A dynamic flight information management method, device and medium based on machine learning

Abstract

The application discloses a kind of dynamic flight information management method, equipment and medium based on machine learning, it is related to civil aviation transport management technical field, including, based on comprehensive flight information dataset, construct flight space-time relation graph, and extract space-time feature and time dynamic feature, generate feature dataset;Based on feature dataset, through allocation rule formulation gate assignment scheme, and in combination with flight dispatch factor, with the optimization target of minimizing total taxi time and maximizing bridge rate, optimize gate assignment scheme by particle swarm optimization algorithm;Collect real-time monitoring data flow, in combination with the gate assignment scheme optimized, predict the risk of flight delay by random forest model, and in combination with early warning threshold, identify the flight exceeding early warning threshold, generate flight resource adjustment suggestion.The application realizes multidimensional feature mining to flight operation state, improves flight state feature representation accuracy and comprehensiveness.

Description

Technical Field

[0001] This invention relates to the field of civil aviation transport management technology, and in particular to a method, equipment and medium for dynamic flight information management based on machine learning. Background Technology

[0002] In the field of civil aviation operations management, efficient processing of flight information and optimized allocation of resources have always been core research topics. Traditional methods usually rely on pre-defined flight plans and static resource allocation schemes, using rule engines or linear programming models to arrange resources such as gates and boarding gates. Traditional methods are mainly based on historical operational data and fixed rules, which can ensure the orderly operation of airports to a certain extent. Existing technical solutions mostly adopt a phased processing mode, that is, first complete data integration and cleaning, and then make resource allocation decisions based on the processed dataset. This serial processing architecture has a relatively wide application foundation in the civil aviation field.

[0003] Existing technologies are ill-suited to the dynamic changes that occur during flight operations. When unexpected situations such as flight delays or sudden weather changes occur, there is a lack of dynamic adjustment mechanisms that are linked to real-time operational status, leading to reduced resource utilization efficiency. In addition, prediction models trained on historical data often lack online learning capabilities and cannot continuously optimize model parameters based on real-time feedback data. This results in discrepancies between prediction results and actual operational status, affecting the accuracy and timeliness of flight resource management. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a machine learning-based dynamic flight information management method to solve the problems of insufficient resource allocation adaptability and decision bias in prediction models.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a dynamic flight information management method based on machine learning, comprising,

[0008] Collect and preprocess flight information from multiple sources to generate a comprehensive flight information dataset;

[0009] Based on a comprehensive flight information dataset, a flight spatiotemporal relationship graph is constructed, and spatiotemporal features and time dynamic features are extracted to generate a feature dataset.

[0010] Based on the feature dataset, an aircraft stand allocation scheme is formulated through allocation rules. Combined with flight scheduling factors, the scheme aims to minimize total taxiing time and maximize docking rate. The particle swarm optimization algorithm is then used to optimize the aircraft stand allocation scheme.

[0011] Collect real-time monitoring data streams, combine them with the optimized gate allocation scheme, predict the risk of flight delays using a random forest model, and identify flights that exceed the warning threshold by combining the warning threshold with the warning threshold, and generate flight resource adjustment suggestions.

[0012] According to the flight information sharing protocol, real-time dynamic data and flight resource adjustment suggestions are shared in real time through a standardized interface, and real-time feedback is recorded to dynamically optimize the random forest model.

[0013] As a preferred embodiment of the machine learning-based dynamic flight information management method of the present invention, the multi-source flight information includes flight plan data, real-time status data, flight departure information, air traffic status data, and environmental information.

[0014] The preprocessing includes data cleaning, missing value imputation, and format conversion.

[0015] As a preferred embodiment of the machine learning-based dynamic flight information management method of the present invention, the steps of constructing a flight spatiotemporal relationship graph based on a comprehensive flight information dataset, extracting spatiotemporal features and time dynamic features, and generating a feature dataset are as follows.

[0016] The feature information of flights is extracted from the comprehensive flight information dataset as flight nodes, the relationships between flight nodes are used as relationship edges, and a spatiotemporal relationship graph of flights is constructed.

[0017] Based on the historical flight spatiotemporal relationship graph, the loss function is minimized through backpropagation algorithm, and the convolution kernel weights of the graph convolutional network are optimized to generate the trained graph convolutional network.

