A calculation method for civil aviation flight pre-arrival data based on ADS-B

Abstract

The present invention relates to a calculation method for civil aviation flight pre-arrival data based on ADS-B. This method integrates ADS-B data, meteorological satellite data, airport runway status monitoring data, and air traffic control real-time instruction data through multi-source data fusion, and dynamically adjusts the weights of each data source according to the flight phase of the flight. A dynamic adjustment time synchronization interval mechanism is adopted, setting a shorter synchronization interval during takeoff and landing phases to capture rapidly changing data, and extending the synchronization interval during the cruise phase to optimize computing resources. At the same time, a deep learning model based on the Transformer architecture uses the multi-head self-attention mechanism and position encoding to enhance feature capture ability, and reduces the computational burden through a sparse activation strategy. The system is equipped with a real-time monitoring layer, integrating log recording and visualization analysis tools to ensure the interpretability and traceability of the model. It supports efficient aviation management decision-making, improving the efficiency and safety of overall air traffic management.

Description

Technical Field

[0001] The present invention belongs to the technical field of aerospace, and particularly relates to a calculation method for civil aviation flight pre-arrival data based on ADS-B. Background Art

[0002] With the rapid development of the civil aviation industry, the number of flights has been increasing day by day, and the complexity and challenges of air traffic management have also increased accordingly. Accurately predicting the Estimated Time of Arrival (ETA) of flights is of great significance for optimizing flight schedules, improving airport operation efficiency, reducing flight delays, and enhancing passenger satisfaction. Existing technologies mainly use methods based on historical data and real-time monitoring data for flight ETA prediction, but there are still deficiencies in data fusion, real-time performance, and prediction accuracy.

[0003] Chinese invention patent CN112862171B discloses a method for predicting flight arrival time based on a spatio-temporal neural network, belonging to the technical field of civil aviation. This method first collects historical flight tracks as samples for a certain period of historical flights, and obtains a historical flight track set after processing; clusters the historical flight track set, each cluster represents a flight track pattern, and selects the representative flight track of each flight track pattern. Then, given a target flight, resample the flight track that the target flight has flown in the terminal area to obtain the existing flight track of the target flight, and calculate the predicted flight track of the target flight through the flight track pattern. Construct an arrival time prediction model, and use the cross-validation method to train the arrival time prediction model. Finally, input the predicted flight track of the target flight into the trained arrival time prediction model to output the estimated arrival time of the target flight. This invention improves the accuracy of the prediction results by considering the comprehensiveness of the factors affecting the flight arrival time and introducing a spatio-temporal attention mechanism. However, this method mainly relies on the matching of historical flight track patterns, has limited response ability to real-time dynamic changes, and there is still room for improvement in multi-source data fusion and real-time performance.

[0004] Chinese invention patent CN117272236B discloses a multi-source predicted arrival time fusion method and system based on flight time series. The system includes: a raw data receiving and parsing module for receiving and parsing raw data; a message queue processing module for acting as a message queue processor to receive data by establishing an activemq queue; a multi-data source comprehensive screening module for comprehensive screening of multi-data sources; and a business system server for processing the business and verifying the signed information. The multi-source predicted arrival time fusion system based on flight time series provided by the invention has higher stability, reliability, accuracy and real-time performance, and has good scalability. Through an error correction mechanism and the predicted arrival time integrating multiple data sources, the system meets the requirements for the analysis and fusion of predicted arrival time in terms of stability and data reliability. However, the system fails to fully consider the dynamic adjustment of data source weights, the optimization of deep learning models, and anomaly monitoring and traceability, which limits its application effect and system reliability in complex air traffic environments.

[0005] The above two existing technologies each have their own advantages in flight ETA prediction, but there are still the following deficiencies:

[0006] CN112862171B mainly relies on historical flight track patterns for prediction, lacking the dynamic fusion and weight adjustment of multi-source real-time data, resulting in insufficient prediction accuracy and response speed when facing real-time flight state changes. Although CN117272236B realizes the fusion of multi-source data, it lacks advanced algorithm support in the dynamic adjustment and optimization of data source weights and is difficult to achieve the best weight allocation.

[0007] Both of the two existing technologies do not adopt advanced deep learning architectures (such as Transformer) for feature capture and prediction optimization, which limits the model's ability to understand and process complex data relationships.

