A method and system for staged coastdown time prediction and coastdown control

Abstract

The application provides a staged taxiing time prediction and taxiing control method and system, and relates to the field of air transportation.A staged taxiing time prediction and taxiing control method comprises the following steps: collecting flight data of a plurality of departure taxiing processes, each of which is divided into a plurality of taxiing stages; extracting flight arrival and departure features and departure taxiing time of each taxiing stage by using the preprocessed flight data; selecting part of the flight arrival and departure features of each taxiing stage to obtain a plurality of interaction features; training a departure taxiing time prediction model by using all the flight arrival and departure features, the interaction features and the departure taxiing time of each taxiing stage; calculating taxiing time slots of each taxiing stage to obtain actual taxiing time for optimizing the model and obtaining predicted departure taxiing time of each taxiing stage.The application improves the prediction accuracy of taxiing time under peak period and different taxiing stages, thereby accelerating the overall operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of air transport, and more specifically, to a method and system for predicting taxiing time in stages and for taxiing control. Background Technology

[0002] With the rapid development of the air transport industry, global airport operations are facing increasing pressure. Against this backdrop, airlines and airport management, while adhering to safety standards, need to focus on improving flight punctuality and overall operational efficiency. Therefore, the prediction and control of aircraft taxiing time has received widespread attention from academia and industry in recent years, becoming an important research direction for improving airport operational efficiency. Traditional taxiing time prediction typically treats the entire taxiing process as a whole, making it difficult to effectively address the complexity of different stages (such as pushback and runway maneuvering). This results in insufficient prediction accuracy of taxiing time during peak periods and at different taxiing stages, thus affecting overall operational efficiency. Currently, there is a need for a phased taxiing time prediction and control method and system that can improve the prediction accuracy of taxiing time during peak periods and at different taxiing stages, thereby accelerating overall operational efficiency. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for predicting and controlling taxiing time in stages, which can improve the accuracy of taxiing time prediction during peak periods and at different taxiing stages, thereby accelerating overall operating efficiency.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] A phased taxiing time prediction and taxiing control method includes: collecting flight data for multiple departure taxiing processes, each departure taxiing process being divided into multiple taxiing stages; extracting multiple flight arrival and departure features and departure taxiing times for each taxiing stage using preprocessed flight data; repeatedly selecting some flight arrival and departure features from each taxiing stage to obtain interaction features; preprocessing the multiple interaction features for each taxiing stage; training a departure taxiing time prediction model using all flight arrival and departure features, the preprocessed interaction features, and the departure taxiing times for each taxiing stage; optimizing the departure taxiing time prediction model by calculating the actual taxiing time slots for each taxiing stage; and using the optimized departure taxiing time prediction model to obtain the predicted departure taxiing time for each taxiing stage during the departure taxiing process.

[0006] Each of the aforementioned departure taxiing processes is divided into multiple taxiing phases, including: from the time of traction taxiing to the time of autonomous taxiing, from the time of autonomous taxiing to the time of entering the runway, and from the time of entering the runway to the time of takeoff, each of the aforementioned departure taxiing processes is divided into three taxiing phases.

[0007] The above-mentioned optimized departure taxiing time prediction model is used to obtain the predicted departure taxiing time for each of the above-mentioned taxiing stages during the departure taxiing process, including: the taxiing stages from the time of traction taxiing to the time of autonomous taxiing, and the taxiing stages from the time of autonomous taxiing to the time of entering the runway, respectively, and the taxiing time is obtained by selecting the corresponding departure taxiing time prediction model; the taxiing stage from the time of entering the runway to the time of takeoff is obtained by using the median of statistical time; the taxiing times of the three taxiing stages are summed to obtain the predicted departure taxiing time for the entire departure taxiing process.

[0008] The above-mentioned interactive features are obtained by selecting partial flight arrival and departure characteristics from each of the aforementioned taxiing phases. Specifically, these features include: obtaining the runway and taxiway usage frequency and space requirements for runways and taxiways based on the number of arrivals; recording the runway and taxiway idle time based on the number of arrivals and departures or the time interval between the most recent arrivals and departures; calculating the arrival and departure throughput based on the number of arrivals and departures, and obtaining a comparison result between the aforementioned arrival and departure throughput and peak hour throughput; and using dummy variable encoding for the combination of parking stands and runways, mapping each combination to a binary feature.

[0009] Preprocessing the above flight data includes: filtering flights that undergo de-icing during taxiing, filtering and deleting outliers, and filling missing values ​​with the median using any one or more of the following methods;

[0010] Preprocessing the above interactive features includes any one or more of the following: using label encoding for categorical variable features, employing target encoding methods for parking positions and runway features, and standardizing numerical features using standard scaling.

