Delay warning methods and storage media for the entire transit flight process

Abstract

This invention provides a delay early warning method and storage medium for the entire process of transit flights. The method includes: obtaining the estimated time of preset stages and the estimated time of preset key nodes in the arrival process of the target flight based on actual airport operation data; balancing the actual airport operation data and identifying delayed flights based on the estimated time of the preset stages and the estimated time of the preset key nodes; and issuing an early warning for delayed flights based on the identification results. This approach, in addition to identifying and issuing early warnings for delayed flights, provides more reference options and more execution time for taking remedial measures.

Description

Technical Field

[0001] This invention relates to the field of air traffic control decision-making technology, and in particular to a delay early warning method and storage medium for transit flights throughout the entire process. Background Technology

[0002] Compared to originating and post-flight flights, transit flights involve more steps and require higher levels of airport support capabilities as well as better coordination among airports, airlines, air traffic control, and other stakeholders.

[0003] However, existing technologies for predicting delayed transit flights are mainly static and cannot provide decision support after the delayed transit flights are identified.

[0004] Therefore, there is an urgent need to provide a delay early warning method that can provide decision support for transit delay flights. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a delay early warning method for the entire process of transit flights. On the basis of identifying and warning of transit delayed flights, it provides more reference options and more execution time for taking remedial measures for transit delayed flights.

[0006] This invention provides a delay warning method for the entire process of transit flights, including:

[0007] Based on actual airport operation data, the estimated time for the preset steps of the target flight's arrival process and the estimated time of the preset key nodes are obtained;

[0008] Based on the estimated time of preset steps and the estimated time of preset key nodes, the actual airport operation data is balanced and delayed flights are identified.

[0009] Warnings are issued for delayed flights based on the identification results.

[0010] Optionally, the preset steps include: approach step, taxiing step, and station transfer step.

[0011] Optionally, the estimated time for the preset steps in the arrival process of the target flight can be obtained by establishing a key node moment prediction model.

[0012] Optionally, establishing the prediction model for key node moments specifically includes:

[0013] Obtain the set of factors affecting the time taken in the preset steps;

[0014] Based on the set of influencing factors, obtain the feature set of the key node moment prediction model;

[0015] A key node moment prediction model is established based on the feature set.

[0016] Optionally, the target flight's arrival process includes preset key nodes, including:

[0017] Arrival time, landing time, wheel chock engagement time, wheel chock disengagement time, takeoff time.

[0018] Optionally, obtaining the estimated times of preset key nodes in the target flight's arrival process specifically includes:

[0019] Obtain the target flight's experienced taxiing time;

[0020] Based on the experienced taxiing time of the target flight and the estimated time of the preset steps in the target flight's arrival process, the estimated time of the preset key nodes in the target flight's arrival process is obtained.

[0021] Optionally, obtaining the target flight's experienced taxiing time specifically includes:

[0022] The actual airport operation data is divided into two groups according to a preset ratio, one group is used as a data mining group and the other group is used as a result verification group.

[0023] Based on the data mining group, and by performing aggregate analysis on the data mining group according to the runway, aircraft position and time period, candidate values ​​representing empirical taxiing time are determined;

[0024] Substitute the candidate values ​​into the result verification group to calculate the mean absolute error value;

[0025] The target flight's experienced taxiing time is determined based on the mean absolute error value.

[0026] Optionally, obtaining the estimated times of preset key nodes in the target flight's arrival process based on the target flight's experienced taxiing time and the estimated time of preset stages in the target flight's arrival process includes:

[0027] The arrival process of the target flight is divided into three stages:

[0028] Phase 1: The starting time is the time when the target flight passes through the arrival point;

[0029] Phase Two: Starting from the landing time of the target flight;

[0030] Phase 3: Starting from the time when the target flight hits the wheel chock;

[0031] The above three stages simultaneously achieve the acquisition of the predicted times of key nodes in the target flight's arrival process.

[0032] Optionally, an imbalanced data classification model can be established to balance the actual airport operation data and identify delayed flights.

[0033] This invention provides a storage medium storing a computer program or instructions, which, when executed, implements the delay warning method described in any of the above embodiments.

