An XGBoost-based flight delay cause prediction method

Abstract

The application discloses a flight delay reason prediction method based on XGBoost, first, flight plan data, flow control data and meteorological data are collected, flight delay reason categories are divided according to the flow control data, various flight delay reason categories are associated with flight plan data and weather data of the flight, associated sample data is obtained, and the sample data is balanced to obtain balanced sample data; then, a flight delay reason prediction model based on XGBoost is constructed, characteristic factors of the flight plan in the balanced sample data and meteorological characteristic factors related to the flight are taken as input characteristic variables of the model, and the flight delay reason categories are taken as target variables of the model prediction output. Finally, the characteristic factors of the flight plan of the flight to be predicted and the meteorological characteristic factors related to the flight are input into the trained flight delay reason prediction model, and the target variable output by the model is the flight delay reason category.

Description

Technical Field

[0001] This invention relates to the field of air traffic management technology, and in particular to a method for predicting the causes of flight delays based on XGBoost. Background Technology

[0002] my country's civil aviation industry is currently experiencing rapid development. With the explosive growth in flight volume, and given the limited overall resources available to various civil aviation units, flight punctuality faces unprecedented challenges. The Civil Aviation Administration of China (CAAC) has significantly improved flight punctuality by actively exploring potential solutions and increasing technological investment to address flight delays. However, because flight operations involve multiple stakeholders—air traffic control, airlines, and airports—and the overall operational process is complex, involves numerous steps, and is subject to significant external disturbances, it is difficult to clearly identify the responsible party for flight delays and predict their causes. This has a substantial impact on flight punctuality management. Accurately predicting the causes of flight delays is crucial for improving flight management. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for predicting the causes of flight delays based on XGBoost.

[0004] This invention is achieved through the following technical solution:

[0005] A method for predicting flight delay causes based on XGBoost includes the following steps:

[0006] S1: Collect flight operation data, including flight plan data, air traffic control data, and meteorological data;

[0007] S2: Based on the flow control data, classify the causes of flight delays into categories, and correlate these categories with the flight plan data and weather data to obtain correlated sample data;

[0008] S3: Based on the classification of flight delay reasons, count the number of delayed flights associated with each type of flight delay reason, and use the ADASYN algorithm to perform data balancing on the sample data obtained in step S2 to obtain balanced sample data.

[0009] S4: Construct a flight delay cause prediction model based on XGBoost. Use the characteristic factors of flight plans and meteorological characteristic factors related to flights in the balanced sample data obtained in step S3 as the input feature variables of the model, and use the flight delay cause category as the target variable of the model prediction output. Divide the balanced sample data in step S3 into training set and validation set to train and validate the flight delay cause prediction model to obtain the trained flight delay cause prediction model.

[0010] S5: Input the characteristic factors of the flight plan of the flight to be predicted and the meteorological characteristic factors related to the flight into the flight delay cause prediction model trained in step S4. The target variable output by the model is the flight delay cause category.

[0011] In the above technical solution, step S1 includes flight plan data including: flight number, registration number, aircraft type, flight date, departure airport, landing airport, planned departure time, actual departure time, planned landing time, and actual landing time; flow control data includes: flight number, registration number, flow control reason, flow control content, etc.; weather data includes: observation time, weather type, visibility, cloud base height, wind direction, wind speed, etc., where weather type includes thunderstorm, rain, snowfall, and fog.

[0012] In the above technical solution, step S2 involves classifying the causes of flight delays based on the flow control reasons and content in the flow control data.

[0013] In the above technical solution, step S2 categorizes flight delays into five types: weather-related, airline-related, air traffic control-related, airport-related, and other reasons.

[0014] In the above technical solution, step S2 involves associating the flight plan data information of the flight with the classified flight delay reason categories based on the flight number and registration number, and associating the flight plan data of the flight with the corresponding weather data based on the time information, thereby realizing the association between the flight delay reason categories and the flight plan data and weather data.

[0015] In the above technical solution, step S3 includes the following steps:

[0016] S3.1: Based on the reasons for flight delays, count the number of delayed flights associated with each type of flight delay reason:

[0017] The formula for calculating the number of delayed flights (G) associated with various reasons for flight delays is as follows:

[0018] G=(m l -m s )×β

[0019] Where G represents the number of delayed flights associated with each type of flight delay reason, and m l For a large number of data categories, m s For a smaller number of data categories, β is a random number in the range [0,1].