[0008] Neither CN112862171B nor CN117272236B has established a perfect anomaly monitoring and traceability mechanism, resulting in difficulties in timely discovering and locating problems when the model has anomalies or large prediction errors, affecting the system's reliability and user trust.

[0009] Based on the above deficiencies, the present invention proposes a calculation method for civil aviation flight predicted arrival data based on ADS-B, aiming to Summary of the Invention

[0010] Aiming at the above existing technologies, the technical problem to be solved by the present invention is how to solve the deficiencies in the existing technologies through multi-source data fusion and dynamic weight adjustment, deep learning optimization based on Transformer, and a perfect anomaly monitoring and traceability mechanism, and improve the accuracy, real-time performance and system reliability of flight ETA prediction.

[0011] Dynamically fuse multi-source data from ADS-B, meteorological satellites, runway status monitoring, and air traffic control during different flight phases, and dynamically adjust the weights of each data source according to the real-time flight status to improve the accuracy and flexibility of flight ETA prediction. How to adopt an advanced deep learning architecture (such as Transformer) through the multi-head self-attention mechanism and positional encoding to enhance the model's understanding and processing capabilities of complex data relationships and temporal characteristics, and further improve the prediction accuracy.

[0012] Establish a perfect anomaly monitoring and traceability mechanism, record and monitor the running status of the model in real time, detect and respond to abnormal situations in a timely manner, and ensure the reliability of the system and the trust of users.

[0013] Through the sparse activation strategy and the intelligent adjustment of the time synchronization interval mechanism, achieve the efficient utilization of computing resources, reduce the system operation cost, and at the same time provide high-frequency and high-precision data processing and prediction during critical flight phases.

[0014] To solve the above problems, the present invention provides a calculation method for civil aviation flight pre-arrival data based on ADS-B, including the following steps:

[0015] S1. Multi-source data fusion step:

[0016] Integrate multi-source data from ADS-B data, meteorological satellite data, airport runway status monitoring data, and air traffic control real-time instruction data;

[0017] Dynamically adjust the weights of the above-mentioned data sources according to the flight phase of the flight;

[0018] S2. Dynamically adjust the time synchronization interval step:

[0019] Adjust the time synchronization interval according to the flight phase (takeoff, cruise, landing) of the flight;

[0020] Set the time synchronization interval to A seconds during takeoff and landing phases, and set the time synchronization interval to B seconds during the cruise phase;

[0021] Intelligently adjust the length of the time synchronization interval according to the change of the real-time flight status;

[0022] S3. Deep learning optimization step:

[0023] Construct an input sequence, including the fused multi-source data, the dynamically adjusted data source weights, and the dynamically adjusted time synchronization interval;

[0024] Adopt a model based on the Transformer architecture, process the input sequence through the self-attention mechanism, and capture long-range dependence relationships;

[0025] Perform a non - linear transformation on the features using a feed - forward neural network;

[0026] Apply positional encoding to add temporal information to each element in the input data sequence;

[0027] Adopt a sparse activation strategy, only activate some neurons for calculation, and activate all neurons when an anomaly is detected;

[0028] S4. Anomaly monitoring and traceability steps:

[0029] Set up a real - time monitoring layer to record the input, processing process, and output results of the model;

[0030] Reduce the black - box problem of the model through visual monitoring and logging;

[0031] When an anomaly occurs, achieve the traceability of the calculation process to help identify the source of the problem;

[0032] S5. Civil aviation flight estimated time of arrival (ETA) prediction steps:

[0033] Predict the ETA of civil aviation flights based on the optimized data obtained from the above steps S1 to S4.

[0034] In the S1 multi - source data fusion step, the ADS - B data includes flight real - time position, speed, altitude, and heading information. The flight real - time position uses high - precision positioning technology to improve the spatial resolution and accuracy of the data.

[0035] In the S1 multi - source data fusion step, the meteorological satellite data includes wind speed, wind direction, air current, and temperature information. The meteorological data is obtained through a real - time data stream interface, and a data interpolation algorithm is used to complement missing data to ensure the continuity and integrity of the meteorological information.

[0036] In the S1 multi - source data fusion step, the airport runway status monitoring data includes runway usage status, wetness, and snow accumulation. The runway status data is collected in real - time through a ground sensor network, and a machine learning algorithm is used to predict and evaluate the wetness of the runway.

