Data processing method, device and medium for obtaining taxiing time of departing flights

Abstract

The present invention relates to the field of data processing technology, and more particularly to a data processing method, device, and medium for obtaining taxi time for departing flights. This method predicts taxi time by integrating taxi trajectory data and taxi-related data. This method accounts for factors influencing geographic space, temporal dynamics, and environmental interactions, resulting in more accurate taxi time predictions. Furthermore, by combining trajectory data and taxi-related data to extract spatiotemporal features of local paths and introducing an attention mechanism to dynamically identify key local paths, the accuracy of taxi time predictions can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of aviation data processing, and in particular to a data processing method, device and medium for obtaining the taxiing time of a departing flight. Background Art

[0002] The arrival and departure processes of flights are highly dynamic, random, and uncertain. Therefore, accurate and scientific prediction and analysis of ground taxi time is crucial. For example, accurate prediction of departure taxi time can help airports optimize flight pushback sequences, thereby improving airport operational efficiency and enhancing flight punctuality. Current research methods for flight taxi time fall primarily into two categories. The first is prediction based on traditional simulation methods. These methods rely on factors such as airport site characteristics and flight dynamics to model flight time. While they can account for multiple factors, the models are complex and computationally intensive, and they have limitations when handling complex nonlinear relationships. The second is prediction methods based on machine learning, such as random forests, support vector machines, and neural networks, to improve taxi time prediction accuracy. Compared to simulation methods, machine learning methods are better able to handle nonlinear relationships and adapt to complex dynamic environments. However, existing machine learning methods primarily focus on modeling macroscopic features, such as taxi distance and airport traffic volume, while neglecting the fine-grained spatiotemporal characteristics of taxi trajectories, which are crucial for accurate prediction in complex environments. Furthermore, existing models exhibit poor robustness under the influence of multiple interactions, particularly in complex operating environments and dynamic changes, often failing to maintain stable prediction accuracy. Traditional machine learning models often struggle to effectively identify and extract key local spatiotemporal features during taxiing, resulting in weak generalization capabilities in complex environments. Therefore, extracting spatiotemporal features from fine-grained taxi trajectory data while simultaneously accounting for the influence of dynamic external factors has become crucial for improving the precision and accuracy of departure taxi time predictions. Summary of the Invention

[0003] In view of the above technical problems, the technical solution adopted by the present invention is:

[0004] According to a first aspect of the present invention, there is provided a data processing method for obtaining taxiing time of a departing flight, the method comprising the following steps:

[0005] S100, obtaining a data set to be processed; the data set to be processed includes simulated taxiing trajectory data and taxiing-related data of a target flight.

[0006] S200 , encoding the simulated taxiing trajectory data and the taxiing-related data respectively to obtain a feature vector of the simulated taxiing trajectory data and a feature vector of the taxiing-related data.

[0007] S300, using a convolution kernel with a width of k, performs a one-dimensional convolution operation on the feature vector of the simulated taxiing trajectory data to extract m local features to form the spatial features of m local paths; m = n - k + 1, where n is the number of position points in the simulated taxiing trajectory data.

[0008] S400 , obtaining the geometric distance of each local path, performing feature splicing on the spatial feature of each local path and the corresponding geometric distance, and using the splicing result as the enhanced spatial feature of the local path.

[0009] S500 , fusing the enhanced spatial features of each local path with the feature vector of the sliding-related data to obtain a multimodal feature of each local path.

[0010] S600 , extracting the temporal dependency relationship in the multimodal features of each local path to obtain the spatiotemporal features of each local path.

[0011] S700: Use the attention mechanism to perform weighted processing on all spatiotemporal features to obtain global spatiotemporal features.

[0012] S800: Process the global spatiotemporal features to obtain a predicted value of the taxiing time of the target flight.

[0013] According to a second aspect of the present invention, an electronic device is provided, comprising a processor and a memory; the processor is configured to execute the steps of the method according to the first aspect of the present invention by calling a program or instruction stored in the memory.

[0014] According to a third aspect of the present invention, there is provided a computer-readable storage medium storing a program or instructions, wherein the program or instructions enable a computer to execute the steps of the method according to the first aspect of the present invention.