[0018] The flight spatiotemporal relationship graph is input into the trained graph convolutional network, and long-term dependencies are captured through multiple convolutional layers. The embedding vector of each flight node is output as a time dynamic feature and spatiotemporal feature.

[0019] The spatiotemporal features and time dynamic features of all flight nodes are fused into a feature dataset.

[0020] As a preferred embodiment of the machine learning-based dynamic flight information management method of the present invention, the steps for formulating a gate allocation scheme based on a feature dataset and allocation rules are as follows:

[0021] The feature information of flight nodes and the weight matrix of relation edges are extracted from the feature dataset. Based on the predefined allocation rules, the initial gate allocation scheme for flights is formulated.

[0022] Based on the time dynamic features in the feature dataset, the initial gate allocation scheme for flights is dynamically adjusted to generate a complete gate allocation scheme.

[0023] As a preferred embodiment of the machine learning-based dynamic flight information management method of the present invention, the step of optimizing the gate allocation scheme by combining flight scheduling factors and minimizing total taxiing time and maximizing gate utilization rate as optimization objectives, using a particle swarm optimization algorithm, is as follows:

[0024] The fitness function is defined based on the initial gate allocation scheme and in combination with flight scheduling factors;

[0025] The initial camera position allocation scheme is encoded as the initial population of particle swarm optimization. The initial camera position allocation scheme is then iteratively optimized using the particle swarm optimization algorithm to generate the optimized camera position allocation scheme.

[0026] As a preferred embodiment of the machine learning-based dynamic flight information management method of the present invention, the steps of collecting real-time monitoring data streams, combining them with an optimized gate allocation scheme, and predicting the risk of flight delays using a random forest model are as follows.

[0027] Collect real-time monitoring data streams and integrate them with the optimized station allocation scheme to form a real-time operational status dataset;

[0028] Flight operation indicators are extracted from the real-time operational status dataset and weighted into a predictive feature vector;

[0029] Based on a historical comprehensive flight information dataset, multiple decision trees are constructed, and the predictions of all decision trees are aggregated to generate a trained random forest model.

[0030] The predicted feature vectors are input into the trained random forest model, which outputs the probability of flight delays.

[0031] As a preferred embodiment of the machine learning-based dynamic flight information management method of the present invention, the steps of identifying flights exceeding the warning threshold and generating flight resource adjustment suggestions by combining the warning threshold are as follows:

[0032] Based on the warning threshold, identify flights that exceed the warning threshold and generate a list of high-risk flights;

[0033] Based on a list of high-risk flights and combined with real-time monitoring data streams, the main causes of flight delays are identified, and suggestions for adjusting flight resources are generated.

[0034] As a preferred embodiment of the machine learning-based dynamic flight information management method of the present invention, the steps of sharing real-time dynamic data and flight resource adjustment suggestions in real time through a standardized interface according to the flight information sharing protocol, recording real-time feedback, and dynamically optimizing the random forest model are as follows.

[0035] Based on the data format standards defined in the Flight Information Sharing Protocol, real-time monitoring data streams and flight resource adjustment suggestions are encapsulated into standardized data packets;

[0036] Standardized data packets are shared to the dispatch center in real time, flight resource adjustment suggestions are implemented, real-time feedback data is collected, and the parameters of the random forest model are dynamically optimized through incremental learning algorithms.

[0037] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the machine learning-based dynamic flight information management method described in the first aspect of the present invention.

[0038] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the machine learning-based dynamic flight information management method described in the first aspect of the present invention.

[0039] The beneficial effects of this invention are as follows: by constructing a flight spatiotemporal relationship graph and extracting spatiotemporal features and time dynamic features, multi-dimensional feature mining of flight operation status is realized, improving the accuracy and comprehensiveness of flight status feature representation; by dynamically optimizing the parameters of the random forest model, flight delay risk prediction can dynamically adapt to changes in the real-time operating environment, enhancing the timeliness of flight management decisions. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart of a machine learning-based dynamic flight information management method.

[0042] Figure 2 This is a flowchart of data preprocessing and feature extraction.

[0043] Figure 3 A flowchart for optimizing camera station allocation.

