A hub passenger travel change intention prediction method under a journey delay scenario

Abstract

This invention relates to the field of intelligent traffic management and aviation emergency dispatching technology, and discloses a method for predicting the travel change intentions of hub passengers in the event of travel delays. The method first constructs a three-dimensional variable system encompassing passenger travel characteristics, sociodemographic attributes, and perceived service quality, and generates experimental schemes for delay scenarios through orthogonal design. Next, it integrates stated and revealed preferences survey data, and trains a passenger behavior prediction model using a parameter-tuned gradient boosting tree machine learning model. Subsequently, it employs the Shapley value additive interpretation algorithm to calculate the marginal contribution of each variable and identify key influencing features. Finally, it groups passengers according to the feature value ranges and matches differentiated emergency dispatching strategies. This invention can accurately quantify the impact of factors such as ticket prices and delay duration, supporting airports in developing targeted evacuation plans and improving the efficiency and resilience of services in delay scenarios.

Description

Technical Field

[0001] This invention relates to the fields of intelligent traffic management and aviation emergency dispatch technology, and in particular to a method for predicting the travel change intentions of hub passengers in the event of travel delays. Background Technology

[0002] Flight delays are a recurring and unavoidable problem in the air transport system. They not only cause significant economic losses but also severely impact passenger travel experiences and lead to chaos within airports. Currently, most hub airports primarily adopt a passive response mechanism when dealing with flight delays: issuing announcements via broadcasts or displays after a delay occurs and then waiting for passengers to go to the counters to process refunds and rebookings. This traditional management model ignores the significant differences among individual passengers, assuming that all passenger needs are homogeneous, making it difficult to allocate scarce emergency resources to where they are most needed.

[0003] Business travelers may urgently need to rebook to save time, while leisure travelers may be more concerned with financial compensation, and elderly travelers may face difficulties due to operational inconveniences. Existing prediction methods are mostly based on simple statistical regression models or single-dimensional questionnaires. These methods have significant limitations when dealing with high-dimensional data and struggle to capture the complex interactions between traveler characteristics, sociodemographic attributes, and perceived service quality. While traditional econometric models offer some interpretability, their linear assumptions often fail to characterize the non-linear characteristics of traveler decision-making in delay scenarios. For example, ticket prices have significantly different impacts on different income groups, and the tolerance for delay duration interacts in complex ways with age.

[0004] Furthermore, existing technologies lack a systematic approach to transforming behavioral predictions into actionable strategies. Even when passenger behavior can be predicted, it's difficult to develop differentiated emergency response plans. Airport managers often rely solely on experience, lacking quantitative data to support resource allocation decisions, leading to inefficient emergency response and severe service homogenization. Therefore, there is an urgent need for a technological solution that can deeply analyze passenger behavior mechanisms, quantify the weights of various influencing factors, and generate targeted evacuation strategies to address the current technical challenges of low prediction accuracy and severe service homogenization in delay management. Summary of the Invention

[0005] The technical problem to be solved by this invention is to address the shortcomings of the existing technology by providing a method for predicting the travel change intentions of hub passengers in the context of travel delays. This method constructs a three-dimensional variable system covering passenger travel characteristics, sociodemographic attributes, and perceived service quality, and combines it with joint survey data on stated preferences and revealed preferences. It uses a gradient boosting tree model to mine nonlinear decision-making patterns and employs Shapley value attribution analysis to identify key influencing features. Finally, it generates differentiated emergency dispatch strategies for different passenger groups, realizing a shift from passive response to proactive guidance.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] A method for predicting the travel change intentions of hub passengers in a trip delay scenario includes the following steps:

[0008] Step S1: Construct a three-dimensional variable system of influencing factors, including passenger travel characteristics, sociodemographic attributes, and perceived service quality. Based on the variable system, use orthogonal design to generate experimental schemes for delay scenarios, including delay duration, weather conditions, food and accommodation compensation, and rebooking fee losses.

[0009] Step S2: Conduct a statement preference survey based on the delay scenario experimental scheme, and obtain the actual travel records of the respondents by combining the disclosure preference survey. The statement preference survey data and the disclosure preference survey data are jointly processed, and after ordinal encoding and missing value imputation, they are divided into training set and test set according to a preset ratio.

[0010] Step S3: Construct a gradient boosting tree machine learning model, and use the training set to fine-tune and train the gradient boosting tree machine learning model to obtain a passenger behavior prediction model that can output prediction results of four behavioral intentions: changing trains, waiting for the original flight, canceling the trip, and changing to other flights.

[0011] Step S4: Input the test set into the passenger behavior prediction model, perform attribution analysis on the output of the passenger behavior prediction model, calculate the marginal contribution of each variable in the variable system to the four behavioral intentions, and select the top 10% of variables with the highest marginal contribution as key influencing features.

[0012] Step S5: Based on the nonlinear mapping relationship between the key impact features and the four behavioral intentions, generate differentiated emergency dispatch strategies for different passenger groups.

[0013] Furthermore, in step S1, the three-dimensional influencing factor variable system includes:

[0014] The dimensions of passenger travel characteristics include ticket price, travel purpose, estimated flight time, frequency of intercity travel, number of trips via the hub, time of arrival at the airport in advance, and cabin class.

[0015] The sociodemographic attributes include job type, monthly income, number of cars owned, gender, age, and education level.

[0016] The perceived service quality dimensions include airport transfer convenience, rest area satisfaction, indoor environment rating, staff assistance satisfaction, refund and change convenience, and food and beverage price satisfaction.

[0017] Furthermore, in step S1, the delay scenario experiment scheme generates multiple selection tasks through orthogonal combination. Each selection task provides four behavioral intention options: transferring to another train, waiting for the original flight, canceling the trip, and rebooking another flight. The delay duration is set to three levels: less than 2 hours, 2 to 6 hours, and more than 6 hours. The weather conditions are set to two levels: sunny and rainy / snowy. The food and accommodation compensation is set to three levels: free, 50% off, and 30% off. The rebooking fee loss is set to three levels: less than 400 yuan, 400 to 600 yuan, and more than 600 yuan.

[0018] Furthermore, in step S2, the joint processing of the stated preference survey data and the revealed preference survey data specifically includes:

[0019] The respondents’ choice preferences in the hypothetical delay scenario in the stated preference survey data are correlated and matched with the respondents’ historical actual travel records in the revealed preference survey data.

[0020] Ordinal encoding is performed on the categorical variables in the associated data to map text features to... The initial integer sequence;

[0021] according to to The proportion will divide the processed data into the training set and the test set.

[0022] Furthermore, in step S3, the gradient boosting tree machine learning model is implemented using the Extreme Gradient Boosting (XGBoost) algorithm, and the objective function of the XGBoost algorithm is calculated according to the following formula (1):

[0023] (1)

[0024] in, Indicates the number of iterations. Indicates the number of samples. Represents the loss function. Indicates the first The actual travel choices of passengers. Indicates the preceding Predicted value after the next iteration Indicates the first The feature vector of each sample Indicates the first In the nth iteration tree to the first The predicted increment for each sample This represents the regularization term.

