Terminal area flight landing time intelligent prediction method based on flight path deployment characteristics

Abstract

The invention relates to the technical field of air transportation and air control, in particular to a terminal area flight landing time intelligent prediction method based on flight path deployment characteristics, which comprises the following steps: step 1, data collection and preprocessing; 2, clustering the arrival positions of the flights; step 3, track feature extraction; 4, constructing and training a prediction model; 5, performing real-time prediction and complex environment adaptation; and step 6, outputting and feeding back a result. According to the method, the prediction precision of the flight landing time in the complex operation environment is remarkably improved by comprehensively utilizing the static characteristics of the flight, the meteorological information, the airport operation state and the dynamic track characteristics in the terminal area.

Description

Technical Field

[0001] The present invention relates to the technical fields of air transportation and air traffic control. Specifically, it relates to an intelligent prediction method for the landing time of terminal area flights based on flight track allocation characteristics. Background Art

[0002] In the field of modern air transportation, the prediction of flight operation timings is an important link in air traffic control and airspace management. Especially in the airport terminal area, with the continuous increase in flight density and the frequent occurrence of complex weather conditions, the accurate prediction of flight landing times has become an important requirement for improving airspace utilization, optimizing operation efficiency, and enhancing the level of aviation safety. However, there are still certain technical limitations in current research and applications.

[0003] Traditional methods usually predict the landing time of flights based on historical statistical analysis or empirical models. These methods often rely on simple mathematical formulas and mainly consider factors such as the scheduled arrival time of the flight, the use of airport runways, and basic meteorological data (such as wind speed, wind direction, etc.). Although these methods can provide certain references under stable operating conditions, their prediction accuracy significantly decreases in complex operating environments, such as sudden weather changes (such as rainfall, low visibility, or strong wind conditions) or when the flight density in the terminal area is high.

[0004] With the development of artificial intelligence technology, especially the widespread application of machine learning methods, some data-driven flight landing time prediction models have emerged in recent years. These models can better capture flight patterns and rules through the training of historical flight data and provide higher-precision predictions by combining meteorological conditions, airport operating status, and other information. However, existing technologies mostly focus on modeling using relatively simple datasets (such as meteorology, airport information, flight attributes, etc.) and fail to fully explore the importance of dynamic flight characteristics in the terminal area, especially flight track allocation characteristics.

[0005] Flight track allocation characteristics reflect the dynamic changes in flight operations in the terminal area, such as trajectory information like flight altitude, speed, and turning characteristics. These characteristics are crucial in predicting flight landing times. Especially in complex weather conditions (such as rainy days or low visibility environments), the actual flight trajectory of a flight may change significantly due to air traffic control instructions, obstacle avoidance, or route adjustments. However, existing prediction methods fail to combine real-time flight track allocation characteristics with other factors, resulting in obvious deficiencies in the accuracy of prediction results under the above special conditions. In addition, the results of existing methods often lack sufficient interpretability, which is not convenient for air traffic controllers and controllers to conduct further decision-making analysis.

[0006] Therefore, there is an urgent need for a new prediction method that can comprehensively consider multi-dimensional factors such as meteorology, airport operation status, and flight characteristics, and at the same time make full use of the track allocation characteristics in the terminal area to improve the prediction accuracy and have a certain interpretability to adapt to complex operation environments and diverse application requirements. Summary of the Invention

[0007] The content of the present invention is to provide an intelligent prediction method for the landing time of terminal area flights based on track allocation characteristics, which can overcome the deficiencies of the prior art in predicting the landing time of flights, especially in complex weather conditions (such as thunderstorm weather, low visibility, etc.), where frequent allocation between flights and changing time intervals lead to inaccurate predicted landing times.

[0008] An intelligent prediction method for the landing time of terminal area flights based on track allocation characteristics according to the present invention includes the following steps:

[0009] Step 1, data collection and preprocessing;

[0010] Step 2, clustering of flight arrival positions;

[0011] Step 3, extraction of track characteristics;

[0012] Step 4, construction and training of a prediction model;

[0013] Step 5, real-time prediction and adaptation to complex environments;

[0014] Step 6, result output and feedback.

