Aircraft slip-off time prediction method based on differential evolution and XGBoost
By introducing a differential evolution algorithm into the XGBoost model to optimize hyperparameters, the shortcomings of existing aircraft taxiing time prediction models in parameter setting are addressed, improving prediction accuracy and stability, and supporting intelligent and refined airport operation management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-22
- Publication Date
- 2026-03-13
AI Technical Summary
Existing aircraft taxiing time prediction models lack systematic and comprehensive parameter settings, resulting in large prediction deviations and poor stability. They are unable to effectively capture potential data patterns in complex dynamic environments, affecting the decision-making accuracy of airport operation management systems.
The XGBoost model is used as the basic prediction model, and the differential evolution algorithm is introduced to optimize the key hyperparameters. The differential evolution algorithm automatically searches for the optimal combination of hyperparameters to improve the training efficiency and generalization ability of the model.
It improves the accuracy and stability of aircraft taxiing time prediction, enhances the model's adaptability in complex environments, and supports the refinement and intelligence of airport flight operation management.
Smart Images
Figure CN121660146A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of civil aviation airport flight operation optimization technology, and in particular to a method for predicting aircraft taxiing time based on differential evolution and XGBoost. Background Technology
[0002] This section provides only background information relevant to this disclosure and is not necessarily prior art.
[0003] In recent years, with the continuous development of global civil aviation, airports, as important operational hubs, have faced increasingly prominent congestion problems on their ground surfaces. How to effectively improve airport operational capacity, especially by controlling the propagation of delays, improving on-time performance, and enhancing operational efficiency during flight transit, has become one of the core challenges facing large hub airports. As a crucial operational management system for airports, the Airport Collaborative Decision Making (A-CDM) system plays a significant role in improving the operational efficiency of large airports. Effective application of A-CDM systems helps increase runway throughput and improve airport operational efficiency, thereby positively impacting airport operations and the overall effectiveness of the regional air traffic network.
[0004] Within the A-CDM (Aircraft-to-Destination Management) system framework, the taxiing time of aircraft at the airport surface is a crucial basis for optimizing surface flight pushback control and departure sequencing. Inaccurate taxiing time prediction, leading to unreasonable decision-making schemes within the A-CDM system, can further result in fuel waste, airport resource waste, and increased operating costs. Therefore, aircraft taxiing time prediction has attracted significant attention from research institutions and scholars. However, due to the influence of various complex factors on the aircraft's taxiing process, the taxiing time exhibits highly dynamic and strongly nonlinear characteristics, making accurate prediction a considerable challenge.
[0005] To achieve accurate taxiing time prediction, many researchers have attempted to construct aircraft taxiing time prediction models using various machine learning models, such as multiple linear regression, support vector regression, neural network models, and some deep learning models. However, most existing research remains focused on using basic machine learning models to construct taxiing time prediction models, neglecting to optimize and improve these basic models to enhance their predictive performance, as exemplified by the paper "BP Neural Network Prediction Model for Departure Flight Taxiing Time." The predictive performance of various machine learning models largely depends on the setting of their key internal parameters. If these parameters are not properly adjusted, the model may struggle to effectively capture the underlying patterns in the data, leading to large prediction biases, poor stability, and weakened predictive capabilities in actual operation. Currently, many studies still use default parameters or simple grid methods for parameter tuning, lacking systematicity and comprehensiveness, and failing to fully realize the model's potential. Therefore, introducing efficient parameter optimization methods to fine-tune the key parameters of the model not only helps improve the accuracy and stability of taxiing time prediction but also enhances the model's adaptability in complex dynamic operating environments, which is of significant practical importance for supporting airport collaborative decision-making systems and other airport operation management systems. Summary of the Invention
[0006] Purpose of the invention: This invention addresses the shortcomings of existing aircraft taxiing time prediction methods by providing an aircraft taxiing time prediction method based on differential evolution and XGBoost.