[0025] Furthermore, in step S3, in order to optimize the objective function, in the first... In the next iteration, formula (1) is approximated by a second-order Taylor expansion, resulting in formula (2):

[0026] (2)

[0027] in, This represents the first derivative of the loss function with respect to the predicted values. This represents the second derivative of the loss function with respect to the predicted values;

[0028] The regularization term Calculate according to formula (3):

[0029] (3)

[0030] in, This represents the penalty coefficient for the leaf nodes. Indicates the first The number of leaf nodes in a tree. The L2 regularization coefficient represents the weight of the leaf nodes. Indicates the first The weight of each leaf node, This indicates that a penalty is applied to the sum of the squares of the weights of all leaf nodes to prevent overfitting.

[0031] Furthermore, in step S3, the specific steps for parameter optimization include:

[0032] Define a hyperparameter grid including column sampling ratio, learning rate, maximum tree depth, number of estimators, and sample sampling ratio, and set the value range for each hyperparameter;

[0033] The training set is divided into several folds, and a cross-validation method is used to select a portion of the data in each fold as the validation set and the remaining data as the training subset.

[0034] Iterate through all parameter combinations in the hyperparameter grid, train the model on the training subset for each parameter combination, and compute the cross-validation score on the validation set;

[0035] The parameter combination that yields the optimal cross-validation score is selected as the optimal hyperparameter combination.

[0036] Furthermore, in step S4, the attribution analysis is implemented using the SHAP algorithm based on Shapley values, specifically including:

[0037] Construct a SHAP interpreter object to calculate the SHAP value of each variable for each sample in the test set in the passenger behavior prediction model;

[0038] Calculate the average of the absolute values ​​of the SHAP of each variable across all samples, and use this as an indicator of the global importance of that variable.

[0039] All variables are sorted in descending order based on the global importance index.

[0040] Furthermore, in step S5, generating differentiated emergency dispatch strategies for different passenger groups specifically includes:

[0041] Passengers are grouped according to the price range of the key influencing features. Passengers whose prices are below the first preset threshold are classified as price-sensitive groups, and passengers whose prices are between the first preset threshold and the second preset threshold are classified as medium-priced groups.

[0042] Passengers are grouped according to the age range in the key influencing features, and passengers whose age is higher than the third preset threshold are classified as elderly.

[0043] Based on the numerical range of delay duration in the key impact features, delay scenarios are classified into categories, and scenarios with delay duration exceeding the fourth preset threshold are classified as long-term delay scenarios.

[0044] For different combinations of passenger groups and delay scenarios, corresponding emergency dispatch strategies are matched.

[0045] Furthermore, the emergency dispatch strategy includes:

[0046] For the aforementioned elderly group, a strategy is generated to provide customized ground shuttle services to hotels or city centers;

[0047] For the aforementioned price-sensitive group, a strategy is generated to open shuttle bus services to the high-speed rail station and recommend air-rail intermodal transport options;

[0048] For the aforementioned long-delay scenarios, strategies are generated to recommend rebooking on other flights or to provide a comfortable rest area and real-time delay updates;

[0049] For the mid-priced ticket group, a strategy is generated that offers flexible rebooking packages at discounted prices.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] (1) This invention constructs a closed-loop analytical system from multi-dimensional data perception to differentiated strategy generation, breaking the limitations of traditional delay management that relies solely on single statistical data for passive response. By combining the static attributes of passengers with dynamic delay scenarios and using a gradient boosting tree model to deeply mine massive survey data, this invention can accurately predict passengers' true behavioral intentions when faced with flight delays, much like decoding a black box. This complete technical chain, from variable system construction to experimental design, data collection, model training, attribution analysis, and strategy generation, is equivalent to providing airport managers with a high-precision behavioral perspective, transforming traditional extensive passenger flow evacuation into refined proactive guidance based on behavioral prediction, significantly improving the emergency response efficiency and operational resilience of hub airports in the face of large-scale delays.

[0052] (2) This invention innovatively establishes a three-dimensional variable system encompassing passenger travel characteristics, sociodemographic attributes, and perceived service quality. Combined with orthogonal design-based declarative preference experiments and revealing preference surveys, it essentially creates a high-precision behavioral profile for each passenger. This joint survey method effectively solves the problem of potential distortion in hypothetical scenarios or insufficient sample size in real-world scenarios from single data sources. The declarative preference survey captures passengers' decision-making tendencies under various hypothetical delay scenarios, while the revealing preference survey provides real historical behavioral data as anchor points. The orthogonal design method, through the scientific combination of different levels of six attributes such as delay duration, weather conditions, and accommodation compensation, acts like a precise stress test, comprehensively capturing subtle psychological changes in passengers across different dimensions such as ticket price sensitivity and delay tolerance, ensuring the comprehensiveness of the input data and the robustness of the prediction results.

[0053] (3) The gradient boosting tree model used in this invention introduces second-order Taylor expansion approximation and L2 norm regularization constraints during its construction process. This mathematical processing is like installing a high-precision filter and stabilizer for the model. The second-order Taylor expansion allows the model to not only see the direction of descent when searching for the optimal solution, but also to perceive the slope of descent, thereby achieving faster convergence speed and higher fitting accuracy. The regularization term effectively prevents the model from rote memorization of training data by double-penalizing the number of leaf nodes and the sum of squared weights, thus avoiding overfitting. Specifically, by traversing the hyperparameter space through grid search and combining it with cross-validation to calculate the performance score of each parameter combination, this invention can automatically find the optimal hyperparameter configurations such as column sampling ratio, learning rate, and maximum tree depth, ensuring that the model can still maintain extremely high generalization ability and prediction accuracy when facing unknown new passenger data.

[0054] (4) This invention utilizes the Shapley value additive interpretation algorithm to make the prediction results of the black-box model transparent, not only knowing what happened but also why. By constructing an interpreter object to calculate the marginal contribution of each variable in all samples and averaging the absolute values ​​of SHAP to obtain the global importance index, this invention can accurately identify the root causes driving passengers to change their travel plans. This game theory-based attribution method ensures fairness and additivity, which is equivalent to assigning a fair contribution score to each influencing factor. After identifying key influencing features, this invention further groups passengers according to the numerical range of features such as ticket price, age, and delay duration, and matches corresponding emergency dispatch strategies for each group. For example, it provides ground shuttle services for the elderly, recommends air-rail intermodal transport for price-sensitive groups, and provides comfortable rest areas and real-time information updates for long-delay scenarios. This technical path from attribution analysis to grouping and then to strategy matching realizes the leap from broadcast notification to precise group services, enabling airport managers to formulate differentiated emergency plans as if prescribing the right medicine for the right disease. Attached Figure Description

[0055] Figure 1 This is a flowchart of a method for predicting the travel change intentions of hub passengers in a trip delay scenario, provided by an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Unless otherwise specifically stated, the relative arrangement, expressions, and values ​​of components and steps set forth in these embodiments do not limit the scope of the present invention. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0057] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0058] like Figure 1 As shown, this invention provides a method for predicting the travel change intentions of hub passengers in a trip delay scenario, mainly including the following steps:

[0059] step A three-dimensional variable system of influencing factors was constructed, including the dimensions of passenger travel characteristics, socio-demographic attributes, and perceived service quality. Based on the variable system, an orthogonal design method was used to generate experimental schemes for delay scenarios, including delay duration, weather conditions, food and accommodation compensation, and rescheduling fees.

