Scene key time prediction method and system based on gradient boosting decision tree
Through the method of improving the decision tree based on gradient, flights are predicted at key time points in the airport scene, solving the problem of uncertainty in key time nodes of flights and improving the operation efficiency of the airport.
Patent Information
- Application Number
- CN202510334860.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-20
- Publication Date
- 2025-06-27
AI Technical Summary
There is time uncertainty in the critical time nodes of flights at large hub airports operating at airport scenes, resulting in a decrease in airport operation efficiency and making it difficult to coordinate airspace and scene resources in a timely manner.
The scene critical time prediction method based on the gradient enhancement decision tree is adopted. By analyzing real-time data, the aircraft's expected takeoff time, unhindered taxi time and expected taxi time are predicted, and the expected withdrawal time is calculated.
Accurate prediction of key flight time points is achieved, the operation efficiency of the airport is improved, and the effective implementation of scene scheduling strategies is ensured.
Smart Images
Figure CN120218345A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of air traffic control technology, and in particular to a scene critical time prediction method and prediction system based on a gradient boosting decision tree. Background Art
[0002] The key time nodes of flight activities on the airport surface are important basic information for the safe and efficient operation of large hub airports, which can ensure the effective implementation of subsequent surface scheduling strategies.
[0003] However, large hub airports have many activities and interrelated operation links. Under the influence of various factors such as airspace flow control, bad weather, and airport emergencies, there is time uncertainty in the key links of flight operations at the airport. In addition, with the continuous growth of traffic demand, airport surface and airspace resources are becoming increasingly tight. If airport control coordination is not timely during busy hours, it will affect the overall operation, and it is easy to have flight departure delays that deviate from the plan and traffic congestion on the surface. These conditions will further make key time points more difficult to predict, resulting in a significant decrease in the operating efficiency of the airport. Summary of the invention
[0004] In view of the defects in the prior art, the present invention provides a scene critical time prediction method based on a gradient boosting decision tree, which can effectively predict the key time points and improve the operation efficiency of the airport.
[0005] The present application provides a scene critical time prediction method based on a gradient boosting decision tree, the prediction method comprising:
[0006] Collect data from target aircraft on the airport surface to obtain real-time data;
[0007] Integrating the real-time data into a gradient boosting decision tree prediction model to predict the estimated take-off time of the target aircraft;
[0008] Retrieving the previously calculated historical average unobstructed taxiing speeds of aircraft on different taxiing sections, and predicting the unobstructed taxiing time of the target aircraft on the planned taxiing path based on the historical average unobstructed taxiing speeds;
[0009] Integrating the unimpeded taxiing time into the gradient boosting decision tree prediction model to predict the expected taxiing time of the target aircraft at the airport surface;
[0010] The estimated off-block time of the target aircraft is calculated based on the estimated take-off time and the estimated taxiing time.
[0011] In one aspect, before the step of combining the real-time data into the gradient boosting decision tree prediction model, the step includes:
[0012] Collect data based on historical flight operation data, where the historical flight operation data at least includes information such as flight plan data, weather data, flow control data, and flight own attribute data;
[0013] Conduct feature analysis on the collected data to construct model training samples;
[0014] Construct a gradient boosting decision tree basic model and set parameters for the gradient boosting decision tree basic model;
[0015] Train the gradient boosting decision tree basic model with the completed parameter settings according to the training samples to form an initial gradient boosting decision tree model.
[0016] In one aspect, the step of conducting feature analysis on the collected data to form training samples required for model training includes:
[0017] Extract key features from the collected data, analyze the correlation between the key features, and establish a correlation matrix between each of the key features;
[0018] Select important features based on the correlation matrix;
[0019] Construct training samples according to the important features.
[0020] In one aspect, after the step of forming the initial gradient boosting decision tree model, it includes:
[0021] Obtain the predicted takeoff time value of the aircraft according to the initial gradient boosting decision tree model;
[0022] Compare the predicted takeoff time value with the actual takeoff time value, calculate the time error between the two, and form a loss change curve based on the time error;
[0023] Feed back the time error and the loss change curve to the parameter setting, and optimize to obtain a gradient boosting decision tree prediction model.