[0011] The above-mentioned method of obtaining the actual taxiing time by calculating the taxiing time slots of each of the aforementioned taxiing stages includes: calculating the predicted taxiing time error of the aforementioned departure taxiing time prediction model using the predicted departure taxiing time and actual departure taxiing time of each of the aforementioned taxiing stages during the departure taxiing process; obtaining the probability density function and its cumulative distribution function of the aforementioned predicted taxiing time error using the KDE method; using the quantile of the aforementioned cumulative distribution function as the upper bound and the symmetric quantile of the aforementioned cumulative distribution function as the lower bound to obtain the range of predicted taxiing time error; calculating the actual taxiing time using the aforementioned predicted taxiing time error of each of the aforementioned taxiing stages, or summing the range of predicted taxiing time errors of all the aforementioned taxiing stages during the departure taxiing process and distributing it to each of the aforementioned taxiing stages to obtain the actual taxiing time.

[0012] The aforementioned departure taxiing time prediction model adopts the extreme gradient boosting model. When training the aforementioned departure taxiing time prediction model for the corresponding taxiing stage, one or more of the following model parameters are optimized: learning rate, maximum depth, sample sampling ratio, feature sampling ratio, and number of trees.

[0013] A phased taxiing time prediction and taxiing control system includes: a data acquisition module for acquiring flight data for multiple departure taxiing processes, each departure taxiing process being divided into multiple taxiing stages; a feature extraction module for extracting multiple arrival and departure features and departure taxiing times for each taxiing stage using preprocessed flight data; a feature interaction module for selecting partial arrival and departure features from each taxiing stage to obtain interaction features; a model training module for preprocessing the interaction features of each taxiing stage; and training a departure taxiing time prediction model using all arrival and departure features, the preprocessed interaction features, and the departure taxiing times for each taxiing stage; a model optimization module for optimizing the departure taxiing time prediction model by calculating the actual taxiing time for each taxiing stage; and a taxiing segmentation module for obtaining the predicted departure taxiing time for each taxiing stage during the departure taxiing process using the optimized departure taxiing time prediction model.

[0014] An electronic device includes a memory, a processor, and a computer program running on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned phased coasting time prediction and coasting control method.

[0015] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned phased coasting time prediction and coasting control method.

[0016] Compared with the prior art, the present invention has at least the following advantages or beneficial effects:

[0017] This application provides a phased taxiing time prediction and taxiing control method. By dividing the departure taxiing phase into multiple taxiing phases, it extracts multiple arrival and departure features from flight data, as well as the taxiing time of each phase, to construct a departure taxiing time prediction model for different phases. Furthermore, it mines the interactive features of departure taxiing through the extracted features, trains the departure taxiing time prediction model for the corresponding taxiing phase, calculates taxiing time slots based on multiple taxiing phases, and optimizes the model. The total taxiing time can be obtained by summing the predictions of each taxiing phase using the model. This invention combines space requirements, idle time, and throughput to train the model to predict runway and taxiway taxiing times. It can improve the prediction accuracy of taxiing time during peak periods and at different taxiing phases, thereby accelerating overall operational efficiency. This invention comprehensively considers the performance of various models at different taxiing phases, showing a lower error level and stronger generalization ability, providing more reliable and stable prediction results. It has stronger applicability in actual operation and performs well in airport scenarios with high taxiing time uncertainty. The rational setting of taxiing time slots at each stage helps to cope with complex operating conditions and uncontrollable factors, ensuring a smooth taxiing process. This flexible taxiing time slot management method not only provides an optimization direction for the future operation of the airport, but also provides the possibility of dynamic adjustment for taxiing in different seasons, time periods and operating environments, ensuring operational robustness and flexibility. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the phased taxiing time prediction and taxiing control method according to an embodiment of the present invention;

[0020] Figure 2 This is a comparison chart of the predicted and actual values ​​of the departure taxiing time prediction model in this embodiment of the invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Example

[0023] like Figures 1-2 The illustrated embodiment of this application provides a phased taxiing time prediction and taxiing control method, which includes: collecting flight data for multiple departure taxiing processes, each departure taxiing process being divided into multiple taxiing stages; extracting multiple flight arrival and departure features and departure taxiing times for each taxiing stage using the preprocessed flight data; repeatedly selecting some flight arrival and departure features from each taxiing stage to obtain interaction features; preprocessing the multiple interaction features for each taxiing stage; training a departure taxiing time prediction model using all flight arrival and departure features, the preprocessed multiple interaction features, and the departure taxiing times for each taxiing stage; optimizing the departure taxiing time prediction model by calculating the actual taxiing time for each taxiing stage; and using the optimized departure taxiing time prediction model to obtain the predicted departure taxiing time for each taxiing stage during the departure taxiing process.

[0024] Each of the aforementioned departure taxiing processes is divided into multiple taxiing phases, including: from the time of traction taxiing to the time of autonomous taxiing, from the time of autonomous taxiing to the time of entering the runway, and from the time of entering the runway to the time of takeoff, each of the aforementioned departure taxiing processes is divided into three taxiing phases.