[0060] In this embodiment, as shown in the figure As shown, the three-dimensional influencing factor variable system includes a total of The system designs variables from three dimensions to comprehensively capture various factors influencing passenger behavior decisions. These passenger travel characteristic dimensions include ticket price, travel purpose, estimated flight time, frequency of intercity travel, number of trips via this hub, early arrival time at the airport, and cabin class. There are several variables. Among them, ticket prices are divided into less than... Yuan, to Yuan, to Yuan, to Yuan, to Yuan and greater than The budget is divided into six ranges to reflect the differences in travel choices among passengers with varying economic capabilities. Travel purpose is categorized into six types: visiting relatives and friends, tourism, return trip, other, school, and business, to differentiate the impact of different travel motivations on delay tolerance. Estimated flight time is set to less than [amount missing]. Hour, to Hours and greater Three timeframes were used to capture the impact of flight length on passenger decisions. Intercity travel frequency was set at [number] times over the past year. Second-rate, to Second-rate, to Second-rate, to Second-rate, to Second-rate, to Second and greater The seven tiers are used to measure the breadth of a passenger's travel experience. Previously, the number of trips through this hub was categorized as follows: Second-rate, to Second-rate, to Second-rate, to Second and greater Five levels are used to assess passengers' familiarity with the airport facilities. Airport familiarity significantly impacts their cognitive load and decision-making efficiency when adjusting their travel plans during delays. The early arrival time at the airport is set to less than... Hour, to Hour, to Hours and greater The four time periods, one hour at a time, reflect passengers' time planning habits and their preparedness for delays. Cabin classes are divided into three levels: Economy, Business, and First Class, with significant differences in service expectations and affordability among passengers in each class.

[0061] The sociodemographic attributes include occupation type, monthly income, car ownership, gender, age, and education level. Several variables. Job type is subdivided into seven categories: students, workers, private business owners or self-employed, others, corporate employees, public institution staff, and civil servants. Significant differences exist among these occupational groups in travel reimbursement systems, time flexibility, and decision-making autonomy. For example, civil servants must report and approve travel expenses in advance, strictly adhere to budgets and reimbursement standards, and changing to other modes of transportation in a short period requires post-event justification and additional approval procedures, making the process cumbersome and carrying reimbursement risks. Monthly income is categorized as less than... Yuan, to Yuan, to Yuan, to Yuan, to Yuan and greater than The system is divided into six income brackets, with income levels directly influencing passengers' sensitivity to fare changes and additional charges. The vehicle ownership is set at... Vehicles Vehicles Vehicles and larger The number of vehicles is categorized into four levels, indirectly reflecting household economic status and the diversity of travel choices. Gender is divided into male and female categories. Age is categorized into... to age, to age, to age, to Age and Five age groups, aged 30 and above, are considered, with age influencing passengers' acceptance of information technology, physical condition, and decision-making preferences. Education level is divided into three levels: master's degree or above, bachelor's degree or associate's degree, and high school or vocational school or below. Educational background affects passengers' ability to acquire and process information, as well as their ability to cope with complex processes.

[0062] The perceived service quality dimensions include airport transfer convenience, rest area satisfaction, indoor environment rating, staff assistance satisfaction, refund and change convenience, and food and beverage price satisfaction. Specifically, this dimension covers airport transfer convenience, transfer comfort, transfer information provision, transfer guidance services, office convenience, leisure convenience, rest area satisfaction, indoor environment rating, staff assistance satisfaction, ease of use of self-service facilities, check-in convenience, baggage drop-off convenience, refund and change convenience, food and beverage variety, and food and beverage price satisfaction. Item 10 continuous scoring indicators, each indicator using 100% scoring criteria. to The Likert scale was used to score the points, among which The score indicates extreme dissatisfaction. The scores indicate high satisfaction, forming a structured dataset for subsequent model input. These service quality indicators comprehensively reflect passengers' subjective perception of the airport's overall service level, which directly influences their behavioral decision-making tendencies in delay scenarios.

[0063] In the design of the experimental scheme for the delay scenario, as shown in the figure As shown, an orthogonal efficiency design method is used to scientifically combine multiple attributes of a delay scenario. The delay scenario experimental scheme generates multiple choice tasks through orthogonal combination. Each choice task provides four behavioral intention options: transferring to another train, waiting for the original flight, canceling the trip, and rebooking another flight. The delay scenario attributes and levels are defined as follows: delay duration is set to less than... Hour, to Hours and greater The three levels of the hourly rate are based on a common understanding within the aviation industry. Delays of less than one hour are considered acceptable short-term delays. Delays exceeding one hour can severely disrupt passengers' travel plans. Weather conditions are set at two levels: sunny and rain / snow. Weather factors not only affect the likelihood of flight delays but also passengers' understanding and tolerance of the reasons for delays. The originally scheduled departure time is divided into daytime periods. hour Distributed to hour Divided and night hour Distributed to the next day hour The system is divided into two time slots. The departure time affects the feasibility of passengers changing their tickets or switching to other modes of transportation. For example, when there are delays at night, there are fewer alternative transportation options, and passengers are more fatigued. Accommodation and meals are provided free of charge. Fold and The compensation measures are divided into three levels, with different degrees affecting passengers' psychological acceptance of delays and their willingness to continue waiting. The rebooking fee is set to be less than... Yuan, to Yuan and greater than The cost of rebooking is a significant economic factor influencing whether passengers choose to rebook. The delay time that the rebooking option can reduce is set to less than [amount missing]. Hours or none to Hours and greater The three-level system of hours reflects the time advantage of the rebooked flight compared to the original flight, and is the core basis for passengers to evaluate the value of rebooking.

[0064] By using orthogonal design methods, the above By scientifically combining different levels of each attribute, a generation Select tasks and randomly divide them evenly. Groups, each group contains One task, respondents must complete at least one. The system includes several tasks to ensure data sufficiency and diversity. Each task selection interface presents respondents with a specific description of the delay scenario, such as "Your flight is expected to be delayed..." The current weather is rainy and snowy. The original scheduled departure time was during the day. The airline provides... Compensation will be provided for meals and accommodation; additional fees will be charged for rebooking on other flights. The option to rebook flights can reduce delays to [amount missing]. The timeframe was set at "hours," and respondents were then asked to choose the travel mode that best suited their needs from four options: transferring to a different train, waiting for their original flight, canceling their trip, or rebooking on another flight. The orthogonal design method comprehensively captures passengers' decision-making preferences under different combinations of delay conditions.

[0065] step Based on the aforementioned delay scenario experimental scheme, a statement preference survey was conducted, and the actual travel records of respondents were obtained in conjunction with a disclosure preference survey. The statement preference survey data and the disclosure preference survey data were jointly processed, and after ordinal encoding and missing value imputation, they were divided into training set and test set according to a preset ratio.