[0024] In one aspect, the step of extracting key features from the collected data includes:
[0025] Preprocess the historical operation data;
[0026] For rerouted flights and cancelled flights, remove the flight plan data corresponding to the rerouting and cancellation;
[0027] Screen duplicate flights and retain only unique flight data;
[0028] Define the data format and align and match all data forms according to the defined data format.
[0029] In one aspect, the step of retrieving the historical average unobstructed taxiing speed of the aircraft on different taxiing sections includes:
[0030] Obtain the road network data of the airport surface, and establish a topological network with a specified moving direction based on the road network data. Among them, the topological network includes multiple measurement nodes, and the measurement nodes are at least one of the surface conflict hot spots, road intersections, parking positions, and runway entrances;
[0031] According to the historical radar trajectory data of the surface, screen the unobstructed taxiing flights, match the taxiing trajectory points of the unobstructed taxiing flights with the measurement nodes in the topological network, and output the flight taxiing route composed of the measurement node sequences;
[0032] Divide the flight taxiing route into a route including at least one of a straight section, a turning section, and a fast taxiing section;
[0033] Respectively statistically obtain the historical average unobstructed taxiing speed of flights on the different types of taxiing sections in the historical surface operation.
[0034] In one aspect, the step of predicting the unobstructed taxiing time of the target aircraft on the planned taxiing path based on the historical average unobstructed taxiing speed includes:
[0035] Obtain the planned taxiing path of the target aircraft on the airport surface, and the planned taxiing path includes several planned taxiing intervals;
[0036] Based on the historical average unobstructed taxiing speed of different sections, combined with the length of the planned taxiing section and the speed limit of the aircraft, estimate the unobstructed taxiing time of the aircraft in each planned taxiing interval, and accumulate the unobstructed taxiing time of the planned taxiing interval to obtain the unobstructed taxiing time of the target aircraft on the planned taxiing path.
[0037] In one aspect, the step of combining the unobstructed taxiing time into the gradient boosting decision tree prediction model to predict the estimated taxiing time of the target aircraft on the airport surface includes:
[0038] Obtain the planned information and variable information corresponding to the target aircraft;
[0039] Based on the gradient boosting decision tree prediction model, combined with the planned information and variable information of the target aircraft, and the unobstructed taxiing time on the planned taxiing path, predict the estimated taxiing time of the target aircraft on the airport surface.
[0040] In one aspect, the planned information at least includes one of a flight plan, a historical radar track of an aircraft's surface taxiing, airport road network structure information, and an airline, and the variable information includes at least one of weather, visibility, instantaneous in-and-out surface traffic flow, cumulative in-and-out surface traffic flow, and the number of flights queuing on the runway.
[0041] In addition, to solve the above problems, the present application also provides a surface critical time prediction system based on a gradient boosting decision tree. The surface critical time prediction system based on a gradient boosting decision tree includes:
[0042] An acquisition module, which is used to collect data of target aircraft on the airport surface to obtain real-time data;
[0043] A prediction module, which is used to combine the real-time data into a gradient boosting decision tree prediction model to predict the estimated takeoff time of the target aircraft; retrieve the historical average unobstructed taxiing speed of the aircraft on different taxiing sections in advance, and predict the unobstructed taxiing time of the target aircraft on the planned taxiing path according to the historical average unobstructed taxiing speed; combine the unobstructed taxiing time into the gradient boosting decision tree prediction model to predict the estimated taxiing time of the target aircraft on the airport surface; calculate the estimated pushback time of the target aircraft according to the estimated takeoff time and the estimated taxiing time.
[0044] The beneficial effects of the present invention are reflected in: it can predict three key time points. Specifically, it analyzes the collected real-time data through a gradient boosting decision tree prediction model to predict the estimated takeoff time of the aircraft. By the historical average unobstructed taxiing speed of the aircraft, it analyzes and predicts the unobstructed taxiing time of the aircraft on the planned taxiing path. Then, it combines the unobstructed taxiing time into the gradient boosting decision tree prediction model to predict the estimated taxiing time of the target aircraft on the airport surface. The estimated pushback time of the aircraft is calculated through the estimated takeoff time and the estimated taxiing time. It can be seen that through the gradient boosting decision tree prediction model, the prediction of the estimated takeoff time, the estimated taxiing time, and the estimated pushback time can be accurately completed, effectively predicting the key time points and improving the operation efficiency of the airport. Description of the Drawings
[0045] 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 use in the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.
