Method for estimating additional taxiing time of flight based on prediction of surface traffic situation characteristics

Abstract

The present invention belongs to the technical field of taxi-out time estimation for surface operation optimization, and specifically relates to a method for estimating the additional taxi time of flights based on the prediction of surface traffic situation characteristics. The method includes: calculating the unperturbed taxi-out time; preprocessing airport flight data to construct complex multi-runway surface traffic situation characteristics and flight basic information characteristics; constructing a traffic situation characteristic based on prediction; constructing a dataset of traffic situation and flight basic information based on prediction; constructing a NOA-XGBOOST combined model; and verifying the accuracy of the NOA-XGBOOST combined model. Multiple factors are introduced as input features, and the NOA model, XGBOOST model, and NOA-XGBOOST combined model are respectively constructed and compared with the XGBOOST, RF, and SVR models; and it is verified that the XGBOOST combined model has high estimation accuracy, the estimation errors MAE and RMSE are reduced, and the accuracies of ±3 min, ±5 min, and +5-10 min are improved. It is beneficial to improve the overall operation efficiency of the airport and reduce the congestion during ground taxiing.

Description

Technical Field

[0001] The present invention belongs to the technical field of taxi-out time estimation for airport surface operation management, and particularly relates to a method for estimating the additional taxiing time of flights based on the prediction of surface traffic situation characteristics. Background Art

[0002] With the continuous increase of flight delay problems, accurately estimating the taxi-out time of aircraft on the airport ground has become particularly important. Estimating the taxi-out time of flights not only helps improve the operational efficiency of airports and reduce ground congestion, but also optimizes flight scheduling, improves on-time performance, reduces fuel consumption, and reduces carbon emissions, promoting the development of green airports.

[0003] The prior art for estimating the taxi-out time first focuses on the overall taxi-out time. However, the taxi-out time is divided into two parts: the unperturbed taxi-out time and the additional taxi-out time. The unperturbed taxi-out time refers to the taxi-out time of a flight when some delay factors are not significant. The additional taxi-out time is the difference between the overall taxi-out time and the unperturbed taxi-out time, which is a key indicator for surface operation management and reflects the operational efficiency of the surface. When optimizing the taxi-out time, the additional taxi-out time is the focus of optimization. Secondly, in terms of feature input, it focuses on the overall traffic flow of the surface and ignores the traffic flow related to the micro-structure, such as the concourse traffic flow. And for the traffic situation characteristics of the surface, it is based on historical traffic flow information, which does not conform to the actual application situation. Finally, in terms of the estimation method, it only simply uses machine learning algorithms or combines them, and the research on optimizing the parameters of existing machine learning methods has not been effectively carried out. Summary of the Invention

[0004] The object of the present invention is to provide a method for estimating the additional taxiing time of flights based on the prediction of surface traffic situation characteristics.

[0005] To solve the above technical problems, the present invention provides a method for estimating the additional taxiing time of flights based on the prediction of surface traffic situation characteristics, including: calculating the unperturbed taxi-out time; preprocessing airport flight data to construct complex multi-runway surface traffic situation characteristics and flight basic information characteristics; constructing traffic situation characteristics based on prediction; constructing a dataset of traffic situation based on prediction and flight basic information; constructing a NOA-XGBOOST combined model; and verifying the accuracy of the NOA-XGBOOST combined model.

[0006] Advantages of the Present Invention

[0007] (1) Focus on the estimation of the additional taxi-out time instead of studying the taxi-out time as a whole;

[0008] (2) In terms of estimating the input factors of the additional taxi-out time, the influence of the corridor flow is considered, and three new indicators, namely the corridor departure flow, the corridor arrival flow, and the departure flow, are established, and the characteristics of the complex multi-runway surface traffic situation are established.

[0009] (3) In terms of characteristics, based on the calculation time nodes in the ACDM system, the traffic situation characteristics based on prediction are constructed and combined with the basic information characteristics of flights.