[0015] The data processing method for obtaining taxi times for departing flights provided in an embodiment of the present invention predicts taxi times by integrating taxi trajectory data and taxi-related data. This method accounts for influencing factors such as geographic space, temporal dynamics, and environmental interactions, resulting in more accurate taxi time predictions. Furthermore, by combining trajectory data and taxi-related data to extract spatiotemporal features of local paths and introducing an attention mechanism to dynamically identify key local paths, the accuracy of departure taxi time predictions can be improved, providing data support for airports to optimize flight rollout sequences and resource allocation decisions.

[0016] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A flow chart of a data processing method for obtaining taxiing time of departing flights provided in an embodiment of the present invention;

[0019] Figure 2 A structural block diagram of a glide time prediction model provided by an embodiment of the present invention;

[0020] Figure 3 A schematic diagram illustrating the effect of the glide time prediction model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0023] It should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the steps as sequential processes, many of the steps can be performed in parallel, concurrently, or simultaneously. In addition, the order of the steps can be rearranged. A process can be terminated when its operation is completed, but can also have additional steps not included in the accompanying drawings. A process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0024] The embodiment of the present invention provides a data processing method for obtaining the taxiing time of a departing flight. Figure 1 As shown, the method includes the following steps:

[0025] S100: Obtain a data set to be processed.

[0026] In an embodiment of the present invention, the data set to be processed includes simulated taxiing trajectory data and taxiing-related data of the target flight, and the taxiing-related data includes flight attribute information, operating period identification value, environmental data, and flight interaction data.

[0027] In the embodiment of the present invention, the simulated taxiing trajectory data in the data set to be processed is taxiing trajectory data obtained by simulation, and the specific simulation method may be the existing technology.

[0028] In an embodiment of the present invention, the target flight of the dataset to be processed is the departing flight corresponding to the dataset currently to be processed, that is, the dataset to be processed is the dataset corresponding to the target flight, that is, the data subject corresponding to the dataset. In an embodiment of the present invention, one departing flight corresponds to one aircraft. In an embodiment of the present invention, flight attribute information may include: a minute timestamp based on the planned take-off time, aircraft type, aircraft ID, taxiing distance, etc. The taxiing distance refers to the distance from a specified starting position to a specified end position. In an embodiment of the present invention, the specified starting position may be a parking space or a boarding gate, and the specified end position is the take-off point of the aircraft.

[0029] In this embodiment of the present invention, the minute timestamp based on the scheduled departure time refers to the minute timestamp corresponding to the scheduled departure time. For example, if the scheduled departure time format is "2025-5-8 8:42", the minute timestamp based on the scheduled departure time is: 522 = (8 × 60 + 42). The aircraft ID refers to the unique identifier of the aircraft, such as the aircraft registration number.

[0030] In the embodiment of the present invention, the operating time period identifier value refers to the time period identifier value of the flight departure taxiing time. The traffic flow of the airport will change over time, and the flight departure taxiing time will be different in different time periods.

[0031] In an embodiment of the present invention, the operating time period can be determined based on the average number of flights per hour. In an exemplary embodiment, the present invention divides the entire day into three operating time periods based on the average number of flights per hour by referring to the actual operation of a certain airport, namely: the first operating time period to the third operating time period, wherein the time period of the first operating time period is [0:00-8:00], the time period of the second operating time period is [8:00-10:00, 22:00-24:00], and the time period of the third operating time period is [10:00-22:00]. Different operating time periods can be represented by different operating time period identification values.

[0032] In an embodiment of the present invention, the environmental data may be weather data of an airport, which may include visibility, wind speed, abnormal weather flags, etc. The abnormal weather flags may include a thunderstorm flag indicating thunderstorm weather and a precipitation flag indicating whether precipitation is occurring.

[0033] In an embodiment of the present invention, the flight interaction data corresponding to the target flight is the number of flights active on the airport surface during the taxiing process of the target flight, that is, the number of flights in a taxiing state on the airport surface, which may specifically include the number of flights in a taxiing state on the runway, taxiway and apron during the taxiing process of the target flight, including the number of arriving flights and departing flights. In a busy airport, arriving and departing flights may share part of the taxiing route, which may cause traffic congestion and thus extend the taxiing time of departing flights. The impact of interaction between flights on taxiing is a complex process. Especially when analyzing the taxiing time of departing flights, it is necessary to fully consider the impact of interactions between different types of flights.

[0034] In an embodiment of the present invention, during the taxiing process of the target flight, the number of flights in 12 different taxiing states may be included. That is, the flight interaction data may include 12 types of interaction data, with each type of interaction data corresponding to the number of flights in each taxiing state. In one exemplary embodiment, the 12 types of interaction data may specifically include five types of inbound flights and seven types of outbound flights during the taxiing process of the target flight.