[0046] Figure 1Schematic diagram of the prediction process of the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0047] Figure 2 Schematic diagram of the process steps of the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0048] Figure 3 Schematic diagram of the process steps of constructing the initial model of the gradient boosting decision tree in the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0049] Figure 4 Schematic diagram of the process steps of forming training samples in the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0050] Figure 5 Schematic diagram of the process steps of tuning to obtain the gradient boosting decision tree prediction model in the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0051] Figure 6 Schematic diagram of the process steps of preprocessing historical operation data in the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0052] Figure 7 Schematic diagram of the process steps of obtaining the historical average unobstructed taxiing speed of different taxiing sections in the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0053] Figure 8 Schematic diagram of the process steps of obtaining the unobstructed taxiing time in the method for predicting critical times on the airfield based on gradient boosting decision trees in this application;
[0054] Figure 9 Schematic diagram of the process steps of predicting the expected taxiing time of the aircraft in the method for predicting critical times on the airfield based on gradient boosting decision trees in this application. Detailed implementation manners
[0055] Hereinafter, embodiments of the technical solution of the present invention will be described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and thus are only examples and cannot be used to limit the protection scope of the present invention.
[0056] It should be noted that unless otherwise specified, the technical terms or scientific terms used in this application should have the ordinary meanings understood by those skilled in the art to which the present invention belongs.
[0057] Such as Figure 1 And Figure 2As shown in the figure, the present application provides a method for predicting critical times on the airport surface based on gradient boosting decision trees. In this embodiment, the critical times include the expected take-off time of the aircraft, the expected taxiing time on the airport surface, and the expected wheel chock removal time. The wheel chock removal time refers to the time point when the aircraft is separated from the ground equipment during the pre-takeoff preparation. Specifically, after the wheel chocks are removed, the aircraft can be towed out of the stand and start taxiing. The wheel chock removal time is an important reference time point for determining whether a flight is considered delayed.
[0058] In the present application, the method for predicting critical times on the airport surface based on gradient boosting decision trees includes:
[0059] Step S10, collect real-time data for the target aircraft on the airport surface; the real-time data includes flight plan data, weather data, flow control data, and flight self-attributes. The flight plan data includes information such as timestamp, planned take-off time, calculated take-off time, calculated wheel chock removal time, planned arrival time, flight push-back time, start taxiing time, end taxiing time, actual take-off time, actual arrival time, etc. The weather data includes visibility, precipitation type, etc. The flow control data includes flow control type, airspace restrictions, etc. The flight self-attributes include flight number, airline, aircraft type, departure airport, arrival airport, boarding gate, runway, taxi route, international / domestic flight attribute. The above data can be obtained from airlines, airport management departments, meteorological departments, etc., or each department can upload the data to the cloud and obtain real-time data from the cloud.
[0060] Step S20, combine the real-time data into the gradient boosting decision tree prediction model to predict the expected take-off time of the target aircraft; by providing the real-time data to the gradient boosting decision tree prediction model, the gradient boosting decision tree prediction model can output the expected take-off time of the aircraft.
[0061] Step S30, retrieve the historical average unobstructed taxiing speed of the aircraft on different taxiing sections, and predict the unobstructed taxiing time of the target aircraft on the planned taxiing path based on the historical average unobstructed taxiing speed; use the historical average unobstructed taxiing speed to exclude unpredictable abnormal events, calculate the time under normal passage, and calculate the unobstructed taxiing time of the aircraft on the airport surface through the historical average unobstructed taxiing speed.
[0062] Step S40, combine the unobstructed taxiing time into the gradient boosting decision tree prediction model to predict the expected taxiing time of the target aircraft on the airport surface; provide the unobstructed taxiing time to the gradient boosting decision tree prediction model, and the gradient boosting decision tree prediction model outputs the expected taxiing time of the aircraft. It can be seen from this that the gradient boosting decision tree prediction model can output the expected take-off time and expected taxiing time of the aircraft.
[0063] Step S50: Calculate the estimated engine start time of the target aircraft based on the estimated takeoff time and the estimated taxiing time. For example, subtract the estimated taxiing time from the estimated takeoff time to calculate the estimated engine start time of the aircraft by reverse deduction.
