical reliability. Based on the analysis results, meteorological elements that are strongly correlated with historical wind shear factors and whose statistical significance meets the preset requirements were selected as key input features for the subsequent construction of a dynamic threshold model.
[0034] In some embodiments of the present invention, the process of training the dynamic threshold model specifically includes: Training data preparation: Strongly correlated meteorological data, wind shear factor and pilot report real labels are extracted from historical data at the same time to form a training dataset, in which the pilot report real labels are used as annotation labels for model training. Base learner creation: Create a logistic regression tree model as a gradient boosting base learner to initially fit the nonlinear relationship between meteorological features and the probability of wind shear occurrence; the configuration parameters of the logistic regression tree model include: a maximum number of splits of 80, a minimum number of samples per leaf node of 5, and each tree using all input features; Ensemble Iterative Training: Based on the aforementioned logistic regression tree-based learner, a gradient boosting decision tree algorithm is used for iterative training. The prediction residuals of the preceding learners are fitted through multiple iterations, and finally, the weighted sum of multiple weak learners is obtained to obtain a dynamic threshold model. The configuration parameters for the ensemble iterative training include: using least squares boosting (LSBoost) as the boosting method, 200 iterations, a learning rate of 0.1, real-time meteorological data as the model input, and an adaptive wind shear discrimination threshold for the current meteorological background as the output.
[0035] The threshold decision is specifically as follows: the wind shear factor calculated in the current period is compared with the adaptive wind shear discrimination threshold. If the wind shear factor exceeds the adaptive wind shear discrimination threshold, the current indicator status is determined to be abnormal and an alarm result is output; otherwise, the current indicator status is determined to be normal and no alarm result is output.
[0036] The online update of the dynamic threshold model specifically includes: adding the real-time radial wind speed data collected in the current period, real-time meteorological data, the calculated wind shear factor for the current period, and the decision result label obtained from the threshold decision to the historical training set; re-executing the feature selection and model training steps based on the updated historical training set to complete the incremental update of the dynamic threshold model. The decision result is confirmed in conjunction with subsequent pilot reports.
[0037] The method for calculating the wind shear factor in this invention is existing technology. Depending on the actual application scenario, lidar performance indicators, and computing resources, those skilled in the art can arbitrarily choose a suitable existing calculation method, such as the least squares method, linear fitting method, gradient method, etc., which will not be elaborated here. As a specific embodiment of this invention, a preferred two-dimensional synthetic wind shear factor calculation procedure is given below to more accurately reflect the combined effects of radial and tangential wind shear, as follows: Step 1: Acquire the radial velocity data of the lidar in PPI mode, select an M×N "fitting window", and determine the parameters used to calculate each two-dimensional synthetic wind shear factor. C S The dataset size is m×n, where M and N are the dimension parameters of the fitting window, and m and n are the radial and tangential distances of the dataset, respectively. Step 2: Based on the m radial distance data in the dataset, calculate the one-dimensional radial shear at distance (i, j). C Rs A schematic diagram of the radial distance library for the fitting window in radial shear calculation is attached. Figure 5 As shown, C Rs The calculation formula is:
[0038] In the formula: Let be the radial velocity of the j-th distance in the fitting window, in m / s; The distance of the j-th radial distance library in the fitting window relative to the lidar is expressed in meters. These are the distances from the center point of the storage facility to the 1st and mth points, in meters. The unit length of the distance library, i.e., the radial resolution of the lidar, is in meters; Step 3: Based on the n tangential distance data in the dataset, calculate the one-dimensional tangential shear at distance (i, j). C As A schematic diagram of the tangential distance library for the fitting window in tangential shear calculation is attached. Figure 6 As shown,C As The calculation formula is as follows:
[0039] In the formula: The radial velocity of the i-th distance in the tangential direction within the fitting window is expressed in m / s. The angle between the i-th tangential distance library in the fitting window and the lidar is expressed in degrees. This represents the angular resolution of the lidar, in degrees. The total tangential angle of the fitting window is expressed in degrees. The distance of the i-th tangential distance library in the fitting window relative to the lidar is expressed in meters. Step 4: Based on the one-dimensional radial shear C Rs and one-dimensional tangential shear C As The two-dimensional synthetic wind shear factor at the distance from the source (i, j) is calculated. C S The calculation formula is as follows:
[0040] Step 5: Repeat steps 1-4 to traverse all distance libraries and obtain the two-dimensional composite wind shear factor distribution of the entire region.