[0025] The above-mentioned optimized departure taxiing time prediction model is used to obtain the predicted departure taxiing time for each of the above-mentioned taxiing stages during the departure taxiing process, including: the taxiing stages from the time of traction taxiing to the time of autonomous taxiing, and the taxiing stages from the time of autonomous taxiing to the time of entering the runway, respectively, and the taxiing time is obtained by selecting the corresponding departure taxiing time prediction model; the taxiing stage from the time of entering the runway to the time of takeoff is obtained by using the median of statistical time; the taxiing times of the three taxiing stages are summed to obtain the predicted departure taxiing time for the entire departure taxiing process.

[0026] The above-mentioned interactive features are obtained by selecting partial flight arrival and departure characteristics from each of the aforementioned taxiing phases. Specifically, these features include: obtaining the runway and taxiway usage frequency and space requirements for runways and taxiways based on the number of arrivals; recording the runway and taxiway idle time based on the number of arrivals and departures or the time interval between the most recent arrivals and departures; calculating the arrival and departure throughput based on the number of arrivals and departures, and obtaining a comparison result between the aforementioned arrival and departure throughput and peak hour throughput; and using dummy variable encoding for the combination of parking stands and runways, mapping each combination to a binary feature.

[0027] Preprocessing the aforementioned flight data includes any one or more of the following methods: filtering flights that undergo de-icing during taxiing, filtering and deleting outliers, and filling missing values ​​with the median; preprocessing the aforementioned interactive features includes any one or more of the following methods: using label encoding for categorical variable features, employing target encoding methods for parking positions and runway features, and standardizing numerical features using standard scaling.

[0028] The above-mentioned method of obtaining the actual taxiing time by calculating the taxiing time slots of each of the aforementioned taxiing stages includes: calculating the predicted taxiing time error of the aforementioned departure taxiing time prediction model using the predicted departure taxiing time and actual departure taxiing time of each of the aforementioned taxiing stages during the departure taxiing process; obtaining the probability density function and its cumulative distribution function of the aforementioned predicted taxiing time error using the KDE method; using the quantile of the aforementioned cumulative distribution function as the upper bound and the symmetric quantile of the aforementioned cumulative distribution function as the lower bound to obtain the range of predicted taxiing time error; calculating the actual taxiing time using the aforementioned predicted taxiing time error of each of the aforementioned taxiing stages, or summing the range of predicted taxiing time errors of all the aforementioned taxiing stages during the departure taxiing process and distributing it to each of the aforementioned taxiing stages to obtain the actual taxiing time.

[0029] The aforementioned departure taxiing time prediction model adopts the extreme gradient boosting model. When training the aforementioned departure taxiing time prediction model for the corresponding taxiing stage, one or more of the following model parameters are optimized: learning rate, maximum depth, sample sampling ratio, feature sampling ratio, and number of trees.

[0030] This invention divides the departure taxiing process of a civil aircraft into three stages, specifically defined as follows: Stage 1: From the moment of traction taxiing to the moment of autonomous taxiing, i.e., from the moment the wheel chocks are removed to the moment the aircraft begins autonomous powered taxiing. After the tractor pulls the aircraft to the designated position, the aircraft brakes and starts its engines, and then the tractor leaves. After completing this series of operations, the pilot performs an inspection and issues a taxiing request to ground control. Upon receiving taxiing clearance, the pilot begins the next stage. Stage 2: From the moment of autonomous taxiing to the moment of runway entry, i.e., from the moment autonomous powered taxiing begins to the moment the aircraft enters the runway. This stage includes autonomous powered taxiing on the taxiway and queuing to enter the runway. Stage 3: From the moment the aircraft enters the runway to the moment of actual takeoff, i.e., from the moment the aircraft enters the runway to the moment of actual takeoff. In this stage, the aircraft remains on the runway, awaiting takeoff instructions, and finally completes takeoff.

[0031] From an air traffic control perspective, each stage of the departure taxiing process has clearly defined control responsibilities and requirements. In stage one, the apron controller must ensure that there are no other aircraft or ground vehicles interfering with the aircraft after it reaches the designated position. The apron controller must communicate with the pilot in a timely manner to ensure safety. In stage two, the apron controller must continuously monitor the aircraft's taxiing position and status to ensure it moves along the designated taxiway. When the aircraft reaches the junction of the main taxiway and the apron, it will be handed over to the tower control ground station for continued control monitoring. In stage three, the tower control must monitor the runway environment in real time to ensure the safety of the aircraft during takeoff, and after confirming safety, issue clearance for takeoff.

[0032] Preprocessing flight data mainly includes filtering out flights undergoing de-icing during taxiing, filtering and deleting outliers, and filling missing values ​​with the median. Furthermore, the flight data is reprocessed to obtain features such as the number of arrivals and departures during taxiing. When constructing a departure taxiing time prediction model, a random sampling method can be used to divide the data into training and test sets, selecting 80% of the data as the training set and the remaining 20% ​​as the test set.