[0035] Furthermore, in an embodiment of the present invention, the number of inbound flights during the taxiing process of the target flight may include the following five types of flight numbers: (1) the number of inbound flights whose landing time is earlier than the off-block time of the target flight and whose on-block time is during the taxiing process of the target flight; (2) the number of inbound flights whose landing time is earlier than the off-block time of the target flight and whose on-block time is later than the take-off time of the target flight; (3) the number of inbound flights whose landing time is later than the off-block time of the target flight and whose on-block time is earlier than the take-off time of the target flight; (4) the number of inbound flights whose landing time is during the taxiing process of the target flight and whose on-block time is later than the take-off time of the target flight; (5) the number of inbound flights in the taxiing process of the target flight, specifically the number of inbound flights whose landing time is earlier than the take-off time of the target flight and whose on-block time is later than the off-block time of the target flight.

[0036] In an embodiment of the present invention, the number of departing flights during the taxiing process of the target flight may include the following seven types of flight numbers: (1) the number of departing flights whose off-block timing is earlier than the off-block timing of the target flight and whose take-off timing is during the taxiing process of the target flight; (2) the number of departing flights whose off-block timing is earlier than the off-block timing of the target flight and whose take-off timing is later than the take-off timing of the target flight; (3) the number of departing flights whose off-block timing is later than the off-block timing of the target flight and whose take-off timing is earlier than the take-off timing of the target flight; (4) the number of departing flights whose off-block timing is during the taxiing process of the target flight and whose take-off timing is later than the take-off time of the target flight; 5) The number of flights in taxiing for departure during the departure of the target flight, specifically the number of flights whose off-block time is earlier than the take-off time of the target flight and whose take-off time is later than the off-block time of the target flight; (6) The number of flights in taxiing for departure during the off-block time of the target flight, specifically the number of flights whose off-block time is earlier than the off-block time of the target flight and whose take-off time is later than the off-block time of the target flight; (7) The number of flights in take-off state during the departure taxi of the target flight, specifically the number of flights whose take-off time is between the off-block time and the take-off time of the target flight.

[0037] Those skilled in the art should understand that the above-mentioned number of inbound flights and the number of outbound flights that interact with the target flight during taxiing can be preliminarily screened based on the estimated time in the flight schedule and obtained by calibration in combination with real-time monitoring data (such as radar and ADS-B data).

[0038] S200 , encoding the simulated taxiing trajectory data and the taxiing-related data respectively to obtain a feature vector of the simulated taxiing trajectory data and a feature vector of the taxiing-related data.

[0039] In an embodiment of the present invention, the simulated taxiing trajectory data may include position information of n position points, and the position information may be longitude and latitude coordinates.

[0040] In an embodiment of the present invention, the feature vector of the simulated taxiing trajectory data can be obtained by performing nonlinear mapping processing on the position information of each position point in the simulated taxiing trajectory data. Specifically, the feature vector of the simulated taxiing trajectory data satisfies the following conditions:

[0041] FL={Local1, Local2,…, Local u ,……,Local n};

[0042] Among them, FL is the characteristic vector of the simulated sliding trajectory data, Local uis the mapping feature obtained by performing nonlinear mapping on the longitude and latitude coordinates of the u-th position point, and the value of u ranges from 1 to n.

[0043] In the embodiment of the present invention, Local u Meet the following conditions: Local u =tanh(W local •concat(Lon u , Lat u )), Lon u is the longitude coordinate of the u-th position point, Lat u is the latitude coordinate of the u-th position point, concat represents the connection operation, W local is the learnable weight matrix and tanh() is the tanh activation function.

[0044] In an embodiment of the present invention, the characteristic vector of the taxiing-related data is obtained by the following steps:

[0045] S201 , pre-processing taxiing-related data to obtain initial feature vectors corresponding to flight attribute information, operating period identification values, environmental data, and flight interaction data.

[0046] In an embodiment of the present invention, preprocessing the taxiing-related data may specifically include:

[0047] (1) For discrete variables in taxi-related data, such as minute timestamps, aircraft types, aircraft IDs, and abnormal weather indicators, an embedding layer is used to convert the features of each categorical attribute into a low-dimensional vector to obtain the initial feature value of each discrete variable. The specific embedding method can be existing technology.