[0041] This invention also provides a civil aviation low-altitude wind shear detection system adapted to weather conditions, which is used to perform the above-described civil aviation low-altitude wind shear detection method adapted to weather conditions. The detection system includes: The feature preprocessing module is used to acquire historical meteorological data, historical wind shear factors, and pilot report real labels, and to perform correlation analysis on the historical meteorological data and historical wind shear factors to obtain a feature set. The model training module is used to train a dynamic threshold model consisting of a logistic regression tree and a gradient boosting decision tree (GBDT) based on the feature set and the true labels reported by the pilots. The real-time data processing module is used to start the detection cycle, collect real-time radial wind speed data and real-time meteorological data, and calculate the wind shear factor for the current cycle based on the real-time radial wind speed data. The threshold generation module is used to input the current period wind shear factor and real-time meteorological data to complete the training of the dynamic threshold model and generate an adaptive wind shear discrimination threshold. The anomaly judgment module is used to make a threshold judgment on the current indicator status based on the adaptive wind shear discrimination threshold, and output an alarm or no alarm result. The incremental update module is used to add new data from the current period to the training set and update the dynamic threshold model online.
[0042] In some embodiments of the present invention, the feature preprocessing module is specifically used to perform the correlation analysis using Pearson correlation coefficient or Spearman correlation coefficient to screen and obtain a feature set that meets the correlation requirements; The model training module is specifically used to jointly train the gradient boosting decision tree (GBDT) and the logistic regression tree to obtain a dynamic threshold model that can output an adaptive wind shear discrimination threshold based on the input features.
[0043] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the above-described adaptive weather background civil aviation low-altitude wind shear detection method are implemented.
[0044] Example 1 This embodiment is based on real historical data from a domestic airport, as well as data transmitted from automatic weather stations and laser wind radar during actual operation. The detection range is 500–600 meters above the glide slope. The proposed adaptive weather background civil aviation low-altitude wind shear detection method is employed, and its process is as follows: Figure 1 As shown, the specific steps include: (1) Collection and preprocessing of historical data First, a historical meteorological and wind shear database was established. Meteorological data was collected through the airport's automatic weather station, specifically including four elements: temperature, humidity, air pressure, and rainfall. The data sampling frequency was matched with the subsequent monitoring cycle.
[0045] Wind shear data is obtained from airport laser wind radar: the radial wind speed data of a designated airspace (such as a certain area of the aircraft glide slope) detected by the radar is inverted and calculated to extract the wind shear factor that can characterize the intensity of wind shear.
[0046] All data are aligned by timestamp to form a raw dataset containing multi-source meteorological features and corresponding wind shear factors.
[0047] (2) Correlation analysis between meteorological elements and wind shear To identify key meteorological features that significantly influence wind shear, statistical analysis was performed on the historical data collected in step (1). The specific steps are as follows: a. Normality test: Normality tests were performed on historical sample data of four meteorological elements (temperature, humidity, air pressure, and rainfall) and historical sample data of wind shear factor; the results are attached. Figure 2 As shown.
[0048] Figure 2 In the histogram, the actual probability density distribution of the variables is represented, the curve is the fitted normal distribution curve, and the right vertical axis is the cumulative probability distribution. The p-values for the normality test of each variable are as follows: temperature (p=0.5000), humidity (p=0.3581), air pressure (p=0.5000), and wind shear (p=0.5000) are all greater than the significance level of 0.05, which conforms to a normal distribution; rainfall (p=0.0010) is less than 0.05, which does not conform to a normal distribution.
[0049] b. Correlation coefficient calculation: Based on the normality test results, the correlation analysis method is selected as follows: Temperature, humidity, air pressure and wind shear all conform to normal distribution, so Pearson correlation coefficient is used to analyze their correlation with wind shear; Rainfall does not conform to normal distribution, so Spearman rank correlation coefficient is used to analyze its correlation with wind shear.