[0033] This article will select one or more of the following flight arrival and departure characteristics: date, flight number, VIP identification, aircraft type, parking position, runway, SID (departure procedure), ETD time, PUS time, TAX time, aircraft runway entry time, ATD time, de-icing apron entry time, and initial sector. Other flight arrival and departure characteristics include runway, taxiway, parking position, number of arrivals and departures, arrival and departure time interval, peak hourly throughput, airline, flight month, congestion assessment result, aircraft type, departure procedure, initial sector, and VIP flights.

[0034] Optionally, the specific arrival and departure characteristics of multiple flights can be selected, as shown in Table 1 below:

[0035]

[0036]

[0037]

[0038] Table 1

[0039] To further enhance the model's predictive power, interactive features are introduced. Interactive features refer to combining multiple features together to generate new features, thereby capturing more complex relationships in the data.

[0040] Optionally, the interaction features can be preprocessed, as follows:

[0041] (1) Categorical variable encoding: Label encoding was used for multiple categorical features, such as aircraft type, airline, SID (Standard Departure Procedure), initial sector, and flow / congestion judgment. Label encoding can convert categorical features into integers, so that these features can be directly used by the model, while ensuring that the order relationship between categories is preserved.

[0042] (2) Target Encoding: To further enhance the expressive power of categorical features, target encoding was employed for the parking positions and runway features. This method converts categorical variables into numerical features by calculating the average value between each category and the target variable. This conversion not only improves the model's ability to capture subtle relationships between categories but also effectively enhances the discriminative performance of categorical variables.

[0043] (3) Numerical Feature Standardization: The numerical features were standardized using a standard scaler, adjusting the mean of the feature values ​​to 0 and the variance to 1. Standardization can eliminate differences between different feature units and avoid certain features from having an unbalanced impact on model training. In the embodiments of this application, features such as the number of arrivals, departures, and arrivals / departures were all standardized to ensure that all numerical features participate in model training at the same scale, thereby improving the convergence and prediction performance of the model.

[0044] Optional, the interaction features are shown in Table 2 below:

[0045]

[0046] Table 2

[0047] The specific analysis process of the principle behind the above-mentioned interaction features is as follows:

[0048] Arrivals and Departures: Increased arrivals may lead to higher runway and taxiway usage, while increased departures may increase the space requirements for takeoff preparation. The combined effect of these two factors can impact runway and taxiway utilization, especially during periods of high traffic density. By using the interaction feature "Arrivals * Departures," the model can identify the impact on taxiing time when both factors change simultaneously, rather than simply their effects on individual variables.

[0049] Arrivals and Departures and Nearest Interval: During taxiing, runway usage frequency and flight intervals are key factors affecting taxiing time. As the number of arriving and departing flights increases, runway utilization becomes more strained, especially within short periods. Changes in flight intervals directly impact runway idle time; shorter intervals mean higher runway utilization, potentially increasing taxiing wait times. Analyzing the ratio of "arrivals and departures / nearest interval" reveals the runway's busyness during taxiing, particularly regarding short-term resource usage.

[0050] Peak hourly throughput and number of arrivals and departures: The relationship between peak hourly throughput (i.e., the airport's maximum flight arrival and departure capacity per unit hour) and the number of arrivals and departures reflects the linkage between the number of arrivals and departures and the airport's resource demand and maximum throughput capacity during peak hours. By analyzing the interaction term between the number of arrivals and departures and peak hourly throughput, we can assess how the airport's throughput capacity and resource allocation affect overall flight taxiing time within a certain time frame. Differences in the number of arrivals and departures can influence airport resource allocation, thereby affecting throughput and taxiing time.

[0051] Parking stands and runways: Different combinations of parking stands and runways mean different taxiing distances from the parking stand to the runway, thus affecting taxiing time. By capturing these features, the model can better handle the efficiency differences between different parking stand and runway combinations, thereby more accurately predicting taxiing time. Through one-hot dummy encoding, each parking stand and runway combination is mapped to a binary feature, allowing the model to learn the impact of different combinations on taxiing time. The generated features are dummy variable columns corresponding to each "parking stand-runway combination," helping the model better identify the potential impact of different parking stand and runway combinations on taxiing time.

[0052] Extreme Gradient Boosting (XGBoost) is an efficient implementation based on gradient boosting trees that enhances the model's prediction accuracy by integrating multiple weak learners (decision trees). This ensemble approach not only improves model performance but also optimizes the combination of individual weak learners through the gradient boosting framework, thereby achieving higher accuracy and efficiency in prediction tasks. Traditional taxiing time prediction tends to handle the entire taxiing process, while this model, by breaking down the taxiing process into multiple stages, allows for independent adjustment and optimization of each stage. Taxiing time is divided into three stages, with stages one and two accounting for a larger proportion of the total taxiing time and requiring higher prediction accuracy. Therefore, when determining whether the proportion of taxiing time in each of these stages exceeds a preset threshold, based on past experience, the Extreme Gradient Boosting model can be selected to predict stages one and two separately. Stage three accounts for a smaller proportion of the taxiing time, so the median statistical time of stage three is directly used as the prediction result for stage three in the model. The sum of the prediction results for stages one, two, and three is the predicted value for the entire taxiing time.

