A method for predicting flight runway taxiing time at civil aviation airports

By building a variety of machine learning models and considering a variety of influencing factors, the problem of poor prediction of flight runway taxiing time in the existing technology has been solved, and more accurate and widely applicable prediction has been achieved, improving airport operation efficiency and service quality.

CN114239970BActive Publication Date: 2024-11-12NANJING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111563857.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-20
Publication Date
2024-11-12
Estimated Expiration
2041-12-20

AI Technical Summary

Technical Problem

When predicting the taxiing time of civil aviation airport flights, the prior art ignores key influencing factors, resulting in poor prediction results, and a single machine learning model is difficult to adapt to the scenarios and needs of different airports, resulting in flight delays and increased fuel costs.

Method used

By acquiring and preprocessing airport historical flight data, a variety of machine learning models are built, and more influencing factors are considered, such as taxi paths, weather, and flight locations, and the most suitable prediction model is selected to improve prediction accuracy.

Benefits of technology

It has achieved more accurate and widely applicable flight runway taxiing time prediction, improved flight operation efficiency and service quality, and provided a more reliable decision-making basis for airport operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114239970B_ABST
    Figure CN114239970B_ABST
Patent Text Reader

Abstract

The present invention is a flight runway taxi time prediction method for civil aviation airports. First, the airport historical flight data is obtained and preprocessed; then the airport historical flight data is divided into two categories: arrival and departure, the taxi time of each flight arrival and departure is counted and the abnormal taxi time data is deleted; the key feature data of each flight is constructed, all feature data constitute a sample data set and are divided into training samples and test samples; then: the training samples are used as input, multiple machine learning methods are selected for training to obtain multiple flight runway taxi time prediction models; finally: the test samples are used to evaluate the performance of each prediction model, and the model with the best prediction effect and meeting the actual application requirements is selected. The present invention analyzes various factors that affect the runway taxi time of civil aviation airport flights, refines and constructs key feature indicators, and constructs a suitable prediction model, which is conducive to improving the airport flight operation efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of civil aviation information, and in particular to a method for predicting flight runway taxiing time at a civil aviation airport. Background Art

[0002] With the rapid development of my country's civil aviation industry, the number of take-off and landing flights is increasing, which has also led to a continuous decline in flight punctuality. The decline in punctuality is not only due to the limitation of airspace capacity, but also has an inseparable relationship with airport operations. Many civil aviation airports have complex layouts and flight runway taxiing time varies greatly, which directly affects the airport's operational efficiency.

[0003] At present, the methods used by domestic airports to predict flight runway taxiing time have the following defects:

[0004] (1) Using average taxi time to predict aircraft runway taxi time ignores key influencing factors such as the number of aircraft taxiing on the runway and the runway taxi path. The prediction effect is poor and it is easy to cause flight delays and increased fuel costs.

[0005] (2) Most prediction methods using machine learning models do not consider the influencing factors comprehensively, have too little feature data, and use a single prediction model, where only the model parameters change in different situations. For different airports, the correlation of feature data is different, and a single model is difficult to adapt to different scenarios and needs. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a flight runway taxiing time prediction method for civil aviation airports. The prediction method analyzes various factors affecting the runway taxiing time of flights at civil aviation airports, refines and constructs key characteristic indicators, and constructs a suitable prediction model, which is conducive to improving the airport flight operation efficiency.

[0007] In order to achieve the above object, the present invention is achieved through the following technical solutions:

[0008] The present invention is a method for predicting flight runway taxiing time at a civil aviation airport, and the method specifically comprises the following steps:

[0009] Step 1: Obtain airport historical flight data and preprocess it;

[0010] Step 2: Split the airport's historical flight data into two categories: arrival and departure, count the taxiing time of each flight's arrival and departure, and delete abnormal taxiing time data;

[0011] Step 3: Construct the key feature data of each flight, form a sample data set with all the feature data and divide it into training samples and test samples;

[0012] Step 4: Using the training samples as input, select multiple machine learning methods for training to obtain multiple flight runway taxi time prediction models;

[0013] Step 5: Use the test samples to evaluate the performance of each prediction model and select the model with the best prediction effect and that meets the actual application requirements.