[0066] During the data acquisition phase, as shown in the figure As shown, this embodiment involves conducting a questionnaire survey at a major hub airport. During the survey, respondents' disclosure preference data was simultaneously collected, namely their historical travel records, including their actual choices when encountering flight delays in the past, and their actions based on steps... The declarative preference data is generated through a hypothetical scenario experiment scheme. Revealing preference data records respondents' past travel choices in real-world delay scenarios; this data is reliable but typically has a small sample size and incomplete scenario coverage. Declarative preference data, on the other hand, obtains respondents' choices under various possible combinations of delay conditions through hypothetical scenario experiments; this type of data can systematically cover multiple combinations of delay conditions. The combined use of these two types of data effectively overcomes the limitations of a single data source. The declarative preference survey captures travelers' decision-making tendencies under various hypothetical delay scenarios, while the revealing preference survey provides real historical behavioral data as an anchor.

[0067] The specific steps for jointly processing the stated preference survey data and the revealed preference survey data include: firstly, associating and matching the respondents' choices in the stated preference survey data under the hypothetical delay scenario with the respondents' historical actual travel records in the revealed preference survey data to ensure that the two types of data for the same respondent can corroborate and complement each other. In the data quality control stage, the returned questionnaires are rigorously screened, and invalid questionnaires with excessively short completion times, obvious logical contradictions in the answers, or a large number of missing values ​​are removed. After screening, this embodiment retains... Valid questionnaires, corresponding Each scene was recorded, with an average of approximately [number] responses per interviewee. These are valid scenario data. This valid data covers passenger groups of different ages, incomes, occupations, and travel characteristics, ensuring the representativeness and diversity of the sample.

[0068] In the data preprocessing stage, ordinal encoding is performed on the categorical variables in the associated data, mapping text features to... The initial integer sequence. For example, encoding "student" in the job type variable as... Worker code Private business owners or self-employed individuals are coded as follows: Other codes are Employee codes of enterprises are The personnel code for public institutions is Civil servant code This process is repeated for all categorical variables. For data records with missing values, median imputation is used to handle missing values ​​for numerical variables, and mode imputation is used to handle missing values ​​for categorical variables, ensuring data integrity. After coding and imputation are completed, the data is then processed according to... to The proportion will divide the processed data into the training set and the test set. Specifically, from Randomly selected from each scene record That is, One record is used as the training set for model parameter learning and optimization; the remaining records... That is, A set of records was used as a test set to evaluate the model's generalization ability and prediction accuracy.

[0069] step A gradient boosting tree machine learning model is constructed, and the parameters of the gradient boosting tree machine learning model are tuned and trained using the training set to obtain a passenger behavior prediction model that can output prediction results of four behavioral intentions: transferring to another railway, waiting for the original flight, canceling the trip, and rebooking other flights.

[0070] During the model building phase, as shown in the figure As shown, this embodiment employs extreme gradient boosting. The algorithm implements a gradient boosting tree machine learning model. Before model training, the training set is first prepared and its features are engineered to ensure that all categorical variables have been ordinally encoded, numerical variables have no outliers or missing values, and the data format fully conforms to the specifications. The algorithm's input requirements. Data preparation also includes feature normalization, which unifies numerical features of different dimensions to the same scale range, preventing certain features from dominating model training due to excessively large numerical ranges. The objective function of the algorithm is as follows: calculate:

[0071]

[0072] in, Indicates the number of iterations. Indicates the number of samples. Represents the loss function. Indicates the first The actual travel choices of passengers. Indicates the preceding Predicted value after the next iteration Indicates the first The feature vector of each sample Indicates the first In the nth iteration tree to the first The predicted increment for each sample This represents the regularization term. In this embodiment, The range of values ​​is to These represent four behavioral intentions: transferring to another train, waiting for the original flight, canceling the trip, and rebooking another flight. It is a collection The feature vectors of each influencing factor cover the steps. All variables defined in the code. The objective function minimizes the weighted sum of prediction error and model complexity in each iteration, thereby ensuring prediction accuracy while avoiding overfitting caused by an overly complex model.

[0073] To optimize the objective function, in the first... In the next iteration, the formula By performing a second-order Taylor expansion approximation, we obtain the formula. :

[0074]

[0075] in, This represents the first derivative of the loss function with respect to the predicted values. This represents the second derivative of the loss function with respect to the predicted values. The second-order Taylor expansion allows the model to determine the descent direction at each iteration not only using gradient information (i.e., the first derivative) but also... The matrix information, specifically the second derivative, determines the descent step size, thereby achieving faster convergence and higher fitting accuracy when analyzing complex passenger decision-making behavior. This mathematical approach enables the model to find the optimal solution more accurately and efficiently in the high-dimensional parameter space. Compared to traditional gradient boosting methods that only use the first derivative, the second-order Taylor expansion significantly reduces the number of iterations required to reach convergence.

[0076] The regularization term According to the formula calculate:

[0077]

[0078] in, This represents the penalty coefficient for the leaf nodes. Indicates the first The number of leaf nodes in a tree. The L2 regularization coefficient represents the weight of the leaf nodes. Indicates the first The weight of each leaf node, This indicates a penalty applied to the sum of squared weights of all leaf nodes to prevent overfitting. The regularization term effectively prevents the model from overfitting the training data by double-penalizing both the number of leaf nodes and the sum of squared weights, ensuring the model can adapt to different training conditions. Questionnaire By learning universal decision-making patterns from scenario records, rather than relying on noise or exceptions in individual samples, the system can maintain robust predictive performance even when faced with new passenger data. (First item) To control the complexity of the tree and prevent the generation of too many leaf nodes from making the model overly refined, the second item... This constrains the magnitude of the weights of the leaf nodes to prevent the weights of some leaf nodes from becoming too large, which could cause the model to become overly sensitive to specific samples.

[0079] During the parameter tuning phase, as shown in the figure As shown, this embodiment uses a grid search combined with cross-validation to find the optimal hyperparameter combination. The specific steps include: first, defining a hyperparameter grid including column sampling ratio, learning rate, maximum tree depth, number of estimators, and sample sampling ratio; and then setting the value range for each hyperparameter. The column sampling ratio is... The candidate values ​​for the parameter are set to , , and This parameter controls the proportion of features randomly selected during the training of each tree; a smaller value can increase the model's randomness and robustness. The learning rate is... The candidate values ​​for the parameter are set to , , and This parameter controls the weight of each tree's contribution to the final prediction result. A smaller learning rate requires more trees to achieve the same fitting effect but usually yields better generalization performance. The maximum depth of the tree is... The candidate values ​​for the parameter are set to , , and This parameter controls the maximum number of layers in each tree. Deeper trees can capture more complex feature interactions but are also more prone to overfitting. The number of estimators is... The candidate values ​​for the parameter are set to , , and This parameter determines the total number of decision trees to be generated. More trees generally improve model performance but also increase computational cost and training time. The sample sampling ratio is... The candidate values ​​for the parameter are set to , , and This parameter controls the proportion of samples randomly selected during the training of each tree. Introducing sample sampling can reduce the risk of overfitting and improve training speed.