[0007] To address the aforementioned technical problems, this invention discloses a method for predicting aircraft taxiing time based on differential evolution and XGBoost, comprising the following steps:
[0008] Step 1: Analyze the taxiing motion of aircraft during departure at civil aviation airports, screen data features that affect aircraft taxiing time, and construct a data feature set for predicting aircraft taxiing time;
[0009] Step 2: Based on the aircraft taxiing time prediction data feature set, collect the corresponding input feature data and the actual taxiing time, and perform data preprocessing to construct the aircraft taxiing time prediction dataset;
[0010] Step 3: Divide the aircraft taxiing time prediction dataset into a training set and a test set;
[0011] Step 4: Use the XGBoost model as the aircraft taxiing time prediction model, select the XGBoost model hyperparameters to be optimized, and set the model hyperparameter search range;
[0012] Step 5: Use the differential evolution optimization algorithm to optimize the selected XGBoost model hyperparameters to obtain the optimal model hyperparameter set;
[0013] Step 6: Using the optimal model hyperparameter set, retrain the XGBoost model to build an aircraft taxiing time prediction model;
[0014] Furthermore, the data features that affect the aircraft taxiing time mentioned in step 1 include: month, day, week of the week, flight number, destination, taxiing distance, planned takeoff time, temperature, humidity, wind speed, wind direction, and air pressure.
[0015] Furthermore, the data preprocessing described in step 2 includes the following steps:
[0016] Step 2-1: Use the label encoding method to convert all non-numerical data into numeric data;
[0017] Step 2-2: Check if any feature data is missing in each data sample. If it is missing, remove the data sample from the dataset.
[0018] Furthermore, in step 3, the aircraft taxiing time prediction dataset is divided into a training set and a test set. Specifically, the aircraft taxiing time prediction dataset is randomly divided into a training set and a test set in a ratio of 8:2, with 80% of the samples used for training and 20% of the samples used for testing.
[0019] Furthermore, in step 4, five hyperparameters of the XGBoost model were selected for optimization, meaning the variable dimension of the optimization problem is D=5. The specific hyperparameters to be optimized are: number of learners (n_estimators), learning rate (learning_rate), maximum tree depth (max_depth), subsample ratio (subsample), and colsample_bytree (colsample_bytree). The XGBoost model possesses strong nonlinear fitting capabilities and good anti-overfitting properties, therefore it was selected as the basic prediction model. The five selected hyperparameters have a crucial impact on the training efficiency, fitting ability, and generalization performance of the XGBoost model; therefore, their optimization is a key focus.
[0020] Furthermore, the specific steps in step 4 for setting the model hyperparameter search range are as follows: the number of learners is an integer within the range of [50, 500], the learning rate is within the range of [0.01, 0.3], the maximum tree depth is an integer within the range of [3, 15], and both the sample sampling ratio and the feature sampling ratio are within the range of [0.5, 1.0]. This search range covers the typical value range of the XGBoost model in practical applications, balancing model performance and search speed.
[0021] Furthermore, step 5 involves optimizing the selected XGBoost model hyperparameters using the differential evolution algorithm, specifically including the following steps:
[0022] Step 5-1: Initialize the population for the differential evolution algorithm. Set the population size NP to NP = 4 * D, where D is the number of hyperparameters. Each individual in the population is encoded as a set of hyperparameter combinations of the XGBoost model. The hyperparameter combinations are generated by random sampling within a pre-set search range for model hyperparameters.
[0023] Step 5-2: Substitute the hyperparameters of each individual in the population into the XGBoost model, train the model using the training set, and calculate the root mean square error (RMSE) of the model on the test set as the fitness function value for that individual. The formula for calculating RMSE is:
[0024]
[0025] Where n is the number of samples in the test set. To determine the actual slide-out time corresponding to the i-th sample in the test set, The predicted slip-out time for the i-th sample in the test set, obtained using the XGBoost model.
[0026] Step 5-3: Perform mutation operations on the population. For each individual in the population... (i=1,2,…,NP), randomly select three distinct individuals. , , The corresponding mutation vector is calculated according to the following mutation strategy. :
[0027]
[0028] Where F is the scaling factor, used to control the perturbation strength of the difference vector, and is uniformly set to F=0.6; the scaling factor F is mainly used to control the search capability of the differential evolution algorithm. The larger F is, the stronger the global exploration capability of the algorithm, and the smaller F is, the stronger the local fine search capability of the algorithm.