[0037] In the S1 multi - source data fusion step, the air traffic control real - time instruction data includes the relative position of the flight with other flights, landing queue, and runway change information. The real - time instruction data is obtained through the air traffic management system interface, and a priority sorting algorithm is used to process the instructions to optimize the flight landing sequence.

[0038] The dynamic data source weight adjustment in the S1 multi - source data fusion step uses a weighted average or weighted aggregation algorithm, where the weights change dynamically according to the flight phase and the real - time reliability of the data source. The weight adjustment is based on the Bayesian optimization algorithm to achieve optimal weight allocation.

[0039] In the step of dynamically adjusting the time synchronization interval of S2, the intelligent adjustment mechanism is based on real-time flight data and flight status. When detecting airflow changes or sudden control instructions, it automatically shortens the time synchronization interval and uses the sliding window technology to dynamically manage the time synchronization interval, so as to improve the real-time performance and accuracy of prediction.

[0040] In the step of optimizing by deep learning of S3, the Transformer model includes multiple layers of self-attention mechanisms and fully connected feed-forward neural networks. The self-attention mechanism uses multi-head attention to enhance the model's ability to capture different features, and the position encoding is generated by sine and cosine functions to provide rich temporal information.

[0041] In the step of optimizing by deep learning of S3, the sparse activation strategy selectively activates neurons through a dynamic gating mechanism, only activates a specific proportion of neurons in the normal state to reduce the computational amount, and when detecting anomalies, triggers the activation of all neurons through a global signal to ensure the overall response ability of the model in case of emergencies.

[0042] In the step of anomaly monitoring and traceability of S4, the real-time monitoring layer integrates a log recording module and a visualization analysis tool to record and display the input data, weight adjustment, time synchronization process and output results of the model at each time step in real time. The log recording uses distributed storage technology to ensure the high availability and integrity of data, and the visualization analysis tool supports dynamic query and backtracking of multi-dimensional data to improve the interpretability of the model and the decision-making support ability of operators.

[0043] In summary, the present application has the following beneficial effects:

[0044] 1. By integrating ADS-B data, meteorological satellite data, runway status monitoring data and air traffic control instruction data, comprehensively considering various key factors affecting the predicted arrival time of flights, significantly improving the accuracy of predicted arrival time, dynamically adjusting the weights of each data source according to different flight phases, ensuring that the most relevant data source contributes maximally to the prediction model in each phase, and further optimizing the prediction results.

[0045] 2. The intelligent adjustment mechanism based on real-time flight data and flight status enables the system to flexibly respond to emergencies during flight (such as airflow changes, sudden control instructions), enhancing the system's adaptability to dynamic changes. Using the sliding window technology to dynamically manage the time synchronization interval, ensuring the continuity and real-time performance of data processing in different flight states, and enhancing the overall flexibility of the system.

[0046] 3. Extend the time synchronization interval during the stable cruise phase of flight, reduce the calculation frequency, and lower the system resource consumption; while shorten the time synchronization interval during key phases such as takeoff and landing to achieve high-frequency data processing and prediction. Through the dynamic gating mechanism, only part of the neurons are activated under normal conditions to reduce the calculation amount and energy consumption; all neurons are activated in case of anomalies to ensure the efficient use of computing resources while maintaining prediction accuracy.

[0047] 4. Adopt high-precision positioning technology to obtain ADS-B data, and use data interpolation algorithms to complement meteorological data to ensure the data quality and continuity of each data source, providing reliable data support for the prediction model. Real-time monitor and predict the runway wetness through the ground sensor network and machine learning algorithms to ensure the accuracy and timeliness of runway status data.

[0048] 5. Set up a real-time monitoring layer, integrate the log recording module and visualization analysis tools, comprehensively record the input, processing process, and output results of the model, enhance the transparency of the system, reduce the black-box problem. Through the dynamic query and backtracking of multi-dimensional data, help operators deeply analyze the model behavior and performance, quickly locate and solve abnormal problems, and ensure the reliability and controllability of system operation.

[0049] 6. Adopt the multi-head self-attention mechanism to enhance the model's ability to capture the dependence relationships between different features and different positions, improve the model's understanding and processing ability of complex flight data. Through the positional encoding generated by sine and cosine functions, rich temporal information is provided to ensure that the model can accurately understand the time sequence and dependence relationships of the data, further improving the prediction accuracy.