[0048] (2) For the non-discrete variables in the taxiing-related data, such as visibility, wind speed, flight interaction data, etc., standardization processing is used to obtain the initial eigenvalue corresponding to each non-discrete variable.

[0049] In an embodiment of the present invention, the initial eigenvalue of each non-discrete variable is obtained by Z-score normalization, i.e., the difference between the numerical value of the non-discrete variable and the corresponding mean is divided by the corresponding standard deviation. The mean and standard deviation corresponding to each non-discrete variable are calculated based on historical data statistics, such as those obtained from a training set.

[0050] S202 , performing feature splicing on the initial feature vectors corresponding to the runtime identification value and the environment data, and using the obtained splicing result as the fused environment feature vector.

[0051] S203, based on the fusion environment feature vector, obtain the weight vector corresponding to the initial feature vector corresponding to the flight interaction data, and based on the obtained weight vector, perform weighted processing on the initial feature vector corresponding to the flight interaction data to obtain the corresponding weighted processing result as the weighted interaction feature vector.

[0052] In this embodiment of the present invention, the weighted interactive feature vector satisfies the following conditions:

[0053] FW=(λ1f1,λ2f2,…,λ r f r , ..., λ Q f Q ).

[0054] Among them, FW is the weighted interaction feature vector, f r is the initial eigenvector corresponding to the rth interaction data, λ r is the weight corresponding to the rth interaction data. r The value of is 0 to 1, the value of r is 1 to Q, Q is the number of interaction data in the aircraft interaction data, and in the embodiment of the present invention, Q=12.

[0055] In the embodiment of the present invention, λ r Based on the fusion environment feature vector and f r Determine. Further, r The following conditions must be met:

[0056] λ r =σ(W e ·E env +W f ·f r +b);

[0057] Among them, W e is the environmental feature weight matrix, E env is the fusion environment feature vector, W f f r The corresponding weight matrix, b is the bias term, σ( ) is the Sigmoid activation function, and · represents the dot product.

[0058] S204 , performing feature splicing on the initial feature vector, the fused environment feature vector, and the weighted interaction feature vector corresponding to the flight attribute information, and using the obtained splicing result as the feature vector of the taxiing-related data.

[0059] In an embodiment of the present invention, since the feature vector of the taxiing-related data takes into account the weighted interaction features obtained based on the environmental data, that is, the weighted interaction feature vector is obtained through a dynamic weighting mechanism of environmental perception, it is possible to achieve scene-adaptive feature screening and fusion, improve the prediction accuracy in busy or harsh scenarios, reduce computational complexity, and enhance the collaborative efficiency with subsequent time series modeling modules.

[0060] S300 , using a convolution kernel with a width of k, performs a one-dimensional convolution operation on the feature vector of the simulated gliding trajectory data to extract m local features to form spatial features of m local paths.

[0061] In an embodiment of the present invention, a one-dimensional convolutional neural network can be used to perform a convolution operation on the local path of the feature vector of the simulated taxiing trajectory data. The purpose of using a convolutional neural network to extract features from the feature vector of the local path of the simulated taxiing trajectory data is to extract local features from the position sequence through a one-dimensional sliding window, helping the model identify turning patterns, path characteristics, and other complex geographical patterns in the path.

[0062] It is known to those skilled in the art that the number of partial paths is m=n-k+1. In the embodiment of the present invention, each partial path is composed of k consecutive position points. The spatial features corresponding to each partial path meet the following conditions:

[0063] LP i conv =σ conv (W conv LP i +b conv ).

[0064] Among them, LP i conv is the spatial feature of the i-th local path obtained by the convolutional neural network, W conv is the learnable weight matrix of the convolutional neural network, LP i is the feature vector corresponding to the ith local path, i.e., the feature vector formed by the mapping features from the ith position point to the i-k+1th position point, b conv is the bias term of the convolutional neural network.

[0065] S400 , obtaining the geometric distance of each local path, performing feature splicing on the spatial feature of each local path and the corresponding geometric distance, and using the splicing result as the enhanced spatial feature of the local path.

[0066] In the embodiment of the present invention, the geometric distance of the i-th local path , i ranges from 1 to m. Among them, is the geometric distance between the g-1th position point and the gth position point on the i-th local path, and the value of g is 2 to k.

[0067] S500 , fusing the enhanced spatial features of each local path with the feature vector of the sliding-related data to obtain a multimodal feature of each local path.