[0053] This invention fine-tunes the model's parameters. The main parameters include: learning_rate, max_depth, subsample, colsample_bytree, and n_estimators. After multiple rounds of testing and parameter tuning, some key parameters for the Stage 1 and Stage 2 models are shown in Table 3 below.

[0054]

[0055] Table 3

[0056] The application of taxiing time slots may differ at different phases of aircraft taxiing. For example, during pushback, taxiing time slots may be used as redundancy to ensure the aircraft safely pushes off the parking stand and onto the taxiway. During autonomous powered taxiing, taxiing time slots may be used to address traffic congestion or situations requiring transfers between multiple taxiways. During runway holding, dynamic taxiing time slots may be used to address changes in takeoff sequence or other temporary airport operational restrictions. When calculating taxiing time slots in Phase 3, the time slot range is smaller and less significant, so it will not be considered. Therefore, taxiing time slots will only be calculated for Phases 1 and 2, namely pushback taxiing time slots and autonomous taxiing time slots on the taxiway.

[0057] Kernel density estimation is a nonparametric method for estimating probability density functions. To ensure the accuracy of kernel density estimation, this embodiment uses a Gaussian kernel function to fit the error distribution.

[0058] First, the error in predicting flight taxiing time is:

[0059] e i =t pre-taxi，i -t taxi，i (1)

[0060] Among them, t pre-taxi，i t is the predicted departure taxiing time for the i-th flight; taxi，i It is the actual departure taxiing time of the i-th flight.

[0061] The probability density function f(e) and its cumulative distribution function F(e) of the error can be obtained using the KDE method:

[0062]

[0063] Where n is the number of data points, and x represents a certain value of the error. i is the error of the actual observation, h is the bandwidth parameter, and K is the Gaussian kernel function.

[0064] The method for calculating taxiing time slots involves defining an error range to ensure that adding the predicted taxiing time to a value within that range results in an on-time rate p for the actual taxiing time. In other words, the goal is to find two boundaries e within the error range. low and e high It meets the following conditions:

[0065] P(e low ≤e≤e high )=p (4)

[0066] Upper boundary e high These are the quantiles of the cumulative distribution function:

[0067] F(e high )=p (5)

[0068] lower bound e low These are the symmetric quantiles of the cumulative distribution function:

[0069] F(e low )=1-p (6)

[0070] Through the inverse function F of the cumulative distribution function -1 (p) yields:

[0071] e high =F -1 (p) (7)

[0072] e low =F -1 (1-p) (8)

[0073] Based on the error distribution obtained through KDE (K Desktop Environment), it can be deduced that the allowable deviation range for departure time when arriving at the runway with an on-time rate p is: [-e high -e low ].

[0074] Flight data is collected based on different taxiing stages of multiple departure taxiing processes. The number of flights includes arrival and departure data, as shown in Tables 4 and 5 below. Table 4 shows departure flight data, and Table 5 shows arrival flight data.

[0075] 2021 / 02 / 28 CBJ5315 / A320 155 11L DOT01D 2021 / 02 / 28 CES9724 / B737 122 35R PEG01D 2021 / 02 / 28 CBJ5351 / A333 153 11L ELK01D

[0076] 22:30:00 22:17:34 22:19:18 22:26:18 22:53:20 22:35:00 22:40:00 22:26:31 22:29:03 22:41:32 23:24:14 22:46:00 22:45:00 22:44:10 22:45:44 22:55:31 23:45:37 22:45:00

[0077] 22:54:37 22:26:57 EDLV 23:26:40 22:47:06 WDLV 23:46:47 22:56:58 EDLV

[0078] Table 4

[0079]

[0080] Table 5

[0081] Flight data includes: date, flight number, VIP identification, aircraft type, parking position, runway, SID (departure procedure), ETD time, PUS time, TAX time, aircraft runway approach time, ATD time, de-icing apron entry time, initial sector, number of arrivals and departures during taxiing, nearest interval, and airport peak hourly throughput, among other information. TAX time includes, for example: taxi out time (min), taxi_phase1 time, taxi_phase2 time, and taxi_phase3 time.

