Dynamic planning allocation method and system for civil airport boarding application based on multi-business data

Abstract

The present invention belongs to the technical field of data processing for passenger boarding applications for civil aviation flights, and discloses a method and system for dynamically planning and allocating passenger boarding applications for civil airports based on multi-business data. The method includes data collection and cleaning processing, feature selection and model training. After data collection and cleaning processing, indicators or attributes that can effectively reflect the intrinsic characteristics of the historical data are selected from the historical data, a prediction model is constructed using a deep learning algorithm, and the obtained indicators or attributes reflecting the intrinsic characteristics of the historical data are used as input information to perform prediction model training; data calculation and release, as well as output verification and optimization. The present invention learns the priority protection rules in the passenger boarding application process through analysis and calculation based on historical operating data, combined with relevant data such as air traffic control, without the need for complex rule maintenance and configuration, and realizes dynamic planning processing of future flight passenger boarding applications, thereby improving airport protection efficiency and the rationality of resource allocation.

Description

Technical Field

[0001] The present invention belongs to the technical field of civil aviation flight boarding application data processing, and in particular relates to a civil airport boarding application dynamic planning allocation method and system for multi-business data. Background Art

[0002] With the development of the air transport industry, airports are facing increasing challenges, especially during peak periods such as the summer travel rush, air traffic control restrictions, and widespread flight delays. Flight delays and stranded passengers often lead to catastrophic consequences. Efficiently processing passenger boarding requests to ensure timely boarding and on-time departure has become a crucial aspect of airport operations.

[0003] Traditional methods of processing passenger boarding applications rely on manual labor, lack flexibility and intelligence, and often suffer from the following problems:

[0004] Reliance on manual experience: Traditionally, passenger pick-up applications are handled by dispatchers based on their own experience and intuition. This approach can lead to unscientific and unreasonable task allocation, and can easily lead to deviations or omissions.

[0005] Inefficient dispatching: Dispatching efficiency is greatly affected by the need to repeatedly confirm progress, especially during busy periods. Dispatchers may need to frequently use walkie-talkies or other communication tools to coordinate work, which not only wastes time but also increases communication costs.

[0006] Difficulty in fully understanding on-site conditions: Due to the ever-changing on-site conditions, it is difficult for dispatchers to fully understand the on-site operation status in real time. This may lead to insufficient information when making dispatch decisions, thus affecting the quality of dispatch.

[0007] Lack of flexibility: Traditional methods have limited room for adjustment when faced with emergencies and are unable to adapt quickly to changes. For example, when flight delays or cancellations occur, resources need to be re-coordinated, and traditional methods often seem powerless in such situations.

[0008] With the continuous advancement of information technology, big data and artificial intelligence technologies such as deep learning have been widely applied in various fields, providing new approaches and solutions to addressing problems existing in traditional approaches. In the civil aviation sector, big data technologies can help collect and process large amounts of flight data, airport operations data, weather information, and route information, providing decision makers with more comprehensive and accurate information support. Deep learning technology can automatically extract useful features from massive amounts of data through training models, and use this information to make predictions and decisions, thereby improving the operational efficiency and management level of the civil aviation system. Summary of the Invention

[0009] To overcome the problems existing in related technologies, the disclosed embodiments of the present invention provide a method and system for dynamically planning and allocating civil airport boarding applications based on multiple business data. Specifically, they relate to a method for dynamically planning and processing civil aviation flight boarding applications based on big data and deep learning. The present invention aims to provide a method for dynamically planning and processing airport boarding applications based on multiple business data, aiming to improve the efficiency and quality of boarding application processing through automated means, thereby meeting the growing needs of the civil aviation industry.

[0010] The technical solution is as follows: A method for dynamically planning and allocating civil airport boarding applications with multiple business data, comprising:

[0011] S1, data collection and cleaning processing: collect historical data from multiple data sources, perform data cleaning and data transformation;

[0012] S2, Feature Selection and Model Training: Select indicators and attributes that reflect the intrinsic characteristics of historical data from historical data, use deep learning algorithms to build a prediction model, and use the obtained indicators and attributes as input information to train the prediction model;

[0013] S3, Data Calculation and Release: Predictive models are trained using historical data and applied to flight boarding application production scenarios. Real-time dynamic flight information and air traffic control (ATC) information are collected and input into the predictive model. The predictive model, combined with historical data analysis, processes boarding applications and predicts the optimal boarding time and order to ensure flight on-time performance.

[0014] S4, output verification and optimization: Use statistical methods to automatically screen out results that do not conform to the normal pattern, mark them, and continuously update the data resource pool to eliminate interference items.

[0015] Furthermore, historical data from various data sources is collected and data cleaning and transformation are performed, including:

[0016] Collect historical data from different data sources, which are obtained through database extraction, API interface calls, and web crawling;

[0017] Data cleaning, including: removing redundant information, consistency checking, outlier detection and correction;

[0018] Data transformation, including: creating new features by performing mathematical operations on existing attributes; adjusting the value range through standardization and normalization.

[0019] Furthermore, historical data includes: flight number, aircraft type, aircraft number, route, region, mission, VIP, parking space, boarding gate, support completion time, planned landing time, actual landing time, planned take-off time, actual take-off time, boarding start time, boarding deadline time, door opening and closing time, boarding time, COBT, CTOT, weather information, route restrictions, flow control, and delay information.