[0050] 7. Through the real-time anomaly detection and response mechanism, the system can quickly identify and respond to abnormal situations during operation to ensure the accuracy of prediction results and the stable operation of the system. In case of anomalies, the system can trace back the specific calculation process, identify the source of the problem, quickly locate and solve the problem, improving the reliability and stability of the overall system.

[0051] 8. Provide high-precision and real-time prediction of the estimated time of arrival, providing reliable data support for the aviation management department to optimize flight scheduling and resource allocation, reducing flight delays and resource waste. Optimize the flight landing sequence through the priority sorting algorithm to improve the airport operation efficiency, reduce the waiting time of flights on the ground and in the air, and enhance the efficiency and safety of overall air traffic management.

[0052] 9. Through detailed log recording and visualization monitoring, reduce the black-box problem of the model, enhance the transparency of the system, and increase users' trust in the prediction results. The anomaly detection and calculation process backtracking mechanism ensures that the system can still maintain efficient operation and accurate prediction in complex flight environments and emergencies, improving the usability and user satisfaction of the system.

[0053] 10. By reasonably allocating computing resources and optimizing the data processing flow, the system operation cost is reduced, the overall operation efficiency is improved, and with accurate predicted arrival time and optimized flight scheduling mechanism, flight delays and resource waste are reduced, and the operation efficiency and service quality of the airline are enhanced. Description of the Drawings

[0054] Figure 1 Workflow of this application Figure One ;

[0055] Figure 2 Workflow of this application Figure Two ;

[0056] Figure 3 Formula diagram of weight assignment for this application;

[0057] Figure 4 Formula diagram of Bayesian optimization objective function for this application;

[0058] Figure 5 Formula diagram of self-attention mechanism for this application;

[0059] Figure 6 Formula diagram of multi-head attention for this application;

[0060] Figure 7 Position encoding diagram for this application;

[0061] Figure 8 Formula diagram of feed-forward neural network for this application;

[0062] Figure 9 Formula diagram of sparse activation strategy for this application;

[0063] Figure 10 Formula diagram of loss function for this application;

[0064] Figure 11 Formula diagram of Adam optimization algorithm update for this application;

[0065] Figure 12 Formula diagram of distributed storage technology for this application;

[0066] Figure 13 Formula diagram of multi-dimensional data dynamic query and backtracking for this application;

[0067] Figure 14 Formula diagram of detailed log and report generation for this application;

[0068] Figure 15 Formula diagram of anomaly detection for this application;

[0069] Figure 16Abnormal detection formula diagram for this application;

[0070] Figure 17 Distributed storage technology formula diagram for this application;

[0071] Figure 18 Multi-dimensional data dynamic query and backtracking formula diagram for this application;

[0072] Figure 19 Detailed log and report generation formula diagram for this application;

[0073] Figure 20 Abnormal detection formula diagram for this application;

[0074] Figure 21 Backtracking step formula diagram for this application. Detailed implementation mode

[0075] In order to make the objectives, technical solutions and beneficial effects of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings for the convenience of those skilled in the art to understand.

[0076] Embodiment 1:

[0077] As Figures 1 to 2 shown, this embodiment details a calculation method for civil aviation flight pre-arrival data based on ADS-B, with a particular focus on the fusion of multi-source data and the dynamic weight adjustment mechanism to improve the accuracy of flight pre-arrival time prediction and the adaptability of the system. The specific steps are as follows:

[0078] Step S1: Multi-source data fusion step

[0079] Data integration

[0080] This step involves integrating data from multiple data sources to construct comprehensive flight status information. Specifically, it includes:

[0081] ADS-B data: Collect real-time position, speed, altitude, and heading information of the flight. The real-time position of the flight is obtained through high-precision positioning technology to ensure the spatial resolution and accuracy of the data. The ADS-B data is transmitted to the data fusion module in real time through a dedicated communication link.

[0082] Meteorological satellite data: Obtain meteorological information including wind speed, wind direction, air current, and air temperature. The meteorological data is obtained through a real-time data stream interface, and data interpolation algorithms are used to fill in possible missing data to ensure the continuity and integrity of the meteorological information. Data interpolation algorithms such as linear interpolation and spline interpolation can be selected according to the actual situation.

