A method and apparatus for runway occupancy time prediction during an aircraft takeoff phase
By combining scalar LSTM and matrix LSTM with a hybrid model of Transformer, the problem of accuracy and efficiency in predicting runway occupancy time during aircraft takeoff is solved, achieving high-precision DROT prediction and runway scheduling optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CIVIL AVIATION FLIGHT UNIV OF CHINA
- Filing Date
- 2026-02-09
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies lack effective methods for predicting runway occupancy time (DROT) during the takeoff phase of aircraft. Traditional models cannot handle complex temporal relationships involving multiple variables, nonlinearity, and strong coupling, and fail to effectively uncover the deep correlation between static and dynamic features, resulting in low prediction accuracy.
A Transformer model combining a scalar LSTM and a matrix LSTM with a global correlation module is adopted. Features are selected through Pearson correlation analysis, global feature dependencies are captured by multi-head self-attention mechanism, and input data is optimized by feature engineering, forming an xLSTM-Transformer hybrid architecture.
It significantly improves prediction accuracy and generalization ability, with errors within ±5 seconds, enhances model adaptability and computational efficiency, optimizes runway scheduling, and improves single-runway operating efficiency by 12%-15%.
Smart Images
Figure CN121682228B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aviation, and more specifically, to a method and device for predicting runway occupancy time during the takeoff phase of an aircraft. Background Technology
[0002] Existing research primarily focuses on predicting runway occupancy time (ROT) during the landing phase, leaving a significant gap in research on predicting runway occupancy time (DROT) during the takeoff phase. The operational modes of takeoff and landing differ greatly; directly applying landing models to predict takeoff leads to biased predictions, low accuracy, and impracticality. Furthermore, current prediction models are typically traditional machine learning models (such as BP neural networks and random forests) and single-time-series models (such as LSTMs), which have weak long-range dependency capture capabilities and struggle to handle the complex, multivariate, nonlinear, and strongly coupled temporal relationships in DROT prediction. While Transformer models can capture global dependencies, they are insensitive to subtle local temporal changes, are complex, prone to overfitting to small to medium-sized datasets, and consume significant computational resources.
[0003] Furthermore, it failed to effectively explore and integrate the deep correlation between existing static features (such as airport elevation and runway layout) and dynamic features (such as takeoff speed and taxi trajectory) in QAR data. Feature engineering failed to closely integrate the physical meaning of runway occupancy time (DROT) during takeoff, resulting in insufficient utilization of the data.
[0004] Therefore, a network that combines static and dynamic features is needed to make predictions more accurate. Summary of the Invention
[0005] The purpose of this invention is to provide a method and device for predicting runway occupancy time during aircraft takeoff, in order to solve the aforementioned problems existing in the prior art.
[0006] In a first aspect, embodiments of the present invention provide a method for predicting runway occupancy time during the takeoff phase of an aircraft, including:
[0007] Acquire aircraft data from multiple categories; the aircraft data represents flight data during the pre-takeoff phase of the aircraft.
[0008] Pearson correlation analysis was used to filter the aircraft data, resulting in strongly correlated and weakly correlated data.
[0009] Based on strongly correlated data from multiple time points, a high-level feature sequence is obtained through a dual-branch memory structure; the high-level feature sequence represents the extracted local temporal features.
[0010] The dual-branch memory structure includes scalar LSTM and matrix LSTM;
[0011] The global association module mines the global dependency between time points and feature dimensions based on the high-level feature sequence to obtain the predicted occupancy time length; the predicted occupancy time length represents the predicted runway occupancy time during the aircraft takeoff phase.
[0012] Optionally, the high-level feature sequence obtained from the strongly correlated data at multiple time points, through a dual-branch memory structure, includes:
[0013] The strongly correlated data is preprocessed to extract features and obtain preprocessed feature vectors; multiple categories of strongly correlated data correspond to multiple preprocessed feature vectors.
[0014] The strongly correlated data is divided into numerical data and high-dimensional event data; the preprocessed feature vectors corresponding to the numerical data are used as numerical feature vectors; and the preprocessed feature vectors corresponding to the high-dimensional event data are used as high-dimensional event feature vectors.
[0015] Numerical feature vectors from multiple time points are input into a scalar LSTM to identify local details of aircraft flight and obtain the first detail feature.
[0016] High-dimensional event-type feature vectors from multiple time points are input into the matrix LSTM to detect high-dimensional or event-type features and obtain the second detail features.