[0020] In step S2, feature selection involves: based on civil aviation domain knowledge and exploratory data analysis, eliminating obviously irrelevant or noisy data fields, calculating the Pearson coefficient between variables and between each variable and the target variable, and identifying strong correlations; and iteratively removing unimportant features using a machine learning algorithm until the ideal scale is reached. The formula for calculating the Pearson coefficient is:

[0021]

[0022] Where, is the Pearson coefficient, are the values ​​of the two variables in the sample, is the mean value of the two variables in the sample, is the sample size;

[0023] The Pearson coefficient calculation formula is used to calculate the degree of linear correlation between two variables, and its value is between -1 and 1; when When it is close to 1 or -1, it indicates that there is a strong linear relationship between the two variables; when When it is close to 0, it means that there is almost no linear relationship between the two variables; a positive value indicates a positive correlation, and a negative value indicates a negative correlation.

[0024] In step S2, a prediction model is constructed using a deep learning algorithm, and the obtained indicators and attributes are used as input information to train the prediction model, including:

[0025] (a) Dataset partitioning: The dataset is divided into training set, validation set, and test set in a ratio of 8:1:1;

[0026] (b) Constructing a Transformer Model: We selected the Transformer model as the underlying architecture for the deep learning algorithm. We converted historical flight features (e.g., flight number, departure and arrival airports, departure and arrival times, COBT, weather conditions, and traffic information) into embedding vectors for use in the Transformer encoder. We constructed a six-layer encoder with eight heads per layer to form a deep Transformer model network. We used a self-attention mechanism to capture dependencies in the historical flight data, thereby capturing complex features.

[0027] (c) Reinforcement Learning Policy Optimization: Using Transformer as the basic architecture, reinforcement learning elements RL are introduced for policy optimization;

[0028] (d) Output verification and optimization: After the prediction model training is completed, the output results of the model are verified.

[0029] In step (c), we use the Transformer as the infrastructure and introduce reinforcement learning elements for policy optimization, including:

[0030] Define the state space, including the number of passengers on the current flight, passenger waiting time, and flight delays. The action space consists of the decision to allow the flight to start boarding passengers. The reward function rewards or penalizes passengers by evaluating the passenger waiting time and the flight release rate.

[0031] A deep Q-network is designed as an intelligent agent to learn the optimal boarding decision strategy and maximize the cumulative reward by optimizing the boarding process. An experience replay mechanism is used to collect and store flight boarding experience tuples in an experience replay buffer for training the deep Q-network.

[0032] The RL module works in conjunction with the Transformer module, which processes flight boarding application data. Through feature extraction, standardization, and time series data processing, it models the boarding sequences of different flights, captures long-range dependencies between input sequences, and performs big data predictions.

[0033] In step (d), the output of the model is verified, including checking the model's fit to the historical data through accuracy, mean absolute error, and mean square error.

[0034] In step S3, the calculated result includes the boarding application result information, boarding time, boarding location, and boarding sequence; the result set is output to the airport management system, air traffic control system, and airline system for user query reference.

[0035] In step S4, after the prediction model is launched, a trial run period is set up, during which a manual review mechanism is retained to compare the differences between the two. Statistical methods are used to automatically screen out result items that do not conform to the normal pattern and mark them for review and confirmation. The data resource pool is continuously updated, the internal knowledge graph content is continuously improved, and interference items are removed.

[0036] Another object of the present invention is to provide a system for dynamically planning and allocating civil airport boarding applications based on multiple business data. The system implements the method for dynamically planning and allocating civil airport boarding applications based on multiple business data. The system comprises:

[0037] Data collection and cleaning processing module, used for data collection and preliminary cleaning and processing of data;

[0038] Feature selection and feature extraction module, which is used to select the most influential subset of historical data for the prediction target, and extract new feature representations that can better reflect the internal laws by transforming or mapping historical data;

[0039] The data modeling and model training module is responsible for training the prediction model based on the extracted feature values. It uses deep learning algorithms to build a prediction model based on the flight boarding application business scenario. Through analysis and training of historical data, it makes plans for future flight boarding applications.

[0040] The decision support or predictive analysis module applies the trained prediction model to the actual production process. By comprehensively analyzing flight history records, passenger flow changes, and weather conditions, it predicts possible delays or congestion in advance and generates the optimal scheduling plan accordingly.

[0041] Combining all of the above technical solutions, the present invention has the following beneficial effects: Targeting the field of civil aviation airport operations management technology, the present invention uses analysis and calculation based on historical operational data, combined with relevant data such as air traffic control, to learn priority support rules during the passenger boarding application process. This eliminates the need for complex rule maintenance and configuration, enabling dynamic planning and processing of future flight boarding applications, thereby improving airport support efficiency and rational resource allocation. This is specifically reflected in the following aspects:

[0042] Speeding up passenger clearance: By accurately predicting the boarding order of flights during peak hours and adjusting the number of open security checkpoints based on real-time data, passengers' waiting time is effectively reduced.

[0043] Optimize flight scheduling processes: Leverage advanced predictive algorithms to proactively identify factors that may impact flight punctuality (such as weather changes, aircraft maintenance requirements, and flow control issues). This allows for the development of flexible scheduling plans to ensure on-time takeoff and landing.

[0044] Improve baggage handling capacity: Based on in-depth analysis of historical data, we can more accurately estimate the baggage volume required for each flight, and then rationally allocate relevant equipment and personnel to avoid baggage backlogs.

[0045] Enhanced emergency response capabilities: When faced with emergencies (such as severe weather causing a large number of flight cancellations or delays), the system can quickly generate emergency plans and coordinate resources from all parties for efficient handling, minimizing the scope of negative impact.