[0015] Preferably, in step 1, the collected data includes:

[0016] Static features: flight number, airline, aircraft type, scheduled arrival time of the flight, departure airport, destination airport;

[0017] Meteorological information: wind speed, wind direction, visibility, temperature, corrected sea level pressure, weather phenomena;

[0018] Airport operation status: runway usage, number of incoming flights, number of outgoing flights, runway occupancy;

[0019] Passenger information: number of seats, seat occupancy rate, number of adult and child passengers, quantity and weight of luggage, weight of goods, weight of mail;

[0020] Dynamic track information: time series data during the flight of the flight, including flight altitude, speed, longitude and latitude, flight direction.

[0021] Preferably, in step 1, the data preprocessing is specifically:

[0022] Missing value processing: filling in missing values;

[0023] Outlier detection: Eliminate the outliers in the track data;

[0024] Normalization: Normalize the numerical data to eliminate the dimension difference;

[0025] Time series arrangement: Arrange the dynamic track data into a unified format according to time steps to generate the input for time series modeling.

[0026] Preferably, in step 2, specifically:

[0027] Using the inbound latitude and longitude data of flights, adopt a clustering algorithm to divide the inbound positions of all flights into 6 categories to form a regional division in the geographical space; when a flight arrives at the airport, determine its category through the clustering model according to its current latitude and longitude coordinates; subsequently, find the nearest landed flight in this category and use its actual flight time as a new feature.

[0028] Preferably, in step 3, the track feature extraction specifically includes:

[0029] 3.1) Time series feature extraction;

[0030] Time series modeling: Use the time series model in deep learning to process the dynamic track information data; the input is the track time series of the current flight, and the output is a one-dimensional feature vector with a fixed length, representing the temporal characteristics of the track change;

[0031] 3.2) Dimensionality reduction and feature compression;

[0032] Adopt a fully connected layer or pooling operation to reduce the dimension of the output features of the time series model and compress them into a one-dimensional vector to ensure that the feature dimension is consistent with other static data; in the output one-dimensional vector, each value represents a certain potential pattern of the track dynamic features in the time series;

[0033] 3.3) Nearest flight track feature extraction;

[0034] Extract the track data of the nearest landed flights: Obtain the time series track features of the 30 nearest landed flights in the terminal area, and input the data of each flight into the time series model for processing to generate the corresponding one-dimensional feature vectors;

[0035] Aggregate features: Statistically aggregate the feature vectors of the nearest 30 flights to generate the overall features of the current terminal area track allocation;

[0036] 3.4) Merge the track features with the static data;

[0037] Merge the track features of the current flight, the aggregated features of the tracks of the last 30 flights, and other static features to form the final model input features. The track features of the current flight are one-dimensional vectors generated by a time series model, and the other static features are meteorological and passenger information.

[0038] Preferably, in step 4, the construction and training of the prediction model specifically include:

[0039] 4.1) Construct a prediction model;

[0040] Model architecture: Use a machine learning algorithm as the prediction model, take the multi-source features generated in step 3 as input, and predict the time for the flight to land from the terminal area boundary to the runway; the machine learning algorithm includes XGBoost, LightGBM or random forest; the multi-source features include static features and one-dimensional track features;

[0041] Feature fusion: Automatically learn the complex relationship between dynamic features and static features through the model to ensure the prediction accuracy of the model;

[0042] 4.2) Model training and validation;

[0043] Training set and test set division: Use historical flight data to construct a training set and a test set to ensure the consistency of time series features and static features in the input;

[0044] Cross-validation: Use the cross-validation method to evaluate the model performance and optimize the parameters;

[0045] 4.3) Model optimization and result interpretation;

[0046] Introduce SHAP value analysis: Use the SHAP tool to interpret the trained prediction model and analyze the contribution of each input feature to the prediction result; the input features include time series features and static features;

[0047] The specific outputs include:

[0048] The impact of dynamic track features on the prediction of landing time;

[0049] The importance ranking of static features;

[0050] Analysis of the synergy between time series and static features.