[0017] The first and second detail features are used to construct a high-level feature sequence.
[0018] Optionally, the step of mining the global dependency between time points and feature dimensions based on the high-level feature sequence through the global association module to obtain the predicted time duration includes:
[0019] The high-level feature sequence is input into the global association module, and a multi-head self-attention mechanism is used to capture global feature dependencies to obtain the global feature vector.
[0020] Based on the global feature vector, the runway occupancy time is predicted, and the predicted occupancy time length is obtained.
[0021] Optionally, the aircraft data is filtered through Pearson correlation analysis to obtain strongly correlated and weakly correlated data, including:
[0022] Pearson correlation analysis was used to obtain strong and weak correlation categories; the strong and weak correlation data were obtained during training.
[0023] Based on the strongly correlated and weakly correlated categories, the aircraft data is filtered to obtain strongly correlated data and weakly correlated data.
[0024] Optionally, the Pearson correlation analysis is used to obtain strong correlation categories and weak correlation categories, including:
[0025] Acquire training aircraft data for multiple categories from multiple tests and the corresponding marked runway occupancy time lengths; the marked runway occupancy time lengths represent the runway occupancy time lengths during the aircraft's takeoff phase in one test.
[0026] Based on the training aircraft data from multiple tests and the corresponding marked runway occupancy time, a time correlation coefficient is obtained; the time correlation coefficient represents the correlation between the training aircraft data and the runway occupancy time during the aircraft takeoff phase.
[0027] Multiple time correlation coefficients were obtained for training aircraft data from multiple categories;
[0028] If the time correlation coefficient is greater than the strong correlation threshold, the category corresponding to the aircraft data is considered a strongly correlated category; if the time correlation coefficient is less than or equal to the strong correlation threshold, the category corresponding to the aircraft data is considered a weakly correlated category.
[0029] Optionally, the time correlation coefficient can be calculated as follows:
[0030]
[0031] Among them, the This represents training aircraft data for one category. Representing variables The mean; the This indicates the length of time the track has been occupied. Representing variables The mean; The time correlation coefficient;
[0032] Where i represents the number of times aircraft data of one category is detected.
[0033] Optionally, dynamic thresholding and attention mechanisms can be introduced into the input gate, forget gate, and output gate of scalar LSTM and matrix LSTM.
[0034] The dynamic threshold and attention mechanism in the forget gate are used to filter noisy data.
[0035] Optional training methods for the bi-branch memory structure and the globally associative module include:
[0036] The model was trained using an early stopping mechanism and a model checkpoint saving strategy, with a learning rate set to 0.001.
[0037] Secondly, embodiments of the present invention provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a method for predicting runway occupancy time during aircraft takeoff.
[0038] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for predicting runway occupancy time during aircraft takeoff.
[0039] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:
[0040] This invention achieves the technical effects of significantly improving prediction accuracy, enhancing generalization and robustness, directly improving operating efficiency, and balancing computational efficiency with accuracy.
[0041] The significant improvement in prediction accuracy is reflected in the fact that the MSE of the hybrid model on the test set decreased to 130.19, a reduction of 27.7% compared to the best single model (Transformer), and the MAE was 6.76 seconds. 94% of the prediction errors fell within ±5 seconds, fully meeting the stringent requirements of civil aviation operations for time prediction accuracy.
[0042] The generalization and robustness enhancements are demonstrated by the fact that the model's error was less than 1.2 seconds in the field tests at six different airports, proving that it has good adaptability and stability for different airport elevations and runway layouts.
[0043] The direct improvement in operational efficiency is reflected in the fact that this predictive model can be integrated into the airport's intelligent decision support system, providing controllers with accurate takeoff time predictions, thereby optimizing departure sequences and runway scheduling. This indicates a potential improvement of single-runway operational efficiency by 12%-15%, which has significant practical value for increasing the capacity of small and medium-sized airports.
[0044] The balance between computational efficiency and accuracy is achieved through feature selection and a carefully designed model structure. While ensuring the highest accuracy, it avoids the over-parameterization problem of pure global correlation module (Transformer) models. The training efficiency is improved by about 35% compared to a single model, making it more suitable for engineering deployment applications. Attached Figure Description
[0045] Figure 1 This is a flowchart of a method for predicting runway occupancy time during the takeoff phase of an aircraft, provided by an embodiment of the present invention.