[0064] In this embodiment, three key time points can be predicted. The real-time data collected is mainly analyzed through the gradient boosting decision tree prediction model to predict the estimated takeoff time of the aircraft. The unobstructed taxiing time of the aircraft on the planned taxiing path is analyzed and predicted based on the historical average unobstructed taxiing speed of the aircraft. Then, the unobstructed taxiing time is combined with the gradient boosting decision tree prediction model to predict the estimated taxiing time of the target aircraft on the airport surface. The estimated engine start time of the aircraft is calculated based on the estimated takeoff time and the estimated taxiing time. It can be seen that through the gradient boosting decision tree prediction model, the prediction of the estimated takeoff time, the estimated taxiing time, and the estimated engine start time can be accurately completed, effectively predicting the key time points and improving the operation efficiency of the airport.
[0065] As Figure 3 shown, in an embodiment of the present application, before the step of combining the real-time data with the gradient boosting decision tree prediction model, it includes:
[0066] Step S01: Collect data based on the historical operation data of the flight. The historical operation data of the flight includes at least flight plan data, weather data, flow control data, and the attributes of the flight itself; the content of the historical operation data of the flight refers to the explanation of the real-time data in step S10. Among them, the historical operation data of the flight can be stored in the flight information collection library, which is convenient for subsequent retrieval and also facilitates the rapid formation of training samples to improve the training efficiency of the model.
[0067] Step S02: Conduct feature analysis on the collected data, extract the key feature information in the data, and thus construct a model training sample.
[0068] Step S03: Construct a gradient boosting decision tree basic model and set parameters for the gradient boosting decision tree basic model; the gradient boosting decision tree basic model can be understood as the basic structure of the model. Parameter setting can be understood as setting parameters such as the depth of the gradient boosting decision tree basic model, the number of weak classifiers, the number of leaf nodes, the maximum number of features considered during partitioning, the minimum number of samples for internal node partitioning, the minimum sample weight of leaf nodes, the weight coefficient of the regularization term, and the learning rate.
[0069] Step S04: Train the gradient boosting decision tree base model with the parameter settings completed using the training samples to form an initial gradient boosting decision tree model. Use the training samples as the input sequence and import them into the gradient boosting decision tree base model for training. The output sequence is the predicted flight departure time. Through repeated training and optimization of the gradient boosting decision tree base model, a gradient boosting decision tree prediction model is formed. The takeoff time and taxiing time of the aircraft can be predicted using the gradient boosting decision tree prediction model.
[0070] As Figure 4 shown, in an embodiment of the present application, the steps of performing feature analysis on the collected data to form the training samples required for model training include:
[0071] Step S021: Extract key features from the collected data, analyze the correlation between the key features, and establish a correlation matrix between each key feature; calculate the correlation between the selected features and other input features, and establish a correlation coefficient matrix between each feature, thereby enabling a more intuitive view of the impact of each other input feature on the predicted takeoff time.
[0072] Step S022: Select important features based on the correlation matrix; for example, the departure and arrival airports, as well as the planned departure and arrival time information, are mapped to the flight plan flight duration feature; the parking position, taxiway, and runway information are mapped to the planned taxiing duration feature; the scene congestion situation is statistically analyzed according to historical flight plan data to obtain the number of aircraft taxiing on the scene and the number of aircraft queuing at the runway during a fixed time period of a day as features; the aircraft type is classified according to heavy, medium, and light; the time stamp of the flight is classified according to whether it is a holiday; the flight type is classified according to international and domestic flights, and the traffic control information is classified according to traffic restrictions, weather impacts, military exercises, and airspace restrictions, and is expressed as classification features.
[0073] Step S023: Construct training samples based on the important features. Combine these important features to complete the construction of the training samples. The initial training of the prediction model can be completed by inputting the training samples into the gradient boosting decision tree base model.
[0074] As Figure 5 shown, in an embodiment of the present application, in order to improve the accuracy of the model's prediction of time points, the model is optimized. Specifically, after the step of forming the initial gradient boosting decision tree model, it includes:
[0075] Step S05: Obtain the predicted takeoff time value of the aircraft according to the initial gradient boosting decision tree model; through the initial gradient boosting decision tree model, first predict the predicted takeoff time value of the aircraft, and then evaluate the accuracy of the takeoff time prediction.