[0046] Optimized resource allocation: By analyzing historical data, we can better understand the usage of various airport facilities in different time periods, thereby achieving refined management of key resources such as parking spaces and boarding gates, and avoiding idle resources or overcrowding.

[0047] Promoting multi-party collaboration: By establishing a unified data sharing platform, information exchange and cooperation between airlines, air traffic control departments and ground service providers are strengthened, forming an efficient and collaborative working mechanism.

[0048] Continuous improvement mechanism: As more real-time feedback data accumulates, the system will have the ability to self-learn and continuously adjust and improve the algorithm logic to ensure that high-level service quality standards are maintained in the long term. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure;

[0050] Figure 1 This is a flow chart of a method for dynamically planning and allocating civil airport boarding applications for multi-service data provided by an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of a method for dynamically planning and allocating multi-service data for civil airport boarding applications provided by an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of a dynamic planning and allocation system for civil airport boarding applications based on multi-service data provided by an embodiment of the present invention;

[0053] Figure 4 is a schematic diagram of a Transformer model provided by an embodiment of the present invention;

[0054] In the figure: 1. Data acquisition and cleaning processing module; 2. Feature selection and feature extraction module; 3. Data modeling and model training module; 4. Decision support or predictive analysis module. DETAILED DESCRIPTION

[0055] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0056] The innovation of the present invention lies in that it comprehensively considers a variety of business information in the civil airport boarding application data, such as flight information, resource usage, flow control, delays, etc., filling the gaps in the domestic civil aviation field in terms of multi-source data fusion, real-time processing, intelligent decision-making, real-time monitoring and feedback. Through dynamic planning and optimization allocation, the boarding process is flexibly adjusted to maximize resource utilization and improve efficiency. Moreover, by combining big data and deep learning technology, intelligent decisions are made based on historical data and real-time conditions, and personalized solutions are provided, which not only improves passenger satisfaction, but also improves airport operation assessment indicators. In addition, the system has a real-time monitoring function, which can track the boarding process at any time and respond quickly and adjust dynamically to ensure smooth boarding. In summary, this method brings a new solution to improving boarding efficiency and passenger satisfaction, and promotes civil airport operations to move towards intelligence, real-time and personalization.

[0057] In Example 1, the method for dynamically planning and allocating multi-service data for civil airport boarding applications provided in the embodiment of the present invention is based on the calculation and analysis of a large amount of historical data. Through deep learning, a dynamic planning processing method for flight boarding applications is proposed, including: data collection and cleaning, feature selection and model training, data calculation and release, and output verification and optimization.

[0058] like Figure 1 As shown, specifically including:

[0059] S1, data collection and cleaning processing: collect historical data from multiple data sources, perform data cleaning and data transformation;

[0060] In order to ensure accurate dynamic planning and automatic processing of passenger boarding applications, a large amount of historical data analysis and calculation is required. The data comes from a wide range of sources, including but not limited to flight number, aircraft model, aircraft number, route, region, mission, VIP, parking space, boarding gate, guarantee completion time, planned landing time, actual landing time, planned take-off time, actual take-off time, boarding start time, boarding deadline, door opening and closing time, boarding time, COBT, CTOT, weather information, route restrictions, flow control, delay information, etc. These data come from various data sources such as the air traffic control system, the national traffic system, and the airport command and dispatch system. After the initial collection, the crucial cleaning and processing stage follows, which includes but is not limited to data cleaning (such as removing redundant information, filling missing values, format conversion), consistency checking (ensuring no duplicates, uniqueness constraints), outlier detection and correction (outlier detection, outlier correction), and data transformation. Data cleaning and processing includes the following tasks:

[0061] Data cleaning: The process of improving data quality by identifying and correcting erroneous, incomplete, or incorrectly formatted records. This process typically includes removing redundant information, filling missing values, and standardizing formats. Using the MurmurHash hash function with a threshold set to 0.8, each record is mapped to a unique hash value through the hash function. Two records are compared. If the hash values ​​are greater than the threshold of 0.8, the two records are considered similar or duplicates, and the duplicate records are marked or deleted. For missing values, an appropriate filling strategy is selected based on the data type, such as the mean or median for numeric data and the mode for categorical data. Furthermore, date and time fields are converted to a unified standard date and time format (such as yyyy-MM-dd HH:mm) to ensure data format consistency. Consistency check: Identify the fields that need to remain unique and set uniqueness constraints for the relevant fields in the database table or dataset. Outlier detection: A box plot is used to identify data points that deviate from the norm, using the formula IQR = Q_3 - Q_1. Use the Apache Commons Math library to calculate the quartiles and IQR, where (Q_1) and (Q_3) are the first quartile and the third quartile respectively. Any value less than (Q_1-1.5*IQR) or greater than (Q_3+1.5*IQR) is considered an outlier. The outlier can be replaced with the mean, median or boundary value according to the business situation, or the outlier record can be deleted directly. Data transformation: Create new features by performing mathematical operations on existing attributes, or adjust the numerical range through standardization and normalization to improve the performance of subsequent predictive models. For example, multiply the flight time by the number of passengers to obtain an interaction term, and create this new feature to represent the relationship between flight time and the number of passengers. And through the normalization method , normalize the route restriction information in order to balance the weight of route restriction in the calculation process, where, is the normalized data, This step aims to improve the overall data quality, eliminate potential noise interference and irregularities, and lay a solid foundation for subsequent variable screening and model training.

[0062] S2, Feature Selection and Model Training: Select indicators and attributes that reflect the intrinsic characteristics of historical data from historical data, use deep learning algorithms to build a prediction model, and use the obtained indicators and attributes as input information to train the prediction model;

[0063] The indicators or attributes that reflect the intrinsic characteristics of historical data are used as input information to train the prediction model.