[0044] Figure 4 This is a flowchart for risk prediction and sharing. Detailed Implementation

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0046] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0047] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0048] Reference Figures 1-4 This is one embodiment of the present invention, which provides a dynamic flight information management method based on machine learning, including the following steps:

[0049] S1. Collect multi-source flight information and preprocess it to generate a comprehensive flight information dataset;

[0050] Multi-source flight information includes flight schedule data, real-time status data, flight departure information, air traffic status data, and environmental information;

[0051] Furthermore, flight plan data, including scheduled departure and arrival times, flight numbers, and airlines, is obtained from the flight plan systems provided by the Civil Aviation Administration of China and airlines; real-time status data (including actual flight status, whether there are delays, and actual departure and arrival times) is collected from flight tracking services; flight departure information is collected from the airport's departure scheduling system; air traffic status data (such as route congestion, flight flow, and flight altitude) is collected from the air traffic control center; and environmental information (such as weather conditions, wind speed, and temperature) is collected from airport meteorological monitoring equipment.

[0052] Preprocessing includes data cleaning, missing value imputation, and format conversion;

[0053] Furthermore, invalid and duplicate data are removed from multi-source flight information to ensure that the information for each flight is complete and conflict-free; for missing flight information, linear interpolation is used to take known data points before and after the missing value, calculate the value of the missing point proportionally, and supplement the missing data during the suspension period; flight information from different data sources is unified into a standard format, for example, the date and time format is unified into UTC time, and all values ​​are unified into the same unit.

[0054] The preprocessed multi-source flight information is integrated into comprehensive flight information data;

[0055] Furthermore, the preprocessed multi-source flight information from different sources is merged according to flight number, date, and time, and overlapping and inconsistent flight information is filled in by linear interpolation within the time window.

[0056] S2. Based on the comprehensive flight information dataset, construct a flight spatiotemporal relationship graph, extract spatiotemporal features and time dynamic features, and generate a feature dataset;

[0057] The feature information of flights is extracted from the comprehensive flight information dataset as flight nodes, the relationships between flight nodes are used as relationship edges, and a spatiotemporal relationship graph of flights is constructed.

[0058] Furthermore, feature information of flights is extracted from the comprehensive flight information dataset as nodes (including flight number, take-off and landing time, flight status, air traffic status, and environmental information). Based on the spatiotemporal relationship between flights, the relationship between flight nodes is defined as an edge (for example, if two flights belong to the same airline or the same airport, then there is an edge between them). The attributes of the edge are set according to the correlation between flights (such as the same flight number, the same airline, and the same airport). The relationship between all flight nodes and related flight nodes is associated according to flight number, take-off and landing time, and correlation to generate a flight spatiotemporal relationship graph model.

[0059] Based on the historical flight spatiotemporal relationship graph, the loss function is minimized through backpropagation algorithm, and the convolution kernel weights of the graph convolutional network are optimized to generate the trained graph convolutional network.

[0060] Furthermore, flight nodes, relationship edges, and historical time series features are extracted from the historical flight spatiotemporal relationship graph and input into the initial graph convolutional network for training. The difference between the predicted and actual values ​​is calculated using the mean squared error loss function, and a loss function is defined. The convolutional kernel weights of the network are adjusted using the backpropagation algorithm to minimize the error between the predicted and actual flight states. Through multiple iterations, the convolutional kernel weights are continuously optimized until the preset number of iterations is reached, generating the trained graph convolutional network.

[0061] It should be noted that the historical flight time series dataset refers to a dataset of detailed operation records of historical flights, including flight take-off and landing times, delay times, flight status, weather conditions, and air traffic conditions; the number of iterations is usually set according to the task complexity and training speed, and the exemplary value is usually 50 to 200 times; the backpropagation algorithm is an optimization method used to train neural networks. It calculates the predicted value through forward propagation, compares it with the actual value to calculate the mean square error, and calculates the gradient of each neuron layer by layer from the output layer based on the mean square error, and updates the weights and biases in the network according to the gradient.

[0062] The flight spatiotemporal relationship graph is input into the trained graph convolutional network, and long-term dependencies are captured through multiple convolutional layers. The embedding vector of each flight node is output as a time dynamic feature and spatiotemporal feature.