[0076] Step S06: Compare the predicted takeoff time value with the actual takeoff time value, calculate the time error between the two, and form a loss change curve based on the time error; record the predicted takeoff time values output by the initial model of the gradient boosting decision tree in each iteration process, compare the predicted takeoff time with the corresponding actual takeoff time value, record the time error, and form a loss change curve through multiple iterations.
[0077] Step S07: Feed back the time error and the loss change curve to the parameter settings, and optimize to obtain the gradient boosting decision tree prediction model; through the loss function, such as setting evaluation indicators of types such as root mean square error RSME and mean absolute error MAE for evaluating the training performance of the model. By tuning one by one or in groups, compare the training results of each tuning of different parameters, and save the model with the best output result as the gradient boosting decision tree prediction model. The gradient boosting decision tree prediction model obtained through tuning can improve the accuracy of time prediction.
[0078] As Figure 6 shown, in an embodiment of the present application, the steps of extracting key features from the collected data include:
[0079] Step S0211: Preprocess the historical operation data; preprocessing means removing abnormal flight data, screening duplicate data, aligning and matching data formats, etc.
[0080] Step S0212: For rerouted flights and cancelled flights, remove the invalid flight plan data before rerouting and the flight plan data corresponding to the cancelled flights; abnormal flight data includes rerouted flights, duplicate flights, cancelled flights, etc. For rerouted flights and cancelled flights, find the data pattern to identify the rerouted flights and remove the invalid flight plan data before rerouting.
[0081] Step S0213: Screen the duplicate flights and retain the unique data of the flights; screen the duplicate data and retain the unique data of the corresponding flights.
[0082] Step S0214: Define the data format, and align and match all data forms according to the defined data format. Aligning and matching the data format is a unified data format, aligning and matching all data forms to obtain a complete data set including the unique identifier of each aircraft, aircraft attributes, location information, taxiing time, taxiing trajectory, and meteorology, etc. Facilitating further processing of the data through a unified data format. For example, perform one-hot encoding on features such as flight type, holidays, flow control, and aircraft type characteristics, and express them in the form of 0-1 numerical values; for continuous features such as actual takeoff time and planned takeoff time, perform timestamp conversion and uniformly express them in the form of numerical values of minutes within a day.
[0083] As Figure 7As shown, in an embodiment of the present application, the step of retrieving the historical average unobstructed taxiing speed of an aircraft on different taxiing sections includes:
[0084] Step S310, obtain the road network data of the airport surface, and establish a topological network with a specified moving direction based on the road network data. Among them, the topological network includes multiple measurement nodes, and the measurement nodes are at least one of the surface conflict hot spots, intersection of taxiways, parking positions, and runway entrances; the measurement nodes can also be understood as the positions where congestion or accidents are likely to occur.
[0085] Step S320, based on the historical radar track data of the surface, screen unobstructed taxiing flights, match the taxiing track points of the unobstructed taxiing flights with the measurement nodes in the topological network in terms of longitude and latitude positions, and output the flight taxiing route composed of a sequence of measurement nodes; in this way, calculations can be performed on each measurement node in the flight taxiing route.
[0086] The flight information set library includes flight plans, surface radar track data, and airport road network data; the surface radar track data includes field surveillance track data, longitude and latitude, time, and motion state information data of the aircraft taxiing track; the airport road network data includes the longitude and latitude and numbers of each node, the numbers of taxiways and their included node sequence connection relationships, parking positions, and runway numbers, etc.
[0087] Step S330, divide the flight taxiing route into a route including at least one of a straight section, a turning section, and a fast taxiing section; by dividing the flight taxiing route, the speed of each section can be calculated better. Moreover, different sections have different speed limits.
[0088] Step S340, respectively statistically obtain the historical average unobstructed taxiing speed of flights on different types of taxiing sections during historical surface operations. Since different sections have different speed magnitude limits, the technical solution of the present application can increase the accuracy of confirming the entire taxiing speed and improve the accuracy of the historical average unobstructed taxiing speed by dividing different sections.
[0089] As Figure 8 shown, in an embodiment of the present application, the step of predicting the unobstructed taxiing time of a target aircraft on a planned taxiing path based on the historical average unobstructed taxiing speed includes:
[0090] Step S350, obtain the planned taxiing path of the target aircraft on the airport surface, and the planned taxiing path includes several planned taxiing intervals; different flights have different planned taxiing paths on the airport surface. The planned taxiing path is split to obtain multiple different planned taxiing intervals. The planned taxiing intervals can be divided into straight sections, turning sections, or fast taxiing sections, etc. according to the section attributes.