[0064] For example, in the dynamic planning process of flight boarding application, the indicators that may be involved include flight number, aircraft type, route, area, VIP, boarding gate, guarantee completion time, planned landing time, actual landing time, planned take-off time, actual take-off time, boarding start time, boarding deadline, cabin door opening and closing time, boarding time, COBT, CTOT, weather information, route restrictions, flow control, delay information and other indicators.

[0065] After feature selection and extraction, the prediction model is constructed. This system utilizes deep learning technology as its core algorithm, particularly the Transformer model and reinforcement learning (RL). The Transformer model performs well when processing long sequences and is capable of parallel computation. Its use of a self-attention mechanism better captures global dependencies. Reinforcement learning (RL) not only makes predictions based on existing data but also continuously optimizes these predictions. Combined with techniques such as the Deep Q-Network (DQN) and policy gradients, it enables effective scheduling in dynamic environments. This enables the system to dynamically plan future flight boarding requests and conduct in-depth analysis to support more scientific and rational decision-making.

[0066] S3, Data Calculation and Release: Predictive models are trained using historical data and applied to flight boarding application production scenarios. Real-time dynamic flight information and air traffic control (ATC) information are collected and input into the predictive model. The predictive model, combined with historical data analysis, processes boarding applications and predicts the optimal boarding time and order to ensure flight on-time performance.

[0067] For example, after receiving data from the airport operation management system, air traffic control system, national traffic system, and airline system, the system predicts and processes flight boarding applications based on the prediction model to ensure the accuracy and real-time nature of the opening and sequence of boarding applications.

[0068] The calculation results may include information about the boarding application results, boarding time, boarding location, and boarding sequence. The result set can be exported to relevant platforms such as airport management systems, air traffic control systems, and airline systems for user reference.

[0069] S4, output verification and optimization: Utilize statistical methods to automatically screen out results that do not conform to conventional patterns, mark them, and continuously update the data resource pool to eliminate interference items;

[0070] For example, after the prediction model is first launched, a trial period is set up, during which the traditional manual review mechanism is retained to compare the differences between the two. principle), automatically screen out result items that do not conform to the normal pattern, mark them for subsequent review and confirmation, continuously update the data resource pool, continuously improve the internal knowledge graph content, and eliminate interference items.

[0071] For example, Figure 2 This is the principle of a method for dynamically planning and allocating civil airport boarding applications for multiple business data provided by an embodiment of the present invention.

[0072] Dataset: The entire process begins with a dataset containing all relevant data. This can be obtained through database extraction, API calls, web scraping, and other methods. In this example, the dataset includes various features of historical flights, such as flight number, departure and arrival airports, departure and arrival times, COBT (estimated time of departure), weather conditions, and traffic information. Feature engineering: The process of converting raw data into features suitable for modeling. During this stage, the raw data is cleaned, integrated, processed, and converted into a format suitable for model input. In particular, the features of historical flights are converted into embedding vectors, a required input format for the Transformer model. Data partitioning: The dataset is divided into three subsets: training, validation, and test. The training set is used to train the model, the validation set is used to evaluate model performance during training and prevent overfitting, and the test set is used to evaluate the final model performance after training. Model training: The Transformer model was selected as the underlying architecture. It consists of six encoder layers, each with eight heads, forming a deep Transformer model network. The Transformer model is a deep learning model that excels at processing sequential data. Through its unique self-attention mechanism, the Transformer model captures dependencies within the input historical flight data regardless of sequence order, making it highly effective for processing complex features. During model training, the Transformer model uses self-attention to capture dependencies within the historical flight data. It converts historical flight features (such as flight number, departure and arrival airports, departure and arrival times, COBT, weather conditions, and traffic information) into embedding vectors, adding positional encodings to each embedding to preserve positional information within the sequence. This embedding is then processed through a multi-layer encoder, each layer comprising a multi-head self-attention mechanism and a feedforward neural network, ultimately generating a prediction output for the flight boarding request time and sequence. During model training, the generated prediction output is compared with the true labels from the actual historical flight data. A loss function, such as mean squared error (MSE) or mean absolute error (MAE), is used to quantify the difference between the predicted output and the true labels. The loss value calculated by this loss function is used to evaluate the model's performance in the current iteration. A smaller loss value indicates that the model's prediction is closer to the true value. Through continuous iterative training, the model is continuously optimized, improving its accuracy and reliability in real-world predictions. Model Evaluation: During training, the model is periodically evaluated on the validation set. This helps monitor the model's learning progress and allows adjustments to the model's parameters or structure based on its performance on the validation set. Model Prediction: Once the model is trained and performs well on the validation set, it can be evaluated on the test set for a final evaluation. This step verifies that the model is ready to handle new, unseen data.Prediction Results: Based on the model's performance on the test set, final prediction results are obtained, including information about boarding requests, boarding time, boarding location, and boarding sequence. The result set can be used to guide dynamic planning and allocation of boarding requests at civil airports, thereby achieving more efficient resource management and services, and improving airport operational efficiency. Example 2, for example. Figure 3 As shown, the civil airport boarding application dynamic planning and allocation system for multiple business data provided by the embodiment of the present invention includes:

[0073] Data collection and cleaning module 1 is mainly responsible for data collection and preliminary cleaning of data to ensure the accuracy of subsequent prediction model training. The main steps are:

[0074] (1) Data collection: Collect historical data (raw data) from different data sources, which may include database extraction, API interface calls, web crawling, etc.