[0034] The delay warning method for transit flights in this embodiment of the invention obtains the estimated time of preset stages and the estimated time of preset key nodes in the arrival process of the target flight based on the actual airport operation data. Based on the estimated time of preset stages and the estimated time of preset key nodes, the actual airport operation data is balanced and delayed flights are identified. Based on the identification results, a warning is issued for delayed flights. It can simultaneously identify the delay of the target flight and predict the time of subsequent key nodes based on different key nodes. Therefore, based on the identification and warning of delayed flights, it provides more reference options and more execution time for the implementation of delay rescue measures. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0036] Figure 1 The diagram illustrates the steps of a delay warning method for the entire transit flight process according to an embodiment of the present invention.

[0037] Figure 2 A flowchart illustrating an influencing factor analysis and extraction method according to an embodiment of the present invention is shown;

[0038] Figure 3 A schematic flowchart of an empirical gliding time statistics method according to an embodiment of the present invention is shown;

[0039] Figure 4 This diagram illustrates the implementation process of a key node moment prediction model in an embodiment of the present invention.

[0040] Figure 5 This diagram illustrates a first-stage delay warning process according to an embodiment of the present invention.

[0041] Figure 6 This diagram illustrates the implementation process of an imbalanced data classification model according to an embodiment of the present invention.

[0042] Figure 7 A schematic diagram of a delay warning process according to an embodiment of the present invention is shown;

[0043] Figure 8This invention illustrates a schematic diagram of the ROC curves for the three stages of flight arrival in an embodiment of the present invention.

[0044] Figure 9 This diagram illustrates a comparison between the recognition rate of delayed flights based on the key node time prediction model in the three stages of flight arrival according to an embodiment of the present invention and the recognition rate of delayed flights after introducing an imbalanced data classification model. Detailed Implementation

[0045] As described in the background section, existing technologies for predicting transit delays are mainly static and cannot provide effective remedial measures after identifying transit delays.

[0046] To address the aforementioned issues, this invention provides a delay warning method for the entire transit flight process. This method can identify and warn of delayed transit flights, and provide more reference options and more execution time for taking remedial measures for delayed transit flights.

[0047] To enable those skilled in the art to better understand and implement the embodiments of the present invention, the following describes the concept, scheme, principle, and advantages of the embodiments of the present invention in detail with reference to the accompanying drawings and through specific application examples.

[0048] Reference Figure 1 This invention provides a delay warning method for the entire process of transit flights, which includes the following steps:

[0049] S1: Based on actual airport operation data, obtain the estimated time of the preset steps in the arrival process of the target flight and the estimated time of the preset key nodes;

[0050] Specifically, a key node timing prediction model is constructed to predict the estimated time of the preset steps in the arrival process of the target flight and the estimated time of the preset key nodes.

[0051] The key node time prediction model is described as follows: For the target transit flight, the XGBoost model is used to predict the time of the preset links in the pre-arrival stage, providing feature input for the imbalanced data classification model, and the estimated time of the subsequent preset key nodes is calculated based on the starting point time to support the implementation of intervention measures for delayed flights.

[0052] The time taken for each stage in the later stage of the flight is obtained through an empirical taxiing time statistics method designed based on data mining technology. This provides feature input for the unbalanced data classification model. Based on the starting time and the prediction results of the previous stage, the estimated time of subsequent preset key nodes is calculated, which provides support for the implementation of intervention measures for delayed flights.

[0053] S1.1: Obtain the estimated time for the preset steps in the arrival process of the target flight;

[0054] Specifically, refer to Figure 2 The diagram illustrates a flowchart of an influencing factor analysis and extraction method. First, a set of influencing factors needed to establish a prediction model for key node moments is obtained. Specifically, the preset stages may include: approach, taxiing, and transit. Potential influencing factors are extracted from these three stages. For example, influencing factors for the approach stage may include: arrival point, arrival time, and landing runway; influencing factors for the taxiing environment may include: landing runway, aircraft stand, and surface traffic flow; and influencing factors for the transit stage may include: aircraft stand, support time, and surface traffic flow. The final set of influencing factors is determined by quantitatively analyzing and filtering the average time of these influencing factors in the three environments.

