Flight landing time prediction method based on machine learning

Abstract

The invention discloses a flight landing time prediction method based on machine learning, and relates to the technical field of flight pre-judgment, and the method comprises the steps: carrying out the preprocessing of data based on historical flight data and historical meteorological data, and outputting feature sample data; performing data cleaning and standardization processing on the feature sample data, and outputting a training data set and a test data set in combination with actual landing and planned landing moment information of historical flights; establishing a landing prediction model, and performing optimization by combining reinforcement learning and a generative adversarial network; using the average absolute percentage error to test the landing prediction model and output a trained landing prediction model; and outputting a dynamically updated landing prediction model by adopting a concept drift detection algorithm, online gradient lifting and adaptive weighted fusion. Through combination of the deep Q network and multi-agent reinforcement learning, the prediction result is optimized in real time, and dynamic adjustment can be performed according to changes of actual flights.

Description

Technical Field

[0001] This invention relates to the field of flight prediction technology, and in particular to a method for predicting flight landing time based on machine learning. Background Technology

[0002] With the rapid development of the air transport industry, accurately predicting flight arrival times has gradually become crucial for improving aviation efficiency, optimizing scheduling, and enhancing passenger experience. Traditional forecasting methods are mostly based on statistical models. While these methods are simple and easy to use, they are poorly adaptable to changes in data. Especially when faced with uncertainties such as sudden weather changes and air traffic control, the forecast results can be highly inaccurate, leading to a series of adverse effects.

[0003] In recent years, the rise of machine learning technology has brought new evolution to this field. In particular, the application of advanced algorithms such as deep learning, reinforcement learning, and generative adversarial networks has gradually improved prediction accuracy and system adaptability. However, current flight arrival time prediction technology still has some significant shortcomings. Many models still rely solely on a single data source, failing to fully integrate multi-source data such as meteorological information and historical flight records. Furthermore, in dynamic environments, the real-time update capability of existing models is relatively weak, hindering accurate predictions. Additionally, many models' optimization processes are based on static training data, unable to flexibly adapt to changes in data distribution over time. This causes the models to become unsuitable for real-world conditions, thus affecting prediction accuracy. Summary of the Invention

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

[0005] Therefore, this invention provides a flight landing time prediction method based on machine learning to solve the problems of poor adaptability to dynamically changing data and low real-time prediction accuracy of existing flight landing 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 flight landing time prediction method based on machine learning, which includes preprocessing historical flight data and historical meteorological data to output feature sample data.

[0008] The feature sample data is cleaned and standardized, and training and test datasets are output by combining the actual and planned landing times of historical flights.

[0009] Establish a landing prediction model and optimize it by combining reinforcement learning and generative adversarial networks;

[0010] The landing prediction model is tested using the mean absolute percentage error, and the trained landing prediction model is output.

[0011] Combining deep Q-networks with multi-agent reinforcement learning outputs real-time optimized predictions;

[0012] A landing prediction model is developed by employing a concept drift detection algorithm, online gradient boosting, and adaptive weighted fusion to output dynamically updated data.

[0013] As a preferred embodiment of the machine learning-based flight landing time prediction method of the present invention, the step of preprocessing historical flight data and historical meteorological data to output feature sample data includes the following specific steps.

[0014] Convert historical flight data and historical meteorological data into a unified time format and output data with unified dimensions;

[0015] Perform missing value imputation, outlier detection, and robust standardization on uniform dimension data to output standardized data;

[0016] The feature selection algorithm is used to accept standardized data and output feature sample data.

[0017] As a preferred embodiment of the machine learning-based flight landing time prediction method of the present invention, the steps of cleaning and standardizing the feature sample data, and combining historical actual and planned flight landing time information to output training and test datasets are as follows:

[0018] Use the median to fill numerical features and use the median to fill categorical features to output complete data.

[0019] Use outlier detection to inspect the complete data, remove outliers from the complete data to output clean data, scale the clean data to the same scale, and process it using robust standardization.

[0020] By utilizing dataset partitioning and time series characteristics, and combining historical flight actual and planned landing time information, training and test datasets are output.

[0021] As a preferred embodiment of the machine learning-based flight landing time prediction method of this invention, the specific steps for establishing the landing prediction model and optimizing it using reinforcement learning and generative adversarial networks are as follows.