[0046] Figure 2 This is a multivariate relationship matrix diagram used to display the correlation of features in a method for predicting runway occupancy time during the takeoff phase of an aircraft, provided in an embodiment of the present invention.
[0047] Figure 3 This is a schematic diagram of the xLSTM structure in an aircraft takeoff phase runway occupancy time prediction method provided in an embodiment of the present invention.
[0048] Figure 4 This is a schematic diagram of the Transformer structure in a method for predicting runway occupancy time during the takeoff phase of an aircraft, provided in an embodiment of the present invention.
[0049] Figure 5 This is a structural diagram of the xLSTM-Transformer hybrid model in an aircraft takeoff phase runway occupancy time prediction method provided in an embodiment of the present invention.
[0050] Figure 6 This is a comparison of the error between the predicted value and the actual value of the xLSTM output in a method for predicting runway occupancy time during the takeoff phase of an aircraft, provided in an embodiment of the present invention.
[0051] Figure 7 This invention provides a method for predicting runway occupancy time during the takeoff phase of an aircraft, which compares the error between the predicted value and the actual value using only the Transformer output in the embodiment of the invention.
[0052] Figure 8 This is a comparison of the error between the predicted value and the actual value of the xLSTM-Transformer output in an aircraft takeoff phase runway occupancy time prediction method provided in an embodiment of the present invention.
[0053] Figure 9 This is a training curve and error distribution diagram of the xLSTM-Transformer model in an aircraft takeoff phase runway occupancy time prediction method provided in an embodiment of the present invention.
[0054] Figure 10 This is the xLSTM-Transformer error distribution histogram in a method for predicting runway occupancy time during aircraft takeoff provided in an embodiment of the present invention.
[0055] Figure 11 This is a radar chart showing the errors of the xLSTM, Transformer, and xLSTM-Transformer models in a method for predicting runway occupancy time during the takeoff phase of an aircraft, provided in an embodiment of the present invention. Detailed Implementation
[0056] The present invention will now be described in detail with reference to the accompanying drawings.
[0057] Example 1
[0058] like Figure 1As shown in the figure, this embodiment of the invention provides a method for predicting runway occupancy time during the takeoff phase of an aircraft. The method includes:
[0059] S101: Acquire multiple categories of aircraft data; the aircraft data represents flight data during the takeoff phase before takeoff.
[0060] Among them, the multiple categories of aircraft data are data acquired by a fast storage recorder. In this embodiment, the data includes strongly correlated data such as airport elevation, runway height, takeoff weight, distance from runway threshold to runway entrance, runway occupancy time, ground speed, whether the aircraft turns around, wind speed, and takeoff weight.
[0061] The relationships between the various categories of aircraft data are as follows: Figure 2 As shown.
[0062] S102: Through Pearson correlation analysis, the aircraft data is filtered to obtain strongly correlated data and weakly correlated data.
[0063] In this context, "strongly correlated data" refers to data strongly correlated with runway occupancy time during aircraft takeoff, while "weakly correlated data" refers to data weakly correlated with runway occupancy time during takeoff. In this embodiment, airport elevation, ground speed, and whether the aircraft turns around are considered strongly correlated data, while wind speed and takeoff weight are considered weakly correlated data.
[0064] S103: Based on strongly correlated data from multiple time points, a high-level feature sequence is obtained through a dual-branch memory structure; the high-level feature sequence represents the extracted local temporal features.
[0065] The dual-branch memory structure includes scalar LSTM and matrix LSTM;
[0066] S104: Through the global association module, based on the high-level feature sequence, the global dependency relationship between time points and feature dimensions is mined to obtain the predicted occupied time length; the predicted occupied time length represents the predicted runway occupancy time length during the aircraft takeoff phase.
[0067] Optionally, the high-level feature sequence obtained from the strongly correlated data at multiple time points, through a dual-branch memory structure, includes:
[0068] The strongly correlated data is preprocessed to extract features and obtain preprocessed feature vectors; multiple categories of strongly correlated data correspond to multiple preprocessed feature vectors.
[0069] The preprocessing includes handling missing values, smoothing noisy data, and identifying and processing outliers.
[0070] This process involves transforming preprocessed strongly correlated data into features and adding them to a feature set. This is done using np.array.
[0071] The function transforms a set of features into a predicted feature vector.