[0020] S3.2: Based on the number of samples of the majority class in the nearest neighbors, calculate the proportion of the class with a larger number of samples in the nearest neighbors; the formula for calculating the proportion is as follows:

[0021] ri =Δ i / K

[0022] Where, r i Δ represents the proportion of the more numerous categories among the nearest neighbors. i For x i The number of majority class samples among the K nearest neighbors, i = 1,...,m s x i For the i-th sample data in a relatively small number of categories, x i There are K nearest neighbors;

[0023] S3.3: The proportion r of the more numerous categories among the nearest neighbors i Standardization is performed, and the standardization formula is as follows:

[0024]

[0025] in, The proportion of a larger number of categories in the standardized nearest neighbors;

[0026] S3.4: Based on the weight of the smaller number of categories, calculate the number of new data to be generated for each smaller number of categories, using the following formula:

[0027]

[0028] Among them, g i For each sample x in the minority category i The number of synthetic samples to be generated;

[0029] S3.5: Calculate the number of new data to be generated for each of the smaller quantity categories, one by one;

[0030] The method for generating the new data is as follows:

[0031]

[0032] in, For synthetic data, u = 1,...,g i x zi For x i A small number of class samples are randomly selected from the K nearest neighbors, where λ is a random number in the range [0,1]; the synthesized data Combined with the original data, the resulting balanced sample data is formed.

[0033] In the above technical solution, the characteristic factors of the flight plan include: flight number, aircraft type, flight date, and planned departure time; the meteorological characteristic factors related to the flight include: visibility, cloud base height, wind direction, wind speed, thunderstorm, rainfall, snowfall, and fog.

[0034] In the above technical solution, the method for constructing a flight delay cause prediction model based on XGBoost is as follows:

[0035] First, define the objective function of the initial model:

[0036] Among them, Obj (t) Let be the objective function of the model, t be the t-th tree, and N be the number of samples. Let Ω(f) be the error between the predicted and actual values ​​of the target variable in the data sample of the cause of flight delay for the nth flight. t ) is the complexity function of the tree;

[0037] The complexity function expression for a tree is as follows:

[0038]

[0039] Where γ is the complexity parameter, T is the number of leaf nodes in the tree, η is a fixed coefficient, and w j Let be the score of the j-th leaf node;

[0040] By combining the bias and variance functions, and using an additive strategy that takes the error between the previous prediction and the actual value from the reference model as input when building the next tree, the objective function is derived. The expression for the objective function is as follows:

[0041]

[0042] We choose the second-order Taylor expansion method to implement gradient descent, which reduces the model error. The specific formula is as follows:

[0043]

[0044] make Substituting this into the above objective function expression, we get:

[0045]

[0046] In the formula, d n The gradient value h for each flight delay sample. n This is the second derivative for each flight delay sample.

[0047] The advantages and beneficial effects of this invention are as follows:

[0048] The XGBoost algorithm of this invention has a good effect on predicting delay causes in multiple categories and has a strong ability to distinguish them. When judging each specific type of flight delay cause, it can make relatively accurate predictions and effectively improve the early warning and management capabilities of flight delays. Attached Figure Description

[0049] Figure 1 The diagram shows the overall architecture of the XGBoost-based method for predicting flight delay causes according to the present invention.

[0050] For those skilled in the art, other related figures can be obtained from the above figures without any creative effort. Detailed Implementation

[0051] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0052] A method for predicting flight delay causes based on XGBoost, see [link to relevant documentation]. Figure 1 The steps of the method include:

[0053] S1: Collect flight operation data.

[0054] The flight operation data includes flight plan data, air traffic control data, and meteorological data.

[0055] The flight plan data includes: flight number, registration number, aircraft type, flight date, departure airport, arrival airport, planned departure time, actual departure time, planned arrival time, and actual arrival time; the flow control data includes: flight number, registration number, reason for flow control, and flow control content; the weather data includes: observation time, weather type, visibility, cloud base height, wind direction, and wind speed, among which the weather type includes thunderstorms, rain, snowfall, and fog.

[0056] S2: Classify the causes of flight delays based on flow control data, and associate the classified flight delay causes with flight plan data and weather data to obtain associated sample data.

[0057] Specifically, based on the reasons and content of flow control data, flight delay reasons are categorized. In this invention, flight delay reasons are divided into five categories: weather-related, airline-related, air traffic control-related, airport-related, and other reasons. These five categories are then linked to flight plan data and weather data. That is, flight plan data can be associated with the identified flight delay reason categories based on flight number and registration number, and flight plan data can be associated with corresponding weather data based on time information, thus achieving a correlation between flight delay reason categories and flight plan and weather data.

[0058] S3: Analyze the number of flight delays based on the categorized reasons for flight delays. Since the number of flight delays corresponding to different categories can vary significantly, the sample data may be imbalanced, affecting the training effect of the subsequent prediction model. Therefore, this invention uses the ADASYN algorithm (Adaptive Synthetic) to balance the sample data obtained in step S2, resulting in balanced sample data.