[0082] Preprocessing the raw flight data mainly includes filtering out flights undergoing de-icing during taxiing, filtering and deleting outliers, and filling missing values ​​with the median. Arrival and departure features, as well as departure taxiing times, are extracted from the flight data for each taxiing stage of the departure process; then, some arrival and departure features are selected to further obtain interaction features. A random sampling method is used to divide the data into training and testing sets, with 80% of the data selected as the training set and the remaining 20% ​​as the testing set. The arrival, departure, and interaction features are as described in the examples above; specific arrival and departure features are shown in Table 1, and interaction features are shown in Table 2. Arrival, departure, and interaction features of different taxiing stages within the same departure taxiing process can be partially identical, thus simultaneously predicting the taxiing times of multiple taxiing stages of the departure process. The arrival, departure, and interaction features of each taxiing stage are incorporated into the departure taxiing time prediction model for training, allowing the model to predict the taxiing times of each stage based on the characteristics of each flight's data.

[0083] Table 6 below compares the predicted and actual taxiing times of several randomly selected flights.

[0084] 18.52 17.59 19.53 21.75 15.53 16.15 15.57 14.49 16.6 17.34 33.93 31.38

[0085] Table 6

[0086] To visually evaluate the performance of the departure taxiing time prediction model, 200 predicted values ​​were randomly selected and compared with the actual values. The results are as follows: Figure 2 As shown, the blue line represents the actual taxiing time, and the red line represents the model-predicted taxiing time. The model's predictions generally align with the actual taxiing time trend, especially within the medium taxiing period (10-45 minutes), where the predictions are quite accurate.

[0087] Furthermore, a comprehensive evaluation was conducted on four models, including Random Forest, Support Vector Regression, Extreme Gradient Boosting (XGBoost), and our own model. The evaluation used mean absolute error (MAE), mean squared error (MSE), and coefficient of determination (R²) to measure these parameters. 2 The evaluation metrics were used to compare the model's performance on the training and test sets. The evaluation results of the departure taxiing time prediction model are shown in Table 7.

[0088]

[0089] Table 7

[0090] The model prediction accuracy is shown in Table 8 below:

[0091]

[0092]

[0093] Table 8

[0094] The predicted departure taxiing time differs from the actual value by ±1, ±2, ±3, ±4, and ±5 minutes. The extreme gradient boosting model outperforms our model by several percentage points across all error ranges, especially at ±5 minutes, where our model achieves an accuracy of 96%.

[0095] Dynamic taxiing time slot control employs two strategies for allocating time slots. The first strategy calculates taxiing time slots independently for each stage. This method sets an independent error range for each stage and derives the allowable taxiing time slot based on the stage's prediction time and error. Each stage can independently meet the set on-time rate throughout the entire taxiing process without affecting others. The second strategy calculates the overall taxiing time slot range and allocates it to each taxiing stage according to its time proportion. This method first determines an overall allowable taxiing error range and then allocates time slots based on the proportion of taxiing time in each stage to ensure the overall on-time rate. The taxiing time slot calculation results for Strategy 1 are shown in Table 9 below:

[0096]

[0097]

[0098] Table 9

[0099] The calculation results for the glide slot under Strategy 2 are shown in Table 10 below:

[0100]

[0101] Table 10

[0102] In actual operation, CTOT (Calculated Takeoff Time) is an important reference standard for analyzing taxiing time slots. According to the Civil Aviation Administration of China's "Rules for Air Traffic Control Flow Management (Trial Implementation)," CTOT time slot tolerances are divided into two categories: Category I tolerance (-5, +10 minutes) and Category II tolerance (-3, +3 minutes). Therefore, when designing taxiing time slots, it is necessary to ensure that taxiing can ultimately be completed within this time window.

[0103] In Strategy 1, the advance and delay times for both taxiing phase one and taxiing phase two increase with the on-time rate. Specifically, Strategy 1 allows advance times of 209 seconds and 267 seconds at 80% and 85% on-time rates, respectively, and delay times of 238 seconds and 290 seconds, both meeting the CTOT Type I tolerance (-5, +10 minutes). However, at a 90% on-time rate, the allowable advance time of 356 seconds does not meet the CTOT Type I tolerance. Therefore, the CTOT Type I tolerance should be used as the boundary standard, and the allowable advance times for each phase should be reduced to a sum of 300 seconds. The advantage of this strategy is that it can precisely control the on-time rate of each taxiing phase, ensuring that the start time is within the error range, especially suitable for situations where there are large differences in taxiing time and error characteristics. However, independent calculation may lead to cumulative errors, and its time cost is relatively high. In comparison, Strategy 2 has advance times of 105 seconds at 80% on-time rate, 133 seconds at 85%, and 176 seconds at 90%. Meanwhile, the delay times at 80%, 85%, and 90% are 120 seconds, 142 seconds, and 178 seconds respectively, all meeting the higher requirements of CTOT Type II tolerance (-3, +3 minutes). The advantage of this strategy is that it considers the entire error range, helping to reduce accumulated errors, and it is simple to implement, requiring no stage-by-stage calculations. However, if the taxiing time distribution varies significantly across stages, higher errors in specific stages may lead to unreasonable allocations, failing to effectively meet the on-time requirements of each stage. In summary, Strategy 1 provides stable time control under high on-time performance, but its time cost is relatively high. In contrast, Strategy 2, while meeting high on-time performance requirements, can more effectively control taxiing time and improve the efficiency of flight scheduling.