[0014] A further improvement of the present invention is that the airport historical flight data in step 1 mainly includes airline, mission category, arrival runway, departure runway, arrival aircraft position, departure aircraft position, actual arrival time, actual departure time, braking time, brake release time and weather.

[0015] A further improvement of the present invention is that in step 1, the airport historical flight data is preprocessed, and the specific process is:

[0016] Step 1-1: Clean up missing data. In the acquired airport historical flight data, if any of the flight data's actual arrival time, actual departure time, braking time, or brake release time is missing, the flight data needs to be removed;

[0017] Step 1-2: Process abnormal data, mainly including flight data with brake release time later than the actual departure time and brake time earlier than the actual arrival time.

[0018] A further improvement of the present invention is that in step 2, the data is divided into two categories: arrival and departure, wherein the flight arrival data mainly includes the flight's airline, mission category, arrival runway, arrival aircraft position, actual arrival time, braking time, weather, actual departure time, brake release time and departure runway; the flight departure data mainly includes the flight's airline, mission category, departure runway, departure aircraft position, actual departure time, brake release time, weather, actual arrival time, braking time and arrival runway.

[0019] A further improvement of the present invention is that in step 2, the taxiing time of each flight arriving and departing is counted, and the specific calculation formula is as follows:

[0020] ST in =L in -RT in

[0021] ST out =L out -RT out

[0022] Among them, ST in Indicates the taxiing time of the flight, ST out Represents the taxiing time of the flight departure, L in Represents the flight's approach braking time, RTin Represents the actual arrival time of the flight, L out Represents the actual departure time of the flight, RT out Indicates the flight's departure brake release time.

[0023] A further improvement of the present invention is that in step 2, abnormal taxi time data is deleted, mainly including flight data with departure taxi time exceeding 60 minutes or less than 3 minutes and flight data with arrival taxi time exceeding 30 minutes or less than 1 minute, and the time can be changed according to the actual situation of the airport.

[0024] A further improvement of the present invention is that in step 3, the key feature data of each flight is constructed by:

[0025] Step 3-1: Calculate the instantaneous surface flow of flights arriving and departing;

[0026] Step 3-2: Calculate the surface cumulative flow of flight arrivals and departures;

[0027] Step 3-3: Calculate the number of aircraft queued at the runway gate for the departure flight;

[0028] Step 3-4: Digitize the characteristic data corresponding to each flight.

[0029] A further improvement of the present invention is that: the instantaneous flow of the flight arrival and departure is calculated, wherein the instantaneous flow of the flight arrival is divided into two parts: the instantaneous flow of the departure scene and the instantaneous flow of the arrival scene, D_SIFI is defined as the instantaneous flow of the departure scene, A_SIFI is the instantaneous flow of the arrival scene, T on is the landing time of the flight, T out The launch time for the flight departure.

[0030] SIFI (instantaneous flow rate) when the flight arrives at the airport in The calculation formula is as follows:

[0031] SIFI in =D_SIFI in +A_SIFI in

[0032] Where D_SIFI in Point to T on Previously launched, in T on The number of flights departing after that, A_SIFI in Point to T on Before landing, in T on The number of flights that will be available later.

[0033] Flight departure instantaneous flow D_SIFI out The calculation formula is as follows:

[0034] SIFI out =D_SIFI out +A_SIFI out

[0035] Where D_SIFI out Point to T out Previously launched, in T out The number of flights departing after that, A_SIFI out Point to T out Before landing, in T out The number of flights that will be available later.

[0036] A further improvement of the present invention is that: the cumulative flow of arrival and departure of the flight is calculated, wherein the cumulative flow of the surface is divided into the cumulative flow of the departure surface and the cumulative flow of the arrival surface, and D_SCFI is defined to represent the cumulative flow of the departure surface, A_SCFI is defined to represent the cumulative flow of the arrival surface, T on and T in T is the landing time of the flight and the arrival time of the parking space, respectively. out and T off They are the push-out time and take-off time of the flight departure respectively.