[0059] The ADASYN algorithm is used to address the data imbalance problem. This involves assigning appropriate weights to different minority class samples, resulting in varying numbers of samples to achieve a relatively balanced model data. Specifically, this includes the following steps:

[0060] S3.1: Based on the reasons for flight delays, count the number of delayed flights associated with each type of flight delay reason:

[0061] The formula for calculating the number G of delayed flights associated with each type of flight delay reason is as follows:

[0062] G=(m l -m s )×β

[0063] Where G represents the number of delayed flights associated with each type of flight delay reason, and m l For a large number of data categories, m s For a smaller number of data categories, β is a random number in the range [0,1].

[0064] S3.2: Calculate the proportion of the majority class among the nearest neighbors based on the number of samples of the majority class among the nearest neighbors.

[0065] The formula for calculating the ratio is as follows:

[0066] r i =Δ i / K

[0067] Where, r i Δ represents the proportion of the more numerous categories among the nearest neighbors. i For x i The number of majority class samples among the K nearest neighbors, i = 1,...,m s x i For the i-th sample data in a relatively small number of categories, x i There are K nearest neighbors.

[0068] S3.3: The proportion r of the more numerous categories among the nearest neighbors i Standardize it.

[0069] The standardized formula is as follows:

[0070]

[0071] in, The proportion of a larger number of categories in the standardized nearest neighbors.

[0072] S3.4: Calculate the number of new data to be generated for each of the smaller number categories based on their weights.

[0073] The formula for calculating the number of new data is as follows:

[0074]

[0075] Among them, g i For each sample x in the minority category i The number of synthetic samples to be generated.

[0076] S3.5: Calculate the number of new data to be generated for each of the smaller number categories.

[0077] The method for generating the new data is as follows:

[0078]

[0079] in, For synthetic data, u = 1,...,g i x zi For x i A small number of class samples are randomly selected from the K nearest neighbors, where λ is a random number in the range [0,1]; the synthesized data Combined with the original data, the resulting balanced sample data is formed.

[0080] S4: Construct an XGBoost-based model for predicting flight delay causes. Use the characteristic factors of flight plans and related meteorological factors from the balanced sample data in step S3 as input variable features, and the category of flight delay causes as the target variable for the model's prediction output. Divide the balanced sample data from step S3 into training and validation sets to train and validate the flight delay cause prediction model.

[0081] Specifically, the characteristic factors of the flight plan include: flight number, aircraft type, flight date, and scheduled departure time; the meteorological characteristic factors related to the flight include: visibility, cloud base height, wind direction, wind speed, thunderstorms, rainfall, snowfall, and fog.

[0082] The method for constructing a flight delay cause prediction model based on XGBoost is as follows:

[0083] First, define the objective function of the initial model:

[0084] Among them, Obj (t) Let be the objective function of the model, t be the t-th tree, and N be the number of samples. Let Ω(f) be the error between the predicted and actual values ​​of the target variable in the data sample of the cause of flight delay for the nth flight. t ) is the complexity function of the tree.

[0085] The complexity function expression for a tree is as follows:

[0086]

[0087] Where γ is the complexity parameter, T is the number of leaf nodes in the tree, η is a fixed coefficient, and w j Let be the score of the j-th leaf node.

[0088] By combining the bias and variance functions, and using an additive strategy that takes the error between the previous prediction and the actual value from the reference model as input when building the next tree, the objective function is derived. The expression for the objective function is as follows:

[0089]

[0090] We choose the second-order Taylor expansion method to implement gradient descent, which reduces the model error. The specific formula is as follows:

[0091]

[0092] make Substituting this into the above objective function expression, we get:

[0093]

[0094] In the formula, d n The gradient value h for each flight delay sample. n This is the second derivative for each flight delay sample.

[0095] After the flight delay cause prediction model is built, the training set and validation set are used to classify and predict the cause of the delay. The prediction effect of the model is verified by precision, recall and F1-score. Finally, a flight delay cause prediction model with satisfactory accuracy is obtained.

[0096] S5: Input the characteristic factors of the flight plan of the flight to be predicted and the meteorological characteristic factors related to the flight into the flight delay cause prediction model trained in step S4. The target variable output by the model is the flight delay cause category.

[0097] The present invention has been described above by way of example. It should be noted that any simple modifications, alterations or other equivalent substitutions that can be made by those skilled in the art without creative effort without departing from the core of the present invention fall within the protection scope of the present invention.