[0068] In the embodiment of the present invention, fusing the enhanced spatial features of each local path and the feature vector corresponding to the sliding association data is performing feature splicing on the enhanced spatial features of each local path and the feature vector corresponding to the sliding association data.

[0069] S600: Extract the temporal dependency relationship in the multimodal features of each local path to obtain corresponding spatiotemporal features.

[0070] In an embodiment of the present invention, the temporal dependencies in the multimodal features of each local path can be extracted through two gated recurrent units connected in series. Specifically, the temporal dependencies in the multimodal features of each local path can be extracted through a first gated recurrent unit and a second gated recurrent unit connected in sequence.

[0071] In an embodiment of the present invention, the first gated recurrent unit is used to perform preliminary temporal modeling on the input features to capture the short-term and long-term dependency information in the sequence, and the second gated recurrent unit is used to further model the features processed by the first gated recurrent unit to extract more complex temporal patterns to more accurately describe the changes in the gliding time.

[0072] Those skilled in the art know that the specific structure and working principle of the gated recurrent unit may be prior art.

[0073] S700: Use the attention mechanism to perform weighted processing on all spatiotemporal features to obtain global spatiotemporal features.

[0074] In this embodiment of the present invention, the global spatiotemporal features can be obtained by the following steps:

[0075] S701: Perform a linear transformation on the feature vector of the taxiing-related data to obtain a corresponding query vector.

[0076] In this embodiment of the present invention, the query vector satisfies the following conditions:

[0077] q=σ att (W q •F c +b q ).

[0078] Where q is the query vector, F c is the characteristic vector of the sliding correlation data, Wq is the linear transformation weight matrix, b q is the bias term. att ( ) is a nonlinear activation function. In an exemplary embodiment, σ att ( ) can be the tanh activation function.

[0079] S702: Obtain the interaction strength between the spatiotemporal features of each local path and the query vector.

[0080] In this embodiment of the present invention, the interaction strength z(i) corresponding to the i-th local path satisfies the following conditions:

[0081] z(i)=(q T •h i ) / (d k ) 1 / 2 .

[0082] Among them, h i is the spatiotemporal characteristics of the i-th local path, q T is the transpose of q, d k is the dimension of q;.

[0083] S703: Normalize the m interaction intensities to obtain the attention weight corresponding to the spatiotemporal features of each local path.

[0084] In this embodiment of the present invention, the attention weight α corresponding to the spatiotemporal feature of the i-th local path is i The following conditions must be met:

[0085] , e is a natural constant, and j ranges from 1 to m.

[0086] S704: Perform weighted summation on the spatiotemporal features of all local paths and the corresponding attention weights to obtain the global spatiotemporal features.

[0087] In this embodiment of the present invention, the global spatiotemporal feature Hatt satisfies the following conditions: .

[0088] S800: Process the global spatiotemporal features to obtain a predicted value of the taxiing time of the target flight.

[0089] In an embodiment of the present invention, the global spatiotemporal features can be processed using a residual fully connected network to obtain a predicted value of the glide time corresponding to the current dataset to be processed. Those skilled in the art will appreciate that the specific structure and operating principle of the residual fully connected network may be conventional.

[0090] The data processing method for obtaining taxi times for departing flights provided in an embodiment of the present invention predicts taxi times by integrating taxi trajectory data and taxi-related data. This method accounts for influencing factors such as geographic space, temporal dynamics, and environmental interactions, resulting in more accurate taxi time predictions. Furthermore, by combining trajectory data and taxi-related data, spatiotemporal features are extracted from local paths, and critical local paths are dynamically identified. This improves taxi time prediction accuracy and provides data support for airports optimizing flight rollout sequences and resource allocation decisions.

[0091] In an embodiment of the present invention, the data processing method for obtaining the taxiing time of a departing flight provided in an embodiment of the present invention can be implemented based on a trained taxiing time prediction model.

[0092] like Figure 2 As shown, the glide time prediction model may include: a feature encoding module 1, a spatial feature extraction module 2, a feature enhancement module 3, a multimodal feature fusion module 4, a temporal modeling module 5, an attention weighting module 6, and a prediction output module 7. The spatial feature extraction module 2, the feature enhancement module 3, the multimodal feature fusion module 4, the temporal modeling module 5, the attention weighting module 6, and the prediction output module 7 are sequentially connected, and the data processing module 1 is respectively connected to the spatial feature extraction module, the multimodal feature fusion module, and the attention weighting module.