[0037] SCFI (surface cumulative flow) when the flight arrives in The calculation process is as follows:

[0038] D_SCFI in =FT out1 +BT out1 ①

[0039] A_SIFI in =FT in1 +BT in1 ②

[0040] SCFI in =D_SCFI in +A_SCFI in ③

[0041] Among them, FT out1 Point to T on Previously launched and in T on The number of flights departing after that, BT out1 Point to T on After that and in T in The number of flights launched before, FT in1 Point to T on Landed before and in T on The number of flights that will be available later, BT in1 Point to T on After that and in Tin The number of flights that have landed previously.

[0042] The calculation process of the cumulative flow of the flight at the departure time is as follows:

[0043] D_SCFI out =FT out2 +BT out2 ①

[0044] A_SIFI out =FT in2 +BT in2 ②

[0045] SCFI out =D_SCFI out +A_SCFI out ③

[0046] Among them, FT out2 Point to T out Previously launched and in T out The number of flights departing after that, BT out2 Point to T out After that and in T off The number of flights launched before, FT in2 Point to T out Landed before and in T out The number of flights that will be available later, BT in2 Point to T out After that and in T off The number of flights that have landed previously.

[0047] A further improvement of the present invention is that: the number of aircraft queuing at the runway gate for the flight departure is calculated, wherein the number of aircraft queuing at the runway gate is divided into the number of runway departure aircraft and the number of runway arrival aircraft. Define D_AQLI to represent the number of runway departure aircraft, A_AQLI to represent the number of runway arrival aircraft, T out and T off They are the push-out time and take-off time of the flight departure respectively.

[0048] The calculation formula for the number of aircraft queued at the runway gate when the flight departs is as follows:

[0049] AQLI=D_AQLI+A_AQLI

[0050] Where D_AQLI refers to the out Before or after the launch, in T out and T off The number of flights taking off between T out After that, T off The number of flights that have landed previously.

[0051] A further improvement of the present invention is that the specific process of digitizing the characteristic data corresponding to each flight is as follows:

[0052] Step 3-4-1: Classify the weather into three categories: almost no effect on the taxiing time, some effect on the taxiing time, and great effect on the taxiing time, represented by natural numbers 1 to 3.

[0053] Step 3-4-2: Divide the time periods. The peak hours for flights are from 6 to 9 a.m. every day. After 9 a.m., the number of flights arriving and departing gradually reaches a balance. It is best to divide the time periods by natural hours, represented by natural numbers from 0 to 23.

[0054] Step 3-4-3: Digitize the parking positions, airlines, runways and aircraft types. Different parking positions, airlines, runways and aircraft types are represented by different numbers. Whether to cross the runway needs to be determined according to the actual situation of different airports, and the numbers 0 and 1 represent crossing the runway and not crossing the runway respectively.

[0055] A further improvement of the present invention is that in step 3, all feature data are formed into a sample data set, wherein the arrival feature data mainly include the instantaneous flow rate at the scene, the cumulative flow rate at the scene, the parking space, the weather, the time period, the aircraft type, whether to cross the runway, the airline and the runway number, and the departure feature data mainly include the instantaneous flow rate at the scene, the cumulative flow rate at the scene, the parking space, the weather, the time period, the aircraft type, the airline, the number of aircraft queuing at the runway gate and the runway number.

[0056] A further improvement of the present invention is that in step 3, the sample data set is divided into training samples and test samples, wherein the optimal division ratio is training sample size: test sample size = 7:3, which can be appropriately changed according to the model effect after training.

[0057] A further improvement of the present invention is that in step 4, multiple machine learning methods are selected for training to obtain multiple flight runway taxi time prediction models, and the specific process is as follows:

[0058] Step 4-1: Select M1, M2, M3...M n There are N types of machine learning methods, mainly including random forest, boosted tree, linear regression, decision tree, naive Bayes, etc.