[0051] Preferably, in step 5, it is specifically:

[0052] 5.1) Real-time prediction;

[0053] Organize the static data, meteorological information, and dynamic track data of the current flight into a unified input format, input it into the trained prediction model, and generate a predicted value for the flight landing time;

[0054] 5.2) Adaptation to complex environments;

[0055] In thunderstorm weather and low visibility environments, dynamically update time series features and meteorological data, and adjust the model input in real time to adapt to complex environmental changes.

[0056] Preferably, in step 6, specifically:

[0057] 6.1) Output of prediction results;

[0058] Output the predicted landing time value and the analysis result of SHAP value, intuitively display the specific contributions of each feature to the prediction, and assist air traffic controllers and dispatchers in making decisions;

[0059] 6.2) Error feedback and model update;

[0060] Compare the prediction result with the actual landing time, record the error and perform offline update on the model to further optimize the prediction accuracy.

[0061] By comprehensively utilizing the static features of flights, meteorological information, airport operation status, and dynamic flight track features in the terminal area, the present invention significantly improves the prediction accuracy of flight landing time in complex operating environments. The specific technical effects are as follows:

[0062] 1) Significantly improve the prediction accuracy of flight landing time;

[0063] Compared with the prior art, by introducing the flight track allocation feature and combining time series modeling technology and machine learning algorithms, the present invention effectively captures the dynamic operation characteristics of flights in the terminal area. The test results show that:

[0064] Under normal weather conditions, the mean squared error (MSE) of the model's prediction error is reduced by more than 5%;

[0065] In thunderstorm weather or low visibility environments, due to the flight track feature being able to dynamically reflect the real-time changes of flight operations, the mean squared error (MSE) of the prediction error is reduced by more than 10%.

[0066] This improvement in accuracy makes the present invention more reliable in actual operation and can effectively meet the high-precision requirements of air traffic control and management for landing time prediction.

[0067] 2) Improve the adaptability under complex weather conditions;

[0068] The present invention specifically addresses the problem of frequent flight allocation and large fluctuations in time intervals among flights in complex environments such as thunderstorm weather and low visibility. By real-time extracting dynamic flight track features and the trajectory data of the recently landed flights, it realizes the adaptive ability to complex environments.

[0069] The introduction of dynamic flight track features enables the prediction model to capture track adjustments caused by air traffic control or obstacle avoidance, significantly reducing prediction errors in complex environments;

[0070] In addition, the model can adjust the prediction results in real time according to the current operating status in the terminal area (such as flight density, runway occupancy), improving the stability and robustness of the model.

[0071] 3) Enhance the interpretability and practicality of the model;

[0072] By introducing the SHAP value analysis tool, the present invention realizes the interpretability analysis of the prediction results, including:

[0073] The importance ranking of dynamic flight track features, meteorological features, and static features for landing time prediction;

[0074] Under specific complex environments, the specific impact of each feature on the prediction error.

[0075] This interpretability greatly enhances the practicality of the model, enabling air traffic controllers and dispatchers to clearly understand the source of the prediction results and make reasonable judgments on the key influencing factors in complex operating environments, assisting in dispatch decisions.

[0076] 4) Achieve a double improvement in social and economic benefits;

[0077] Improve airspace utilization efficiency

[0078] Through more accurate flight landing time prediction, the present invention effectively reduces runway waiting or airspace congestion problems caused by prediction errors, improving the utilization efficiency of the terminal area airspace and airport runways.

[0079] Reduce aviation operation costs

[0080] More accurate prediction can reduce flight waiting time, fuel consumption, and airline operation costs. Especially under complex weather conditions, the prediction ability of the present invention can reduce unnecessary delays and holding times.