[0104] This invention divides the departure taxiing process into multiple stages based on the autonomous taxiing process and the taxiway. It uses the Extreme Gradient Boosting (XGBoost) model to predict the taxiing time for each stage, and then sums the results. This method overcomes the limitation of traditional models that treat the entire taxiing process as a single process, making the prediction more targeted and accurate. The performance of four machine learning models (random forest, support vector regression, extreme gradient boosting, and the proposed model) in predicting departure taxiing time was evaluated. The results show that the proposed model has a mean absolute error of 1.796 on the test set, which is better than random forest, support vector regression, and extreme gradient boosting. The mean squared error is significantly lower than other models, indicating smaller prediction error fluctuations. The coefficient of determination is 0.870, indicating its strong explanatory power for taxiing time variation trends. The cross-validation MAE is 1.277±0.016, showing superior stability. In terms of prediction accuracy, the proposed model achieves an accuracy of 96% within ±5 minutes, which is higher than other models.

[0105] Furthermore, this paper proposes a taxiing time slot control method based on chance constraints. By fitting the taxiing time prediction error, the taxiing time slots for each stage are calculated, enabling airports to achieve precise control of the taxiing process. This time slot control method provides airports with a more flexible time management strategy, ensuring on-time performance during peak periods and improving operational efficiency during off-peak periods. With the application of big data and artificial intelligence technologies, future taxiing control will rely more heavily on accurate taxiing time prediction and analysis.

[0106] In summary, the embodiments of this application provide a phased taxiing time prediction and taxiing control method and system:

[0107] This invention divides the departure taxiing phase into multiple taxiing stages, extracts arrival and departure features from multiple flight data points, and analyzes the taxiing time for each stage to construct a departure taxiing time prediction model for different stages. Furthermore, it mines the interactive features of departure taxiing by extracting these features, trains the corresponding departure taxiing time prediction model for each taxiing stage, calculates taxiing time slots based on multiple taxiing stages, optimizes the model, and uses the model's predictions of each taxiing stage to sum up the total taxiing time. This invention combines space requirements, idle time, and throughput to train the model to predict runway and taxiway taxiing times, improving prediction accuracy during peak periods and at different taxiing stages, thereby accelerating overall operational efficiency. This invention comprehensively considers the performance of various models at different taxiing stages, demonstrating a lower error level and stronger generalization ability, providing more reliable and stable prediction results. It has stronger applicability in actual operations and performs well in airport scenarios with high taxiing time uncertainty. The reasonable setting of taxiing time slots at each stage helps to cope with complex operating conditions and uncontrollable factors, ensuring a smooth taxiing process. This flexible taxiing time slot management method not only provides optimization direction for future airport operations but also offers the possibility of dynamic adjustments to taxiing in different seasons, time periods, and operating environments, ensuring operational robustness and flexibility. This invention exhibits higher accuracy and stronger generalization ability, providing more relevant features for each taxiing phase while maintaining prediction accuracy. It adapts to complex and ever-changing airport operating environments, providing reliable support for flight scheduling and operations management.

[0108] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting and controlling phased taxiing time, characterized in that, include: Flight data for multiple departure taxiing processes are collected. Each departure taxiing process is divided into multiple taxiing stages, including: from traction taxiing to autonomous taxiing, from autonomous taxiing to runway entry, and from runway entry to takeoff, each departure taxiing process is divided into three taxiing stages. Using the preprocessed flight data, extract multiple arrival and departure characteristics of each flight in each taxiing phase, as well as departure taxiing time; Interactive features are obtained by repeatedly selecting partial flight arrival and departure characteristics from each taxiing phase, including: obtaining the runway and taxiway usage frequency and space requirements based on the number of arrivals; recording the runway and taxiway idle time based on the number of arrivals and departures or the time interval between the most recent arrivals and departures; calculating the arrival and departure throughput based on the number of arrivals and departures, and obtaining a comparison result between the arrival and departure throughput and the peak hour throughput; and performing dummy variable encoding on the combination of parking stands and runways, mapping each combination to a binary feature. The multiple interaction features of each taxiing phase are preprocessed; using all the flight arrival and departure features of each taxiing phase, the preprocessed multiple interaction features, and the departure taxiing time, a departure taxiing time prediction model is trained. The departure taxiing time prediction model is optimized by calculating the actual taxiing time in each taxiing phase. The optimized departure taxiing time prediction model is used to obtain the predicted departure taxiing time for each stage of the departure taxiing process, including: For the taxiing phases from traction taxiing to autonomous taxiing and from autonomous taxiing to runway entry, the corresponding departure taxiing time prediction models are selected to obtain the taxiing time; for the taxiing phase from runway entry to takeoff, the statistical median time is used to obtain the taxiing time; the summation of the taxiing times of the three taxiing phases yields the predicted departure taxiing time for the entire departure taxiing process. Preprocessing the interactive features includes any one or more of the following: using label encoding for categorical variable features, employing target encoding methods for parking positions and runway features, and standardizing numerical features using standard scaling. The step of obtaining the actual taxiing time by calculating the taxiing time slots of each taxiing phase includes: The predicted taxiing time error of the departure taxiing time prediction model is calculated using the predicted and actual departure taxiing times for each taxiing stage during the departure taxiing process. The probability density function and cumulative distribution function of the predicted taxiing time error are obtained using the KDE method. The quantile of the cumulative distribution function is used as the upper bound, and the symmetric quantile of the cumulative distribution function is used as the lower bound to obtain the range of predicted taxiing time error. The actual taxiing time is calculated using the predicted taxiing time error for each taxiing stage, or the actual taxiing time is obtained by summing the predicted taxiing time error ranges for all taxiing stages during the departure taxiing process and then distributing them to each taxiing stage. The departure taxiing time prediction model adopts the extreme gradient boosting model. When training the departure taxiing time prediction model for the corresponding taxiing stage, it optimizes any one or more of the following model parameters: learning rate, maximum depth, sample sampling ratio, feature sampling ratio, and number of trees.