[0059] Step 4-2: Train these N types of machine learning methods using the training data set as input to obtain multiple flight runway taxi time prediction models.

[0060] A further improvement of the present invention is that in step 5, the performance of each prediction model is evaluated using the test sample, and the model with the best prediction effect and meeting the actual application requirements is selected. The specific process is as follows:

[0061] Step 5-1: Use mean square error to evaluate each prediction model. The specific formula for calculating mean square error is as follows:

[0062]

[0063] Where n is the number of sample data, y i is the actual taxiing time of the flight. The runway taxi time of the flight predicted by the model;

[0064] Step 5-2: Select the prediction model with the smallest MSE_Pre value and that meets the actual application requirements and apply it to the actual scenario.

[0065] The beneficial effects of the present invention are:

[0066] The present invention has comprehensive considerations, accurate predictions, and wide applicability. Compared with existing technical methods, it takes into account more and more important factors that affect the runway taxiing time of flights and uses a variety of machine learning models to predict the taxiing time. It can be widely applied to civil aviation airports, improve the predictability of flight surface operations, and provide a decision-making basis for surface optimization scheduling such as aircraft stand support, departure sorting, and push-out time control, thereby improving surface operation efficiency and service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart of the present invention.

[0068] Figure 2 Schematic diagram for calculating surface instantaneous flow, surface cumulative flow and the number of aircraft queuing at the runway gate. DETAILED DESCRIPTION

[0069] The following will disclose the embodiments of the present invention with drawings. For the purpose of clear description, many practical details will be described together in the following description. However, it should be understood that these practical details should not be used to limit the present invention. That is to say, in some embodiments of the present invention, these practical details are not necessary.

[0070] like Figure 1As shown, the present invention is a flight runway taxi time prediction method for civil aviation airports, and the method mainly includes four stages: data preprocessing, sample data generation, prediction model training, and prediction model evaluation and determination. In the data preprocessing stage, the historical operation data of airport flights is obtained and cleaned; in the sample data generation stage, characteristic data samples of incoming and outgoing flights are constructed respectively and divided into training samples and test samples; in the prediction model training stage, a variety of machine learning methods are selected for training to obtain multiple runway taxi time prediction models; in the prediction model evaluation and determination stage, the prediction model performance is evaluated using test samples, and the model with the best prediction effect and meeting the actual application needs is selected. Specifically: the prediction method includes the following steps:

[0071] Step 1: Obtain airport historical flight data and preprocess it.

[0072] The airport's historical flight data mainly includes airline, mission category, arrival runway, departure runway, arrival parking position, departure parking position, actual arrival time, actual departure time, braking time, brake release time and weather. The specific content is shown in Table 1:

[0073] Table 1 Airport historical flight data type and description

[0074]

[0075]

[0076] The basic flight data is stored in a table. At the same time, based on the weather information data provided by the official website of the provincial meteorological bureau, the corresponding weather conditions for each flight arriving and departing are recorded.

[0077] The final exported flight basic information is shown in Table 2.

[0078] Table 2 Basic information of some imported flights

[0079] ID …… Airlines Task Category Actual arrival time Actual departure time 1 …… CZ airliner 2019 / 1 / 16 22:15 2019 / 1 / 17 0:26 2 …… CZ airliner 2019 / 1 / 16 19:40 2019 / 1 / 17 0:36 3 …… CZ airliner 2019 / 1 / 16 23:59 2019 / 1 / 18 9:29 4 …… CZ airliner 2019 / 1 / 16 15:53 2019 / 1 / 17 11:44 5 …… PN airliner 2019 / 1 / 16 22:24 2019 / 1 / 16 23:58 6 …… MS airliner 2019 / 1 / 16 16:58 2019 / 1 / 16 23:50

[0080] Then, the missing and abnormal data are processed. For any flight, if any of the data of actual arrival time, actual departure time, braking time or brake release time is missing, it needs to be eliminated. At the same time, the flight data whose brake release time is later than the actual departure time or whose braking time is earlier than the actual arrival time also needs to be eliminated. For example, as shown in Table 3, the data of flights 2 and 3 are missing key information, and the braking time of flight 4 is earlier than the actual arrival time, so it needs to be eliminated.