[0063] Furthermore, flight nodes, relationship edges, and time-series features are extracted from the flight spatiotemporal relationship graph and input into the trained graph convolutional network. The graph convolutional network learns the spatiotemporal dependencies between flights by aggregating the feature information of each flight node and its neighboring nodes. Each layer of graph convolution calculates the feature information of flight nodes through the convolution kernel and weights and merges it with the feature information of neighboring nodes. As the number of network layers increases, more complex spatiotemporal patterns are gradually captured. At the same time, the graph convolutional network processes time-series data through multiple convolutional layers, gradually capturing the dynamic features of flight status changes over time, such as flight delays, changes in air traffic flow, and weather changes. Finally, an embedding vector is generated for each flight node, which encodes spatiotemporal features and time dynamic features.

[0064] The spatiotemporal features and time dynamic features of all flight nodes are fused into a feature dataset;

[0065] Furthermore, the spatiotemporal features and time dynamic features of all flight nodes are converted into spatiotemporal feature vectors and time dynamic feature vectors, and then connected sequentially in terms of dimensions to directly merge them into a feature dataset.

[0066] S3. Based on the feature dataset, a gate allocation scheme is formulated through allocation rules, and combined with flight scheduling factors, with the optimization objectives of minimizing total taxiing time and maximizing gate utilization rate, the gate allocation scheme is optimized through particle swarm optimization algorithm;

[0067] The feature information of flight nodes and the weight matrix of relation edges are extracted from the feature dataset. Based on the predefined allocation rules, the initial gate allocation scheme for flights is formulated.

[0068] Furthermore, the spatiotemporal features of each flight node (such as flight number, take-off and landing time, and air traffic status) are extracted from the feature dataset. At the same time, the weight matrix of the relationship edges between flight nodes is extracted, including the spatiotemporal dependencies between flight nodes, such as flights from the same airline or taking off and landing at the same airport. Combined with predefined allocation rules, an initial gate allocation scheme is generated.

[0069] It should be noted that the allocation rules are a series of rules formulated based on the characteristics, priority, airport resource constraints and actual operational needs of flights. The allocation rules are based on the spatiotemporal characteristics of flights and are usually set through historical comprehensive flight information datasets and airport scheduling experience. The allocation rules include prioritizing the allocation of gates for important and long-haul flights, ensuring sufficient time intervals between flights to avoid delayed flights occupying gates for too long, and reasonably allocating gate types (such as large aircraft and small aircraft).

[0070] Based on the time dynamic features in the feature dataset, the initial gate allocation scheme for flights is dynamically adjusted to generate a complete gate allocation scheme.

[0071] Furthermore, the time dynamic features of each flight node are extracted from the feature dataset, and the flight status changes over time. Based on the time dynamic features, the initial gate allocation scheme is dynamically adjusted, gates are reallocated, and a complete gate allocation scheme is generated. For example, for flights with significant delays, the gate occupancy time is increased or gates closer to the boarding gate are prioritized. For flights that are significantly affected by weather or air traffic, the take-off and landing order of the flights and the delayed gate allocation are adjusted.

[0072] The fitness function is defined based on the initial gate allocation scheme and in combination with flight scheduling factors;

[0073] Furthermore, based on the initial gate allocation scheme and combined with flight scheduling factors, a fitness function is defined by comprehensively considering flight priority, delay time, air traffic flow changes, and weather conditions. The fitness function aims to minimize flight delays and maximize gate utilization efficiency. The weight of flight delay time is based on economic cost, considering that flight delays lead to additional fuel, maintenance, and passenger compensation costs for airlines, resulting in significant socio-economic losses. This is achieved by quantifying the average cost per unit of delay (the higher the delay cost, the greater the weight is usually assigned). The weight of gate utilization efficiency considers that gate allocation can improve the overall capacity and throughput of the airport and reduce aircraft taxiing time. It is typically determined by simulating the utilization efficiency (e.g., "actual usage time / planned available time") under different allocation schemes. During peak flight periods, higher weights are assigned to maximize gate turnover. Flight priority weights differentiate the commercial value and urgency of different flights (e.g., in cargo, "same-day delivery" has higher priority than "next-day delivery"; in passenger transport, international trunk lines, routes using large aircraft, and flights operated by high-value airlines are given higher priority). Priorities are assigned based on route value and passenger spending power, and are set accordingly. Weather penalty weights are based on historical meteorological data and flight delay data, quantified by the probability and average duration of delays caused by weather conditions (the higher the weather risk, the higher the weight is usually assigned). Weather penalties are numerical values ​​that quantify the negative impact of weather conditions on normal flight operations, based on real-time wind speed, temperature, precipitation, and visibility. Weather penalties and their weights together constitute the penalty item.