[0093] In the embodiment of the present invention, the feature encoding module 1 is used to obtain the feature vector of the glide trajectory data and the feature vector of the glide association data.

[0094] The spatial feature extraction module 2 is used to perform a convolution operation on the feature vector corresponding to the sliding trajectory data to obtain the spatial features corresponding to each local path. The spatial feature extraction module can be a one-dimensional convolutional neural network.

[0095] The feature enhancement module 3 is used to perform feature splicing on the spatial features and geometric distances corresponding to each local path, and use the splicing result as the enhanced spatial feature of the local path.

[0096] The multimodal feature fusion module 4 is used to fuse the enhanced spatial features of each local path and the feature vector of the sliding-related data to obtain the multimodal features of each local path.

[0097] The temporal modeling module 5 is used to extract the temporal dependency in the multimodal features of each local path to obtain the spatiotemporal features of each local path. The temporal modeling module 5 may include a first gated recurrent unit and a second gated recurrent unit.

[0098] The attention weighting module 6 is used to use the attention mechanism to perform weighted processing on all spatiotemporal features to obtain global spatiotemporal features.

[0099] The prediction output module 7 processes the global spatiotemporal features to obtain a predicted value of the taxiing time of the target flight. The prediction output module 7 may be a residual fully connected network.

[0100] In a specific embodiment of the present invention, the convolution kernel size of the one-dimensional convolutional neural network is 3, and the number of convolution kernels is 32. The hidden layer size of the gated recurrent unit is 128. The number of residual fully connected layers is 4, and the size is 128. The optimization used in training is SGD, with a learning rate of 0.001, a batch size of 32, and 500 epochs.

[0101] In this embodiment of the present invention, the trained glide time prediction model can be obtained by the following steps:

[0102] S10, obtaining the original dataset and dividing the original dataset into a training set, a validation set, and a test set.

[0103] In the embodiment of the present invention, the ratio of the training set, the validation set, and the test set can be set based on actual needs.

[0104] In this embodiment of the present invention, the raw data set is obtained through aircraft Automatic Dependent Surveillance-Broadcast (ADS-B), Airport Collaborative Decision-Making (A-CDM) systems, and Aeronautical Meteorological Information Reports (METAR). METAR reports are generated every 30 to 60 minutes, correlating the aircraft's departure time with the most recent report and matching weather conditions with flight attributes, operating time identifiers, and flight interaction data.

[0105] Each data item in the original dataset may include the taxi trajectory data, taxi time, and taxi-related data for the corresponding departing flight. Furthermore, some features are derived through calculation and processing of the original data, such as the flight's operating time zone identifier in the taxi-related data and features corresponding to the flight interaction data.

[0106] S20, preprocessing the original data set to obtain a preprocessed original data set.

[0107] In an embodiment of the present invention, preprocessing the original data set may include:

[0108] (1) Standardize the original data set using the mean and standard deviation.

[0109] (2) Fill in missing values and remove extreme outliers.

[0110] In the embodiment of the present invention, the 3σ principle can be used to eliminate outliers. If the data follows a normal distribution, outliers are defined as values that deviate from the mean by more than three times the standard deviation. They will be considered outliers and deleted from the data set.

[0111] In the embodiment of the present invention, the mean and standard deviation of each non-discrete variable in the to-be-processed data set in the aforementioned content are the mean and standard deviation of the corresponding non-discrete variable in the original data set.

[0112] S30 , training the initial taxiing time prediction model using the training set and the validation set to obtain a trained taxiing time prediction model, and evaluating the trained taxiing time prediction model using the test set.

[0113] In an embodiment of the present invention, the initial taxiing time prediction model is trained using the training set and the validation set to obtain a trained taxiing time prediction model, which may specifically include the following steps:

[0114] S31, set the iteration counter C=1; set the initial value of the validation loss in the validation loss record set to infinity, and set the optimal model parameter storage path.

[0115] S32, if C≤C0, execute S33; otherwise, execute S50; C0 is a preset iteration number threshold, which can be an empirical value.

[0116] S33: Input the current batch of training data into the current feature encoding module for processing, obtain the feature vector of the glide trajectory data and the feature vector of the glide association data corresponding to each training data, and send them to the spatial feature extraction module and multimodal feature fusion module of the current glide time prediction model.

[0117] In the embodiment of the present invention, the method for obtaining the characteristic vector of the taxiing trajectory data and the characteristic vector of the taxiing association data corresponding to each training data can be referred to the above content. The initial value of the current taxiing time prediction model is the initialized taxiing time prediction model.