[0081] Table 3 Schematic diagram of missing and abnormal data

[0082] ID …… Actual arrival time Actual departure time Braking time Brake release time 1 …… 2019 / 7 / 1 23:17 2019 / 7 / 2 8:54 2019 / 7 / 1 23:25 2019 / 7 / 2 8:40 2 …… 2019 / 7 / 2 00:07 NULL 2019 / 7 / 2 0:15 2019 / 7 / 2 6:29 3 …… 2019 / 7 / 1 22:50 NULL 2019 / 7 / 1 23:14 NULL 4 …… 2019 / 7 / 2 00:21 2019 / 7 / 3 0:45 2019 / 7 / 2 0:17 2019 / 7 / 3 0:25

[0083] Step 2: Split the airport's historical flight data into two categories: arrival and departure; count the taxi time of each flight's arrival and departure and delete abnormal taxi time data; construct key feature data for each flight; construct a sample data set with all feature data and divide it into training samples and test samples.

[0084] The airport's historical flight data is divided into two categories: arrival and departure. The arrival data mainly includes the flight's airline, mission type, arrival runway, arrival aircraft position, actual arrival time, braking time, weather, actual departure time, brake release time, and departure runway; the departure data mainly includes the flight's airline, mission type, departure runway, departure aircraft position, actual departure time, brake release time, weather, actual arrival time, braking time, and arrival runway.

[0085] Then, the taxi time of each flight arriving and departing is counted and the abnormal taxi time data is deleted. The specific calculation formula is as follows:

[0086] ST in =L in -RT in

[0087] ST out =L out -RT out

[0088] Among them, ST in Indicates the taxiing time of the flight, ST out Represents the taxiing time of the flight departure, L in Represents the flight's approach braking time, RT in Represents the actual arrival time of the flight, L out Represents the actual departure time of the flight, RT out Indicates the flight's departure brake release time.

[0089] Some calculation results are shown in Table 4.

[0090] Table 4 Schematic diagram of flight taxiing time calculation results

[0091]

[0092] Then, the abnormal taxiing time data is processed. The taxiing time for departure exceeding 60 minutes or less than 3 minutes can be eliminated; the taxiing time for arrival exceeding 30 minutes or less than 1 minute can be eliminated. Table 5 shows some of the data to be eliminated.

[0093] Table 5 Abnormal taxiing time of flights

[0094]

[0095] Step 3: Then calculate the instantaneous flow of flight departure and arrival, the cumulative flow of the surface and the number of aircraft queued at the runway gate when the flight departs, and digitize the characteristic data corresponding to each flight.

[0096] Define D_SIFI as the instantaneous flow rate of departure, A_SIFI as the instantaneous flow rate of arrival, T on is the landing time of the flight, T out The launch time for the flight departure.

[0097] SIFI (instantaneous flow rate) when the flight arrives at the airport in The calculation formula is as follows:

[0098] SIFI in =D_SIFI in +A_SIFI in

[0099] Where D_SIFI in Point to T on Previously launched, in T on The number of flights departing after that, A_SIFI in Point to T on Before landing, in T on The number of flights that will be available later.

[0100] Flight departure instantaneous flow D_SIFI out The calculation formula is as follows:

[0101] SIFI out =D_SIFI out +A_SIFI out

[0102] Where D_SIFI out Point to T out Previously launched, in T out The number of flights departing after that, A_SIFI out Point to T out Before landing, in T out The number of flights that will be available later.

[0103] The surface cumulative flow is divided into the departure surface cumulative flow and the arrival surface cumulative flow. D_SCFI is defined to represent the departure surface cumulative flow, A_SCFI is defined to represent the arrival surface cumulative flow, T on and T in T is the landing time of the flight and the arrival time of the parking space, respectively. out and T off They are the push-out time and take-off time of the flight departure respectively.