[0074] The fitness function is expressed as follows:

[0075] ;

[0076] in, The output is a fitness value that quantifies the merits of the current machine station allocation scheme; To adjust the weighting of flight delay times; To adjust the weighting of camera station utilization efficiency; To adjust the weighting of flight priority; To control the weighting of weather penalties; For flights Actual delay time; For camera position The efficiency of its use; For flights Priority; For flights The length of time the machine space is occupied; Weather penalties incurred by a flight in adverse weather conditions (typical values ​​are between 0 and 1).

[0077] The initial camera position allocation scheme is encoded as the initial population of particle swarm optimization. The initial camera position allocation scheme is then iteratively optimized using the particle swarm optimization algorithm to generate the optimized camera position allocation scheme.

[0078] Furthermore, the initial gate allocation scheme is encoded as multiple particles in a particle swarm. Each particle represents a gate allocation scheme, including flight takeoff and landing times, priorities, gate utilization efficiency, and flight delay times. The position of each particle in the search space represents the specific gate allocation scheme, and the velocity represents the particle's movement direction in the search space. A fitness function is defined with the optimization objective of minimizing total taxiing time and maximizing docking ratio. The fitness value of each particle is calculated based on the fitness function, and the position and velocity of the particles are updated through a particle swarm optimization algorithm. The optimal solution is searched through multiple iterations. In each iteration, the particles adjust their velocities based on their historical best position and global best position, gradually approaching the optimal gate allocation scheme. As the iteration progresses, the fitness value of the particles continuously increases. After reaching a preset number of iterations, an optimized gate allocation scheme is generated.

[0079] It should be noted that the search space refers to the set of all possible machine position allocation schemes in the particle swarm optimization algorithm; the number of iterations is usually set according to the complexity of the problem and the convergence speed, and the exemplary value range is usually between 50 and 200.

[0080] S4. Collect real-time monitoring data streams, combine them with the optimized gate allocation scheme, predict the risk of flight delays through a random forest model, and combine them with early warning thresholds to identify flights that exceed the early warning thresholds and generate flight resource adjustment suggestions.

[0081] Collect real-time monitoring data streams and integrate them with the optimized station allocation scheme to form a real-time operational status dataset;

[0082] Furthermore, real-time monitoring data streams are collected from airport, air traffic control, and meteorological monitoring equipment, including real-time flight status, real-time weather data, and real-time air traffic flow. The real-time monitoring data is then matched with the optimized gate allocation scheme based on flight number and takeoff and landing time to generate a real-time operational status dataset.

[0083] Flight operation indicators are extracted from the real-time operational status dataset and weighted into a predictive feature vector;

[0084] Furthermore, flight operation indicators are selected from the real-time operational status dataset, including actual take-off and landing times, delay times, air traffic flow, and weather conditions. Different weights are assigned based on the degree of influence of each flight operation indicator on flight delay prediction, and the flight operation indicators are weighted into a prediction feature vector.

[0085] Based on a historical comprehensive flight information dataset, multiple decision trees are constructed, and the predictions of all decision trees are aggregated to generate a trained random forest model.

[0086] Furthermore, historical flight operation indicators are extracted from the historical comprehensive flight information dataset, and multiple decision trees are constructed. The training data is continuously iterated based on the mean squared error loss function, and the predicted values ​​of the leaf nodes in the decision trees are adjusted. When the number of leaf node samples is less than the training stopping threshold, training is stopped and multiple trained decision trees are output. The trained decision trees are aggregated, and the predicted values ​​of each decision tree are summarized through a voting mechanism to generate a trained random forest model.

[0087] It should be noted that the training stopping threshold is usually set based on the amount of training data and the complexity of the features. For example, for tasks with a large amount of data, the minimum number of samples is set to be small (such as 5 to 10), and the exemplary range is usually 5 to 20. The voting mechanism is an ensemble method in random forests, which selects the most frequently occurring class label as the prediction value by statistically analyzing the classification labels of each decision tree.