[0055] As a specific example, the final set of influencing factors for the three preset stages determined in this embodiment is shown in Table 1:

[0056] Table 1 Final Influencing Factors Set

[0057]

[0058]

[0059] Furthermore, based on the final set of influencing factors, features required for establishing a prediction model for key node times are extracted from the actual airport operation data to construct a feature set. Character information that the model cannot process in the feature set, such as airline, aircraft type and runway number, is converted into numerical information.

[0060] As a specific example, this embodiment takes the slide-in stage as an example, and the established feature set is shown in Table 2:

[0061] Table 2 Feature Set of Slide-in Phase Time Prediction Model

[0062]

[0063] Furthermore, a prediction model for flight approach time, taxiing time, and transit time is constructed based on the XGBoost algorithm.

[0064] Furthermore, the prediction model is optimized using a grid search method on the training feature set to obtain the optimal values ​​of the three key parameters of the time prediction model, and predictions are made on the test set to obtain the prediction results of approach time, slip-in time and transit time.

[0065] S1.2: Obtain the target flight's experienced taxiing time;

[0066] Specifically, the empirical taxiing time is defined as follows: a representative value of the taxiing time for a fixed "aircraft stand-runway" combination obtained through statistical analysis of historical airport operation data. This value is used as the taxiing time corresponding to this "aircraft stand-runway" combination during daily operation.

[0067] Reference Figure 3 The flowchart of the empirical taxiing time statistics method shown first divides the actual airport operation data into two groups in a 7:3 ratio, namely the data mining group and the result verification group, referring to the method of dividing the training set and test set in machine learning.

[0068] Then, based on the data mining team, the data was aggregated and grouped by runway, aircraft position and time period, and candidate values ​​representing empirical taxiing time were determined;

[0069] Next, the obtained candidate values ​​are substituted into the result validation group to calculate the mean absolute error (MAE).

[0070] Finally, the final value of the empirical gliding time was determined based on the MAE comparison results.

[0071] S1.3: Obtain the estimated time of preset key nodes after the target flight arrives;

[0072] Specifically, refer to Figure 4 The flowchart of the key node timing prediction model shown in Figure 3 and the abbreviations in the flowchart shown in Table 3, where the preset key nodes are as follows: Figure 4 As shown, it can include: arrival time (PAT), landing time (LDT), wheel chock time (IBT), wheel chock removal time (OBT), and takeoff time (TOT), dividing the target flight's arrival process into three stages: stage one starts with the target flight's actual arrival time (APAT), stage two starts with the target flight's actual landing time (ALDA), and stage three starts with the target flight's actual wheel chock time (AIBT).

[0073] During the flight arrival process, delayed flights are identified and subsequent key milestones are predicted simultaneously in the three stages mentioned above. The specific steps are as follows: Based on the key milestone prediction model, during the arrival process of the target flight, the estimated times of each key milestone after the target flight arrives are calculated by combining the results obtained in steps S1.1 and S1.2 at the times of passing the arrival point, landing, and engaging wheel chocks.

[0074] Table 3. Explanation of Abbreviations in the Flowchart of the Prediction Model for Key Node Moments

[0075]

[0076]

[0077] The following uses the first stage as an example, referring to... Figure 5 The flowchart for the first-stage delay warning shown below details the calculation method for the estimated time of each key node:

[0078] The estimated arrival time (ELDT) is obtained from the actual arrival time (APAT) of the flight based on the predicted approach time (PAT), as shown in equation (1):

[0079] APAT+PAT=ELDT (1)

[0080] The estimated shift point (EIBT) is obtained based on the empirical slip-in time (ETI), and the calculation method is shown in equation (2):

[0081] ELDT+ETI=EIBT (2)

[0082] The estimated time to remove wheel chocks (EOBT) is obtained based on the planned transit time (SCT), and the calculation method is shown in equation (3):

[0083] EIBT+SCT=EOBT (3)

[0084] The estimated takeoff time (ETOT) is obtained based on the empirical taxiway time (ETO), and the calculation method is shown in equation (4):

[0085] EOBT+ETO=ETOT (4)

[0086] S2: Balance the actual airport operation data and identify delayed flights;

[0087] Specifically, a feature set is constructed based on the extracted influencing factors related to flight delays and the estimated time of the preceding steps. Data is extracted based on the feature set and the delayed flights are identified after the data is balanced using an imbalanced data classification model.