[0072] The strongly correlated data is divided into numerical data and high-dimensional event data; the preprocessed feature vectors corresponding to the numerical data are used as numerical feature vectors; and the preprocessed feature vectors corresponding to the high-dimensional event data are used as high-dimensional event feature vectors.
[0073] Numerical feature vectors from multiple time points are input into a scalar LSTM to identify local details of aircraft flight and obtain the first detail feature.
[0074] High-dimensional event-type feature vectors from multiple time points are input into the matrix LSTM to detect high-dimensional or event-type features and obtain the second detail features.
[0075] The first and second detail features are used to construct a high-level feature sequence.
[0076] The first and second detailed features are combined and mapped by a linear layer (nn.Linear function) through linear transformation to generate more expressive feature representations, highlighting certain important feature patterns while weakening some noise or unimportant information.
[0077] Optionally, the step of mining the global dependency between time points and feature dimensions based on the high-level feature sequence through the global association module to obtain the predicted time occupancy includes:
[0078] The high-level feature sequence is input into the global association module, and a multi-head self-attention mechanism is used to capture global feature dependencies to obtain a global feature vector.
[0079] Among them, the multi-head self-attention mechanism in the global association (Transformer) module has the characteristic of capturing global feature dependencies.
[0080] Among them, the global feature dependencies include how the features of the airport elevation category interact with the features of the ground speed category throughout the takeoff process.
[0081] Based on the global feature vector, the runway occupancy time is predicted, and the predicted occupancy time length is obtained.
[0082] In this embodiment, the predicted runway occupancy time is output after calculation by the fusion layer.
[0083] The output of the dual-branch memory structure serves as the input to the global association module.
[0084] The training uses aircraft data and corresponding labeled runway occupancy time to train a dual-branch memory structure and a global association module.
[0085] Optionally, the aircraft data is filtered through Pearson correlation analysis to obtain strongly correlated and weakly correlated data, including:
[0086] Pearson correlation analysis was used to obtain strong and weak correlation categories; the strong and weak correlation data were obtained during training.
[0087] Based on the strongly correlated and weakly correlated categories, the aircraft data is filtered to obtain strongly correlated data and weakly correlated data.
[0088] Optionally, the aircraft data is filtered through Pearson correlation analysis to obtain strongly correlated and weakly correlated data, including:
[0089] Acquire training aircraft data for multiple categories from multiple tests and the corresponding marked runway occupancy time lengths; the marked runway occupancy time lengths represent the runway occupancy time lengths during the takeoff phase of an aircraft in one test.
[0090] One test corresponds to one sample b number.
[0091] The training aircraft data refers to aircraft data for which the runway occupancy time is known during training.
[0092] Based on the training aircraft data from multiple tests and the corresponding marked runway occupancy time, a time correlation coefficient is obtained; the time correlation coefficient represents the correlation between the training aircraft data and the runway occupancy time during the aircraft takeoff phase.
[0093] Multiple time correlation coefficients were obtained for aircraft data from multiple categories;
[0094] in, Indicates a positive correlation. Indicates a negative correlation. Indicates that it is irrelevant;
[0095] If the time correlation coefficient is greater than the strong correlation threshold, the category corresponding to the aircraft data is considered a strongly correlated category; if the time correlation coefficient is less than or equal to the strong correlation threshold, the category corresponding to the aircraft data is considered a weakly correlated category.
[0096] The strong correlation threshold is 0.4.
[0097] Optionally, the time correlation coefficient can be calculated as follows:
[0098]
[0099] Among them, the This represents training aircraft data for one category. Representing variables The mean; the This indicates the length of time the track has been occupied. Representing variables The mean; This is the time correlation coefficient.
[0100] Where i represents the number of times aircraft data of one category is detected.
[0101] Optionally, dynamic thresholding and attention mechanisms can be introduced into the input gate, forget gate, and output gate of scalar LSTM and matrix LSTM.
[0102] The dynamic threshold and attention mechanism in the forget gate are used to filter noisy data.
[0103] For example, the forget gate can more intelligently filter out noise data such as "wind speed" that has a relatively small impact on runway occupancy time (DROT) during takeoff.
[0104] The above methods improve the intelligence of feature selection.
[0105] Optional training methods for the bi-branch memory structure and the globally associative module include:
[0106] The model was trained using an early stopping mechanism and a model checkpoint saving strategy, with a learning rate set to 0.001.
[0107] The optimizer is Adam.