[0080] The training set is then divided into several folds, and cross-validation is used to select a subset of data from each fold as the validation set, with the remaining data serving as the training subset. This embodiment employs... Cross-validation, which involves dividing the training set into equal parts... One portion, take turns selecting one of them One set is used as the validation set, and the rest are used as the validation set. A subset of samples is used as the training set for model training. For each hyperparameter combination, in In folded cross-validation, the validation set performance is calculated separately, and then the results are taken. The average of the validation results is used as the cross-validation score for this parameter combination. Next, all parameter combinations in the hyperparameter grid are traversed. For each parameter combination, the model is trained on the training subset, and the cross-validation score is calculated on the validation set. Finally, the parameter combination that yields the optimal cross-validation score is selected as the optimal hyperparameter combination. After grid search and cross-validation, the optimal hyperparameter combination determined in this embodiment is: column sampling ratio of... The learning rate is The maximum depth of the tree is The number of estimators is The sample sampling ratio is Through this configuration, the model obtains The optimal cross-validation score indicates that the model achieved a good balance between fitting performance and generalization ability during training.

[0081] In another embodiment, parameter tuning can also employ Bayesian optimization instead of grid search. By constructing a probabilistic surrogate model between hyperparameters and model performance, Bayesian inference is used to gradually narrow the hyperparameter search space. In each iteration, the hyperparameter combination with the best chance of improving model performance is selected for trial, thereby finding a near-optimal hyperparameter configuration with fewer iterations. Bayesian optimization determines the optimal parameter combination by minimizing the cross-validation loss of the objective function. Compared to exhaustive grid search, Bayesian optimization can explore the hyperparameter space more intelligently, making it particularly suitable for scenarios with high hyperparameter dimensionality or long single model training times.

[0082] After completing parameter tuning, use the training set data and the optimal hyperparameter combination to... The model is then formally trained. The model employs a multi-class classification approach. The objective function is designed to handle classification and prediction tasks for four behavioral intentions simultaneously. It outputs the probability distribution of each sample belonging to one of the four categories and ultimately selects the category with the highest probability as the prediction result. During training, the model continuously reduces the prediction error through iterative optimization. Each iteration adds a new decision tree to fit the residual of the current model until a preset number of iterations is reached or the loss function converges. The final result is a passenger behavior prediction model that can output prediction results for four behavioral intentions: transferring to another train, waiting for the original flight, canceling the trip, and rebooking another flight.

[0083] step The test set is input into the passenger behavior prediction model, and attribution analysis is performed on the output of the passenger behavior prediction model. The marginal contribution of each variable in the variable system to the four behavioral intentions is calculated, and the variables with the highest marginal contribution are selected as key influencing features.

[0084] During the model evaluation phase, as shown in the figure... As shown, the test set data is input into the trained passenger behavior prediction model, and the model outputs the most likely behavioral intention category for each test sample. The model's performance on the test set is then calculated. The model in this embodiment uses scores to evaluate its overall performance. Score reached This indicates that the model performs well in balancing precision and recall, and can accurately predict passengers' behavioral intentions in delay scenarios. The score is the harmonic mean of precision and recall, and its value ranges from 100 to 100. to The closer the value is The better the model performance.

[0085] The attribution analysis employed a Shapley value-based approach. Algorithm implementation. Specific steps include: building... The interpreter object calculates the variables for each sample in the test set within the passenger behavior prediction model. value. Value based on game theory Value theory fairly allocates the contribution of each feature to the model's prediction results. For each sample, the algorithm calculates the marginal impact on the prediction result when the feature is added or removed, obtaining the value of each feature by iterating through all possible feature combinations. Value. Specifically, The algorithm fairly distributes the difference between the model's predictions and the baseline predictions according to the marginal contribution of each feature, ensuring that all features are equally valuable. The sum of the values ​​equals the difference between the actual predicted value and the baseline value; this additive property makes... The values ​​have good interpretability and consistency. Then, the values ​​for each variable in all samples are calculated. The average of the absolute values ​​is used as the global importance index for the variable, and all variables are then sorted in descending order based on this global importance index. The formula for calculating the global importance index is as follows:

[0086]

[0087] in The total number of samples, Indicates the first In the nth sample 1 variable The absolute value is taken to comprehensively consider the influence of the variable on the prediction results in different directions, without considering the positive or negative directions.

[0088] To more intuitively demonstrate the attribution analysis results, this embodiment further utilizes... Visualization tools generate various charts. First, utilize... Dependency graphs reveal the non-linear relationship between variables and behavioral intentions. A dependency graph displays the values ​​of a feature and their corresponding values ​​in the form of a scatter plot. The relationship between the values ​​is shown, with each point representing a sample. The horizontal axis represents the actual value of the feature, and the vertical axis represents the value of the feature in that sample. The value, or color, of a point is often used to indicate the value of another related feature, thus revealing the interaction between the two features. Through... Dependency graphs clearly show the non-linear mapping relationship between features and prediction results; for example, the impact of ticket prices on transfer intentions across different ranges exhibits a non-monotonic trend. Secondly, based on... Value sorting, before filtering As key variables for policy intervention, these variables have the most significant impact on passenger behavior and decision-making, and are crucial for formulating precise emergency strategies. Finally, through The summary figure shows the distribution of feature importance, where each point represents a sample, red indicates high feature values, blue indicates low feature values, and the horizontal position shows... The vertical axis represents the direction and intensity of the influence of the value on the prediction, arranged from top to bottom according to the importance of the feature. Abstract plots can simultaneously present the importance ranking of all features, the direction and strength of each feature's influence on different samples, and the distribution of feature values ​​in a single chart.

[0089] pass Attribution analysis in this embodiment successfully identified the key features that have the greatest impact on passenger behavioral intentions. For the intention to transfer trains, the top... Key influencing factors include ticket price, airport staff assistance satisfaction, and job type. The analysis shows that ticket price is located in... to Passengers traveling in the mid-range are less likely to switch to rail, as airfares still offer a time advantage over rail travel at this price level. In contrast, purchasing a low-priced airfare (i.e., a ticket price lower than...) Passengers with budgets of around 100 yuan are more cost-sensitive, and the relative price difference between discounted airfares and high-speed rail is smaller, so they are more inclined to choose rail for their transfers. Passengers dissatisfied with the assistance of airport staff are more likely to seek alternatives such as rail, while those with higher satisfaction tend to wait for their original flight or rebook on another flight. Government employees are less likely to choose rail for their transfers because their travel expenses require prior reporting and approval, and changing to other modes of transportation in a short period of time requires additional approval procedures, which are cumbersome and carry the risk of reimbursement issues.