[0104] SCFI (surface cumulative flow) when the flight arrives in The calculation process is as follows:

[0105] D_SCFI in =FT out1 +BT out1 ①

[0106] A_SIFI in =FT in1 +BT in1 ②

[0107] SCFI in =D_SCFI in +A_SCFI in ③

[0108] Among them, FT out1 Point to T on Previously launched and in T on The number of flights departing after that, BT out1 Point to T on After that and in T in The number of flights launched before, FT in1 Point to T on Landed before and in T on The number of flights that will be available later, BT in1 Point to T on After that and in T in The number of flights that have landed previously.

[0109] The calculation process of the cumulative flow of the flight at the departure time is as follows:

[0110] D_SCFI out =FT out2 +BT out2 ①

[0111] A_SIFI out =FT in2 +BT in2 ②

[0112] SCFI out =D_SCFI out +A_SCFI out ③

[0113] Among them, FT out2 Point to T out Previously launched and in T out The number of flights departing after that, BT out2 Point to T out After that and in T off The number of flights launched before, FT in2 Point to Tout Landed before and in T out The number of flights that will be available later, BT in2 Point to T out After that and in T off The number of flights that have landed previously.

[0114] The number of aircraft queuing at the runway gate is divided into the number of runway departure aircraft and the number of runway arrival aircraft. Define D_AQLI to represent the number of runway departure aircraft, A_AQLI to represent the number of runway arrival aircraft, T out and T off They are the push-out time and take-off time of the flight departure respectively.

[0115] The calculation formula for the number of aircraft queued at the runway gate when the flight departs is as follows:

[0116] AQLI=D_AQLI+A_AQLI

[0117] Where D_AQLI refers to the out Before or after the launch, in T out and T off The number of flights taking off between T out Afterwards, T off The number of flights that have landed previously.

[0118] Take the flights in a certain period of time in the embodiment as an example. Figure 2 shown. Figure 2 d1 to d4 represent departing aircraft, the starting point of the black line represents the departure aircraft being pushed out from the parking position, and the end point represents the arrival at the runway for takeoff; a1 to a4 represent arriving aircraft, the starting point of the yellow line represents the arrival aircraft landing on the runway, and the end point represents the taxiing to the parking position, d0 and a0 represent the aircraft under study, the green line represents the process of the aircraft being pushed out from the parking position to taxiing on the runway until taking off, and the red line represents the process of the aircraft being braked and taxiing from landing on the runway to the parking position. out is the launch time of d0, t off is the take-off time of d0, t on is the time a0 enters the runway, t in It is the time from a0 to the parking stand.

[0119] The formula can be used to calculate the instantaneous flow rate at the surface. The calculation results of the instantaneous flow rate at the surface when aircraft a0 arrives are shown in Table 6. The surface cumulative flow rate, the settlement results of the surface cumulative flow rate when aircraft a0 arrives are shown in Table 7. The number of aircraft queuing at the runway gate, the settlement results of the number of aircraft queuing at the runway gate when aircraft d0 departs are shown in Table 8.

[0120] Table 6 Schematic diagram of scene instantaneous flow calculation and processing

[0121]

[0122]

[0123] Table 7 Schematic diagram of scene cumulative flow calculation and processing

[0124]

[0125] Table 8 Schematic diagram of calculation and processing of the number of aircraft queuing at the runway gate

[0126]

[0127] Then the characteristic data corresponding to each flight is digitized. The process is:

[0128] a) Classify the weather conditions into three categories: almost no impact on taxiing time, some impact on taxiing time, and great impact on taxiing time, represented by natural numbers 1 to 3.

[0129] b) Divide the time periods. The peak hours for flights are from 6 to 9 a.m. every day. After 9 a.m., the number of flights arriving and departing gradually reaches a balance. It is best to divide the time periods by natural hours, represented by natural numbers from 0 to 23.