[0075] (2) Data cleaning: The process of improving data quality by identifying and correcting erroneous, incomplete, or incorrectly formatted records. This process typically includes removing redundant information, filling missing values, and unifying formats. Using the MurmurHash hash function, with a threshold set to 0.8, each record is mapped to a unique hash value through the hash function. The hash values ​​of two records are compared. If the hash values ​​are greater than the threshold of 0.8, the two records are considered similar or duplicate, and the duplicate records are marked or deleted. For missing values, an appropriate filling strategy is selected based on the data type, such as the mean or median for numerical data and the mode for categorical data. At the same time, the date and time fields are converted to a unified standard date and time format (such as yyyy-MM-dd HH:mm format) to ensure data format consistency.

[0076] (3) Consistency check: Identify the fields that need to maintain uniqueness and set uniqueness constraints for related fields in the database table or data set.

[0077] (4) Outlier Detection: A box plot method is used to identify data points that deviate from the norm. The formula is IQR = Q_3 - Q_1. The Apache Commons Math library is used to calculate the quartiles and IQR, where (Q_1) and (Q_3) are the first and third quartiles, respectively. Any value less than (Q_1 - 1.5 * IQR) or greater than (Q_3 + 1.5 * IQR) is considered an outlier. Depending on the business situation, the outlier can be replaced with the mean, median, or boundary value, or the outlier record can be directly deleted.

[0078] (5) Data transformation: Create new features by performing mathematical operations on existing attributes, or adjust the value range through normalization and standardization to improve the performance of subsequent prediction models. For example, multiply the flight time by the number of passengers to obtain an interaction term, and create this new feature to represent the relationship between the flight time and the number of passengers. And through the normalization method , normalize the route restriction information in order to balance the weight of route restriction in the calculation process, where, is the normalized data, It is the original data.

[0079] Feature Selection and Extraction Module 2: This module is primarily responsible for selecting the most influential subset of historical data for the prediction target, thereby reducing dimensionality, improving prediction model performance, and enhancing interpretability. Furthermore, by transforming or mapping historical data, it generates new feature representations that better reflect inherent patterns, thereby improving the algorithm's learning ability and achieving dimensionality reduction, thereby improving the accuracy and efficiency of the overall prediction model. The main steps are:

[0080] 1) Feature Selection: Based on civil aviation domain knowledge and exploratory data analysis, we eliminated data fields that were clearly irrelevant or noisy. We also calculated the Pearson Correlation Coefficient between variables and between them and the target variable to identify strong correlations. We constructed multiple decision trees using the random forest method, setting the number of trees to 200, the maximum depth to 10, and the threshold to 30%. We iteratively modified the configuration parameters to avoid overfitting and underfitting. We ranked features based on their importance scores, selecting the top-ranked features and removing the least important ones until we reached the ideal scale.

[0081] For example, the Pearson coefficient measures the degree of linear correlation between two variables. Based on civil aviation knowledge and exploratory data analysis, it eliminates obviously irrelevant or noisy data fields, calculates the Pearson coefficient between variables and between each variable and the target variable, and identifies strong correlations. It then iteratively removes unimportant features using a machine learning algorithm until the ideal scale is reached. The formula for calculating the Pearson coefficient is:

[0082]

[0083] Where, is the Pearson coefficient, are the values ​​of the two variables in the sample, is the mean value of the two variables in the sample, is the sample size;

[0084] The Pearson coefficient calculation formula is used to calculate the degree of linear correlation between two variables, and its value is between -1 and 1; when When it is close to 1 or -1, it indicates that there is a strong linear relationship between the two variables; when When it is close to 0, it means that there is almost no linear relationship between the two variables; a positive value indicates a positive correlation, and a negative value indicates a negative correlation.

[0085] 2) Feature extraction: After feature selection, the feature values ​​that are closely related to the prediction model are extracted for use in prediction model training. In terms of dynamic planning processing of flight boarding applications, relevant features may include, but are not limited to, time information: flight departure time, flight arrival time at the airport, boarding time, COBT, CTOT, cabin door opening and closing time, and guarantee completion time. Flight information: airline, flight number, region, mission, VIP, aircraft number, aircraft model, destination, flight duration, and passenger capacity. Resource information: terminal, aircraft stands, boarding gates, baggage carousels, security checkpoints, and runways. Route information: route restrictions, flow control information, and weather conditions.

[0086] Data Modeling and Model Training Module 3: This module is primarily responsible for training the prediction model based on the extracted feature values. It uses a deep learning algorithm to build a prediction model based on the flight boarding application business scenario. The prediction model primarily consists of a Transformer module and an RL module. The two modules are combined to form a hybrid model that combines the sequence modeling capabilities of the Transformer model with the decision-making optimization capabilities of the RL module. The different modules collaborate with each other to complete the prediction task, enabling the model to better handle complex prediction problems and achieve higher performance and generalization capabilities. This module aims to plan for future flight boarding applications through analysis and training of historical data. The main steps of this module are:

[0087] (a) Dataset partitioning: The dataset is divided into training set, validation set, and test set in a ratio of 8:1:1 to ensure the generalization ability of the prediction model.

[0088] (b) Build a Transformer model (such as Figure 4 ): The present invention selects the Transformer model as the underlying architecture of the deep learning algorithm, and converts the characteristics of historical flights such as flight number, take-off and landing airports, take-off and landing times, COBT, weather conditions, traffic information, etc. into embedding vectors for use by the Transformer encoder. At the same time, a 6-layer encoder is set up with 8 heads in each layer to form a deep Transformer model network. The self-attention mechanism is used to capture the dependencies in the historical flight data to capture complex features.