[0050] c. Significance test: Calculate the coefficient of determination (R²) simultaneously with the correlation coefficient. 2 The statistical significance level (P-value) was used to assess the strength of the correlation and statistical reliability between each meteorological element and historical wind shear factor; the results are as follows: Figure 3 As shown.
[0051] Figure 3 Figure A shows the correlation analysis between wind shear factor and meteorological elements; the results show that humidity is strongly negatively correlated with wind shear factor, temperature and precipitation are weakly positively correlated, and air pressure is extremely weakly correlated with wind shear factor.
[0052] Figure 3 Plot B shows the statistical significance (p-value) results of the correlation between various meteorological elements and wind shear factors. The results indicate that the correlations between temperature, humidity, and precipitation are statistically significant, while the correlation between air pressure and precipitation is not statistically significant.
[0053] Figure 3 In the figure, C represents the coefficient of determination (R²) of each meteorological element on the variation of wind shear factor. 2 The results are shown in the figure. The results indicate that humidity has the strongest explanatory power for the wind shear factor, while temperature and precipitation have weaker explanatory power, and air pressure has almost no explanatory power. Based on the above analysis, humidity is confirmed as the main meteorological factor affecting the wind shear factor, showing a significant strong negative correlation with it and explaining approximately 65% of the wind shear factor variation; temperature and precipitation are secondary influencing factors, showing a significant weak positive correlation with the wind shear factor; air pressure has no significant effect on the wind shear factor.
[0054] d. Feature selection: Based on the calculation results, meteorological elements that show a strong correlation with historical wind shear factors and whose statistical properties meet the preset requirements are selected as key input features for the subsequent construction of dynamic threshold models.
[0055] (3) Training of dynamic threshold models for weather background This step aims to establish a dynamic threshold model that reflects the occurrence patterns of wind shear under different meteorological backgrounds. Using the aforementioned feature set as model input and the true labels reported by the pilots as model output labels, the dynamic threshold model, composed of a logistic regression tree and a gradient boosting decision tree (GBDT), is trained. The specific implementation process is as follows: a. Training data preparation: The training dataset is composed of strongly correlated meteorological data, wind shear factor and pilot report real labels extracted from historical data at the same time. The pilot report real labels are derived from the exact records of wind shear encounters in the pilot air reports (PIREP) and are used as annotation labels for model training. b. Model building and training: Base learner creation: A logistic regression tree model is created as a base learner for gradient boosting to initially fit the nonlinear relationship between meteorological features and the probability of wind shear occurrence. In some embodiments of the present invention, the configuration parameters of the logistic regression tree model include: a maximum number of splits of 80, a minimum number of samples per leaf node of 5, and each tree using all input features.
[0056] Integrated Iterative Training: Based on the aforementioned logistic regression tree-based learner, a gradient boosting decision tree algorithm is used for iterative training. The prediction residuals of the preceding learners are fitted through multiple iterations, and finally, the weighted sum of multiple weak learners is obtained to obtain a dynamic threshold model. In some embodiments of this invention, the configuration parameters of the integrated iterative training include: using least squares boosting (LSBoost) as the boosting method, 200 iterations, a learning rate of 0.1, real-time meteorological data as the model input, and an adaptive wind shear discrimination threshold for the current meteorological background as the output. The training process is as follows... Figure 4 As shown.
[0057] in Figure 4 Figure A shows the adaptive threshold variation. The adaptive wind shear discrimination threshold fluctuates dynamically within the range of 4 to 6.2 over time, while the fixed threshold remains stable at approximately 7.7. The adaptive wind shear discrimination threshold is always lower than the fixed threshold, demonstrating the characteristic of the present invention that the threshold can be dynamically adjusted according to the environment.
[0058] Figure 4Figure B shows the changes in meteorological elements. Humidity fluctuates most dramatically, ranging from 0 to 120, and is the main environmental influencing factor. Temperature fluctuates slightly and steadily within the range of 0 to 40. Rainfall values are close to 0, with only sporadic pulses. The threshold remains near 0 as a reference. Overall, the fluctuation pattern of the adaptive wind shear discrimination threshold matches the changing trend of meteorological elements (especially humidity), indicating that this invention can dynamically adjust the threshold according to the real-time meteorological environment. Compared with a fixed threshold, it is more in line with the needs of actual scenarios, improving the rationality and sensitivity of monitoring or control.