[0081] Improve passenger satisfaction

[0082] By reducing flight delays and optimizing flight operation scheduling, the present invention can improve flight punctuality rates, thereby enhancing passengers' travel experiences and satisfaction.

[0083] 5) Support from experimental results;

[0084] The present invention has been experimentally verified through a large amount of historical flight data and real-time operation data, and the specific results are as follows:

[0085] In the historical data of the terminal area operation, when the present invention is compared with the existing prediction models based on traditional machine learning methods, the results show that:

[0086] Under normal weather conditions, the MSE value of the present invention is reduced by 5.3% on average;

[0087] Under thunderstorm weather or low visibility environment, the MSE value of the present invention is reduced by 12.7% on average.

[0088] In actual tests, the present invention shows high prediction accuracy and robustness under different operating states. Especially under dense flight or complex weather conditions, it can significantly reduce the prediction error.

[0089] 6) Comprehensive technical effects;

[0090] In summary, by introducing the track allocation feature, combining the time series deep learning method and the traditional machine learning model, the present invention significantly improves the accuracy of flight landing time prediction, especially showing strong adaptability and robustness in complex environments. By improving the prediction accuracy, enhancing the model interpretability and adaptability, the present invention not only solves the deficiencies of the existing technologies, but also realizes significant social, economic and technical benefits, providing strong technical support for the optimization of terminal area air operations. Description of the Drawings

[0091] Figure 1 It is a flowchart of an intelligent prediction method for terminal area flight landing time based on track allocation features in Embodiment 1. Detailed Embodiments

[0092] To further understand the content of the present invention, the present invention will be described in detail in combination with the drawings and embodiments. It should be understood that the embodiments are only for explaining the present invention rather than limiting it.

[0093] Embodiment 1

[0094] As Figure 1 shown, this embodiment provides an intelligent prediction method for terminal area flight landing time based on track allocation features, which includes the following steps:

[0095] Step 1, data collection and preprocessing;

[0096] Collect multi-dimensional data related to flight operations. The collected data includes:

[0097] Static features: flight number, airline, aircraft type, flight planned arrival time, departure airport, destination airport, etc.;

[0098] Meteorological information: wind speed, wind direction, visibility, temperature, corrected sea level pressure, weather phenomena (such as thunderstorms, fog), etc.;

[0099] Airport operating status: runway usage, number of incoming flights, number of outgoing flights, runway occupancy, etc.;

[0100] Passenger information: number of seats, seat occupancy rate, number of adult and child passengers, quantity and weight of luggage, weight of goods, weight of mail, etc.;

[0101] Dynamic flight track information: time series data during the flight of a flight, including flight altitude, speed, longitude and latitude, flight direction, etc.

[0102] Data preprocessing, cleaning and processing the above data to ensure data quality and consistency, specifically:

[0103] Missing value handling: filling in missing values (such as interpolation method, mean filling, etc.);

[0104] Outlier detection: removing abnormal points in the flight track data (such as unreasonable altitude, speed, etc.);

[0105] Normalization: performing normalization processing on numerical data to eliminate the dimension difference;

[0106] Time series arrangement: arranging the dynamic flight track data into a unified format according to time steps (such as one recording point every 2 seconds), generating the input for time series modeling.

[0107] Step 2: Clustering of incoming flight positions;

[0108] Using the incoming longitude and latitude data of flights, all incoming flight positions are divided into 6 categories by using a clustering algorithm, forming a regional division in geographical space; when a flight (such as MU3688) arrives, according to its current longitude and latitude coordinates, its belonging category is determined through the clustering model; subsequently, the nearest landed flight in this category is found, and its actual flight time is used as a new feature. For example, flight MU3688 arrives at 17:33:26 on August 8, 2023, and is classified into category 3. The landing time of the nearest flight in this category is 17:32:26 on August 8, 2023, and its actual flight time in the port is 23 minutes. This time will be used as a new feature for predicting the landing time of flight MU3688.