Claims

1. A method for predicting flight delay causes based on XGBoost, characterized in that, Includes the following steps: S1: Collect flight operation data, including flight plan data, air traffic control data, and meteorological data; S2: Based on the flow control data, classify the causes of flight delays into categories, and correlate these categories with the flight plan data and weather data to obtain correlated sample data; S3: Based on the classification of flight delay reasons, count the number of delayed flights associated with each type of flight delay reason, and use the ADASYN algorithm to perform data balancing on the sample data obtained in step S2 to obtain balanced sample data; flight delay reasons are divided into five categories: weather reasons, airline reasons, air traffic control reasons, airport reasons, and other reasons; S4: Construct a flight delay cause prediction model based on XGBoost. Use the characteristic factors of flight plans and meteorological characteristic factors related to flights in the balanced sample data obtained in step S3 as the input feature variables of the model, and use the flight delay cause category as the target variable of the model prediction output. Divide the balanced sample data in step S3 into training set and validation set to train and validate the flight delay cause prediction model to obtain the trained flight delay cause prediction model. S5: Input the characteristic factors of the flight plan of the flight to be predicted and the meteorological characteristic factors related to the flight into the flight delay cause prediction model trained in step S4. The target variable output by the model is the flight delay cause category.

2. The method for predicting flight delay causes based on XGBoost according to claim 1, characterized in that: Step S1: Flight plan data includes: flight number, registration number, aircraft type, flight date, departure airport, arrival airport, planned departure time, actual departure time, planned landing time, and actual landing time; flow control data includes: flight number, registration number, reason for flow control, and flow control content information; weather data includes: observation time, weather type, visibility, cloud base height, wind direction, and wind speed information, where weather type includes thunderstorms, rain, snowfall, and fog.

3. The method for predicting flight delay causes based on XGBoost according to claim 1, characterized in that: Step S2: Based on the flow control reasons and flow control content in the flow control data, classify the categories of flight delay reasons.

4. The method for predicting flight delay causes based on XGBoost according to claim 1, characterized in that: Step S2: Associate the flight plan data information of the flight with the classified flight delay reason category based on the flight number and registration number, and associate the flight plan data with the corresponding weather data based on the time information, thereby realizing the association between the flight delay reason category and the flight plan data and weather data.

5. The method for predicting flight delay causes based on XGBoost according to claim 1, characterized in that: Step S3 includes the following steps: S3.1: Based on the reasons for flight delays, count the number of delayed flights associated with each type of flight delay reason: The number of delayed flights associated with various reasons for flight delays The calculation formula is as follows: in, The number of delayed flights is associated with various reasons for flight delays. For a large number of data categories, For a smaller number of categories of data, A random number in the range [0,1]. S3.2: Calculate the proportion of the more numerous category among the nearest neighbors based on the number of samples of that category; the formula for calculating the proportion is as follows: in, The proportion of the larger number of categories among the neighbors. for of K The number of samples from a larger number of neighboring categories , For a smaller number of categories of data, the first One sample data, exist K A close neighbor; S3.3: The proportion of the most numerous categories among the nearest neighbors Standardization is performed, and the standardization formula is as follows: in, The proportion of a larger number of categories in the standardized nearest neighbors; S3.4: Based on the weights of the fewer categories, calculate the number of new data to be generated for each of the fewer categories, using the following formula: in, For each sample in a smaller number of categories The number of synthetic samples to be generated; S3.5: Calculate the number of new data to be generated for each of the fewer categories, based on the number of categories to be generated; The method for generating the new data is as follows: in, For synthesized data, , for of A small number of sample data from each class are randomly selected from the nearest neighbors. The synthesized data is a random number in the range [0,1]. Combined with the original data, the resulting balanced sample data is formed.

6. The method for predicting flight delay causes based on XGBoost according to claim 1, characterized in that: The flight plan features include: flight number, aircraft type, flight date, and scheduled departure time; the meteorological features related to the flight include: visibility, cloud base height, wind direction, wind speed, thunderstorms, rainfall, snowfall, and fog.

7. The method for predicting flight delay causes based on XGBoost according to claim 1, characterized in that: The method for constructing a flight delay cause prediction model based on XGBoost is as follows: First, define the objective function of the initial model: in, The objective function of the model is... For the first A tree, For the number of samples, For the first n The error between the predicted and actual values ​​of the target variable in a sample of flight delay cause data. Let be the complexity function of the tree; The complexity function expression for a tree is as follows: in, For complexity parameters, The number of leaf nodes in the tree. For fixed coefficients, For the first j The scores of each leaf node; By combining the bias and variance functions, and using an additive strategy that takes the error between the previous prediction and the actual value from the reference model as input when building the next tree, the objective function is derived. The expression for the objective function is as follows: We choose the second-order Taylor expansion method to implement gradient descent, which reduces the model error. The specific formula is as follows: make , , , Substituting this into the above objective function expression, we get: In the formula, The gradient value for each flight delay sample. This is the second derivative for each flight delay sample.

Images