[0088] The imbalanced data classification model is described as follows: The SMOTE algorithm is used to balance the data, which solves the problem of poor classification model accuracy caused by sample imbalance. Then, a classification model based on the CatBoost algorithm is used to identify delayed flights.

[0089] S2.1: Obtain the input dataset for the imbalanced data classification model;

[0090] Specifically, based on the feature set of each stage of flight arrival extracted by the influencing factor analysis and the estimated time of each link obtained by the key node time prediction model, an influencing factor set of the imbalanced data classification model for the three stages of flight arrival is constructed. The feature set extraction is based on existing research papers. By analyzing the flight arrival process and combining it with the actual operating experience of airport controllers, factors affecting flight regularity are obtained. The three stages are the first stage starting from the arrival point time, the second stage starting from the landing time, and the third stage starting from the wheel chock time. The feature set of the model is extracted from the corresponding fields in the original data according to the influencing factor set.

[0091] S2.2: Train an imbalanced data classification model based on the input dataset;

[0092] Specifically, refer to Figure 6 The diagram illustrates the implementation process of the imbalanced data classification model. The input dataset is divided into two parts: a training set and a test set. First, the SMOTE algorithm is used to generate minority class samples on the test set to balance the minority and majority class samples. Then, the CatBoost model is trained on the training set, and the grid search method is used to fine-tune the model parameters to obtain the optimal parameters of the CatBoost model.

[0093] S2.3: Use the CatBoost model to identify delayed flights.

[0094] Specifically, evaluation metrics for the classification model are determined, the classification performance of the model is evaluated, and the results are compared with the remaining time prediction results of transit flights obtained based on the key node time prediction model. The remaining time of the transit flights is calculated according to the following formula:

[0095] Phase 1: Approach time + Taxi entry time + Turnover time + Taxi exit time = Remaining time for the turnover flight;

[0096] Second stage: Taxi entry time + transit time + taxi exit time = remaining time for transit flights;

[0097] Phase 3: Turnover time + taxiing time = remaining time for the turnover flight.

[0098] S3.1: Select evaluation metrics for the classification model;

[0099] Specifically, precision, recall, and ROC curve (Receiver Operating Characteristic curve) can be selected to evaluate the classification performance of the model.

[0100] S3.2: Evaluate the predictive performance of the imbalanced data classification model itself;

[0101] Specifically, the model's ability to identify on-time and delayed flights can be evaluated in detail by calculating precision and recall; the overall performance of the model can also be evaluated by plotting ROC curves.

[0102] S3.3: Compare and evaluate the prediction performance of the imbalanced data classification model;

[0103] Specifically, the prediction results of the remaining time of transit flights obtained by the prediction model based on key node time are compared to compare the recognition rate of delayed flights at different stages.

[0104] S4: Issue warnings for delayed flights based on the identification results.

[0105] Specifically, refer to Figure 7 The flowchart shown is for delay warning.

[0106] To enable those skilled in the art to better understand and implement the embodiments of the present invention, the following describes the concept, scheme, principle, and advantages of the embodiments of the present invention in detail with reference to the accompanying drawings and through specific application scenarios.

[0107] Step A: Obtain actual operational data from an airport. First, preprocess the acquired dataset by removing outliers and imputing missing values. Then, model approach time, taxiing time, and transit time based on the preprocessed dataset. Simultaneously, use grid search to optimize the key parameters of the algorithm. The optimal values ​​of the key parameters for the three preset time prediction models are shown in Table 4, and the prediction results for the three times are shown in Table 5.

[0108] Table 4. Optimal values ​​for key parameters of the three preset time prediction models.

[0109]

[0110] Table 5. Predicted Approach Time, Trace-in Time, and Passage Time.