[0118] S34, using the current spatial feature extraction module to perform a convolution operation on the feature vector of the gliding trajectory data to obtain the spatial features of each local path, and send them to the feature enhancement module; and using the feature enhancement module to perform feature splicing on the spatial features of each local path and the corresponding geometric distance to obtain the enhanced spatial features of each local path, and send them to the multimodal feature fusion module.

[0119] S35: Using a multimodal feature fusion module, the enhanced spatial features of each local path and the feature vector of the sliding associated data are fused to obtain a multimodal feature of each local path, and the multimodal feature is sent to the time series modeling module.

[0120] S36, use the temporal modeling module to extract the multimodal features of each local path, obtain the corresponding spatiotemporal features, and send them to the attention weighting module.

[0121] S37, use the attention weighting module to perform weighted processing on all spatiotemporal features to obtain global spatiotemporal features, and send them to the prediction output module.

[0122] S38: Process the global spatiotemporal features using the prediction output module to obtain a predicted value of the coasting time corresponding to the current batch of training data.

[0123] S39, based on the predicted value of the glide time corresponding to the current batch of training data and the corresponding true value, obtain the loss of the current glide time prediction model as the current loss, and update the parameters of the current glide time prediction model based on the current loss; if the current batch of training data is the last batch of training data in the training set, execute S40, otherwise, use the next batch of training data as the training data of the current batch and execute S33.

[0124] In an embodiment of the present invention, the current loss may be obtained based on a preset loss function, such as a mean absolute error, and the parameters of the current coasting time prediction model may be updated using a stochastic gradient descent method.

[0125] S40, based on the validation set, obtain the average validation loss of the current coasting time prediction model as the current validation loss, and compare the current validation loss with the minimum validation loss in the current validation loss record set. If the current validation loss is less than the minimum validation loss in the current validation loss record, update the minimum validation loss in the current validation loss record set, and save the model parameters of the current coasting time prediction model to the optimal model parameter storage path; set C=C+1, and execute S32.

[0126] S50: The coasting time prediction model corresponding to the minimum verification loss in the current verification loss record set is used as the trained coasting time prediction model.

[0127] In an embodiment of the present invention, an independent test set is used to evaluate the trained taxi time prediction model. The mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination (R²), and different tolerance levels can be used to evaluate the accuracy of the model prediction.

[0128] Figure 3 A comparison chart of the predicted glide time and the actual glide time of the prediction model used in the method provided by the present invention and a diagram of the error distribution are shown. Figure 3The upper part of the figure shows the fit between the actual taxi time of the aircraft and the predicted taxi time. It can be seen that the taxi time curve predicted by the model has a similar trend to the actual taxi time curve, indicating that the model can capture the changing pattern of taxi time to a certain extent. In summary, the data processing method for obtaining taxi time of departing flights provided by the embodiment of the present invention, through deep fusion of multimodal data and deep learning architecture, can solve the technical bottlenecks of traditional taxi time prediction methods in fine-grained feature extraction and dynamic environment adaptability. Specifically, the present invention constructs a collaborative analysis framework for ADS-B trajectory data and multi-dimensional operation data of the A-CDM system, integrating multi-dimensional data such as aircraft taxi trajectory points, operation period identification values, flight attributes, and flight interaction data. The present invention adopts a one-dimensional convolutional neural network and combines it with great circle geometric distance to obtain an enhanced spatial feature map of the local path, realizing refined modeling of key local path spatial features such as airport taxi path intersections and bottleneck areas. At the same time, a two-layer gated recurrent unit network is used to dynamically capture the temporal evolution of the taxi process through a gating mechanism. Furthermore, the present invention introduces an attention mechanism to automatically identify and enhance features of coasting sections that significantly influence the prediction results based on external influencing factors, thereby increasing the model's focus on key sections. A residual fully connected network is used to deeply mine global spatiotemporal features and output coasting time predictions, further enhancing the accuracy of the predicted coasting time.

[0129] An embodiment of the present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are configured to execute the method described in the embodiment of the present invention.

[0130] An embodiment of the present invention further provides a computer-readable storage medium storing computer-executable instructions, wherein the computer instructions are used to execute the method described in the embodiment of the present invention.

[0131] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved. This is not limited herein.