[0108] Example 2
[0109] Based on the above-described method for predicting runway occupancy time during aircraft takeoff, this invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method for predicting runway occupancy time during aircraft takeoff.
[0110] Example 3
[0111] Based on the above-described method for predicting runway occupancy time during aircraft takeoff, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for predicting runway occupancy time during aircraft takeoff.
[0112] Example 4
[0113] This invention aims to solve the aforementioned problems. Its core innovation lies in proposing a novel hybrid deep learning architecture—the xLSTM-Transformer model—specifically designed for predicting aircraft runway occupancy time. This innovation is not a simple stacking of models, but rather achieves a performance breakthrough through a collaborative enhancement mechanism.
[0114] Innovation Point 1: Dedicated Hybrid Architecture Design for DROT (Runway Occupancy Time) Prediction
[0115] Approach: Abandoning the traditional approach of using a single model, we creatively couple and cascade the improved xLSTM with the Transformer to form a collaborative pipeline of "xLSTM (local fine perception) + Transformer (global relational modeling)".
[0116] Specific methods: The xLSTM module, as a primary feature extractor, first processes the input time-series data. Its sLSTM and mLSTM branches accurately capture local details and event-type features such as "changes in ground speed" and "turning operations," respectively. Subsequently, the high-level feature sequence output by the xLSTM is fed into the Transformer module, which utilizes its powerful self-attention mechanism to deeply mine the global dependencies between all time steps and feature dimensions.
[0117] Through the above methods, the architecture possesses both a keen awareness of local events and a strong understanding of the global context, achieving a 1+1>2 effect.
[0118] Innovation Point 2: Improved xLSTM module to enhance temporal feature extraction
[0119] Approach: Make targeted improvements to the traditional LSTM to better suit the characteristics of aviation time series data.
[0120] Specific means:
[0121] Dual-branch memory structure: sLSTM() is introduced to handle numerical state changes, and mLSTM (matrix LSTM) is used to handle high-dimensional or event-based features (such as "whether to turn around"), which enhances the model's ability to express heterogeneous data.
[0122] Enhanced gating mechanism: Dynamic thresholds and attention mechanisms are introduced into the input gate, forget gate, and output gate. For example, the forget gate can more intelligently filter out noisy data such as "wind speed" that has little impact on DROT, thus improving the intelligence of feature selection.
[0123] The above method provides a cleaner and more informative high-order feature representation for the subsequent Transformer module.
[0124] Innovation Point 3: Feature Engineering and Model Input Optimization Based on Correlation Analysis
[0125] Approach: Instead of blindly inputting all available data into the model, feature selection is based on rigorous physical analysis and statistics (Pearson correlation analysis).
[0126] Specific methods: Analysis revealed a strong correlation between "takeoff speed" and "airport elevation" and DROT (correlation coefficient > 0.65), while "takeoff weight" and "wind direction" showed a weak correlation (correlation coefficient < 0.3). Therefore, strongly correlated features were prioritized as model inputs, with weakly correlated features serving as optional supplements.
[0127] The above methods greatly reduce the input dimensionality and model complexity, reduce the risk of overfitting, improve training efficiency and model generalization ability, and make the model more focused on key influencing factors.
[0128] Acquire QAR data for aircraft takeoff; preprocess and feature-engineer the data to extract feature vectors including airport elevation, takeoff speed, and whether a turnaround is required; input the processed data into a pre-trained xLSTM-Transformer hybrid model; the hybrid model extracts local temporal features through the xLSTM module, then captures global feature dependencies through the multi-head self-attention mechanism of the Transformer module, and outputs the predicted runway occupancy time after calculation by the fusion layer.
[0129] The xLSTM module includes an sLSTM branch and an mLSTM branch. The sLSTM branch is used to process scalar time-series features, and the mLSTM branch is used to process matrix features.
[0130] The feature engineering is based on Pearson correlation analysis results, and features with a correlation coefficient higher than 0.4 with takeoff runway occupancy time are preferentially selected as model inputs.
[0131] The hybrid model is trained using an early stopping mechanism and a model checkpoint saving strategy. The optimizer is Adam, and the learning rate is set to 0.001.
[0132] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages is for the purpose of disclosing the best mode of implementation of the invention.
[0133] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.
[0134] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the apparatus according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.