[0111]

[0112] As shown in Table 5, with an average approach time of 23.8 minutes, the predicted MAE is 2.08 minutes, achieving a prediction accuracy of 92.1% within a ±5-minute error range. With an average taxiing time of 12.1 minutes, the predicted MAE is 1.44 minutes, achieving a prediction accuracy of 91.5% within a ±3-minute error range. Under these accuracy conditions, the predicted landing time and predicted wheel chock time can provide decision support for airports to intervene in potentially delayed flights by ensuring the availability of scheduling and support resources. With an average turnaround time of 69.7 minutes, the predicted MAE is 4.75 minutes, achieving a prediction accuracy of 96.5% within a ±15-minute error range. This accuracy also provides more reliable input features for imbalanced data classification models.

[0113] Then, the MAE (Motor-Available Estimated Taxi Time) results obtained using the median and mean as candidate values ​​are shown in Table 6. Table 6 shows that the MAE obtained using the median for both taxiing entry and exit times is significantly smaller than the mean, and the error between the MAE and the average of both is within 18%. This level of accuracy meets the airport's operational requirements; therefore, the median is used as the final value for the empirical taxi time.

[0114] Table 6. Statistical Results of Experienced Glide Time (MAE)

[0115]

[0116] Step B: Next, the impact of airlines, aircraft types, arrival points, arrival / departure runways, and arrival / departure gates on flights is reflected; the impact of traffic flow on flights is reflected by time periods and the time taken for each stage based on the key node time prediction model; the third stage reflects the impact of preceding flight delays on subsequent flights by incorporating arrival delay duration. The feature sets of the three stages of flight arrival are shown in Table 7.

[0117] Table 7 Feature sets of the three-stage imbalanced data classification model

[0118]

[0119]

[0120] Step C: The SMOTE algorithm is then used to balance the dataset, and grid search is used to tune the parameters of the established CatBoost classification model. To increase the model's applicability, the optimal values ​​of the key parameters of the three-stage classification model were standardized. While ensuring the prediction performance at each stage, the optimal values ​​of the key parameters are shown in Table 8. The calculated precision and recall results for the three stages of flight arrival are shown in Table 9, and the ROC curves are shown in... Figure 8 As shown.

[0121] Table 8 Optimal values ​​of key parameters for the CatBoost model

[0122]

[0123] Table 9. Accuracy and Recall Rate Results for the Three Stages of Flight Arrival

[0124]

[0125] As shown in Table 9, during the arrival phase (stage 1), the model's recognition rates for on-time and delayed flights were 88.9% and 83.6%, respectively, both below 90%. However, during the landing phase (stage 2), the model's recognition rates for on-time and delayed flights improved to 93.0% and 90.4%, respectively, reaching a relatively high level compared to similar technologies. During the wheel chock phase (stage 3), the model's recognition rates for on-time and delayed flights reached 96.3% and 96.5%, respectively, with the recognition rate for delayed flights exceeding that of similar technologies by 4%. Figure 8 It can also be visually observed that from Phase 1 to Phase 3, the ROC curve gradually shifts towards the upper left corner, and the AUC (Area Under Curve) value (the area under the curve, ranging between 0 and 1) also gradually increases. Although the identification rate of delayed flights is relatively low in the first phase, unidentified flights can be further identified in subsequent phases. Therefore, this relatively independent yet progressive identification method can minimize the omission of delayed flights, while the implementation of phased intervention measures can also provide airports with greater buffer space.

[0126] Finally, the comparison between the identification rate of delayed flights based solely on the key node time prediction model and the identification rate of delayed flights after introducing an imbalanced data classification model is presented. Figure 9 As shown. From Figure 9 As can be seen, compared with the critical node moment prediction model, the introduction of the imbalanced data classification model has greatly improved the identification rate of delayed flights in all three stages, from the lowest 51.3% in stage one to 83.6%, and from the highest 72.1% in stage three to 96.5%.

[0127] Furthermore, to highlight the effectiveness of the present invention, Table 9 shows a comparison of the prediction accuracy of the present invention with that of the prior art.

[0128] Table 9 Comparison between this paper and existing technologies

[0129]

[0130] This invention provides a storage medium storing a computer program or instructions, which, when executed, implements the delay warning method described in any of the above embodiments.