[0010] (4) In terms of the estimation model, the NOA model, the XGBOOST model, and the NOA-XGBOOST combined model are respectively constructed and compared with the XGBOOST, RF, and SVR models; and it is verified that the NOA-XGBOOST model has higher estimation accuracy than other models, with the MAE and RMSE values reduced, and the R2, ±3-minute accuracy, ±5-minute accuracy, and +5 - 10-minute accuracy improved.

[0011] (5) In terms of case verification, taking the flight operation data of Shanghai Pudong International Airport as an example, it is verified that the NOA-XGBOOST combined model has higher estimation accuracy than the single XGBOOST model, with the estimation error MAE and RMSE values reduced, and the ±3-minute accuracy, ±5-minute accuracy, and +5 - 10-minute accuracy improved.

[0012] (6) The method for estimating the additional taxi-out time of flights based on the prediction of surface traffic situation characteristics of the present invention is beneficial to improving the overall operation efficiency of the airport and reducing the congestion phenomenon during ground taxiing. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 It is a flowchart of the method for estimating the additional taxi-out time of flights based on the prediction of surface traffic situation characteristics in the embodiments of the present invention;

[0015] Figure 2 It is a flowchart of the calculation of the unperturbed taxi-out time in the embodiments of the present invention;

[0016] Figure 3 It is the construction process of the dataset of the method for estimating the additional taxi-out time in the embodiments of the present invention;

[0017] Figure 4Comparison chart of estimation results of different methods in the embodiments of the present invention;

[0018] Figure 5 Comparison chart of estimation errors before and after removing the structural related features in the embodiments of the present invention. Detailed implementation manners

[0019] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] As Figure 1 shown, this embodiment provides a method for estimating the additional taxiing time of flights based on the prediction of surface traffic situation characteristics, including: calculating the unperturbed taxi-out time; preprocessing the airport flight data to construct the complex multi-runway surface traffic situation characteristics and flight basic information characteristics; constructing the traffic situation characteristics based on prediction; constructing the traffic situation based on prediction and flight basic information data set; constructing the NOA-XGBOOST combined model; and verifying the accuracy of the NOA-XGBOOST combined model.

[0021] In this embodiment, calculating the unperturbed taxiing time includes: grouping the flights, screening the arrival and departure flow indicators by using the Pearson correlation coefficient, and establishing a linear regression model for calculation.

[0022] In an application scenario, specifically, as Figure 2 shown, first, 22 aircraft stand groups are divided, and the aircraft stand groups are paired with the departure runways one by one to form 88 groups of data. Then, the arrival and departure flow indicators that have a time interaction with the flight under study are screened, and the one with the largest Pearson correlation coefficient with the taxi-out time is the screened arrival and departure flow indicator. Finally, a regression model is established by using the screened indicators, so that the coefficient of the departure variable in the regression model is 1 and the coefficient of the arrival variable is 0, and the unperturbed taxiing time is calculated.

[0023] Subtract the calculated unperturbed taxi-out time from the overall taxi-out time of each flight to obtain the additional taxi-out time, and use this value as the basis for estimation and comparison.

[0024] In this embodiment, in the steps of preprocessing airport flight data and constructing complex multi-runway surface traffic situation features and flight basic information features, the complex multi-runway surface traffic situation features include departure flow, arrival flow, departure flow ratio, departure flow in the corridor, and arrival flow in the corridor. The flight basic information features include airline, aircraft type, restricted status, and time. The airline features include domestic airlines and foreign airlines; the aircraft type features include Category C flights, Category D flights, Category E flights, and Category F flights; the restricted status features include restricted flights and non-restricted flights; the time features include hours. Among them, the departure flow ratio, departure flow in the corridor, and arrival flow in the corridor are new features proposed in this embodiment.

[0025] The departure flow ratio refers to the ratio of the departure flow to the sum of the arrival flow and the departure flow. The departure flow in the corridor refers to the number of other departing flights that slide from the east (west) control area aircraft position group through the corridor TW3 (TW4) to the west (east) control area departure runway during flight taxiing. Correspondingly, the arrival flow in the corridor is the number of other arriving flights.