[0130] c) The parking stands, airlines, runways and aircraft types are digitized. Different parking stands, airlines, runways and aircraft types are represented by different numbers.

[0131] Taking the embodiment as an example, it is explained that departure and arrival respectively contain 9 major feature data. For each feature data, if the representation form is non-digital, it needs to be digitized. For example, different aircraft models are represented by different numbers, different parking positions are also represented by different numbers, etc. Table 9 lists a set of arrival and departure feature data of a certain flight.

[0132] Table 9 Schematic diagram of arrival and departure characteristic data

[0133]

[0134] Finally, all feature data constitute a sample data set and divide it into training samples and test samples, with the division ratio of training samples: test samples = 7:3.

[0135] Step 4: Using training samples as input, select multiple machine learning methods for training to obtain multiple flight runway taxi time prediction models.

[0136] In the embodiment, five machine learning methods including random forest, linear regression, gradient descent boosted tree, support vector machine and BP neural network are selected and the machine learning methods are trained with training samples as input to obtain multiple flight runway taxi time prediction models.

[0137] Step 5: Use the test samples to evaluate the performance of each prediction model and select the model with the best prediction effect and that meets the actual application requirements.

[0138] MSE_Pre was calculated to evaluate each prediction model. The specific results are shown in Table 10:

[0139] type Machine Learning Methods MSE_Pre Category A Random Forest 31992.923 Category B Linear Regression 23340.303 Category C Gradient Descent Boosted Trees 28569.429 Category D Support Vector Machine 23165.275 Category E BP Neural Network 28603.988

[0140] According to the results in Table 10, the D-type support vector machine method can be selected. Finally, the trained model can be combined with the airport situation to predict the runway taxiing time of flight departure and arrival.

[0141] The flight runway taxiing time prediction method of the present invention improves the predictability of flight field operation, provides a decision-making basis for field optimization scheduling such as aircraft stand guarantee, departure sorting, and push-out time control, and improves field operation efficiency and service quality.