[0022] A landing prediction model is established using a long short-term memory network and a temporal convolutional network.

[0023] The prediction results are optimized by adjusting the flight landing time prediction model through reinforcement learning;

[0024] By using generative adversarial networks, the generator produces predicted landing time samples, the discriminator evaluates the prediction accuracy, and the landing time prediction model is optimized through adversarial training.

[0025] As a preferred embodiment of the machine learning-based flight landing time prediction method of the present invention, the specific steps for testing the landing prediction model using the mean absolute percentage error and outputting the trained landing prediction model are as follows:

[0026] The mean absolute percentage error is used to calculate the percentage error between the predicted value and the actual value, and the error value is output.

[0027] Prediction accuracy based on mean absolute percentage error;

[0028] The trained landing prediction model is output by combining the prediction error value and the prediction accuracy.

[0029] As a preferred embodiment of the machine learning-based flight landing time prediction method of the present invention, the specific steps of combining deep Q-networks and multi-agent reinforcement learning to optimize prediction in real time are as follows:

[0030] We employ adaptive feature selection and data weighting, combined with missing value imputation, outlier detection, and robust standardization to output feature data.

[0031] The extreme gradient boosting algorithm and Bayesian optimization are used to predict the output accuracy.

[0032] Real-time flight data is used as the training input to a deep Q-network, and the output is combined with multi-agent reinforcement learning to optimize predictions in real time.

[0033] As a preferred embodiment of the machine learning-based flight landing time prediction method of the present invention, the specific steps of employing a concept drift detection algorithm, online gradient boosting, and adaptive weighted fusion to output a dynamically updated landing prediction model are as follows:

[0034] A concept drift detection algorithm is used to identify changes in data distribution and output an updated landing prediction model.

[0035] The landing prediction model is updated online using an online gradient boosting method;

[0036] An adaptive weighted fusion strategy is used to output a dynamically updated landing prediction model.

[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 flight landing time prediction 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 flight landing time prediction method described in the first aspect of the present invention.

[0039] The beneficial effects of this invention are as follows: This invention employs a preprocessing method based on historical flight and meteorological data, combined with deep learning, reinforcement learning, and generative adversarial networks to optimize the landing prediction model, significantly improving the accuracy of flight landing time predictions. By combining deep Q-networks with multi-agent reinforcement learning, the prediction results are optimized in real time, allowing for dynamic adjustments based on changes in actual flight schedules. The use of a concept drift detection algorithm and an adaptive weighted fusion strategy enables timely updates and maintains high-precision prediction performance even when data distribution changes. The beneficial effects of this method not only improve prediction accuracy but also enhance adaptability to different environments and data changes, greatly improving the scheduling efficiency and safety of the air transport system. 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 the flight landing time prediction method based on machine learning in Example 1.

[0042] Figure 2 This is a flowchart of the data preprocessing and feature engineering process in Example 1.

[0043] Figure 3 This is a flowchart of the establishment and optimization of the landing prediction model in Example 1. Detailed Implementation

[0044] 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.

[0045] 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.

[0046] 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.

[0047] Example 1, referring to Figures 1-3 This is the first embodiment of the present invention, which provides a flight landing time prediction method based on machine learning, including the following steps:

[0048] S1. Based on historical flight data and historical meteorological data, preprocess the data and output feature sample data.

[0049] Furthermore, historical flight data and historical meteorological data are converted into a unified time format to output data with unified dimensions.

[0050] It should be noted that historical flight data refers to information such as take-off and landing times, routes, and airlines of all flights within a past period. This data provides information on the regular behavior patterns of flights. Historical meteorological data refers to meteorological data related to flight operations, such as wind speed, temperature, air pressure, and precipitation. This data has a significant impact on flight paths and landing times.

[0051] Specifically, historical flight data and historical meteorological data from different sources are converted into a unified time format using Coordinated Universal Time (UTC) to output data with unified dimensions. This allows historical flight data and meteorological data to be compared and processed on the same scale, ensuring consistency between data sources and facilitating subsequent processing and modeling.