[0088] Input the predicted feature vectors into the trained random forest model to output the probability of flight delays;

[0089] Furthermore, the predicted feature vectors are input into the trained random forest model. The predicted value of each decision tree is calculated by weighting each feature. Then, through a voting mechanism among multiple decision trees, the predicted values ​​of each decision tree are combined to output the probability of flight delay.

[0090] Based on the warning threshold, identify flights that exceed the warning threshold and generate a list of high-risk flights;

[0091] Furthermore, by comparing the delay probability of each flight with the warning threshold, flights that exceed the warning threshold are identified and marked as high-risk flights, and all high-risk flights are aggregated into a high-risk flight list.

[0092] It should be noted that the warning threshold is set based on a historical comprehensive flight information dataset, combined with the relationship between the probability of flight delays and actual flight operations. An exemplary value range is between 0.7 and 0.9. For example, if flight delays have a significant impact on other flights and airport resources, a lower warning threshold (e.g., 0.8) can be set.

[0093] Based on a list of high-risk flights and combined with real-time monitoring data streams, the main causes of flight delays are identified, and flight resource adjustment suggestions are generated.

[0094] Furthermore, based on the high-risk flight list, the collected real-time monitoring data stream is compared with historical flight delay data to filter out high-risk flights with the same weather data and air traffic flow as those in the historical flight delay data, and to identify the potential impact on flight operations. For example, when a flight is delayed under specific weather conditions, the impact of weather on delays is identified by filtering other delays under similar weather conditions. Based on flight priority, potential impact on flight operations, and resource usage, flight resource adjustment suggestions are generated. For example, for flights with severe delays, it is recommended to increase gate occupancy time and prioritize gates closer to the boarding gate; for flights with high air traffic flow, it is recommended to adjust the flight path or take-off and landing sequence.

[0095] S5. Based on the flight information sharing protocol, real-time dynamic data and flight resource adjustment suggestions are shared in real time through standardized interfaces, and real-time feedback is recorded to dynamically optimize the random forest model.

[0096] Based on the data format standards defined in the Flight Information Sharing Protocol, real-time monitoring data streams and flight resource adjustment suggestions are encapsulated into standardized data packets;

[0097] Furthermore, based on the data format standards predetermined in the flight information sharing agreement, the real-time monitoring data stream and flight resource adjustment suggestions are matched and encoded, and encapsulated into standardized data packets.

[0098] It should be noted that the data format standard is based on the flight information sharing protocol and typically includes the name, data type, unit, and format requirements of the data fields.

[0099] Standardized data packets are shared to the dispatch center in real time, flight resource adjustment suggestions are implemented, real-time feedback data is collected, and the parameters of the random forest model are dynamically optimized through incremental learning algorithms.

[0100] Furthermore, standardized data packets are shared with the dispatch center in real time, and flight resource adjustment suggestions are implemented. Simultaneously, real-time feedback data (such as the latest flight delay information and air traffic conditions) is collected. Through incremental learning algorithms, the real-time feedback data is merged with the predicted feature vectors to form an updated dataset. For each decision tree, the updated dataset is used for local updates, including adjusting the splitting conditions of each tree and optimizing the predicted values ​​of leaf nodes, and re-evaluating the splitting criteria, such as resetting the training stopping threshold. Each splitting node is fine-tuned through a random forest model, enabling each splitting node to incorporate new data trends and generate a dynamically optimized random forest model.

[0101] It should be noted that incremental learning is a machine learning method that learns and updates by gradually introducing new data without retraining the model, ensuring that the model can continuously learn and adapt as the data changes.

[0102] This embodiment also provides a computer device applicable to the dynamic flight information management method based on machine learning, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the dynamic flight information management method based on machine learning as proposed in the above embodiment.