[0090] Key influencing factors for the intention to wait for the original flight include the frequency of previous trips through the hub airport, the length of the delay, and the amount of delay that could be reduced by rebooking options. Passengers who rarely visit the hub airport are more likely to wait for their original flight because first-time visitors are unfamiliar with the airport layout and experience a higher cognitive load when navigating. Adjusting travel plans requires additional cognitive effort, while waiting for the original flight is the default option, saving mental resources and reducing the psychological cost of decision-making. Passengers experiencing delays of less than 2 hours are more likely to choose alternative modes of transportation rather than wait for their original flight. Prolonged waits not only disrupt subsequent travel plans but also cause physical fatigue and psychological anxiety. Conversely, those who decide to wait are primarily those experiencing shorter delays. Similarly, if the delay time that can be shortened by the rebooking option is less than... If the time difference is not improved, passengers are more likely to choose to wait for the original flight, because the time savings from rescheduling are limited and not enough to compensate for the additional costs and hassles incurred during the rescheduling process.

[0091] Key factors influencing the intention to cancel a trip include expected flight time, age, and satisfaction with food and beverage prices. For long-haul domestic flights (i.e., expected flight time exceeding [a certain value]), [the following factors are relevant]. Passengers with short flight times are more likely to cancel their trips because these flights often fly to remote western destinations with limited availability of alternative flights or rail connections, leaving them with fewer rebooking options. Older passengers are more likely to cancel their original flights due to the complexity of rebooking and refund processes and their limited access to travel-related information services. Regarding food and beverage price satisfaction, the analysis shows that passengers with higher satisfaction with airport food and beverage prices are actually more likely to cancel their trips. This is because passengers who were initially dissatisfied with food and beverage prices perceive free meals or discount coupons during delays as significant compensation, thus reducing their tendency to cancel. However, passengers who were already satisfied with food and beverage prices do not perceive a significant increase in value when receiving free meals, so free meals do not significantly change their willingness to cancel. This metric targets price satisfaction rather than food or beverage quality satisfaction. The difference between the two groups' expectations and actual experiences of compensation measures—the former experiencing a positive bias while the latter did not—explains the association between higher price satisfaction and higher cancellation probability.

[0092] Key factors influencing the intention to change to another flight include ticket price, age, and the cost of changing the flight. Ticket prices are... to Passengers in the mid-to-high price range are more likely to choose to rebook other flights because this group has both sufficient financial means and a demand for service quality and time efficiency, and will actively seek better alternative flights when delays occur. In contrast, fares exceeding [amount missing] Passengers with high-priced tickets, due to their higher economic value, are usually given priority by airlines, which provide them with customized services; therefore, this group has a relatively low demand for rebooking. Younger passengers are more inclined to rebook, which is related to their familiarity with and habits of using information services. Younger people are generally better at using digital tools such as mobile applications and self-service kiosks to quickly complete rebooking. In contrast, most older passengers are less inclined to rebook. Older passengers often feel confused when faced with complex rebooking processes and are therefore more likely to cancel their trips or wait for their original flights. Lower rebooking fees are positively correlated with rebooking behavior, indicating that the current policy of waiving rebooking fees for the same airline or partner airlines effectively encourages passengers to choose flight rebooking. However, higher rebooking costs become a limiting factor for rebooking. Since airline cooperation agreements are not widespread, if passengers wish to choose other flights with non-partner airlines, they may need to cancel their existing tickets and purchase new ones, resulting in additional costs such as fare differences, increasing the actual cost of rebooking.

[0093] Through the above Attribution analysis showed that this embodiment successfully screened out the top contributors by marginal contribution. The variables, as key influencing features, provide a scientific basis and quantitative support for the subsequent generation of differentiated emergency dispatch strategies.

[0094] step Based on the nonlinear mapping relationship between the key impact features and the four behavioral intentions, differentiated emergency dispatch strategies are generated for different passenger groups.

[0095] During the strategy generation phase, as shown in the figure... As shown, passengers are grouped according to the price range in the key influencing features. Passengers with prices below a first preset threshold are classified as price-sensitive, while those with prices between the first and second preset thresholds are classified as medium-priced passengers. In this embodiment, the first preset threshold is set to... Yuan, the second preset threshold is set to Yuan. Therefore, the ticket price is lower than Passengers with fares in the range of [amount missing] are categorized as price-sensitive; these passengers are highly sensitive to travel costs and tend to choose more cost-effective modes of transportation when delays occur. Ticket prices are [missing information]. to Passengers with fares between [price range] are categorized as mid-range ticket buyers; these passengers have sufficient purchasing power and value service quality and time efficiency. Fares higher than [price range]... Passengers paying 10,000 yuan are categorized as high-fare passengers. These passengers typically enjoy higher-level services and are relatively less price-sensitive.

[0096] Passengers are grouped according to the age range among the key influencing characteristics, with those older than a third preset threshold classified as the elderly group. In this embodiment, the third preset threshold is set to... age, that is Passengers aged 60 and above are classified as elderly, while Passengers under the age of 18 are further subdivided into the youth group based on other characteristics. to The 30-year-old and middle-aged groups to Passengers of different ages show significant differences in physical condition, acceptance of information technology, and service needs.

[0097] Delay scenarios are classified according to the numerical range of delay duration in the key impact features, and scenarios with delay durations exceeding a fourth preset threshold are classified as long-delay scenarios. In this embodiment, the fourth preset threshold is set to... hours, that is, a delay of more than 1 hour Scenarios lasting an hour are categorized as long-delay scenarios, with delay durations ranging from [missing information]. to Scenarios with delays between one hour and two hours are classified as moderate delay scenarios, with delays less than [a certain duration]. Scenarios lasting an hour are categorized as short-term delays. Different delay durations have varying degrees of psychological and physiological impact on passengers, requiring different levels of response measures.

[0098] For different combinations of passenger groups and delay scenarios, corresponding emergency dispatch strategies are matched. These strategies include: for the elderly group, generating customized ground shuttle services to hotels or the city center. Since elderly passengers face significant difficulties in rebooking and refund processes and have relatively limited access to information services, providing convenient ground shuttle services can prevent them from spending long periods at the airport, reducing their physical and psychological burden. The airport can arrange dedicated shuttle buses or taxis to safely transport elderly passengers to nearby hotels for rest or directly to the city center, providing clear directions and full-time accompaniment to ensure they can smoothly complete their subsequent travel arrangements. This strategy is particularly suitable for nighttime delay scenarios, as public transportation options are limited at night, and the safety risks for elderly passengers traveling alone are higher.

[0099] For the aforementioned price-sensitive group, a strategy is generated to provide shuttle bus services to the high-speed rail station and recommend air-rail intermodal transport options. This price-sensitive group is highly concerned about travel costs, especially regarding flight delays, particularly those exceeding [a certain duration]. For passengers with delayed flights, the preferred mode of transportation is often the most cost-effective. Airports could offer flexible shuttle services to transport these passengers to nearby high-speed rail stations. The frequency of shuttle departures should be dynamically adjusted based on the number of delayed flights and passenger volume to ensure timely transfers. Simultaneously, suitable air-rail intermodal transport options should be recommended through mobile applications or on-site information desks, providing one-stop services such as high-speed rail ticket availability checks, online ticketing, and seat reservations to help passengers quickly replan their trips. Furthermore, negotiations could be held with railway authorities to offer fare discounts or priority ticketing services; for example, passengers holding delayed flight tickets could enjoy high-speed rail ticket discounts. Discounts will further enhance the attractiveness of this strategy and reduce the economic losses suffered by passengers due to delays.