[0109] Step 3: Extraction of flight track features;

[0110] 3.1) Extraction of time series features;

[0111] Time series modeling: using the time series model in deep learning to process the dynamic flight track information data (such as flight altitude, speed, direction, etc.); the input is the trajectory time series of the current flight, and the output is a one-dimensional feature vector with a fixed length, representing the temporal features of the flight track change.

[0112] The time series model is a long short-term memory network (LSTM), which can better capture the temporal dependence relationship of flight trajectories.

[0113] 3.2) Dimensionality reduction and feature compression;

[0114] Use a fully connected layer or pooling operation to reduce the dimensionality of the output features of the time series model and compress them into a one-dimensional vector to ensure that the feature dimensions are consistent with other static data; in the output one-dimensional vector, each value represents a certain potential pattern of the track dynamic features in the time series.

[0115] 3.3) Extraction of recent flight track features;

[0116] Extract the track data of the recently landed flights: Obtain the time series track features (such as altitude change, speed change, etc.) of the 30 recently landed flights in the terminal area, and input the data of each flight into the time series model for processing to generate the corresponding one-dimensional feature vector.

[0117] Aggregate features: Statistically aggregate the feature vectors of the recent 30 flights (such as mean, variance, maximum, minimum, etc.) to generate the overall features of the current terminal area track allocation.

[0118] 3.4) Merge track features with static data;

[0119] Merge the track features of the current flight (the one-dimensional vector generated by the time series model), the aggregated features of the tracks of the recent 30 flights, and other static features (meteorology, passenger information, etc.) to form the final model input features.

[0120] Step 4, Construction and training of the prediction model;

[0121] 4.1) Construct the prediction model;

[0122] Model architecture: Use a machine learning algorithm as the prediction model, take the multi-source features generated in step 3 as the input, and predict the time for the flight to land from the terminal area boundary to the runway; the machine learning algorithms include XGBoost, LightGBM or random forest; the multi-source features include static features and one-dimensional track features.

[0123] Feature fusion: Automatically learn the complex relationship between dynamic features and static features through the model to ensure the prediction accuracy of the model.

[0124] 4.2) Model training and verification;

[0125] Division of the training set and the test set: Use historical flight data to construct a training set (80%) and a test set (20%) to ensure the consistency of time series features and static features in the input.

[0126] Cross-validation: The cross-validation method is adopted to evaluate the model performance and optimize parameters (such as learning rate, tree depth, etc.).

[0127] 4.3) Model optimization and result interpretation;

[0128] Introduction of SHAP value analysis: The SHAP (SHapley Additive exPlanations) tool is used to explain the trained prediction model and analyze the contribution of each input feature to the prediction result; the input features include time series features and static features.

[0129] Specific outputs include:

[0130] The impact of dynamic flight track features on the prediction of landing time (such as the contribution of speed fluctuations to the prediction result);

[0131] The importance ranking of static features (such as meteorological data, passenger information, etc.);

[0132] Analysis of the synergy between time series and static features.

[0133] Step 5, Real-time prediction and adaptation to complex environments.

[0134] 5.1) Real-time prediction;

[0135] The static data, meteorological information, and dynamic flight track data of the current flight are organized into a unified input format and input into the trained prediction model to generate a predicted value of the flight landing time.

[0136] 5.2) Adaptation to complex environments;

[0137] In thunderstorm weather and low visibility environments, the time series features and meteorological data are dynamically updated, and the model input is adjusted in real time to adapt to complex environmental changes.

[0138] Step 6, Result output and feedback.

[0139] 6.1) Prediction result output;

[0140] Output the predicted landing time value and the SHAP value analysis result, visually display the specific contribution of each feature to the prediction, and assist air traffic controllers and dispatchers in making decisions.

[0141] 6.2) Error feedback and model update;

[0142] Compare the prediction result with the actual landing time, record the error, and perform offline update on the model to further optimize the prediction accuracy.