[0131] While the embodiments of the present invention have been disclosed above, the present invention is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the present invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A delay early warning method for the entire process of transit flights, characterized in that, The method includes the following steps: S1: Based on actual airport operation data, obtain the estimated time for each preset stage of the target flight's arrival process and the estimated time of preset key nodes; specifically: S1.1: Obtain the estimated time for the preset steps in the arrival process of the target flight; First, the set of influencing factors needed to establish a prediction model for key node moments is obtained. Specifically, the preset stages include: approach stage, taxiing stage, and transit stage. The influencing factors for the approach stage include: arrival point, arrival time, and landing runway; the influencing factors for the taxiing stage include: landing runway, aircraft stand, and surface traffic flow; the influencing factors for the transit stage include: aircraft stand, support time, and surface traffic flow. By quantitatively analyzing and filtering the average time of the influencing factors in the above three preset stages, the final set of influencing factors is determined. S1.2: Obtain the target flight's experienced taxiing time; The definition of the empirical taxiing time is as follows: the representative value of the taxiing time for a fixed "stand-runway" combination obtained by statistical analysis of historical airport operation data, and this value is used as the taxiing time corresponding to the "stand-runway" combination during daily operation; First, referring to the method of dividing training and test sets in machine learning, the actual airport operation data was divided into two groups in a 7:3 ratio: a data mining group and a result verification group. Then, based on the data mining team, the data was aggregated and grouped by runway, aircraft position and time period, and candidate values ​​representing empirical taxiing time were determined; Next, the obtained candidate values ​​are substituted into the result verification group to calculate the mean absolute error value; Finally, the final value of the empirical gliding time is determined based on the comparison results of the mean absolute error value. S1.3: Obtain the estimated time of preset key nodes after the target flight arrives; The preset key nodes include: arrival point time (PAT), landing time (LDT), wheel chock time (IBT), wheel chock removal time (OBT), and takeoff time (TOT). The target flight's arrival process is divided into three stages: stage one starts with the target flight's actual arrival point time (APAT), stage two starts with the target flight's actual landing time (ALDA), and stage three starts with the target flight's actual wheel chock time (AIBT). During the flight arrival process, delayed flights are identified and the estimated times of subsequent preset key nodes are simultaneously realized in the above three stages. The specific steps are as follows: Based on the key node time prediction model, during the arrival process of the target flight, the estimated times of each preset key node after the target flight arrives are calculated by combining the results obtained in steps S1.1 and S1.2 at the time of passing the arrival point, the landing time and the time of the wheel chock. S2: Based on the estimated time of preset processes and the estimated time of preset key nodes, balance the actual airport operation data and identify delayed flights; specifically: S2.1: Obtain the input dataset for the imbalanced data classification model; Based on the feature set of each stage of flight arrival extracted by the influencing factor analysis and the estimated time of each stage obtained by the key node time prediction model, an influencing factor set of the imbalanced data classification model for the three stages of flight arrival is constructed, and the feature set of the model is extracted from the corresponding fields of the original data according to the influencing factor set. S2.2: Train an imbalanced data classification model based on the input dataset; The input dataset is divided into two parts: a training set and a test set. First, the SMOTE algorithm is used to generate minority class samples on the test set to balance the minority class and majority class samples. Then, the CatBoost model is trained on the training set, and the grid search method is used to fine-tune the model parameters to obtain the optimal parameters of the CatBoost model. S2.3: Use the CatBoost model to identify delayed flights; S3: Accuracy testing of imbalanced data classification models; S4: Issue warnings for delayed flights based on the identification results.

2. The delay early warning method for the entire process of transit flights according to claim 1, characterized in that, The method for calculating the estimated time of the preset key node is to obtain the estimated landing time ELDT based on the predicted approach time PAT at the actual arrival time APAT of the flight. The calculation method is shown in Equation (1): APAT+PAT=ELDT (1); The expected gear engagement time EIBT is obtained based on the empirical slip-in time ETI, and the calculation method is shown in equation (2): ELDT+ETI=EIBT (2); The estimated wheel chock removal time EOBT is obtained based on the planned transit time SCT, and the calculation method is shown in equation (3): EIBT+SCT=EOBT (3); The estimated takeoff time ETOT is obtained based on the empirical taxiway time ETO, and the calculation method is shown in equation (4): EOBT+ETO=ETOT (4).

3. A storage medium, characterized in that, It stores a computer program or instructions that, when executed, implement the method as described in claim 1.

Images