[0029] Step 5-4: Perform crossover operation on the population. For each individual in the population... (i=1,2,…,NP), generate experimental vectors by performing crossover operations according to the following formula. :
[0030]
[0031] Where j = 1, 2, ..., D represents the dimension index of the individual vector. Representative vector The component in the j-th dimension; Individual vectors The corresponding mutation vector, Representative vector The component in the j-th dimension; Individual vector The component in the j-th dimension; It is a dimension randomly selected within the range [1,D], ensuring that at least one dimension in the crossover operation comes from the mutation vector; CR is the crossover probability, which is fixed at CR=0.5;
[0032] Step 5-5: For each generated trial vector Check whether the values of each dimension of the vector exceed the preset hyperparameter search range. If they do, use the truncation method to correct them to the preset search range.
[0033] Steps 5-6: For each trial vector The model hyperparameters in the vector are substituted into the XGBoost model, the model is trained using the training set, and the root mean square error (RMSE) value corresponding to the model is calculated on the test set as the fitness function value of the test vector. If the test vector... The RMSE value is less than that of the corresponding original individual. The RMSE value is then used. replace Preserve the individual for the next generation of the population; otherwise, retain the original individual. ;
[0034] Steps 5-7: Repeat steps 5-3 to 5-6 until the termination condition is met. Meeting the termination condition means that the number of iterations reaches a certain number. Then, output the hyperparameter combination corresponding to the individual with the best fitness (smallest RMSE value) in the current population as the optimal hyperparameter set for the model. The number of iterations ranges from [30, 60]. Setting a maximum number of iterations as the termination condition balances the optimization effect and computational cost, ensuring efficient convergence of the algorithm.
[0035] Furthermore, the retraining of the XGBoost model in step 6 to construct the aircraft taxiing time prediction model specifically involves substituting the optimal model hyperparameter set into the XGBoost model, and then training the XGBoost model using all data samples from the aircraft taxiing time prediction dataset, thereby forming the aircraft taxiing time prediction model. The data used for retraining the XGBoost model in step 6 is the aircraft taxiing time prediction dataset obtained in step 2.
[0036] Beneficial effects:
[0037] This invention analyzes the aircraft departure taxiing process at airports, filters key features affecting taxiing time, constructs a feature set and collects relevant data, forms a structured dataset after preprocessing, uses XGBoost as the basic prediction model, and introduces differential evolution algorithm to intelligently optimize the key hyperparameters of the model, finally obtaining a high-precision taxiing time prediction model with the optimal parameter combination.
[0038] The advantage of this invention lies in its automatic search for the optimal hyperparameters of the XGBoost model using a differential evolution algorithm. This avoids the subjectivity and inefficiency of traditional manual or simple hyperparameter tuning methods, thus improving the automation and generalization ability of model training. Simultaneously, the XGBoost model itself possesses strong nonlinear fitting capabilities, effectively capturing the influence of various factors on taxiing time under complex and ever-changing airport operating conditions. The overall method improves the accuracy of aircraft taxiing time prediction, contributing to the refinement and intelligence of airport flight operation management and supporting improvements in airport operational efficiency. Attached Figure Description
[0039] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.
[0040] Figure 1 This is a schematic diagram of the overall process of the present invention.
[0041] Figure 2 It represents the change in the root mean square error value corresponding to the optimal individual during the model hyperparameter optimization process.
[0042] Figure 3 This is a schematic diagram of the predicted aircraft taxiing time. Detailed Implementation
[0043] This invention addresses the shortcomings of existing methods for predicting departure aircraft taxiing time by providing a method based on differential evolution and XGBoost. This method uses the XGBoost model as the basic prediction model and introduces the differential evolution algorithm to intelligently optimize the key hyperparameters of the model, ultimately obtaining a high-precision taxiing time prediction model with the optimal parameter combination, providing decision support for airport flight operation optimization.
[0044] Example 1:
[0045] This embodiment provides a method for predicting aircraft taxiing time based on differential evolution and XGBoost. The embodiment of the invention will be described below with reference to the accompanying drawings. The specific implementation process and steps are as follows, and the overall flow is as follows: Figure 1 As shown.