Claims
1. A method for predicting runway occupancy time during aircraft takeoff, characterized in that, include: Acquire multiple categories of aircraft data; the aircraft data represents flight data during the takeoff phase before takeoff; the flight data includes airport elevation, takeoff weight, distance from runway threshold to runway entrance, ground speed, whether the aircraft is turning around, and wind speed; Pearson correlation analysis was used to filter the aircraft data, resulting in strongly correlated and weakly correlated data. Based on strongly correlated data from multiple time points, a high-level feature sequence is obtained through a dual-branch memory structure; the high-level feature sequence represents the extracted local temporal features. The dual-branch memory structure includes scalar LSTM and matrix LSTM; The global association module mines the global dependency between time points and feature dimensions based on the high-level feature sequence to obtain the predicted occupancy time length; the predicted occupancy time length represents the predicted runway occupancy time during the aircraft takeoff phase. The highly correlated data based on multiple time points is used to obtain a high-level feature sequence through a dual-branch memory structure, including: The strongly correlated data is preprocessed to extract features and obtain preprocessed feature vectors; multiple categories of strongly correlated data correspond to multiple preprocessed feature vectors. The strongly correlated data is divided into numerical data and high-dimensional event data; the preprocessed feature vectors corresponding to the numerical data are used as numerical feature vectors; and the preprocessed feature vectors corresponding to the high-dimensional event data are used as high-dimensional event feature vectors. Numerical feature vectors from multiple time points are input into a scalar LSTM to identify local details of aircraft flight and obtain the first detail feature. High-dimensional event-type feature vectors from multiple time points are input into the matrix LSTM to detect high-dimensional or event-type features and obtain the second detail features. The first and second detail features are used to construct a high-level feature sequence.
2. The method for predicting runway occupancy time during aircraft takeoff as described in claim 1, characterized in that, The step of using the global association module to mine the global dependency relationship between time points and feature dimensions based on the high-level feature sequence to obtain the predicted time length includes: The high-level feature sequence is input into the global association module, and a multi-head self-attention mechanism is used to capture global feature dependencies to obtain the global feature vector. Based on the global feature vector, the runway occupancy time is predicted, and the predicted occupancy time length is obtained.
3. The method for predicting runway occupancy time during aircraft takeoff as described in claim 1, characterized in that, The aircraft data is filtered through Pearson correlation analysis to obtain strongly correlated and weakly correlated data, including: Pearson correlation analysis was used to obtain strong and weak correlation categories; the strong and weak correlation data were obtained during training. Based on the strongly correlated and weakly correlated categories, the aircraft data is filtered to obtain strongly correlated data and weakly correlated data.
4. The method for predicting runway occupancy time during aircraft takeoff as described in claim 3, characterized in that, The Pearson correlation analysis yielded strong and weak correlation categories, including: Acquire training aircraft data for multiple categories from multiple tests and the corresponding marked runway occupancy time lengths; the marked runway occupancy time lengths represent the runway occupancy time lengths during the aircraft's takeoff phase in one test. Based on the training aircraft data from multiple tests and the corresponding marked runway occupancy time, a time correlation coefficient is obtained; the time correlation coefficient represents the correlation between the training aircraft data and the runway occupancy time during the aircraft takeoff phase. Multiple time correlation coefficients were obtained for training aircraft data from multiple categories; If the time correlation coefficient is greater than the strong correlation threshold, the category corresponding to the aircraft data is considered a strongly correlated category; if the time correlation coefficient is less than or equal to the strong correlation threshold, the category corresponding to the aircraft data is considered a weakly correlated category.
5. The method for predicting runway occupancy time during aircraft takeoff as described in claim 4, characterized in that, The method for calculating the time correlation coefficient is as follows: Among them, the This represents training aircraft data for one category. Representing variables The mean; the This indicates the length of time the track was occupied. Representing variables The mean; The time correlation coefficient; Where i represents the number of times aircraft data of one category is detected.
6. The method for predicting runway occupancy time during aircraft takeoff as described in claim 1, characterized in that, Dynamic thresholding and attention mechanisms are introduced into the input gate, forget gate, and output gate of scalar LSTM and matrix LSTM; The dynamic threshold and attention mechanism in the forget gate are used to filter noisy data.
7. The method for predicting runway occupancy time during aircraft takeoff as described in claim 1, characterized in that, Training methods for dual-branch memory structures and globally associative modules include: The model was trained using an early stopping mechanism and a model checkpoint saving strategy, with a learning rate set to 0.
001.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.