[0026] In this embodiment, the method for constructing traffic situation features based on prediction includes: using the CTOT, COBT, CLDT, and CIBT values in the ACDM system to predict surface traffic situation features, and these values are usually determined within 30 minutes or 1 hour before the aircraft takes off.

[0027] In this embodiment, the method for constructing a dataset of traffic situation based on prediction and flight basic information includes:

[0028] According to the established traffic situation features based on prediction and flight basic information features, relevant data is selected from the original departure and arrival data, and the original data is preprocessed. For missing values, since the ratio of the missing quantity to the total quantity is small, the corresponding data rows are directly deleted. For outliers, the standard deviation method is used, that is, the data with the departure taxiing time value exceeding its mean plus or minus three standard deviations is deleted. Finally, the preprocessed data is used for calculation to obtain a dataset of traffic situation based on prediction and flight basic information. The dataset construction process is as Figure 3 shown.

[0029] In this embodiment, the method for constructing an additional taxi-out time estimation model combined with NOA and XGBOOST includes: constructing a NOA model; constructing an XGBOOST model; constructing a NOA-XGBOOST combined model.

[0030] The constructed NOA model includes:

[0031] In the global search stage of the foraging and storage strategy stage, the update formula for the position is:

[0032]

[0033] where: is the new position of the i-th nutcracker in the current generation; is the j-th position of the i-th nutcracker in the current generation; U j and L j are vectors including the upper and lower bounds of the j-th dimension in the optimization problem; γ is a random number generated according to the levy row; is the optimal solution of the j-th dimension obtained so far; A, C, and B are three different indices randomly selected from the population to facilitate exploring high-quality food sources; τ1, τ2, r, r1 are random real numbers in the range of [0, 1]; is the mean value of the j-th dimension of all solutions in the current population at the t-th iteration; μ is a number randomly generated between 0 and 1 (τ3) based on the normal distribution (τ4) and levy-flight (τ5);

[0034] In the local exploitation stage of the foraging and storage strategy phase, the position update formula is:

[0035]

[0036] where: λ is a number generated according to the levy flight, τ3 is a random number between 0 and 1; l is a factor that linearly decreases from 1 to 0 for the NOA exploitation behavior;

[0037] In the global search stage of the cache search and recovery strategy, its position update formula is:

[0038]

[0039] where: is the first reference point of the current position / cache of the i-th nutcracker in the current iteration t; r1, r2, τ3, τ4, τ5, τ6 are random numbers between 0 and 1, and C is the index of the solution randomly selected from the population;

[0040] In the local exploitation stage of the cache search and recovery strategy, its position update formula is:

[0041]

[0042] where: is the first reference point of the current cache of the i-th nutcracker at iteration t, is the second reference point of the current cache of the i-th nutcracker at iteration t;

[0043] Adjustment should be made during all the above update processes, that is:

[0044]

[0045] The construction of the XGBoost model includes:

[0046] Initializing the base learner, that is

[0047]

[0048] where is the initial estimate of the i-th sample, y i is the actual value, and n is the number of samples;

[0049] In each round of iteration m, calculate the residual that is, the difference between the true value and the current estimated value:

[0050]

[0051] where is the estimated value of the i-th sample in the (m - 1)-th round of iteration;

[0052] Train the base learner h m (x) to fit the residual Update the base learner:

[0053]

[0054] where η is the learning rate, which controls the contribution of each base learner to the final model;

[0055] Add the new base learner to the model and update the model's estimate:

[0056]

[0057] where M is the total number of iterations.

[0058] The construction of the NOA-XGBoost combined model includes:

[0059] Set the objective function of the NOA model to the RMSE value of the estimation result, set the optimization variables to the hyperparameters of XGBoost, input the optimized hyperparameter values into the XGBoost model for estimation, and finally obtain the estimated value of the additional taxi-out time.

[0060] The steps to execute the NOA-XGBoost combined model include:

[0061] Step 1: Use 80% of the data in the traffic situation prediction-based and flight basic information dataset as training samples and 20% of the data as test samples.

[0062] Step 2: Set the optimization range of XGBOOST hyperparameters, and set the objective function of NOA to the RMSE value.