[0100] For scenarios involving long delays, strategies should be developed to recommend rebooking on alternative flights, provide comfortable rest areas, and offer real-time delay updates. Long delays put passengers' patience and stamina to a significant degree, leading to fatigue and anxiety. For passengers willing to wait for their original flights, airports should provide comfortable rest areas equipped with ample seating, charging facilities, and free [services / resources]. - To ensure passengers can maintain uninterrupted communication and use their electronic devices while waiting, rest areas should be located in relatively quiet and softly lit areas, equipped with air conditioning and air purification equipment to guarantee a comfortable indoor environment. Delay information should be updated in real time through multiple channels such as displays, mobile push notifications, and SMS messages, allowing passengers to stay informed about flight status, including estimated departure time, reasons for delays, and compensation measures, avoiding feelings of insecurity and distrust caused by a lack of transparency. For passengers wishing to rebook, the airport should proactively recommend available alternative flights, selecting the best rebooking options based on passengers' destination, time preferences, and cabin class requirements, prioritizing seating, and simplifying the rebooking process to reduce waiting time and operational complexity. Dedicated rebooking counters or mobile service points, staffed with experienced personnel to provide one-on-one consultation and processing services, can be established to improve rebooking efficiency.

[0101] For the aforementioned mid-fare group, a strategy is developed to offer flexible rebooking packages at discounted rates. This group possesses both purchasing power and high expectations for service quality and time efficiency, making them a key target group for airlines to cultivate customer loyalty. Airports or airlines can offer flexible rebooking packages to this group, allowing them priority rebooking at discounted rates to other flights, including flights with different partner airlines, in the event of delays, without incurring additional fare differences. Specifically, when the delay duration reaches [a certain value], [the strategy is implemented]. to In the case of a moderate delay of one hour, package deals can be pushed to passengers with medium-priced tickets, such as paying a discount on the original ticket price. This grants priority rebooking rights and waives the usual rebooking fees and fare differences. The rebooking target flights cover all flights operated by partner airlines at that airport. This package can be offered as an add-on service at the time of ticket purchase, serving as an insurance-like value-added product, or it can be recommended on-site after a delay, flexibly sold based on the passenger's real-time needs. This approach satisfies the flexible travel needs of mid-priced passengers, reducing their time and financial losses due to delays, while also improving their overall satisfaction with airline services and enhancing brand loyalty.

[0102] Furthermore, based on Based on the attribution analysis, this embodiment also proposes countermeasures and policy recommendations for specific influencing factors. Given the significant impact of airport staff assistance satisfaction on passenger behavior, it is recommended to appropriately increase staff hourly wages during periods of widespread flight delays to incentivize employees to maintain good service attitude and work efficiency under high-intensity working conditions. Simultaneously, deploy mobile personnel to provide additional support, setting up temporary service points in areas with high passenger concentration, equipped with mobile terminals and printing equipment, to provide on-site services such as refunds and changes, flight information inquiries, and answering questions. This will alleviate the workload of fixed counters, improve service attitude, thereby reducing the tendency for passengers to choose other modes of transportation due to service dissatisfaction, and improving the overall service level and passenger retention rate of the airport.

[0103] To address the difficulties elderly travelers face in using information services, it is recommended to further develop ticketing systems and service interfaces tailored to seniors. Specific measures include providing a font magnification function, allowing elderly travelers to adjust the font size according to their eyesight to ensure clear readability; developing a voice navigation function, guiding elderly travelers through various procedures via voice prompts to reduce the burden of reading text instructions; and setting up a one-click human service button, allowing elderly travelers to directly connect to customer service with real-time voice guidance or video assistance when encountering difficulties, preventing them from getting lost in complex self-service interfaces. Additionally, adding dedicated service windows for elderly travelers at airports, staffed with patient and attentive staff, will provide comprehensive assistance, from flight inquiries and rebooking to arranging subsequent transportation, offering a one-stop solution. These measures will help elderly travelers complete rebooking and refunds more conveniently, reducing the proportion of trip cancellations due to operational difficulties and improving the travel experience and satisfaction of the elderly population.

[0104] Given the significant impact of rebooking fees on passenger decisions, it is recommended that regulatory authorities proactively promote the establishment of a mandatory cross-airline rebooking agreement network among airlines operating at hub airports. This network should explicitly stipulate that in the event of force majeure events such as flight delays or cancellations, airlines are obligated to provide cross-airline rebooking services to affected passengers without charging additional rebooking fees or fare differences. The rebooking scope should cover all major airlines operating at the airport. Such a mandatory agreement can break down barriers between airlines, significantly increasing passengers' rebooking options and preventing the predicament of passengers being forced to cancel their trips or experiencing long waits due to a lack of cooperation agreements. Furthermore, it is recommended to levy taxes on routes prone to delays and establish a delay risk fund, with higher taxes levied on routes with higher delay rates. This fund would be specifically used to help offset the costs of rebooking for passengers, such as subsidizing fare differences arising from cross-airline rebooking, providing transportation vouchers, and funding airport improvements to emergency service facilities. This would enhance overall service resilience under flight delays and also provide economic incentives for airlines to improve on-time performance and reduce the frequency of delays.

[0105] Further recommendations for airports and airlines to adopt differentiated strategies include: ticket prices at... to Passengers in this price range tend to rebook other flights rather than transferring to rail, so for tickets in this price range, optional flexible rebooking packages can be offered. Purchasing such a package gives passengers priority rebooking of alternative flights, including across partner airlines, in the event of flight delays. Flight delays exceeding [amount missing] Passengers who are only an hour away are usually unwilling to wait for their original flight, and ticket prices are... Passengers with fares below a certain amount are also more willing to transfer to the railway. Therefore, the airport could offer flexible shuttle services to transport these passengers to the high-speed rail station. Shuttle services should be dynamically scheduled based on the arrival times of delayed flights and the high-speed rail departure timetables. Some passengers, such as the elderly, often cancel their trips altogether. For these passengers, providing shuttle services to transport them to nearby hotels or the city center will help avoid prolonged stays at the airport and disruption to subsequent flights, while also reducing the physical burden and psychological stress on elderly passengers.

[0106] This embodiment involves conducting an on-site investigation at a major hub airport and collecting data. Valid questionnaires correspond Each scenario was recorded, and data from both revealed and stated preferences surveys were used to construct a comprehensive... A three-dimensional influencing factor system for each variable was adopted. Machine learning model implementation Score reached High-precision behavior prediction, and utilize Interpretability tools provide in-depth analysis of the marginal contribution and nonlinear mapping relationships of various influencing factors. Based on these analytical results, this invention offers airports and airlines a complete technical solution encompassing data collection, model training, attribution analysis, and strategy generation. This solution enables a shift from a traditional passive response model to a proactive guidance model based on behavior prediction, significantly improving emergency response efficiency, resource allocation accuracy, and operational resilience in flight delay scenarios. It provides solid technical support and practical guidance for the intelligent and refined management of delays.