[0143] In this embodiment, by comprehensively utilizing the static characteristics of flights, meteorological information, airport operation status, and dynamic track characteristics in the terminal area, the prediction accuracy of flight landing time in a complex operation environment is significantly improved.

[0144] Embodiment 2

[0145] In this embodiment, an effect experiment and data analysis are carried out.

[0146] 1. Experimental process

[0147] 1.1 Experimental environment

[0148] Dataset: Flight operation data in the terminal area of a certain international airport for two months (including sunny days and thunderstorm days).

[0149] Experimental tools: Python language, using TensorFlow to implement the LSTM model and using XGBoost for final prediction.

[0150] 1.2 Experimental design

[0151] The experiment is divided into two groups:

[0152] Baseline method: Only use static features and meteorological data to construct a prediction model;

[0153] Method of the present invention: Introduce dynamic track features (extracted using LSTM) and fuse them with static features to construct a prediction model.

[0154] 1.3 Test metrics

[0155] Prediction mean squared error (MSE): Measure the error between the predicted value and the actual value;

[0156] Interpretability analysis (analyze the key feature contributions of the model through SHAP values).

[0157] 2. Experimental results

[0158] 2.1 Comparison of prediction accuracy on sunny days and rainy days

[0159] Sunny days:

[0160] The MSE of the baseline method is 8.3 seconds;

[0161] The MSE of the method of the present invention is 7.8 seconds (a decrease of 6.02%).

[0162] Rainy days:

[0163] The MSE of the baseline method is 14.7 seconds;

[0164] The MSE of the method of the present invention is 12.9 seconds (a decrease of 12.24%).

[0165] 2.2 SHAP value analysis

[0166] The results of SHAP value analysis show that under sunny conditions, meteorological characteristics (such as wind speed and visibility) contribute the most to the prediction results. Under rainy conditions, the contribution of dynamic flight track characteristics increases significantly and becomes a key factor in prediction.

[0167] The interpretability of this method helps air traffic controllers better understand the impact of flight track changes on landing time under complex weather conditions.

[0168] Comprehensive effect analysis

[0169] Through the above implementation methods, the present invention significantly improves the prediction accuracy of the landing time of terminal area flights, especially the prediction performance under complex weather conditions, verifying the feasibility and effectiveness of the present invention. Its social and economic benefits are significant, providing strong technical support for air traffic control and airspace management.

[0170] The present invention and its implementation manners are schematically described above. The description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Therefore, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent prediction method for flight landing time in the terminal area based on track deployment characteristics, characterized by: The following steps are involved: Step 1: Data collection and preprocessing; Step 2: Clustering of flight arrival locations; Step 3: Extract track features; Step 4: Prediction model construction and training; Step 5: Real-time prediction and adaptation to complex environments; Step 6: Result output and feedback.

2. The method for intelligently predicting flight landing time in the terminal area based on track deployment characteristics according to claim 1 is characterized by: In step 1, the data collected include: Static features: flight number, airline, aircraft type, flight scheduled arrival time, departure airport, destination airport; Meteorological information: wind speed, wind direction, visibility, temperature, corrected sea pressure, weather phenomena; Airport operation status: runway usage, number of inbound flights, number of outbound flights, runway occupancy; Passenger information: number of seats, occupancy rate, number of adult and child passengers, number and weight of luggage, cargo weight, mail weight; Dynamic track information: time series data during the flight, including flight altitude, speed, longitude and latitude, and flight direction.

3. The method for intelligently predicting the landing time of a flight in the terminal area based on the track deployment characteristics according to claim 2 is characterized in that: In step 1, data preprocessing is specifically as follows: Missing value processing: fill in missing values; Outlier detection: remove abnormal points in track data; Normalization: normalize numerical data to eliminate dimensional differences; Time series organization: The dynamic track data is organized into a unified format according to time steps to generate input for time series modeling.