[0063] Step 3: Initialize the parameters of the NOA and XGBOOST models on the training set, and use the NOA algorithm to iteratively update to obtain the optimal XGBOOST hyperparameters.

[0064] Step 4: Use the trained model and the optimal hyperparameters to estimate on the test set to obtain the estimated value of the additional taxi-out time.

[0065] In this embodiment, the maximum number of iterations of the model is set to 100, the population size is set to 50, and the hyperparameters to be adjusted are the number of base learners, maximum depth, learning rate, minimum split loss, subsample ratio, column sampling ratio, L1 regularization coefficient, and L2 regularization coefficient.

[0066] In this embodiment, verifying the accuracy of the NOA-XGBOOST combined model includes: using the mean absolute error (MAE), root mean square error (RMSE), ±3-minute accuracy, ±5-minute accuracy, and +5 - 10-minute accuracy as evaluation indicators; comparing the evaluation indicators estimated by the NOA-XGBOOST combined model with those of different models.

[0067] Optionally, the different models include: XGBOOST, RF, SVR.

[0068] In this embodiment, optionally, using the visualization method and actual operation data to verify the accuracy and rationality of the proposed additional taxi-out time estimation model based on the NOA-XGBOOST model, the specific implementation process includes:

[0069] Using the mean absolute error (MAE), root mean square error (RMSE), ±3-minute accuracy, ±5-minute accuracy, and +5 - 10-minute accuracy as evaluation indicators, and calculating them respectively using the following formulas:

[0070]

[0071] where, y i is the true value, is the estimated value, is the average value of the true values.

[0072] Select 80% of the data in the flight operation dataset as the training sample and 20% of the data as the test sample. Taking Shanghai Pudong International Airport as an example, the data of each error term estimated by different models is shown in Table 1. The comparison of the estimation results of the models is as Figure 4As shown in the figure, the additional taxi-out times of the first 200 flights are presented. It can be found that in terms of estimating the additional taxi-out times, the estimation effect of the NOA-XGBOOST combined model is significantly better than that of the XGBOOST model, SVR model, and RF model, improving the estimation accuracy.

[0073] Table 1 Estimation Errors of NOA-XGBOOST, XGBOOST, RF, and SVR Models

[0074]

[0075] Compared with the XGBOOST, RF, and SVR models, the NOA-XGBOOST model has lower values in terms of the MAE and RMSE metrics, while it has higher values in terms of the accuracies of ±3 min, ±5 min, and +5 - 10 min.

[0076] The results of the ablation experiment of features using the NOA-XGBOOST algorithm are shown in Table 2.

[0077] Table 2 Feature Ablation Experiment of the NOA-XGBOOST Algorithm

[0078]

[0079] It can be found that the conventional traffic features have the greatest impact on the estimation results, followed by the three newly proposed features. After adding the proposed features, the model has lower values in terms of the MAE and RMSE metrics, while it has higher values in terms of the accuracies of ±3 min, ±5 min, and +5 - 10 min. The error comparison chart of the first 200 flights before and after adding the structure-related features is as Figure 5 shown.