[0089] Figure 4The figure shows multiple tasks (T1, T2...TN) and their corresponding inputs (E1, E2...EN). Each business domain (such as flight scheduling, passenger services, and ground operations) is considered a separate task (T1, T2...TN), each with its own unique inputs (E1, E2...EN), such as flight schedules, passenger information, weather information, airport resource usage, and flow control information. These inputs are first fed into a shared representation layer (Trm). This layer plays a central role by transforming the inputs from each task into a common form and extracting common cross-domain features or high-level concepts. These features may include flight frequency patterns, passenger behavior preferences, weather impact, and flow control impact. Following the shared representation layer, each task has its own dedicated representation layer (Trm). These layers further process the outputs from the shared layer to better suit their specific needs. For example, in the flight schedule management task, the model may need to focus on connecting times and idle time periods between flights; whereas, in the flow control task, the model may be more concerned with the heading and distance of outbound flights. In this way, the Transformer-based multi-task learning framework not only effectively integrates and utilizes data from different business domains, but also promotes knowledge sharing and complementarity between different tasks, thereby improving overall prediction accuracy and decision quality. Furthermore, this framework is highly scalable, allowing new tasks to be easily added as new business needs emerge without requiring major modifications to the entire system.

[0090] (c) Reinforcement Learning Policy Optimization: The prediction model uses the Transformer as its infrastructure and incorporates reinforcement learning elements for policy optimization. The reinforcement learning (RL) module is designed to optimize the prediction of flight boarding request times and sequences. First, the state space is defined, including the number of passengers on the current flight, passenger waiting times, and flight delays. The action space consists of decisions about whether to allow a flight to begin boarding, i.e., whether to allow the flight to begin boarding at a specific time. The reward function rewards or penalizes passengers by evaluating metrics such as passenger waiting time and flight clearance rates, enabling the prediction model to make better boarding decisions and update the policy using a deep Q-network (DQN). Through multiple rounds of iterative optimization and improvement of overall performance, the RL module will ultimately work in conjunction with the Transformer component, enabling the system to flexibly respond to different flight scenarios and provide accurate predictions of boarding times and sequences.

[0091] In step (c), the reinforcement learning strategy is optimized, including: the reinforcement learning RL module is used to optimize the prediction of the time and order of flight boarding applications. First, the state space is defined, including the number of passengers on the current flight, the waiting time of passengers, and the flight delays. The action space consists of the decision to allow a flight to start boarding, and the choice of whether to allow the flight to start boarding at a specific time. The reward function rewards or punishes by evaluating the passenger waiting time and the flight release regularity index. For example, successfully processing a boarding application is rewarded with +2 points, the passenger waiting time is less than 1 hour +1 point, and the flight is released and taken off normally is +3 points; failing to process a boarding application is deducted with -2 points, not processing the boarding application after the scheduled time is deducted with -1 point, violating business rules is deducted with -2 points, and flight takeoff is delayed with -3 points. The reward function is adjusted based on user feedback, and the effects of different reward functions are compared to select the optimal solution to incentivize the model to make better boarding decisions and improve the operational efficiency of the whole day.

[0092] A Deep Q-Network (DQN) is designed as an intelligent agent to learn the optimal boarding decision strategy. By optimizing the boarding process, cumulative rewards are maximized, for example, reducing passenger wait times and improving flight departure and departure on-time rates. Incorporating an experience replay mechanism, we collect experience tuples related to flight boarding and store them in an experience replay buffer. These experiences are used to train the Deep Q-Network, thereby learning the optimal boarding decision strategy. A target network is introduced to stabilize the training process and reduce volatility. Mean squared error (MSE) or Huber loss is used to minimize the gap between predicted and target Q-values. Continuous model optimization and training data updates are used to continuously improve model performance, enhancing flight boarding efficiency and flight departure and departure on-time rates. Optimization is continuously performed through multiple rounds of iterations.

[0093] Ultimately, the RL module works in tandem with the Transformer module. The Transformer module processes flight boarding request data. Through feature extraction, normalization, and time-series data processing, it models boarding sequences for different flights, captures long-range dependencies between input sequences, and performs big data predictions. Simultaneously, the RL module, based on a Deep Q-Network (DQN), learns the optimal boarding decision strategy. Through interaction with the environment, it continuously explores and optimizes boarding behavior to maximize cumulative rewards, such as reducing boarding wait times and improving flight departure and clearance rates. By combining techniques such as experience replay, target networks, and loss functions, it stabilizes the training process and gradually optimizes the decision strategy. In this collaborative work, the Transformer module provides predicted boarding demand and trends, providing the RL module with decision-making basis and context. Based on the Transformer module's predictions, the RL module dynamically adjusts the boarding strategy, achieving intelligent decision-making. This collaborative work seamlessly integrates data prediction and decision optimization, improving overall system performance and efficiency, making the system more intelligent and efficient, and delivering enhanced service quality and passenger experience. This method of comprehensively utilizing different technical means can effectively cope with complex passenger boarding scenarios and improve the overall performance and competitiveness of the system.

[0094] (d) Output verification and optimization: After the prediction model training is completed, the output results of the model need to be verified to ensure the accuracy and reliability of the prediction. For example, the prediction accuracy, mean absolute error, mean square error, etc., are used to check the fit of the model on historical data. This helps to identify problems, such as whether the model is too complex (overfitting) or unable to capture effective trends (underfitting). Based on the evaluation results, the present invention may adjust hyperparameters or select different features to improve the prediction effect. Through continuous testing and verification, the present invention ensures that the prediction model can accurately predict the boarding situation of flights in actual applications, thereby helping airlines make better decisions and operational plans.