[0052] Perform missing value imputation, outlier detection, and robust standardization on uniform dimension data to output standardized data;

[0053] Specifically, mean imputation is used to fill in missing values ​​and fill in the blanks in the data. Statistical methods are used to detect and correct extreme values ​​in the data to make the data complete and accurate, avoid the impact of missing values ​​and outliers on the performance of the training model, and improve stability. The median and interquartile range are used for standardization to output standardized data, making the distribution of all data consistent, providing a better data foundation for subsequent training, and reducing training time and complexity.

[0054] The feature selection algorithm is used to accept standardized data and output feature sample data.

[0055] Specifically, filtering methods are used to select the most valuable features for the prediction task, remove redundant or noisy data, and output feature sample data. Through feature selection, complexity is reduced and training speed and accuracy are improved.

[0056] It should be noted that the feature sample data includes numerical features and categorical features. Numerical features are quantifiable and continuous attributes, which are mainly processed through statistical analysis, such as actual flight time, planned flight time, weather temperature, wind speed, air pressure, precipitation, and the operational efficiency of takeoff and landing airports. Categorical features are data that cannot be arithmetically calculated and represent qualitative or categorical information, including flight type, airline, airport, season, day of the week, and flight departure and arrival times.

[0057] S2. Clean and standardize the feature sample data, and output training and test datasets by combining the actual and planned landing times of historical flights.

[0058] Furthermore, the median is used to fill numerical features and to fill categorical features to output complete data;

[0059] Specifically, using the median to impute numerical features is more robust, while using the median to impute categorical features can effectively fill in missing values ​​in the data, avoiding bias in outputting complete data due to missing data.

[0060] Use outlier detection to examine the complete data, remove outliers, and output clean data.

[0061] Specifically, outlier detection methods are used to clean the data, removing outliers from the complete data to ensure the validity of the data. After removing outliers, the true distribution of the data can be better understood, thereby improving prediction accuracy and reducing errors.

[0062] Scaling clean data to the same scale and processing it using robust standardization;

[0063] Specifically, using robust standardization to scale cleaned data to the same scale can reduce the impact of skewed data distribution or outliers, and better understand the intrinsic relationships in the data, thereby improving the stability and convergence speed of the prediction model.

[0064] By utilizing dataset partitioning and time series characteristics, and combining historical flight actual and planned landing time information, training and test datasets are output.

[0065] Specifically, the data is divided into time sequences, so that the training dataset contains historical actual flight landing data and the test dataset contains planned landing time information. This allows the model to learn the trends and patterns of the time series, so that the prediction not only has a reliance on historical data, but also takes into account the changes that future flights may face, thereby improving the timeliness and accuracy of the prediction.

[0066] In particular, this method is applicable to various airlines, airport management entities, and aviation data analysis platforms. It has significant application value, especially in areas such as air traffic control, flight safety, and weather warnings. By accurately predicting flight landing times, it can effectively optimize airport resource allocation, reduce flight delays, improve flight punctuality, and bring economic and social benefits to the aviation industry.

[0067] S3. Establish a landing prediction model and optimize it by combining reinforcement learning and generative adversarial networks.

[0068] Furthermore, a landing prediction model is established using long short-term memory networks and temporal convolutional networks;

[0069] Specifically, by combining long short-term memory networks and temporal convolutional networks, a landing prediction model is constructed to predict flight landing times. This landing prediction model can accurately extract multi-level features of flights based on large-scale data, thereby improving prediction accuracy.

[0070] It should be noted that a landing prediction model is established using a long short-term memory network and a temporal convolutional network;

[0071] Furthermore, we utilize temporal convolutional networks combined with max pooling to extract local temporal features;

[0072] Local temporal features are extracted using a combination of temporal convolutional networks and max pooling.

[0073] Specifically, by combining historical flight data, causal convolution is used to ensure that the prediction at the current moment depends only on past information to prevent the leakage of future information. Extended convolution is used to improve the model's ability to perceive long-term features by introducing intervals into the convolution kernel. Deep temporal information is obtained by stacking multiple convolutional layers while avoiding the gradient vanishing problem. Max pooling is combined to filter key short-term information, reduce feature dimensions, improve model computation efficiency, and enhance the ability to retain key features and output local temporal features.

[0074] Long Short-Term Memory (LSTM) networks are used to capture long-term dependencies and extract deep temporal patterns.

[0075] Specifically, using long short-term networks enables the model to learn over a longer time span, taking into account factors such as fuel consumption, cruising speed changes, and airspace intersections during long-haul flights, thereby improving prediction accuracy and outputting deep time-series patterns by learning long-term dependencies.