[0142] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting flight runway taxiing time at civil aviation airports. Features: The flight runway taxiing time prediction method comprises the following steps: Step 1: Obtain airport historical flight data and preprocess it; Step 2: Split the airport's historical flight data into two categories: arrival and departure, count the taxiing time of each flight's arrival and departure, and delete abnormal taxiing time data; Step 3: Construct the key feature data of each flight, form a sample data set with all the feature data and divide it into training samples and test samples; Step 4: Using the training samples as input, select multiple machine learning methods for training to obtain multiple flight runway taxi time prediction models; Step 5: Use the test samples to evaluate the performance of each prediction model and select the model with the best prediction effect and that meets the actual application requirements. The process of constructing the key feature data of each flight in step 3 includes the following steps: Step 3-1: Calculate the instantaneous surface flow of flights arriving and departing; Step 3-2: Calculate the surface cumulative flow of flight arrivals and departures; Step 3-3: Calculate the number of aircraft queued at the runway gate for the departure flight; Step 3-4: Digitize the characteristic data corresponding to each flight; The specific method for calculating the flight arrival and departure surface instantaneous flow in step 3-1 is as follows: define D_SIFI as the departure surface instantaneous flow, A_SIFI as the arrival surface instantaneous flow, T on is the landing time of the flight, T out is the launch time of the flight departure, SIFI (instantaneous flow rate) when the flight arrives at the airport in The calculation formula is as follows: SIF in =D_SIFI in +A_SIFI in Where D_SIFI in Point to T on Previously launched, in T on The number of flights that took off afterwards, A_SIFI in Point to T on Before landing, in T on The number of flights that will be available later; Flight departure instantaneous flow D_SIFI out The calculation formula is as follows: SIF out =D_SIFI out +A_SIFI out Where D_SIFI out Point to T out Previously launched, in T out The number of flights departing after that, A_SIFI out Point to T out Before landing, in T out The number of flights that will be available later; The step 3-2 of calculating the flight arrival and departure surface cumulative flow is as follows: define D_SCFI to represent the departure surface cumulative flow, A_SCFI to represent the arrival surface cumulative flow, T on and T in T is the landing time of the flight and the arrival time of the parking space, respectively. out and T off The launch time and take-off time of the flight departure are respectively, SCFI (surface cumulative flow) when the flight arrives in The calculation process is as follows: D_SCFI in =FT out1 +BT out1 A_SIFI in =FT in1 +BT in1 SCFI in =D_SCFI in +A_SCFI in Among them, FT out1 Point to T on Previously launched and in T on The number of flights departing after that, BT out1 Point to T on After that and in T in The number of flights launched before, FT in1 Point to T on Landed before and in T on The number of flights that will be seated later, BT in1 Point to T on After that and in T in The number of flights that have landed before, The calculation process of the cumulative flow of the flight at the departure time is as follows: D_SCFI out =FT out2 +BT out2 A_SIFI out =FT in2 +BT in2 SCFI out =D_SCFI out +A_SCFI out Among them, FT out2 Point to T out Previously launched and in T out The number of flights departing after that, BT out2 Point to T out After that and in T off The number of flights launched before, FT in2 Point to T out Landed before and in T out The number of flights that will be seated later, BT in2 Point to T out After that and in T off The number of flights that have landed previously; The number of aircraft queuing at the runway gate in the step 3-3 of calculating the number of aircraft queuing at the runway gate for the flight departure is divided into the number of runway departure aircraft and the number of runway arrival aircraft. Specifically, D_AQLI is defined to represent the number of runway departure aircraft, A_AQLI represents the number of runway arrival aircraft, T out and T off The launch time and take-off time of the flight departure are respectively, The calculation formula for the number of aircraft queued at the runway gate when the flight departs is as follows: AQLI=D_AQLI+A_AQLI Where D_AQLI refers to the out Before or after the launch, in T out and T off The number of flights taking off between T out After that, T off The number of flights that have landed previously; The specific process of digitizing the characteristic data corresponding to each flight in step 3-4 is as follows: Step 3-4-1: Classify the weather into three categories: almost no effect on the taxiing time, some effect on the taxiing time, and great effect on the taxiing time, represented by natural numbers 1 to 3; Step 3-4-2: Divide the time periods: 6-9 a.m. is the peak time for flights arriving and departing. After 9 a.m., the number of flights arriving and departing gradually reaches a balance. It is best to divide the time periods by natural hours, represented by natural numbers from 0 to 23. Step 3-4-3: Digitalize the parking stands, airlines, runways and aircraft types: Different parking stands, airlines, runways and aircraft types are represented by different numbers; In step 5, the performance of each prediction model is evaluated using the test samples, and the specific process of selecting the model with the best prediction effect and meeting the actual application requirements is as follows: Step 5-1: Use mean square error to evaluate each prediction model. The specific formula for calculating mean square error is as follows: Where n is the number of sample data, y i is the actual taxiing time of the flight. The model predicts the flight runway taxi time; Step 5-2: Select the prediction model with the smallest MSE_Pre value and that meets the actual application requirements and apply it to the actual scenario.

2. According to the method for predicting flight runway taxiing time for civil aviation airports according to claim 1, Features: The specific process of selecting multiple machine learning methods for training to obtain multiple flight runway taxi time prediction models in step 4 is as follows: Step 4-1: Select M 1 、M 2 、M 3 ……M n There are N types of machine learning models; Step 4-2: Train these N types of machine learning methods using the training data set as input to obtain multiple flight runway taxi time prediction models.

3. According to the method for predicting flight runway taxiing time for civil aviation airports according to claim 1, Features: In step 1, the airport historical flight data is obtained and preprocessed. The specific process of preprocessing is as follows: Step 1-1: Clean missing data: In the acquired airport historical flight data, if any of the flight data's actual arrival time, actual departure time, braking time, or brake release time is missing, the flight data needs to be removed; Step 1-2: Process abnormal data: mainly including flight data whose brake release time is later than the actual departure time and brake time is earlier than the actual arrival time.