[0095] Among them, the mean absolute error Mean absolute error (MAE) is an important metric for measuring the performance of a prediction model, particularly for regression problems. It evaluates the quality of a model by calculating the average of the absolute differences between the predicted and actual values ​​across all observations. Its formula is as follows:

[0096]

[0097] Where, is the number of samples, To predict flight boarding time, is the actual boarding time of the flight. During the output verification process, based on the historical data model prediction, multiple sets of data are substituted into the formula to calculate the final mean absolute error value;

[0098] Mean square error The squared average of the difference between the predicted value and the true value reflects the overall volatility of the model's prediction results. The formula is as follows:

[0099]

[0100] Accuracy: This represents the proportion of correctly predicted samples to the total number of samples, directly reflecting the model's classification ability. The formula is as follows: TP represents the number of samples correctly predicted as positive, FP represents the number of samples incorrectly predicted as positive, TN represents the number of samples correctly predicted as negative, and FN represents the number of samples incorrectly predicted as negative. Substitute multiple sets of data into the formula to calculate the final accuracy.

[0101]

[0102] Decision Support or Predictive Analysis Module 4: This module applies the trained passenger boarding request prediction model to actual production processes. By comprehensively analyzing factors such as flight history, passenger flow fluctuations, and weather conditions, this module can predict potential delays or congestion in advance and generate an optimal scheduling plan based on this information. The system also dynamically adjusts departure priorities based on real-time updates, ensuring that every aircraft can take off or land smoothly within the optimal time window. This advanced tool allows airport managers to significantly improve operational efficiency, shorten passenger wait times, and significantly enhance overall service quality and user experience.

[0103] The above examples demonstrate that, in terms of passenger boarding application decision-making analysis, with the continuous increase in the number of flights and travelers in recent years, airports have faced major challenges, including inefficient boarding processes, extended passenger wait times, and irrational resource allocation. These challenges not only increase operating costs but also risk declining passenger satisfaction and exacerbating safety hazards. By implementing a dynamic planning approach for passenger boarding applications, combined with historical data analysis and relevant air traffic control clearance restrictions, information such as optimal boarding times, boarding sequences, and resource availability can be pushed to airport and airline systems. This assists airport scheduling operations, improves boarding efficiency, rationalizes resource allocation, and enhances the overall passenger experience.

[0104] In terms of delay warning: we use historical data and real-time information such as control and flow control to conduct comprehensive analysis to determine whether the flight can meet the normal boarding and take-off requirements under the release restrictions, provide early warning of possible delays, and provide managers with scientific decision-making support to help them achieve more reasonable flight scheduling and resource allocation, thereby optimizing overall operational efficiency and service levels, improving passengers' travel experience and controlling potential economic losses and safety risks.

[0105] This invention effectively processes boarding requests, reduces boarding time, optimizes resource utilization, reduces labor costs and energy consumption, thereby improving airport operational efficiency and maximizing economic benefits. It also enhances passenger satisfaction, strengthens customer loyalty, and enhances the airport's positive reputation and brand value. Furthermore, the technical solution of this invention can be applied not only to passenger boarding optimization at civil airports but also to other areas, such as public transportation venues like train stations and ferry terminals, and even to personnel management at large-scale events, expanding its commercial application potential and market prospects.

[0106] The present invention addresses the following issues: ① Traditional boarding optimization algorithms often use static planning or heuristic methods, making it difficult to effectively respond to real-time changes in boarding application requirements and conditions. This invention introduces a reinforcement learning module as the core of the optimization algorithm. Through interactive learning with the environment, the system can continuously adjust its decision-making strategy based on reward signals, filling a gap in the application of RL reinforcement learning in the field of boarding optimization. ② Traditional boarding optimization algorithms often require manual design of features and rules when processing data, making them difficult to adapt to complex and changing boarding environments. This invention utilizes a Transformer module, a deep learning model based on the self-attention mechanism, which can better understand and process input data, improving the system's data learning capabilities and inference accuracy, thus filling a gap in the application of deep learning models in the field of boarding optimization. ③ Traditional boarding optimization algorithms mostly use static planning methods and lack the ability to dynamically adjust to real-time changes. This invention combines RL reinforcement learning with the Transformer module to achieve intelligent decision-making and dynamic adjustment. The system can autonomously learn and optimize based on real-time conditions, filling a technical gap in the integration of dynamic planning and intelligent decision-making in the field of boarding optimization.