[0076] A prediction model is generated by combining temporal convolutional networks and long short-term memory networks and using fully connected layers;

[0077] Specifically, by combining local temporal features and deep temporal patterns, an adaptive weight allocation mechanism is adopted to allow the model to automatically adjust the importance of local temporal features and deep temporal patterns, and a fully connected layer is used to output the prediction model.

[0078] It should be noted that the input layer of the landing prediction model is preprocessed flight time-series data; the temporal convolutional network layer contains local temporal features, which are combined with max pooling to retain key features and reduce data dimensionality; the long short-term memory network layer receives local features extracted by the temporal convolutional network and captures long-term dependencies; the fully connected layer concatenates the output features of the temporal convolutional network and the long short-term memory network and inputs them into the fully connected layer for feature integration.

[0079] In a better way, after the temporal convolutional network and the long short-term memory network each play their roles, the final fully connected layer can fuse the features of the two to form a complete flight landing time prediction model. It can take into account both local and global features, and can respond quickly to short-term anomalies while maintaining long-term stability. By adjusting the weights of the fully connected layer, the model can adapt to the personalized features of different airports and different aircraft types, thereby improving its generalization ability.

[0080] It should be noted that the model loss function expression is as follows:

[0081]

[0082] Where L is the loss function, representing the error between the model's predicted value and the actual value, L≥0, and the smaller the value, the higher the prediction accuracy; N represents the sample size, and N is a positive integer. i Let i be the actual landing time of the i-th flight. This represents the predicted arrival time of the i-th flight. Let be the prediction error for the i-th flight. It can be positive or negative, representing the deviation of the predicted value from the actual value. Let be the square of the prediction error for the i-th flight.

[0083] Specifically, the objective function of this model is to minimize the error between the predicted landing time and the actual landing time. The mean squared error is used as the loss function, which is used to optimize the model training parameters.

[0084] It should be noted that the time constraint means the predicted landing time should fall within a reasonable range of the flight schedule, i.e. Physical constraints require that the prediction results should conform to the basic physical laws of flight operations, such as the landing time not being earlier than the departure time. Data distribution constraints require that the model prediction results should be consistent with the distribution characteristics of historical data to reduce the occurrence of outliers.

[0085] It should be noted that parameter optimization uses the Adam optimizer to update model parameters, and the learning rate is set to adaptive adjustment to make the training process stable; model validation can use cross-validation, adopting the K-fold cross-validation method to make the model stable and reliable on different datasets;

[0086] The prediction results are optimized by adjusting the flight landing time prediction model through reinforcement learning;

[0087] Specifically, the landing prediction model continuously adjusts flight landing times by interacting with the environment through reinforcement learning. The landing prediction model can adaptively adjust the prediction strategy, further reduce prediction errors, optimize the long-term stability of predictions, and improve the prediction accuracy of the landing prediction model.

[0088] By using a generative adversarial network, the generator generates predicted landing time samples, the discriminator evaluates the prediction accuracy, and the landing time prediction model is optimized through adversarial training.

[0089] Specifically, using a generative adversarial network, the generator is responsible for generating predicted landing time samples, and the discriminator evaluates the generator's output to determine whether the generated prediction is close to the actual flight landing time. The generator and discriminator are combined and optimized through adversarial training. The generator continuously generates predictions that are closer to the actual landing time, while the discriminator continuously improves its ability to distinguish between true and false predictions. The two improve their respective abilities through continuous adversarial training, ultimately achieving the effect of generating accurate predictions.

[0090] Ideally, reinforcement learning and generative adversarial networks should be combined. Reinforcement learning adjusts the model's prediction strategy based on real-time data to ensure prediction accuracy, while generative adversarial networks continue to optimize the prediction results to ensure that each generated prediction is as close as possible to the real result.

[0091] S4. Using the mean absolute percentage error, test the landing prediction model and output the trained landing prediction model.

[0092] Furthermore, the mean absolute percentage error is used to calculate the percentage error between the predicted and actual values ​​and output the error value.

[0093] Specifically, using the actual landing time of the flight as the actual value and the model prediction time as the predicted value, the average absolute percentage error between the predicted value and the actual value is calculated, and the prediction error value is output. The average absolute percentage error can quantify the deviation of the landing prediction model's prediction results and can be used to fine-tune the model parameters.