[0046] Step 1: Analyze the taxiing motion of aircraft during departure at civil aviation airports, screen data features that affect aircraft taxiing time, and construct a data feature set for predicting aircraft taxiing time;
[0047] Data characteristics that affect aircraft taxiing time include: month, day, week of the week, flight number, destination, taxiing distance, scheduled takeoff time, temperature, humidity, wind speed, wind direction, and air pressure.
[0048] Step 2: Based on the aircraft taxiing time prediction data feature set, collect the corresponding input feature data and the actual taxiing time, and perform data preprocessing to construct the aircraft taxiing time prediction dataset;
[0049] Data preprocessing includes the following steps:
[0050] Step 2-1: Use the label encoding method to convert all non-numerical data into numeric data;
[0051] Step 2-2: Check if any feature data is missing in each data sample. If a missing feature is found, remove the data sample from the dataset. The data feature type is determined using the aircraft taxiing time prediction data feature set from Step 1.
[0052] The constructed aircraft taxiing time prediction dataset contains 4000 data samples, as shown in Table 1:
[0053] Table 1. Example of a data sample
[0054] Sample number moon day Week Flight number destination Gliding distance Planned departure time temperature humidity wind speed wind direction air pressure Actual slide time 1 10 22 6 2041 36 840 9 55 50 25 15 29.91 21 2 11 3 5 610 28 1475 46 46 58 24 9 30.23 26
[0055] Step 3: Divide the aircraft taxiing time prediction dataset into a training set and a test set. The specific method for dividing the training set and the test set is to randomly divide the aircraft taxiing time prediction dataset into a training set and a test set in a ratio of 8:2, with 80% of the samples used for training and 20% of the samples used for testing.
[0056] Step 4: Use the XGBoost model as the aircraft taxiing time prediction model, select the XGBoost model hyperparameters to be optimized, and set the model hyperparameter search range;
[0057] Five hyperparameters of the XGBoost model were selected for optimization, meaning the variable dimension of the optimization problem is D=5. The specific hyperparameters to be optimized are: number of learners (n_estimators), learning rate (learning_rate), maximum tree depth (max_depth), sample sampling ratio (subsample), and feature sampling ratio (colsample_bytree).
[0058] The specific search range for model hyperparameters is set as follows: the number of learners is an integer within the range of [50, 500], the learning rate is within the range of [0.01, 0.3], the maximum tree depth is an integer within the range of [3, 15], and both the sample sampling ratio and the feature sampling ratio are within the range of [0.5, 1.0].
[0059] Step 5: Use the differential evolution optimization algorithm to optimize the selected XGBoost model hyperparameters to obtain the optimal model hyperparameter set;
[0060] The hyperparameters of the selected XGBoost model are optimized using the differential evolution algorithm. The specific steps include:
[0061] Step 5-1: Initialize the population for the differential evolution algorithm. Set the population size NP to NP=20. Each individual in the population is encoded as a set of hyperparameter combinations of the XGBoost model. The hyperparameter combinations are generated by random sampling within a pre-set search range for model hyperparameters. Setting the population size to 4 times the dimension of the search variables helps to provide diverse candidate solutions and enhances the algorithm's global optimization capability.
[0062] Step 5-2: Substitute the hyperparameters of each individual in the population into the XGBoost model, train the model using the training set, and calculate the root mean square error (RMSE) of the model on the test set as the fitness function value for that individual. The formula for calculating RMSE is:
[0063]
[0064] Where n is the number of samples in the test set. To determine the actual slide-out time corresponding to the i-th sample in the test set, is the predicted slip-out time obtained using the XGBoost model for the i-th sample in the test set; root mean square error is a commonly used fitness metric that can intuitively reflect the degree of deviation between the model's prediction and the true value.
[0065] Step 5-3: Perform mutation operations on the population. For each individual in the population... (i=1,2,…,NP), randomly select three distinct individuals. , , The corresponding mutation vector is calculated according to the following mutation strategy. :
[0066]
[0067] Where F is the scaling factor, used to control the perturbation intensity of the difference vector, and is uniformly set to F=0.6;
[0068] Step 5-4: Perform crossover operation on the population. For each individual in the population... (i=1,2,…,NP), generate experimental vectors by performing crossover operations according to the following formula. :
[0069]
[0070] Where j = 1, 2, ..., D represents the dimension index of the individual vector. Representative vector The component in the j-th dimension; Individual vectors The corresponding mutation vector, Representative vector The component in the j-th dimension; Individual vector The component in the j-th dimension; It is a dimension randomly selected within the range of [1,D] to ensure that at least one dimension in the crossover operation comes from the mutation vector; CR is the crossover probability, which is fixed at CR=0.5; this is beneficial for balancing the diversity of the population and retaining excellent individuals.