4. The method for intelligently predicting the landing time of a flight in the terminal area based on track deployment characteristics according to claim 3 is characterized in that: In step 2, specifically: Using the arrival latitude and longitude data of flights, a clustering algorithm is used to divide the arrival locations of all flights into six categories, forming a regional division in geographical space; when a flight arrives, its category is determined by the clustering model based on its current latitude and longitude coordinates; Then, find the most recent flight that has landed in this category and use its actual flight time as a new feature.

5. The method for intelligently predicting the landing time of a flight in the terminal area based on track deployment characteristics according to claim 4 is characterized in that: In step 3, the track feature extraction specifically includes: 3.1) Time series feature extraction; Time series modeling: Use the time series model in deep learning to process dynamic track information data; the input is the track time series of the current flight, and the output is a one-dimensional feature vector of fixed length, which represents the time series characteristics of track changes; 3.2) Dimensionality reduction and feature compression; The output features of the time series model are reduced in dimension using a fully connected layer or pooling operation, compressed into a one-dimensional vector to ensure that the feature dimension is consistent with other static data; in the output one-dimensional vector, each value represents a potential pattern of the dynamic features of the track in the time series; 3.3) Extraction of recent flight trajectory features; Extract the trajectory data of the most recent landing flight: obtain the time series trajectory characteristics of the 30 flights that have landed most recently in the terminal area, input the data of each flight into the time series model for processing, and generate the corresponding one-dimensional feature vector; Aggregation features: statistically aggregate the feature vectors of the most recent 30 flights to generate the overall features of the current terminal area track deployment; 3.4) Merging track features with static data; The track features of the current flight, the trajectory aggregation features of the most recent 30 flights and other static features are combined to form the final model input features. The track features of the current flight are one-dimensional vectors generated by the time series model, and other static features are weather and passenger information.

6. The method for intelligently predicting the landing time of a flight in the terminal area based on the track allocation characteristics according to claim 5 is characterized in that: In step 4, the prediction model construction and training specifically include: 4.1) Build a prediction model; Model architecture: Use a machine learning algorithm as a prediction model, and use the multi-source features generated in step 3 as input to predict the time it takes for a flight to land on the runway from the terminal area boundary; the machine learning algorithm includes XGBoost, LightGBM, or random forest; the multi-source features include static features and one-dimensional track features; Feature fusion: The model automatically learns the complex relationship between dynamic features and static features to ensure the model prediction accuracy; 4.2) Model training and verification; Training set and test set division: Use historical flight data to build training sets and test sets to ensure that time series features and static features are consistent in input; Cross-validation: Use cross-validation method to evaluate model performance and optimize parameters; 4.3) Model optimization and result interpretation; Introducing SHAP value analysis: Use the SHAP tool to interpret the trained prediction model and analyze the contribution of each input feature to the prediction result; input features include time series features and static features; The output includes: The impact of dynamic track characteristics on landing time prediction; Importance ranking of static features; Synergy analysis between time series and static features.

7. The method for intelligently predicting the landing time of a flight in the terminal area based on track deployment characteristics according to claim 6 is characterized in that: In step 5, specifically: 5.1) Real-time prediction; The static data, weather information and dynamic track data of the current flight are organized into a unified input format, input into the trained prediction model, and generate the predicted value of the flight landing time; 5.2) Adaptation to complex environments; In thunderstorms and low visibility environments, time series characteristics and meteorological data are dynamically updated, and model inputs are adjusted in real time to adapt to complex environmental changes.

8. The method for intelligently predicting the landing time of a flight in the terminal area based on track deployment characteristics according to claim 7 is characterized in that: In step 6, specifically: 6.1) Output of prediction results; Output the predicted landing time value and SHAP value analysis results, intuitively display the specific contribution of each feature to the prediction, and assist controllers and dispatchers in making decisions; 6.2) Error feedback and model updating; Compare the predicted results with the actual landing time, record the errors and update the model offline to further optimize the prediction accuracy.

Images