[0094] It should be noted that the formula for the mean absolute percentage error is as follows:

[0095]

[0096] It should be noted that the smaller the MAPE value, the more accurate the result.

[0097] Prediction accuracy is based on mean absolute percentage error (MAPE); it is most accurate when MAPE equals 0.

[0098] Specifically, prediction accuracy measures the overall accuracy of a prediction model and directly reflects the reliability of the landing time prediction. The higher the value, the more accurate the prediction, which can be used to optimize flight scheduling and airport management.

[0099] It should be noted that the accuracy formula is as follows:

[0100] Prediction accuracy = 100% - MAPE;

[0101] The trained landing prediction model is output by combining the prediction error value and the prediction accuracy.

[0102] Specifically, during the retraining process, the error is reduced by combining the prediction error value and the prediction accuracy through Bayesian optimization, and the trained prediction model is output. After error evaluation and optimization adjustment, the mean absolute percentage error is minimized and the prediction accuracy is maximized.

[0103] It should be noted that the weighted absolute percentage error is used as the core indicator for model performance evaluation. The loss function expression for the performance testing of the prediction model is as follows:

[0104]

[0105] Among them, L eval This represents the evaluation loss value, a quantitative indicator of the model's prediction quality. y represents the actual landing time vector, in minutes. denoted as the predicted landing time vector in minutes, ω is the dynamic weight vector, and ∑ represents the summation operation over all samples.

[0106] S5 combines deep Q-networks with multi-agent reinforcement learning to output real-time optimized predictions.

[0107] Furthermore, adaptive feature selection and data weighting are used, combined with missing value imputation, outlier detection, and robust standardization to output feature data;

[0108] Specifically, adaptive feature selection and data weighting are used to select and weight data features to improve the quality of data input to the model. Adaptive feature selection dynamically selects features that have an important impact on the prediction results based on the correlation between the data and the target variable. Data weighting assigns weights to different features to highlight features that are more useful for prediction. Interpolation is used to fill in missing values, standard deviation scores are used to detect and handle outliers, and the median is used for robust standardization.

[0109] The extreme gradient boosting algorithm and Bayesian optimization are used to predict the output accuracy.

[0110] Using the extreme gradient boosting algorithm to train the landing prediction model can efficiently process structured data. Incorporating Bayesian optimization can automatically adjust the model parameters during training, further improving the model's generalization ability. This optimization enables the landing prediction model to process large amounts of data and output efficient and accurate results in complex prediction environments.

[0111] The deep Q-network is used to train real-time flight data as input, and the output is combined with multi-agent reinforcement learning to optimize predictions in real time.

[0112] Specifically, by using a deep Q-network to input real-time flight data, the system can train a model based on the real-time flight data, enabling it to self-adjust and update its decision-making strategy based on real-time data. By employing multi-agent reinforcement learning and combining real-time data from multiple flights, the landing time predictions of each flight can influence each other, ensuring collaborative operation between different flights and improving the predictive ability and response speed of the entire system.

[0113] In particular, this method can accurately predict flight landing times. By combining adaptive feature selection, data weighting, reinforcement learning algorithms and other technologies, it can provide stable and accurate predictions in a changing aviation environment. This not only improves flight scheduling efficiency but also effectively reduces flight delays and enhances the operational efficiency of the aviation system.

[0114] S6. Employs a concept drift detection algorithm, online gradient boosting, and adaptive weighted fusion to output a dynamically updated landing prediction model.

[0115] Furthermore, a concept drift detection algorithm is used to identify changes in data distribution and output an updated landing prediction model.

[0116] Specifically, by comparing and analyzing historical and real-time data using a concept drift detection algorithm, it can detect whether the statistical distribution of the data has changed. It can dynamically monitor changes in input data, identify changes in data distribution in real time, and automatically output the need for updates. It can cope with constantly changing realities and ensure that the prediction results are always accurate.

[0117] The main advantage of the concept drift detection algorithm is that it can cope with changes in data distribution, especially in flight landing prediction. Changes in external factors such as weather and flight delays can cause changes in data distribution. Through concept drift detection, the model can monitor changes in data distribution in real time and automatically output the need to adjust the prediction model, thereby maintaining high accuracy and enhancing the system's adaptability to dynamic environmental changes. This effectively avoids the problem that traditional static models cannot cope with new situations.