2. The method for predicting and controlling phased taxiing time as described in claim 1, characterized in that, Preprocessing the flight data includes: Select one or more of the following methods: de-icing flights during taxiing, filtering and removing outliers, and filling missing values ​​with the median.

3. A phased taxiing time prediction and taxiing control system, characterized in that, include: Data acquisition module: Collects flight data for multiple departure taxiing processes. Each departure taxiing process is divided into multiple taxiing stages, including: from traction taxiing to autonomous taxiing, from autonomous taxiing to runway entry, and from runway entry to takeoff, each departure taxiing process is divided into three taxiing stages. Feature extraction module: Extracts multiple arrival and departure features of flights in each taxiing phase, as well as departure taxiing time, using the preprocessed flight data; Feature interaction module: This module repeatedly selects partial flight arrival and departure features from each taxiing phase to obtain interactive features, including: obtaining the runway and taxiway usage frequency and space requirements based on the number of arrivals; recording runway and taxiway idle time based on the number of arrivals and departures or the time interval between the most recent arrivals and departures; calculating arrival and departure throughput based on the number of arrivals and departures, and obtaining a comparison between the arrival and departure throughput and peak hourly throughput; and performing dummy variable encoding on the combinations of parking stands and runways, mapping each combination to a binary feature. Model training module: preprocesses multiple interaction features of each taxiing phase; trains a departure taxiing time prediction model using all flight arrival and departure features of each taxiing phase, the preprocessed multiple interaction features, and the departure taxiing time; preprocessing the interaction features includes any one or more of the following: label encoding for categorical variable features, target encoding method for parking positions and runway features, and standard scaling for numerical features; the departure taxiing time prediction model adopts the extreme gradient boosting model, and optimizes any one or more of the following model parameters when training the departure taxiing time prediction model for the corresponding taxiing phase: learning rate, maximum depth, sample sampling ratio, feature sampling ratio, and number of trees; Model optimization module: Optimizes the departure taxiing time prediction model by calculating the actual taxiing time for each taxiing stage; the step of calculating the actual taxiing time for each taxiing stage includes: calculating the predicted taxiing time error of the departure taxiing time prediction model using the predicted departure taxiing time and actual departure taxiing time for each taxiing stage during departure taxiing; obtaining the probability density function and cumulative distribution function of the predicted taxiing time error using the KDE method; using the quantile of the cumulative distribution function as the upper bound and the symmetric quantile of the cumulative distribution function as the lower bound to obtain the predicted taxiing time error range; calculating the actual taxiing time using the predicted taxiing time error for each taxiing stage, or summing the predicted taxiing time error ranges for all taxiing stages during departure taxiing and distributing them to each taxiing stage to obtain the actual taxiing time; Taxiing segmentation module: The optimized departure taxiing time prediction model is used to obtain the predicted departure taxiing time for each taxiing stage during the departure taxiing process, including: the taxiing stages from the end of traction taxiing to the beginning of autonomous taxiing, and the taxiing stages from the beginning of autonomous taxiing to the beginning of runway entry, for which the corresponding departure taxiing time prediction model is selected to obtain the taxiing time; the taxiing stage from the beginning of runway entry to the beginning of takeoff, for which the median of the statistical time is used to obtain the taxiing time; the taxiing times of the three taxiing stages are summed to obtain the predicted departure taxiing time for the entire departure taxiing process.

4. An electronic device comprising a memory, a processor, and a computer program running on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the phased taxiing time prediction and taxiing control method as described in any one of claims 1 to 2.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps of the phased taxiing time prediction and taxiing control method as described in any one of claims 1 to 2.

Images