[0059] (4) Real-time meteorological data acquisition Real-time acquisition of airport meteorological data at the start of the current scanning cycle of the lidar, with the data type consistent with step (2), namely, meteorological data strongly correlated with the wind shear factor.
[0060] (5) Matching and generation of adaptive wind shear discrimination threshold Using the real-time radial wind speed data acquired by the lidar during the current scanning cycle, the real-time wind shear factor is calculated using the same algorithm as in step 1.
[0061] The real-time meteorological data collected in step (4) and the real-time wind shear factor are input into the dynamic threshold model trained in step (3). The dynamic threshold model automatically calculates and outputs a wind shear discrimination threshold that is adapted to the current weather background based on the combination characteristics of the current meteorological elements.
[0062] (6) Wind shear detection and alarm during the current scanning cycle a. Threshold decision: The calculated real-time wind shear factor is compared with the wind shear discrimination threshold generated in step 5, which is adapted to the current weather background.
[0063] b. Output of results: If the real-time wind shear factor exceeds the wind shear discrimination threshold, it is determined that there is a wind shear event in the current scanning period, and the system immediately generates an alarm message and outputs it to the air traffic control terminal.
[0064] (7) Online updating and optimization of the model To maintain the model's long-term adaptability to local microclimates and seasonal changes at the airport, an online learning mechanism is introduced. After the current cycle's detection and decision, the meteorological data at that moment, the calculated wind shear factor, and the final decision result (confirmed in conjunction with subsequent pilot reports) are added to the training set as a new set of sample data. The system can periodically or after reaching a certain amount of data trigger incremental training or retraining of the model, thereby continuously optimizing the accuracy of the dynamic threshold model and enabling it to adaptively track the long-term evolution of the airport's meteorological environment.
[0065] The above description is merely a preferred embodiment of the present invention and is illustrative in nature, not intended to limit the scope of the invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the invention should fall within the protection scope defined by the claims.
Claims
1. A method for detecting low-altitude wind shear in civil aviation that adapts to weather backgrounds, characterized in that, Includes the following steps: Historical meteorological data, historical wind shear factor data, and pilot report real labels are obtained. Correlation analysis is performed on the historical meteorological data and historical wind shear factor to obtain a feature set. Using the feature set as the model input and the true labels of the pilot reports as the model output labels, a dynamic threshold model composed of a logistic regression tree and a gradient boosting decision tree (GBDT) is trained. Initiate the detection cycle, collect real-time radial wind speed data and real-time meteorological data, and calculate the wind shear factor for the current cycle based on the real-time radial wind speed data; The real-time meteorological data and the calculated current period wind shear factor are input into the trained dynamic threshold model. The model automatically calculates and outputs an adaptive wind shear discrimination threshold suitable for the current weather background based on the combination characteristics of the current meteorological elements and the wind shear factor. Based on the adaptive wind shear discrimination threshold, a threshold decision is made on the wind shear factor of the current cycle, and an alarm or no alarm result is output. Add new data from the current period to the training set, update the dynamic threshold model online, and wait for the next detection period.
2. The civil aviation low-altitude wind shear detection method according to claim 1, characterized in that, Meteorological data includes four types of meteorological elements: temperature, humidity, air pressure, and rainfall.
3. The civil aviation low-altitude wind shear detection method according to claim 2, characterized in that, The correlation analysis specifically involves performing normality tests on historical sample data of four meteorological elements (temperature, humidity, air pressure, and rainfall) and historical sample data of wind shear factor. When the historical sample data of a certain type of meteorological element to be analyzed and the historical sample data of the historical wind shear factor both meet the normal distribution condition, the Pearson correlation coefficient is used to calculate the correlation between the meteorological element and the historical wind shear factor. When the historical sample data of a certain type of meteorological element to be analyzed does not meet the normal distribution condition, or the historical sample data of the historical wind shear factor does not meet the normal distribution condition, the Spearman rank correlation coefficient is used to analyze the correlation between the meteorological element and the historical wind shear factor. While calculating the correlation coefficient, also calculate the coefficient of determination R. 2 The statistical significance level (P-value) is used to assess the strength of the correlation between various meteorological elements and historical wind shear factors, as well as the statistical reliability. Based on the analysis results, meteorological elements that are strongly correlated with historical wind shear factors and whose statistical significance meets the preset requirements were selected as key input features for the subsequent construction of a dynamic threshold model.