[0103] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0104] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the machine learning-based dynamic flight information management method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0105] In summary, this invention achieves multi-dimensional feature mining of flight operation status by constructing a flight spatiotemporal relationship graph and extracting spatiotemporal features and time dynamic features, thereby improving the accuracy and comprehensiveness of flight status feature representation; and by dynamically optimizing the parameters of the random forest model, the flight delay risk prediction can dynamically adapt to changes in the real-time operating environment, thereby enhancing the timeliness of flight management decisions.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic flight information management method based on machine learning, characterized in that: include, Collect and preprocess flight information from multiple sources to generate a comprehensive flight information dataset; Based on a comprehensive flight information dataset, a flight spatiotemporal relationship graph is constructed, and spatiotemporal features and temporal dynamic features are extracted to generate a feature dataset. The specific steps are as follows. The feature information of flights is extracted from the comprehensive flight information dataset as flight nodes, the relationships between flight nodes are used as relationship edges, and a spatiotemporal relationship graph of flights is constructed. Based on the historical flight spatiotemporal relationship graph, the loss function is minimized through backpropagation algorithm, and the convolution kernel weights of the graph convolutional network are optimized to generate the trained graph convolutional network. The flight spatiotemporal relationship graph is input into the trained graph convolutional network, and long-term dependencies are captured through multiple convolutional layers. The embedding vector of each flight node is output as a time dynamic feature and spatiotemporal feature. The spatiotemporal features and time dynamic features of all flight nodes are fused into a feature dataset; Based on the feature dataset, a gate allocation scheme is formulated through allocation rules. Combined with flight scheduling factors, the optimization objective is to minimize total taxiing time and maximize gate utilization. The gate allocation scheme is then optimized using a particle swarm optimization algorithm. The specific steps are as follows. The fitness function is defined based on the initial gate allocation scheme and in combination with flight scheduling factors; The initial camera position allocation scheme is encoded as the initial population of particle swarm optimization. The initial camera position allocation scheme is then iteratively optimized using the particle swarm optimization algorithm to generate the optimized camera position allocation scheme. Real-time monitoring data streams are collected, and combined with the optimized gate allocation scheme, a random forest model is used to predict the risk of flight delays. Based on warning thresholds, flights exceeding the warning thresholds are identified, and flight resource adjustment suggestions are generated. The specific steps are as follows: Collect real-time monitoring data streams and integrate them with the optimized station allocation scheme to form a real-time operational status dataset; Flight operation indicators are extracted from the real-time operational status dataset and weighted into a predictive feature vector; Based on a historical comprehensive flight information dataset, multiple decision trees are constructed, and the predictions of all decision trees are aggregated to generate a trained random forest model. Input the predicted feature vectors into the trained random forest model to output the probability of flight delays; According to the flight information sharing protocol, real-time dynamic data and flight resource adjustment suggestions are shared in real time through a standardized interface, and real-time feedback is recorded to dynamically optimize the random forest model.

2. The dynamic flight information management method based on machine learning as described in claim 1, characterized in that: The multi-source flight information includes flight plan data, real-time status data, flight departure information, air traffic status data, and environmental information; The preprocessing includes data cleaning, missing value imputation, and format conversion.

3. The dynamic flight information management method based on machine learning as described in claim 1, characterized in that: The steps for formulating a gate allocation scheme based on feature datasets and allocation rules are as follows: The feature information of flight nodes and the weight matrix of relation edges are extracted from the feature dataset. Based on the predefined allocation rules, the initial gate allocation scheme for flights is formulated. Based on the time dynamic features in the feature dataset, the initial gate allocation scheme for flights is dynamically adjusted to generate a complete gate allocation scheme.

4. The dynamic flight information management method based on machine learning as described in claim 1, characterized in that: The process of combining early warning thresholds to identify flights exceeding the thresholds and generating flight resource adjustment suggestions involves the following steps: Based on the warning threshold, identify flights that exceed the warning threshold and generate a list of high-risk flights; Based on a list of high-risk flights and combined with real-time monitoring data streams, the main causes of flight delays are identified, and suggestions for adjusting flight resources are generated.

5. The dynamic flight information management method based on machine learning as described in claim 4, characterized in that: According to the flight information sharing protocol, real-time dynamic data and flight resource adjustment suggestions are shared in real time through a standardized interface, and real-time feedback is recorded to dynamically optimize the random forest model. The steps are as follows: Based on the data format standards defined in the Flight Information Sharing Protocol, real-time monitoring data streams and flight resource adjustment suggestions are encapsulated into standardized data packets; Standardized data packets are shared to the dispatch center in real time, flight resource adjustment suggestions are implemented, real-time feedback data is collected, and the parameters of the random forest model are dynamically optimized through incremental learning algorithms.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the machine learning-based dynamic flight information management method according to any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the machine learning-based dynamic flight information management method according to any one of claims 1 to 5.

Images