[0083] Airport runway status monitoring data: including runway usage status, wetness level, and snow accumulation. Runway status data is collected in real-time through a ground sensor network, and machine learning algorithms are used to predict and evaluate the wetness level of the runway. The sensor network includes humidity sensors, temperature sensors, and snow detection sensors to collect real-time runway status information.

[0084] Airport runway status monitoring data

[0085] Airport runway status monitoring data includes runway usage status, wetness level, and snow accumulation. Runway status data is collected in real-time through a ground sensor network, which includes humidity sensors, temperature sensors, and snow detection sensors to collect real-time runway status information. To accurately predict and evaluate the wetness level of the runway, the Random Forest machine learning algorithm is adopted. The Random Forest algorithm can effectively process multi-dimensional sensor data and capture complex non-linear relationships by constructing multiple decision trees and combining the prediction results of each tree, thus providing a high-precision prediction of the wetness level. In addition, the Random Forest has good anti-overfitting ability and high computational efficiency, making it suitable for real-time application in the runway status monitoring system.

[0086] Brief description of the implementation steps of the Random Forest algorithm:

[0087] Data preprocessing:

[0088] Collect and organize real-time data from humidity sensors, temperature sensors, and snow detection sensors.

[0089] Perform data cleaning to handle missing values and outliers to ensure data quality.

[0090] Standardize the data to improve the training effect of the algorithm.

[0091] Feature selection and construction:

[0092] Select key features, such as humidity, temperature, snow depth, etc., as input variables for the model.

[0093] Construct feature vectors, combined with time series information, to enhance the model's ability to capture runway status changes.

[0094] Model training:

[0095] Use historical runway wetness level data to train the Random Forest model.

[0096] Optimize model parameters, such as the number and depth of trees, through cross-validation to improve the generalization ability of the model.

[0097] Real-time prediction and evaluation:

[0098] Input the real-time collected data into the trained random forest model to predict the runway wetness level.

[0099] Based on the prediction results, evaluate the runway wetness condition and promptly feedback it to the aviation management system to support flight scheduling and ground operation decisions.

[0100] Air traffic control real-time instruction data: including the relative positions of flights with other flights, landing queue, and runway change information. The real-time instruction data is obtained through the air traffic management system interface, and a priority sorting algorithm is used to process the instructions to optimize the flight landing sequence. The priority sorting algorithm dynamically adjusts the landing sequence according to the importance, urgency, and current queue status of the flights.

[0101] The specific sorting process and basis are as follows:

[0102] The specific sorting process

[0103] Data collection and preprocessing:

[0104] Data acquisition: Real-time obtain the relative positions of each flight, landing queue information, and runway change instructions from the air traffic management system interface.

[0105] Data cleaning: Clean the obtained data to remove noise and outliers to ensure the accuracy and integrity of the data.

[0106] Feature extraction: Extract the key features of each flight, including flight importance, urgency, and current queue position, etc.

[0107] Priority score calculation:

[0108] Weight assignment: According to the preset weight coefficients, score the flight importance, urgency, and queue position respectively.

[0109] Comprehensive score: Perform a weighted sum of each score according to the weights to obtain the comprehensive priority score of each flight.

[0110] Flight sorting:

[0111] Descending order arrangement: Arrange all flights in descending order according to the comprehensive priority score, and the flights with higher priority scores are ranked in the front.

[0112] Dynamic adjustment: When the flight status changes (such as an emergency or a runway change instruction is issued), recalculate the priority scores of the relevant flights and update the sorting results.

[0113] Optimize the landing sequence:

[0114] Determine the landing sequence: Determine the landing sequence of each flight according to the sorting results to ensure that high-priority flights land first.

[0115] Instruction Issuance: Feed the optimized landing sequence back to the air traffic management system to guide the flight to perform the corresponding landing operations.

[0116] Basis for Sorting

[0117] Importance of the Flight:

[0118] Operator Priority: Score according to the importance of the airline or operator. Flights of key operators have a higher priority.

[0119] Flight Type: Flights such as emergency medical flights and government official flights have a higher importance score.

[0120] Degree of Urgency:

[0121] Emergency Situation: Whether there is an emergency on the flight (such as mechanical failure, passenger health problems, etc.). Emergency flights have a higher priority.