[0091] Step S360: Based on the historical average unobstructed taxiing speed of different sections, combined with the length of the planned taxiing section and the speed limit of the aircraft on different sections, estimate the unobstructed taxiing time of the aircraft in each planned taxiing interval, and accumulate the unobstructed taxiing time of the planned taxiing intervals to obtain the unobstructed taxiing time of the target aircraft on the planned taxiing path.
[0092] As Figure 9 shown, in an embodiment of the present application, the step of combining the unobstructed taxiing time into the gradient boosting decision tree prediction model to predict the estimated taxiing time of the target aircraft on the airport surface includes:
[0093] Step S410: Obtain the planned information and variable information corresponding to the target aircraft; the planned information includes at least one of the flight plan, taxiing distance, apron, and airline, and the variable information includes at least one of weather, visibility, instantaneous flow of arrivals and departures on the surface, cumulative flow of arrivals and departures on the surface, and number of flights queuing on the runway. The variable information includes at least one of weather, visibility, instantaneous flow of arrivals and departures on the surface, cumulative flow of arrivals and departures on the surface, and number of flights queuing on the runway.
[0094] Step S420: Based on the gradient boosting decision tree prediction model, combine the planned information and variable information of the target aircraft, as well as the unobstructed taxiing time on the planned taxiing path, to predict the estimated taxiing time of the target aircraft on the airport surface.
[0095] The present application also provides a surface key time prediction system based on the gradient boosting decision tree. The surface key time prediction system based on the gradient boosting decision tree includes: a collection module and a prediction module.
[0096] The collection module is used to collect data of the target aircraft on the airport surface to obtain real-time data;
[0097] The prediction module is used to combine the real-time data into the gradient boosting decision tree prediction model to predict the estimated takeoff time of the target aircraft; retrieve the historical average unobstructed taxiing speed of the aircraft on different taxiing sections in advance, and predict the unobstructed taxiing time of the target aircraft on the planned taxiing path based on the historical average unobstructed taxiing speed; combine the unobstructed taxiing time into the gradient boosting decision tree prediction model to predict the estimated taxiing time of the target aircraft on the airport surface; calculate the estimated wheel chock off time of the target aircraft based on the estimated takeoff time and the estimated taxiing time.
[0098] The prediction system further includes a time plan information display terminal. The display terminal can simultaneously display the estimated takeoff time, the estimated taxiing time, and the estimated wheel chock off time on the display screen, and is for the apron control seat and the tower control seat.
[0099] For the specific embodiments and beneficial effects of the scene critical time prediction system based on gradient boosting decision tree in this application, please refer to the above-mentioned scene critical time prediction method based on gradient boosting decision tree, which will not be elaborated here.
[0100] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.
Claims
1. A scene critical time prediction method based on gradient boosting decision tree, characterized in that: The prediction method comprises: Collect data from target aircraft on the airport surface to obtain real-time data; Integrating the real-time data into a gradient boosting decision tree prediction model to predict the estimated take-off time of the target aircraft; Retrieving the previously calculated historical average unobstructed taxiing speeds of aircraft on different taxiing sections, and predicting the unobstructed taxiing time of the target aircraft on the planned taxiing path based on the historical average unobstructed taxiing speeds; Integrating the unimpeded taxiing time into the gradient boosting decision tree prediction model to predict the expected taxiing time of the target aircraft at the airport surface; The estimated off-block time of the target aircraft is calculated based on the estimated take-off time and the estimated taxiing time.
2. The prediction method according to claim 1, characterized in that: Before the step of combining the real-time data into the gradient boosting decision tree prediction model, it includes: Collecting data based on historical flight operation data, where the historical flight operation data includes at least flight plan data, weather data, flow control data, and flight attributes; Perform feature analysis on the collected data and construct model training samples; Constructing a gradient boosting decision tree basic model, and setting parameters for the gradient boosting decision tree basic model; The gradient boosting decision tree basic model with completed parameter settings is trained according to the training samples to form a gradient boosting decision tree initial model.