[0107] The above description is only a preferred specific implementation method of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with this technical field within the technical scope disclosed by the present invention and within the spirit and principles of the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for dynamically planning and allocating passenger boarding applications at a civil airport based on multiple business data, characterized in that: This approach comprehensively considers various business information in civil airport boarding application data, including flight information, resource usage, flow control, and delays. It fills a gap in the civil aviation field in terms of multi-source data integration, real-time processing, intelligent decision-making, and real-time monitoring and feedback. Through dynamic planning and optimized allocation, the boarding process can be flexibly adjusted to maximize resource utilization and improve efficiency. Furthermore, by combining big data and deep learning technologies, intelligent decisions can be made based on historical data and real-time conditions, providing personalized solutions to improve passenger satisfaction and enhance airport operational performance indicators. This approach includes: S1, Data Collection and Cleaning Processing: Collect historical data from multiple data sources and perform data cleaning and transformation. Historical data includes: flight number, aircraft type, aircraft number, route, region, mission, VIP, gate, support completion time, planned landing time, actual landing time, planned take-off time, actual take-off time, boarding start time, boarding deadline time, door opening and closing time, boarding time, COBT, CTOT, weather information, route restrictions, traffic control, and delay information. S2, Feature Selection and Model Training: Select indicators and attributes that reflect the intrinsic characteristics of historical data from historical data, use deep learning algorithms to build a prediction model, and use the obtained indicators and attributes as input information to train the prediction model; S3, Data Calculation and Release: Predictive models are trained using historical data and applied to flight boarding application production scenarios. Real-time dynamic flight information and air traffic control (ATC) information are collected and input into the predictive model. The predictive model, combined with historical data analysis, processes boarding applications and predicts the optimal boarding time and order to ensure flight on-time performance. S4, output verification and optimization: Utilize statistical methods to automatically screen out results that do not conform to conventional patterns, mark them, and continuously update the data resource pool to eliminate interference items; In step S2, a prediction model is constructed using a deep learning algorithm, and the obtained indicators and attributes are used as input information to train the prediction model, including: (a) Dataset partitioning: The dataset is divided into training set, validation set, and test set in a ratio of 8:1:1; (b) Constructing a Transformer Model: The Transformer model is selected as the underlying architecture of the deep learning algorithm. The historical flight features (flight number, departure and arrival airports, departure and arrival times, COBT, weather conditions, and traffic information) are converted into embedding vectors for use by the Transformer encoder. A six-layer encoder with eight heads per layer is set up to form a deep Transformer model network. The self-attention mechanism is used to capture the dependencies in the historical flight data to capture complex features. (c) Reinforcement learning strategy optimization: Using Transformer as the basic architecture, reinforcement learning elements are introduced for strategy optimization; (d) Output verification and optimization: After the prediction model training is completed, the output results of the model are verified; In step (c), we use the Transformer as the infrastructure and introduce reinforcement learning elements for policy optimization, including: Define the state space, including the number of passengers on the current flight, passenger waiting time, and flight delays. The action space consists of the decision to allow the flight to start boarding passengers. The reward function rewards or penalizes passengers by evaluating the passenger waiting time and the flight release rate. A deep Q-network is designed as an intelligent agent to learn the optimal boarding decision strategy and maximize the cumulative reward by optimizing the boarding process. An experience replay mechanism is used to collect and store flight boarding experience tuples in an experience replay buffer for training the deep Q-network. The reinforcement learning module works in conjunction with the Transformer module. The Transformer module processes flight boarding request data. Through feature extraction, standardization, and time series data processing, it models boarding sequences for different flights, captures long-range dependencies between input sequences, and performs big data predictions. In step S2, feature selection includes: based on civil aviation domain knowledge and exploratory data analysis, eliminating obviously irrelevant or noisy data fields, calculating the Pearson coefficient between variables and between each variable and the target variable, and finding the existence of strong correlations; combining machine learning algorithms to iteratively remove unimportant features until the ideal scale is reached. The calculation formula of the Pearson coefficient is: In the formula, r is the Pearson coefficient, x i ,y i are the values ​​of the two variables in the sample, is the mean of the two variables in the sample, and n is the size of the sample; The Pearson coefficient calculation formula is used to calculate the degree of linear correlation between two variables, and its value ranges from -1 to 1. When r is close to 1 or -1, it indicates that there is a strong linear relationship between the two variables. When r is close to 0, it indicates that there is almost no linear relationship between the two variables. A positive value indicates a positive correlation, and a negative value indicates a negative correlation. In step (d), the output of the model is verified, including: checking the fit of the model to the historical data through accuracy, mean absolute error, and mean square error; In step S3, the calculated result includes the boarding application result information, boarding time, boarding location, and boarding sequence; the result set is output to the airport management system, air traffic control system, and airline system for user query reference.

2. The method for dynamically planning and allocating multi-service data for civil airport boarding applications according to claim 1, characterized in that: Collect historical data from multiple data sources, perform data cleaning and data transformation, including: Collect historical data from different data sources, which are obtained through database extraction, API interface calls, and web crawling; Data cleaning, including: removing redundant information, consistency checking, outlier detection and correction; Data transformation, including: creating new features by performing mathematical operations on existing attributes; adjusting the value range through standardization and normalization.

3. The method for dynamically planning and allocating multi-service data for civil airport boarding applications according to claim 1, characterized in that: In step S4, after the prediction model is launched, a trial run period is set up, during which a manual review mechanism is retained to compare the differences between the two. Statistical methods are used to automatically screen out result items that do not conform to the normal pattern and mark them for review and confirmation. The data resource pool is continuously updated, the internal knowledge graph content is continuously improved, and interference items are removed.

4. A multi-service data civil airport boarding application dynamic planning allocation system, characterized by: The system implements the method for dynamically planning and allocating civil airport boarding applications for multi-service data as described in any one of claims 1 to 3, and the system comprises: Data collection and cleaning processing module (1), used for data collection and preliminary cleaning processing of data; Feature selection and feature extraction module (2) is used to select the most influential subset of historical data for the prediction target, and extract new feature representations that can better reflect the internal laws by transforming or mapping the historical data; The data modeling and model training module (3) is responsible for training the prediction model based on the extracted feature values. It uses deep learning algorithms to build a prediction model based on the flight boarding application business scenario. Through the analysis and training of historical data, it makes plans for future flight boarding applications. The decision support or prediction analysis module (4) applies the trained prediction model to the actual production process. By comprehensively analyzing flight history records, passenger flow changes, and weather conditions, it predicts delays or congestion in advance and generates the optimal scheduling plan accordingly.

Images