[0080] Taking the ideal embodiments of the present invention described above as an inspiration, through the above description, relevant staff can completely make various changes and modifications within the scope not deviating from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for estimating flight extra taxiing time based on the prediction of traffic situation characteristics, characterized in that: include: Calculate the undisturbed slide-out time; Pre-process airport flight data to construct complex multi-runway traffic situation characteristics and flight basic information characteristics; Construct prediction-based traffic situation features; Construct a data set of traffic situation and basic flight information based on prediction; Construct NOA-XGBOOST combination model; Verify the accuracy of the NOA-XGBOOST combined model; in The construction of complex multi-runway traffic situation characteristics and flight basic information characteristics includes: The traffic situation characteristics of the complex multi-runway scene include departure flow, arrival flow, departure flow ratio, corridor departure flow and corridor arrival flow; The basic flight information features include airline, aircraft type, restricted status, and time; The airline characteristics include domestic airlines and foreign airlines; The aircraft type characteristics include Class C flights, Class D flights, Class E flights, and Class F flights; The restricted status characteristics include restricted flights and non-restricted flights; The time characteristics include hours; The construction of the traffic situation features based on prediction includes: Use the CTOT, COBT, CLDT and CIBT values ​​at the calculation time point in the ACDM system to predict the characteristics of the surface traffic situation. These values ​​are usually determined 30 minutes or 1 hour before the aircraft takes off. The method for constructing a traffic situation and flight basic information data set based on prediction includes: According to the traffic situation characteristics based on prediction and the basic information characteristics of flights, relevant data are selected from the original arrival and departure data, the original data are preprocessed, and the preprocessed data are used for calculation to obtain the traffic situation and basic information data set based on prediction; The method for constructing the NOA-XGBOOST combination model comprises: constructing a NOA model, constructing an XGBOOST model and constructing a NOA-XGBOOST combination model; wherein The construction of the NOA model includes two stages: foraging and storage strategy and cache search and recovery strategy, each of which includes global search and local development; In the global search phase of the foraging and storage strategy phase, the position update formula is: Where: is the new position of the i-th nutcracker in the current generation; is the jth position of the i-th nutcracker in the current generation; U j and L j is a vector, including the upper and lower bounds of the j-th dimension in the optimization problem; γ is a random number generated according to the levy row; is the optimal solution of the jth dimension obtained so far; A, C and B are three different indicators randomly selected from the population to facilitate the exploration of high-quality food sources; τ1, τ2, r, r1 are random real numbers in the range of [0, 1]; is the j-th dimension mean of all solutions of the current population in the t-th iteration; μ is a number randomly generated between 0 and 1 (τ3) based on normal distribution (τ4), levy-flight (τ5); In the local development phase of the foraging and storage strategy phase, the update formula for the position is: Where: λ is a number generated based on levy flight, τ3 is a random number between 0 and 1; l is a factor that linearly decreases from 1 to 0 in NOA development behavior; In the global search phase of the cache search and recovery strategy, the position update formula is: Where: is the first reference point of the current position / cache of the ith nutcracker in the current iteration t; r1, r2, τ3, τ4, τ5, τ6 are random numbers between 0 and 1, and C is the index of a randomly selected solution from the population; During the local development phase of the cache search and recovery strategy, the position update formula is: Where: is the first reference point currently cached by the ith Nutcracker at iteration t, is the second reference point currently cached by the i-th Nutcracker at iteration t; All of the above updates require adjustments, namely: The construction of the XGBOOST model includes: Initialize the base learner, that is In the formula is the initial estimate of the ith sample, y i is the actual value, n is the number of samples; In each round of iteration m, the residual is calculated That is, the difference between the true value and the current value: In the formula, is the estimated value of the i-th sample in the m-1-th iteration; Training base learner h m (x) Fitting residual Update the base learner: Where η is the learning rate, which controls the contribution of each base learner to the final model; Add the new base learner to the model and update the model's estimate: Where M is the total number of iterations; The constructed NOA-XGBOOST combined model includes: The objective function of the NOA model is set to the RMSE value of the estimation result, the optimization variable is set to the hyperparameter of XGBOOST, and the optimized hyperparameter value is input into the XGBOOST model for estimation, and finally the estimated value of the additional slip time is obtained.

2. The method for estimating the additional taxiing time of a flight according to claim 1, characterized in that: The calculation of the undisturbed taxi-out time includes: grouping flights, screening inbound and outbound flow indicators using the Pearson correlation coefficient, and establishing a linear regression model for calculation.

3. The method for estimating the additional taxiing time of a flight according to claim 1, characterized in that: The steps to implement the NOA-XGBOOST combined model include: Step 1: Use 80% of the data in the dataset as training samples and 20% of the data as test samples; Step 2: Set the optimization range of XGBOOST hyperparameters and set the objective function of NOA to the RMSE value; Step 3: Initialize the NOA and XGBOOST model parameters on the training set, and iteratively update them using the NOA algorithm to obtain the optimal XGBOOST hyperparameters; Step 4: Use the trained model and optimal hyperparameters to estimate the additional slide time on the test set.

Images