3. The prediction method according to claim 2, characterized in that: The steps of performing feature analysis on the collected data to form training samples required for model training include: Extract key features from the collected data, analyze the correlation between the key features, and establish a correlation matrix between the key features; Selecting important features based on the correlation matrix; Training samples are constructed based on the important features.
4. The prediction method according to claim 2, characterized in that: After the step of forming the initial model of the gradient boosting decision tree, the following steps are included: Obtaining a predicted value of the take-off time of the aircraft according to the gradient boosting decision tree initial model; Comparing the predicted take-off time value with the actual take-off time value, calculating the time error between the two, and forming a loss change curve according to the time error; The time error and the loss change curve are fed back to the parameter setting, and the gradient boosting decision tree prediction model is obtained through tuning.
5. The prediction method according to claim 3, characterized in that: The step of extracting key features from the collected data includes: Preprocessing the historical operation data; For diverted flights and cancelled flights, the invalid flight plan data before the diversion and the flight plan data corresponding to the cancelled flight will be deleted; Filter duplicate flights and retain unique flight data; Define the data format and align and match all data forms according to the defined data format.
6. The prediction method according to claim 1, characterized in that: The step of retrieving the previously counted historical average unobstructed taxiing speed of the aircraft on different taxiing sections includes: Acquire the road network data of the airport surface, and establish a topological network with a specified moving direction according to the road network data, wherein the topological network includes a plurality of metering nodes, and the metering node is at least one of a hotspot location of surface conflict, a road intersection, a parking space, and a runway entrance; According to the historical radar trajectory data of the scene, unobstructed taxiing flights are screened, the taxiing trajectory points of the unobstructed taxiing flights are matched with the longitude and latitude positions of the metering nodes in the topological network, and the flight taxiing route composed of the metering node sequence is output; Dividing the flight taxi route into a route including at least one of a straight section, a turning section and a fast taxi section; The historical average unobstructed taxiing speeds of flights on the different types of taxiing sections during historical field operations are statistically obtained respectively.
7. The prediction method according to claim 6, characterized in that: The step of predicting the unobstructed taxiing time of the target aircraft on the planned taxiing path according to the historical average unobstructed taxiing speed comprises: Acquire a planned taxiing path of the target aircraft on the airport surface, wherein the planned taxiing path includes a plurality of planned taxiing sections; According to the historical average unobstructed taxiing speed of different sections, combined with the length of the planned taxiing section and the speed limit of the aircraft, the unobstructed taxiing time of the aircraft in each planned taxiing interval is estimated, and the unobstructed taxiing time of the planned taxiing interval is accumulated to obtain the unobstructed taxiing time of the target aircraft on the planned taxiing path.
8. The prediction method according to claim 6, characterized in that: The step of combining the unimpeded taxiing time with the gradient boosting decision tree prediction model to predict the expected taxiing time of the target aircraft at the airport surface includes: Obtain the plan information and variable information corresponding to the target aircraft; Based on the gradient boosting decision tree prediction model, combined with the planned information and variable information of the target aircraft, and the unobstructed taxiing time on the planned taxiing path, the estimated taxiing time of the target aircraft at the airport is predicted.
9. The prediction method according to claim 8, characterized in that: The planning information includes at least one of the flight plan, the historical radar trajectory of aircraft taxiing on the ground, the airport road network structure information and the airline, and the variable information includes at least one of the weather, visibility, instantaneous flow of arrival and departure surfaces, cumulative flow of arrival and departure surfaces and the number of flights queued on the runway.
10. A scene critical time prediction system based on gradient boosting decision tree, characterized in that: The prediction system comprises: A collection module, the collection module is used to collect data from target aircraft on the airport surface to obtain real-time data; A prediction module is used to combine the real-time data with a gradient boosting decision tree prediction model to predict the estimated take-off time of the target aircraft; retrieve the pre-stated historical average unobstructed taxiing speed of the aircraft on different taxiing sections, and predict the unobstructed taxiing time of the target aircraft on the planned taxiing path based on the historical average unobstructed taxiing speed; combine the unobstructed taxiing time with the gradient boosting decision tree prediction model to predict the estimated taxiing time of the target aircraft at the airport; and calculate the estimated off-block time of the target aircraft based on the estimated take-off time and the estimated taxiing time.