[0118] It should be noted that the concept drift detection algorithm uses a cumulative sum control chart, which can monitor the cumulative error of the data to detect changes in the data distribution. If the statistical characteristics of the data change significantly, the cumulative sum control chart will detect it and issue a signal. When applied to flight landing time detection, it can detect changes in the distribution of landing time caused by factors such as weather changes and flight delays.

[0119] It should be noted that the function expression for the concept drift detection algorithm is as follows:

[0120]

[0121] Where D represents the drift detection statistic, D≥0, x i For the current data point, s represents the historical window mean. h The standard deviation is the historical window value.

[0122] The landing prediction model is updated online using an online gradient boosting method;

[0123] Specifically, the online gradient boosting method uses incremental learning to adjust model parameters in real time, continuously acquiring new data and updating them. This avoids model obsolescence and continuously improves the model's predictive ability. Especially in flight landing time prediction, online gradient boosting can adjust the model in a timely manner based on the latest flight and meteorological data to ensure the real-time performance and accuracy of the prediction results.

[0124] The superior online gradient boosting method can quickly learn online after receiving new data and gradually update the prediction model, which improves the real-time performance of data processing. It can also avoid the dependence of traditional batch processing methods on large amounts of data, and can respond in real time to external data changes in the flight system and continuously update the prediction model, ensuring the timeliness and accuracy of the model prediction. It is particularly suitable for complex scenarios such as flight delays and sudden weather changes.

[0125] It should be noted that the online gradient boosting expression is as follows:

[0126]

[0127] Where θ represents the current model parameters, η is the learning rate, and 0 < η ≤ 1. y represents the gradient of the loss function with respect to the predicted value f(x; θ), where y is the true value and x represents the input feature.

[0128] An adaptive weighted fusion strategy is used to output a dynamically updated landing prediction model;

[0129] Specifically, the system employs an adaptive weighted fusion strategy combined with a weighted average method, using multiple sub-models. Each model may perform differently under different input data and scenarios. Through the adaptive weighted fusion strategy, the system can dynamically adjust the weight of each sub-model according to the characteristics of the current data. The model will evaluate the accuracy of each sub-model in actual prediction and select the optimal sub-model for output.

[0130] In a better way, through an adaptive weighted fusion strategy, the system can integrate the advantages of each sub-model and make dynamic adjustments according to the characteristics of the current data, ensuring the high accuracy and robustness of the prediction results. This enables the prediction model to make accurate judgments in different situations and avoids errors caused by a single model being insufficient to cope with all situations.

[0131] It should be noted that the expression for the adaptive weighted fusion function is as follows:

[0132]

[0133] in, This represents the final fused prediction value, where α is the dynamic weight of the new model (0 ≤ α ≤ 1). This represents the predicted value of the new model. These are the predicted values ​​from the old model.

[0134] The optimal approach combines concept drift detection, online gradient boosting, and adaptive weighted fusion to provide a dynamic, real-time, and efficient solution for flight landing time prediction. Concept drift detection ensures that the model can promptly identify data changes, online gradient boosting guarantees the model's adaptability in the data stream, and adaptive weighted fusion ensures the model's accuracy under different data scenarios. This not only improves prediction accuracy but also ensures that the system can respond to complex and dynamic changes in the aviation field in real time.

[0135] This embodiment also provides a computer device applicable to the flight landing time prediction 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 flight landing time prediction method based on machine learning as proposed in the above embodiment.

[0136] 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.

[0137] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the flight landing time prediction method based on machine learning as 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.

[0138] In summary, this invention significantly improves the accuracy of flight landing time predictions by employing a preprocessing method based on historical flight and meteorological data, combined with deep learning, reinforcement learning, and generative adversarial networks to optimize the landing prediction model. By combining deep Q-networks with multi-agent reinforcement learning, the prediction results are optimized in real time and dynamically adjusted according to changes in actual flight schedules. The use of a concept drift detection algorithm and an adaptive weighted fusion strategy enables timely updates to maintain high-precision prediction performance even when data distribution changes. The beneficial effects of this method not only improve prediction accuracy but also enhance adaptability to different environments and data changes, greatly improving the scheduling efficiency and safety of air transport systems.