4. The civil aviation low-altitude wind shear detection method with adaptive weather background as described in claim 1, characterized in that, The process of training a dynamic threshold model specifically includes: Training data preparation: Strongly correlated meteorological data, wind shear factor and pilot report real labels are extracted from historical data at the same time to form a training dataset, in which the pilot report real labels are used as annotation labels for model training. Base learner creation: Create a logistic regression tree model as a gradient boosting base learner to initially fit the nonlinear relationship between meteorological features and the probability of wind shear occurrence; Integrated iterative training: Based on the logistic regression tree-based learner, the gradient boosting decision tree algorithm is used for iterative training. The prediction residuals of the preceding learners are fitted through multiple rounds of iteration. Finally, the weighted sum of multiple weak learners is obtained to obtain the dynamic threshold model.
5. The civil aviation low-altitude wind shear detection method according to claim 4, characterized in that, The configuration parameters of the logistic regression tree model include: a maximum number of splits of 80, a minimum number of samples per leaf node of 5, and each tree using all input features; The configuration parameters for the integrated iterative training include: using least squares boosting (LSBoost) as the boosting method, 200 iterations, a learning rate of 0.1, real-time meteorological data as the model input, and an adaptive wind shear discrimination threshold for the current meteorological background as the output.
6. The civil aviation low-altitude wind shear detection method according to claim 1, characterized in that, The threshold decision is specifically as follows: the wind shear factor calculated in the current period is compared with the adaptive wind shear discrimination threshold. If the wind shear factor exceeds the adaptive wind shear discrimination threshold, the current indicator status is determined to be abnormal and an alarm result is output. Otherwise, the current indicator status is determined to be normal and a no-alarm result is output.
7. The civil aviation low-altitude wind shear detection method according to claim 1, characterized in that, The online update of the dynamic threshold model specifically includes: adding the real-time radial wind speed data collected in the current period, real-time meteorological data, the calculated wind shear factor of the current period, and the decision result label obtained from the threshold decision to the historical training set; and re-executing the feature selection and model training steps based on the updated historical training set to complete the incremental update of the dynamic threshold model.
8. A civil aviation low-altitude wind shear detection system that adapts to weather backgrounds, characterized in that, For performing the civil aviation low-altitude wind shear detection method with adaptive weather background as described in any one of claims 1 to 7, the detection system comprises: The feature preprocessing module is used to acquire historical meteorological data, historical wind shear factors, and pilot report real labels, and to perform correlation analysis on the historical meteorological data and historical wind shear factors to obtain a feature set. The model training module is used to train a dynamic threshold model consisting of a logistic regression tree and a gradient boosting decision tree (GBDT) based on the feature set and the true labels reported by the pilots. The real-time data processing module is used to start the detection cycle, collect real-time radial wind speed data and real-time meteorological data, and calculate the wind shear factor for the current cycle based on the real-time radial wind speed data. The threshold generation module is used to input the current period wind shear factor and real-time meteorological data to complete the training of the dynamic threshold model and generate an adaptive wind shear discrimination threshold. The anomaly judgment module is used to make a threshold judgment on the current indicator status based on the adaptive wind shear discrimination threshold, and output an alarm or no alarm result. The incremental update module is used to add new data from the current period to the training set and update the dynamic threshold model online.
9. The civil aviation low-altitude wind shear detection system with adaptive weather background as described in claim 8, characterized in that, The feature preprocessing module is specifically used to perform the correlation analysis using Pearson correlation coefficient or Spearman correlation coefficient to screen and obtain a feature set that meets the correlation requirements; The model training module is specifically used to jointly train the gradient boosting decision tree (GBDT) and the logistic regression tree to obtain a dynamic threshold model that can output an adaptive wind shear discrimination threshold based on the input features.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the civil aviation low-altitude wind shear detection method with adaptive weather background as described in any one of claims 1 to 7.