[0132] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A data processing method for obtaining taxiing time of departing flights, characterized in that: The method comprises the following steps: S100, obtaining a data set to be processed; the data set to be processed includes simulated taxiing trajectory data and taxiing-related data of a target flight; S200, encoding the simulated taxiing trajectory data and the taxiing-related data to obtain a feature vector of the simulated taxiing trajectory data and a feature vector of the taxiing-related data; S300 , using a convolution kernel with a width of k, performs a one-dimensional convolution operation on the feature vector of the simulated taxiing trajectory data to extract m local features to form spatial features of m local paths; m = n - k + 1, where n is the number of position points in the simulated taxiing trajectory data; S400, obtaining the geometric distance of each local path, and performing feature splicing on the spatial features of each local path and the corresponding geometric distance, and using the splicing result as the enhanced spatial feature of the local path; S500, fusing the enhanced spatial features of each local path with the feature vector of the sliding-related data to obtain a multimodal feature of each local path; S600, extracting the temporal dependency relationship in the multimodal features of each local path to obtain the spatiotemporal features of each local path; S700: Use the attention mechanism to perform weighted processing on all spatiotemporal features to obtain global spatiotemporal features; S800: Process the global spatiotemporal features to obtain a predicted value of the taxiing time of the target flight.

2. The method according to claim 1, characterized in that The taxiing-related data includes flight attribute information, operating period identification value, environmental data and flight interaction data; In S200, the characteristic vector of the coasting-related data is obtained by the following steps: S201, pre-processing taxiing-related data to obtain initial feature vectors corresponding to flight attribute information, operating period identification values, environmental data, and flight interaction data; S202, performing feature concatenation on the initial feature vectors corresponding to the runtime identification value and the environment data, and using the concatenated result as a fused environment feature vector; S203, based on the fusion environment feature vector, obtaining a weight vector corresponding to the initial feature vector corresponding to the flight interaction data, and performing weighted processing on the initial feature vector corresponding to the flight interaction data based on the obtained weight vector to obtain a corresponding weighted processing result as a weighted interaction feature vector; S204 , performing feature splicing on the initial feature vector, the fused environment feature vector, and the weighted interaction feature vector corresponding to the flight attribute information, and using the obtained splicing result as the feature vector of the taxiing-related data.

3. The method according to claim 2, characterized in that The flight interaction data includes the number of flights active on the airport surface during the taxiing process of the target flight.

4. The method according to claim 3, characterized in that The weighted interaction eigenvector satisfies the following conditions: FW=(λ1f1,λ2f2,……,λ r f r ,……,l Q f Q ); Among them, FW is the weighted interaction feature vector, f r is the initial eigenvector corresponding to the rth interaction data, λ r is the weight corresponding to the rth interaction data, λ r Based on the fusion environment feature vector and f r Determine that r ranges from 1 to Q, where Q is the number of interaction data in the flight interaction data.

5. The method according to claim 4, characterized in that λ r The following conditions must be met: l r =σ(W e ·E env +W f ·f r +b); Among them, W e is the environmental feature weight matrix, E env is the fusion environment feature vector, W f f r The corresponding weight matrix, b is the bias term, σ( ) is the Sigmoid activation function, and · represents the dot product.

6. The method according to claim 1, characterized in that The global spatiotemporal features are obtained by the following steps: S701, performing a linear transformation on the feature vector of the taxiing-related data to obtain a corresponding query vector; S702, obtaining the interaction strength between the spatiotemporal features of each local path and the query vector, and obtaining m interaction strengths; S703, normalize the m interaction intensities to obtain the attention weight corresponding to the spatiotemporal features of each local path; S704: Perform weighted summation on the spatiotemporal features of all local paths and the corresponding attention weights to obtain the global spatiotemporal features.

7. The method according to claim 6, characterized in that in, The interaction strength z(i) corresponding to the i-th local path satisfies the following conditions: z(i)=(q T •h i ) / (d k ) 1 / 2 , h i is the spatiotemporal feature of the i-th local path, q is the query vector, d k is the dimension of q; T It is the transpose of q, and the value of i ranges from 1 to m.

8. The method according to claim 1, characterized in that In S600 , two gated recurrent units connected in series are used to extract the temporal dependencies in the multimodal features of each local path.

9. An electronic device, characterized in that: including processor and memory; The processor is configured to execute the steps of the method according to any one of claims 1 to 8 by calling the program or instructions stored in the memory.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store a program or instruction, and the program or instruction enables a computer to execute the steps of the method according to any one of claims 1 to 8.

Images