[0071] Step 5-5: For each generated trial vector The algorithm checks whether the values of each dimension of the vector exceed the preset hyperparameter search range. If they do, it uses a truncation method to correct them to the preset search range. The truncation method is a commonly used boundary handling method in the optimization field, which can ensure that all generated hyperparameter combinations are within the set range.
[0072] Steps 5-6: For each trial vector The model hyperparameters in the vector are substituted into the XGBoost model, the model is trained using the training set, and the root mean square error (RMSE) value corresponding to the model is calculated on the test set as the fitness function value of the test vector. If the test vector... The RMSE value is less than that of the corresponding original individual. The RMSE value is then used. replace Preserve the individual for the next generation of the population; otherwise, retain the original individual. Fitness-based selection mechanisms ensure that the population as a whole evolves towards a better direction.
[0073] Step 5-7: Repeat steps 5-3 to 5-6 until the termination condition is met. The termination condition means that the number of iterations reaches 30. Then, output the hyperparameter combination corresponding to the individual with the best fitness (smallest RMSE value) in the current population as the optimal hyperparameter set of the model.
[0074] Figure 2 This paper demonstrates the change in the root mean square error (RMSE) value of the best individual in the population during the iterative optimization of model hyperparameters using the differential evolution algorithm. It can be seen that as the algorithm iterates, the RMSE value of the best individual gradually decreases, indicating that the corresponding model prediction error gradually decreases and the prediction ability gradually improves.
[0075] After completing the iterative optimization of hyperparameters, the optimal set of model hyperparameters obtained is: n_estimators=175, learning_rate=0.21, max_depth=11, subsample=0.53, colsample_bytree=0.73.
[0076] Step 6: The retraining of the XGBoost model described in Step 6 to construct the aircraft taxiing time prediction model involves substituting the optimal model hyperparameter set into the XGBoost model, and then training the XGBoost model using all data samples from the aircraft taxiing time prediction dataset, thus forming the aircraft taxiing time prediction model. Using all data for the final model training fully utilizes data information to improve the model's prediction stability and accuracy in practical applications.
[0077] Figure 3 The results of using the finally constructed aircraft taxiing time prediction model to predict a portion of the data samples are shown. The model's average absolute error is 1.8, which can predict the aircraft taxiing time relatively accurately and has good practicality.
[0078] This invention provides a method and approach for predicting aircraft taxiing time based on differential evolution and XGBoost. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.
Claims
1. A method for predicting aircraft taxiing time based on differential evolution and XGBoost, characterized in that, Includes the following steps: Step 1: Analyze the taxiing motion of aircraft during departure at civil aviation airports, screen data features that affect aircraft taxiing time, and construct a data feature set for predicting aircraft taxiing time; Step 2: Based on the aircraft taxiing time prediction data feature set obtained in Step 1, collect the corresponding input feature data and the actual taxiing time, perform data preprocessing, and construct the aircraft taxiing time prediction dataset. Step 3: Divide the aircraft taxiing time prediction dataset obtained in Step 2 into a training set and a test set; Step 4: Use the XGBoost model as the aircraft taxiing time prediction model, select the XGBoost model hyperparameters to be optimized, and set the model hyperparameter search range; Step 5: Use the differential evolution optimization algorithm to optimize the selected XGBoost model hyperparameters to obtain the optimal model hyperparameter set; Step 6: Using the optimal model hyperparameter set, retrain the XGBoost model to build an aircraft taxiing time prediction model.
2. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 1, characterized in that, The data features that affect the aircraft taxiing time mentioned in step 1 include: month, day, weekday, flight number, destination, taxiing distance, planned takeoff time, temperature, humidity, wind speed, wind direction, and air pressure.
3. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 1, characterized in that, The data preprocessing described in step 2 includes the following steps: Step 2-1: Use the label encoding method to convert all non-numerical data into numeric data; Step 2-2: Check if any feature data is missing in each data sample. If it is missing, remove the data sample from the dataset. The data feature type is determined by the aircraft taxiing time prediction data feature set in Step 1.
4. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 1, characterized in that, The method for dividing the aircraft taxiing time prediction dataset into training and testing sets in step 3 is to randomly divide the dataset into training and testing sets in a ratio of 8:2, with 80% of the samples used for training and 20% used for testing.
5. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 1, characterized in that, The selected XGBoost model hyperparameters to be optimized in step 4 include the number of learners, learning rate, maximum tree depth, sample sampling ratio, and feature sampling ratio, i.e., the variable dimension of the optimization problem is D=5.
6. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 5, characterized in that, The specific steps for setting the model hyperparameter search range in step 4 are as follows: the number of learners is an integer within the range of [50, 500], the learning rate is within the range of [0.01, 0.3], the maximum tree depth is an integer within the range of [3, 15], and both the sample sampling ratio and the feature sampling ratio are within the range of [0.5, 1.0].
7. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 1, characterized in that, Step 5 describes optimizing the selected XGBoost model hyperparameters using the differential evolution algorithm. The specific steps include: Step 5-1: Initialize the population for the differential evolution algorithm. Set the population size NP to NP = 4 * D, where D is the number of hyperparameters. Each individual in the population is encoded as a set of hyperparameter combinations of the XGBoost model. The hyperparameter combinations are generated by random sampling within a pre-set search range for model hyperparameters. Step 5-2: Substitute the hyperparameters of each individual in the population into the XGBoost model, train the model using the training set, and calculate the root mean square error (RMSE) value of the model on the test set as the fitness function value of the individual. The formula for calculating RMSE is: Where n is the number of samples in the test set. To determine the actual slide-out time corresponding to the i-th sample in the test set, The predicted slip-out time for the i-th sample in the test set, obtained using the XGBoost model. Step 5-3: Perform mutation operations on the population. For each individual in the population... (i=1,2,…,NP), randomly select three distinct individuals. , , The corresponding mutation vector is calculated according to the following mutation strategy. : Where F is the scaling factor, which controls the perturbation strength of the difference vector, and F=0.6; Step 5-4: Perform crossover operation on the population. For each individual in the population... (i=1,2,…,NP), generate experimental vectors by performing crossover operations according to the following formula. : Where j = 1, 2, ..., D represents the dimension index of the individual vector. Representative vector The component in the j-th dimension; Individual vectors The corresponding mutation vector, Representative vector The component in the j-th dimension; Individual vector The component in the j-th dimension; It is a dimension randomly selected within the range [1,D], ensuring that at least one dimension in the crossover operation comes from the mutation vector; CR is the crossover probability, which is fixed at CR=0.5; Step 5-5: For each generated trial vector Check whether the values of each dimension of the vector exceed the preset hyperparameter search range. If they do, use the truncation method to correct them to the preset search range. Steps 5-6: For each trial vector The model hyperparameters in the vector are substituted into the XGBoost model, the model is trained using the training set, and the root mean square error (RMSE) value corresponding to the model is calculated on the test set as the fitness function value of the test vector. If the test vector... The RMSE value is less than that of the corresponding original individual. The RMSE value is then used. replace Preserve the individual for the next generation of the population; otherwise, retain the original individual. ; Step 5-7: Repeat steps 5-3 to 5-6 until the termination condition is met. The termination condition is to repeat the iteration a certain number of times and output the hyperparameter combination corresponding to the individual with the best fitness in the current population, that is, the individual with the smallest RMSE value, as the optimal hyperparameter set of the model.
8. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 7, characterized in that, The number of iterations ranges from [30, 60].
9. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 1, characterized in that, The retraining of the XGBoost model in step 6 to construct the aircraft taxiing time prediction model involves substituting the optimal model hyperparameter set into the XGBoost model and training the XGBoost model using all data samples in the aircraft taxiing time prediction dataset to form the aircraft taxiing time prediction model.
10. The aircraft taxiing time prediction method based on differential evolution and XGBoost according to claim 1, characterized in that, The data used to retrain the XGBoost model in step 6 is the aircraft taxiing time prediction dataset obtained in step 2.