[0139] 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 flight arrival time prediction method based on machine learning, characterized in that: include, Based on historical flight data and historical meteorological data, the data is preprocessed to output feature sample data; The feature sample data is cleaned and standardized, and training and test datasets are output by combining the actual and planned landing times of historical flights. Establish a landing prediction model and optimize it by combining reinforcement learning and generative adversarial networks; The landing prediction model is tested using the mean absolute percentage error, and the trained landing prediction model is output. Combining deep Q-networks with multi-agent reinforcement learning outputs real-time optimized predictions; A landing prediction model is developed by employing a concept drift detection algorithm, online gradient boosting, and adaptive weighted fusion to output dynamically updated data.

2. The flight landing time prediction method based on machine learning as described in claim 1, characterized in that: The process involves preprocessing historical flight and meteorological data to output feature sample data. The specific steps are as follows: Convert historical flight data and historical meteorological data into a unified time format and output data with unified dimensions; Perform missing value imputation, outlier detection, and robust standardization on uniform dimension data to output standardized data; The feature selection algorithm is used to accept standardized data and output feature sample data.

3. The flight landing time prediction method based on machine learning as described in claim 2, characterized in that: The process of cleaning and standardizing the feature sample data, and then combining historical flight actual and planned landing time information to output training and test datasets, is detailed below. Use the median to fill numerical features and use the median to fill categorical features to output complete data. Use outlier detection to inspect the complete data, remove outliers from the complete data, and output clean data. Scaling clean data to the same scale and processing it using robust standardization; By utilizing dataset partitioning and time series characteristics, and combining historical flight actual and planned landing time information, training and test datasets are output.

4. The flight landing time prediction method based on machine learning as described in claim 3, characterized in that: The specific steps for establishing the landing prediction model, and optimizing it by combining reinforcement learning and generative adversarial networks, are as follows. A landing prediction model is established using a long short-term memory network and a temporal convolutional network. The prediction results are optimized by adjusting the flight landing time prediction model through reinforcement learning; By using generative adversarial networks, the generator produces predicted landing time samples, the discriminator evaluates the prediction accuracy, and the landing time prediction model is optimized through adversarial training.

5. The flight landing time prediction method based on machine learning as described in claim 4, characterized in that: The step of testing the landing prediction model using the mean absolute percentage error (MASE) and outputting the trained landing prediction model is as follows: The mean absolute percentage error is used to calculate the percentage error between the predicted value and the actual value, and the error value is output. Prediction accuracy based on mean absolute percentage error; The trained landing prediction model is output by combining the prediction error value and the prediction accuracy.

6. The flight landing time prediction method based on machine learning as described in claim 5, characterized in that: The specific steps for combining deep Q-networks with multi-agent reinforcement learning to output real-time optimized predictions are as follows: We employ adaptive feature selection and data weighting, combined with missing value imputation, outlier detection, and robust standardization to output feature data. The extreme gradient boosting algorithm and Bayesian optimization are used to predict the output accuracy. Real-time flight data is used as the training input to a deep Q-network, and the output is combined with multi-agent reinforcement learning to optimize predictions in real time.

7. The flight landing time prediction method based on machine learning as described in claim 6, characterized in that: The landing prediction model, which employs a concept drift detection algorithm, online gradient boosting, and adaptive weighted fusion to output dynamically updated data, follows these steps: A concept drift detection algorithm is used to identify changes in data distribution and output an updated landing prediction model. The landing prediction model is updated online using an online gradient boosting method; An adaptive weighted fusion strategy is used to output a dynamically updated landing prediction model.

8. The flight landing time prediction method based on machine learning as described in claim 7, characterized in that: The landing prediction model is established using a long short-term memory network and a temporal convolutional network. The specific steps are as follows. Local temporal features are extracted using a combination of temporal convolutional networks and max pooling. Long Short-Term Memory (LSTM) networks are used to capture long-term dependencies and extract deep temporal patterns. A prediction model is developed by combining temporal convolutional networks and long short-term memory networks and using fully connected layers.

9. 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 flight landing time prediction method based on machine learning as described in any one of claims 1 to 8.

10. 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 flight landing time prediction method based on machine learning as described in any one of claims 1 to 8.

Images