[0107] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for predicting the travel change intentions of hub passengers in a trip delay scenario, characterized in that, Includes the following steps: Step S1: Construct a three-dimensional variable system of influencing factors, including passenger travel characteristics, sociodemographic attributes, and perceived service quality. Based on the variable system, use orthogonal design to generate experimental schemes for delay scenarios, including delay duration, weather conditions, food and accommodation compensation, and rebooking fee losses. Step S2: Conduct a statement preference survey based on the delay scenario experimental scheme, and obtain the actual travel records of the respondents by combining the disclosure preference survey. The statement preference survey data and the disclosure preference survey data are jointly processed, and after ordinal encoding and missing value imputation, they are divided into training set and test set according to a preset ratio. Step S3: Construct a gradient boosting tree machine learning model, and use the training set to fine-tune and train the gradient boosting tree machine learning model to obtain a passenger behavior prediction model that can output prediction results of four behavioral intentions: changing trains, waiting for the original flight, canceling the trip, and changing to other flights. Step S4: Input the test set into the passenger behavior prediction model, perform attribution analysis on the output of the passenger behavior prediction model, calculate the marginal contribution of each variable in the variable system to the four behavioral intentions, and select the top 10% of variables with the highest marginal contribution as key influencing features. Step S5: Based on the nonlinear mapping relationship between the key impact features and the four behavioral intentions, generate differentiated emergency dispatch strategies for different passenger groups.

2. The method for predicting the travel change intentions of hub passengers in the scenario of travel delay as described in claim 1, characterized in that, In step S1, the three-dimensional influencing factor variable system includes: The dimensions of passenger travel characteristics include ticket price, travel purpose, estimated flight time, frequency of intercity travel, number of trips via the hub, time of arrival at the airport in advance, and cabin class. The sociodemographic attributes include job type, monthly income, number of cars owned, gender, age, and education level. The perceived service quality dimensions include airport transfer convenience, rest area satisfaction, indoor environment rating, staff assistance satisfaction, refund and change convenience, and food and beverage price satisfaction.

3. The method for predicting the travel change intentions of hub passengers in the scenario of travel delay as described in claim 2, characterized in that, In step S1, the delay scenario experiment scheme generates multiple selection tasks through orthogonal combination. Each selection task provides four behavioral intention options: transfer to another train, wait for the original flight, cancel the trip, and rebook another flight. The delay duration is set to three levels: less than 2 hours, 2 to 6 hours, and more than 6 hours. The weather conditions are set to two levels: sunny and rainy / snowy. The food and accommodation compensation is set to three levels: free, 50% off, and 30% off. The rebooking fee loss is set to three levels: less than 400 yuan, 400 to 600 yuan, and more than 600 yuan.

4. The method for predicting the travel change intentions of hub passengers in the scenario of travel delay as described in claim 1, characterized in that, In step S2, the joint processing of the stated preference survey data and the revealed preference survey data specifically includes: The respondents’ choice preferences in the hypothetical delay scenario in the stated preference survey data are correlated and matched with the respondents’ historical actual travel records in the revealed preference survey data. Ordinal encoding is performed on the categorical variables in the associated data to map text features to... The initial integer sequence; according to to The proportion will divide the processed data into the training set and the test set.

5. The method for predicting the travel change intentions of hub passengers in a trip delay scenario according to claim 1, characterized in that, In step S3, the gradient boosting tree machine learning model is implemented using the Extreme Gradient Boosting (XGBoost) algorithm, and the objective function of the XGBoost algorithm is calculated according to the following formula (1): （1） in, Indicates the number of iterations. Indicates the number of samples. Represents the loss function. Indicates the first The actual travel choices of passengers. Indicates the preceding Predicted value after the next iteration Indicates the first The feature vector of each sample Indicates the first In the nth iteration tree to the first The predicted increment for each sample This represents the regularization term.

6. The method for predicting the travel change intentions of hub passengers in the scenario of travel delay as described in claim 5, characterized in that, In step S3, to optimize the objective function, in the first... In the next iteration, formula (1) is approximated by a second-order Taylor expansion, resulting in formula (2): （2） in, This represents the first derivative of the loss function with respect to the predicted values. This represents the second derivative of the loss function with respect to the predicted values; The regularization term Calculate according to formula (3): （3） in, This represents the penalty coefficient for the leaf nodes. Indicates the first The number of leaf nodes in a tree. The L2 regularization coefficient represents the weight of the leaf nodes. Indicates the first The weight of each leaf node, This indicates that a penalty is applied to the sum of the squares of the weights of all leaf nodes to prevent overfitting.

7. The method for predicting the travel change intentions of hub passengers in the scenario of travel delay as described in claim 1, characterized in that, In step S3, the specific steps of parameter tuning include: Define a hyperparameter grid including column sampling ratio, learning rate, maximum tree depth, number of estimators, and sample sampling ratio, and set the value range for each hyperparameter; The training set is divided into several folds, and a cross-validation method is used to select a portion of the data in each fold as the validation set and the remaining data as the training subset. Iterate through all parameter combinations in the hyperparameter grid, train the model on the training subset for each parameter combination, and compute the cross-validation score on the validation set; The parameter combination that yields the optimal cross-validation score is selected as the optimal hyperparameter combination.

8. The method for predicting the travel change intentions of hub passengers in the scenario of travel delay as described in claim 1, characterized in that, In step S4, the attribution analysis is implemented using the SHAP algorithm based on Shapley values, specifically including: Construct a SHAP interpreter object to calculate the SHAP value of each variable for each sample in the test set in the passenger behavior prediction model; Calculate the average of the absolute values ​​of the SHAP of each variable across all samples, and use this as an indicator of the global importance of that variable. All variables are sorted in descending order based on the global importance index.

9. The method for predicting the travel change intentions of hub passengers in a trip delay scenario according to claim 1, characterized in that, In step S5, generating differentiated emergency dispatch strategies for different passenger groups specifically includes: Passengers are grouped according to the price range of the key influencing features. Passengers whose prices are below the first preset threshold are classified as price-sensitive groups, and passengers whose prices are between the first preset threshold and the second preset threshold are classified as medium-priced groups. Passengers are grouped according to the age range in the key influencing features, and passengers whose age is higher than the third preset threshold are classified as elderly. Based on the numerical range of delay duration in the key impact features, delay scenarios are classified into categories, and scenarios with delay duration exceeding the fourth preset threshold are classified as long-term delay scenarios. For different combinations of passenger groups and delay scenarios, corresponding emergency dispatch strategies are matched.

10. The method for predicting the travel change intentions of hub passengers in a trip delay scenario according to claim 9, characterized in that, The emergency dispatch strategy includes: For the aforementioned elderly group, a strategy is generated to provide customized ground shuttle services to hotels or city centers; For the aforementioned price-sensitive group, a strategy is generated to open shuttle bus services to the high-speed rail station and recommend air-rail intermodal transport options; For the aforementioned long-delay scenarios, strategies are generated to recommend rebooking on other flights or to provide a comfortable rest area and real-time delay updates; For the mid-priced ticket group, a strategy is generated that offers flexible rebooking packages at discounted prices.

Images