[0122] Flight Status: Factors such as whether the flight is approaching landing and whether an emergency landing is required affect the degree of urgency score.

[0123] Current Queuing Situation:

[0124] Queuing Position: The position of the flight in the landing queue. Flights at the front of the queue have a higher priority.

[0125] Estimated Landing Time: The estimated landing time predicted based on the current position and flight speed of the flight. Flights with an earlier estimated landing time have a higher priority.

[0126] Dynamic Weight Adjustment

[0127] Based on data integration, dynamically adjust the weights of each data source according to the different flight phases (takeoff, cruise, landing) of the flight to improve the flexibility and accuracy of the prediction model. Specific methods include:

[0128] Weighted Average or Weighted Aggregation Algorithm: Dynamically allocate weights according to the flight phase of the flight and the real-time reliability of each data source. The weight adjustment is based on the Bayesian optimization algorithm, and the weight allocation is continuously optimized through historical data and real-time data to achieve the optimal weight allocation. For example, during the takeoff phase, the weight of the airport runway status monitoring data is larger; during the cruise phase, the weight of the meteorological satellite data increases; during the landing phase, the weight of the air traffic control instruction data is the largest.

[0129] Weight Allocation Formula: Such as Figure 3 shown

[0130] Bayesian Optimization Algorithm: Utilize the Bayesian optimization algorithm to dynamically adjust weights according to the flight phase and the reliability of data sources. Bayesian optimization builds a probability model to evaluate the effects of different weight combinations, thereby selecting the optimal weight allocation scheme. This method can efficiently find the optimal solution within a limited number of evaluations and is suitable for real-time dynamic adjustment requirements.

[0131] Bayesian Optimization Objective Function Formula: As Figure 4 shown

[0132] The following details the specific implementation process of the Bayesian optimization algorithm in weight adjustment.

[0133] Problem Definition

[0134] The goal is to allocate optimal weights W = {W1, W2, W3, W4} for different data sources (ADS-B data, meteorological satellite data, runway status monitoring data, air traffic control instruction data) to minimize the mean squared error (MSE) of flight predicted arrival time. The weights W i need to satisfy the following constraints:

[0135]

[0136] Objective Function Definition

[0137] The core of Bayesian optimization lies in optimizing a black-box objective function. In the present invention, the objective function is defined as the negative mean squared error of the prediction model:

[0138]

[0139] Bayesian Optimization Process

[0140] Bayesian optimization finds the optimal weight combination W ∗ through an iterative process, and the specific steps are as follows:

[0141] Step 1: Initialization

[0142] Initial sample point selection: Randomly select several groups of weights W that satisfy the constraints as the initial sample points.

[0143] Evaluate the initial sample points: For each group of initial weights W, run the prediction model and calculate the corresponding f(W).

[0144] Step 2: Build a surrogate model

[0145] Gaussian Process Regression (GPR):

[0146] Use Gaussian process regression as a surrogate model to fit the evaluated weight combinations

[0147] The relationship between W and the corresponding objective function value f(W).

[0148] The Gaussian process provides an estimate of the mean and uncertainty of the objective function, providing a basis for the next acquisition function.

[0149] Step 3: Select the next evaluation point

[0150] Acquisition Function:

[0151] Commonly used acquisition functions include Expected Improvement (EI), Probability of Improvement (PI), and Knowledge Gradient (KG).

[0152] Taking the expected improvement as an example, select the next weight combination W next , to maximize the expected improvement over the current optimal objective function value:

[0153]

[0154] Step 4: Evaluate the new sample point

[0155] Run the prediction model: Use the newly selected weight combination W next Run the prediction model to calculate

[0156] f(W next ).

[0157] Step 5: Update the surrogate model

[0158] Data update: Add the new weight combination W next and its objective function value f(W next ) to the evaluated sample set.

[0159] Retrain the surrogate model: Retrain the Gaussian process regression model using the updated sample data to update the estimate of the objective function.

[0160] Step 6: Iteration

[0161] Repeat steps 3 to 5: Continuously iterate to select new evaluation points, evaluate the objective function values, and update the surrogate model until the stopping condition is met (such as reaching a predetermined number of iterations or the objective function converges).

[0162] Step 7: Obtain the optimal weight

[0163] Determine the optimal weight combination W ∗ : After the iteration ends, select the weight combination W with the highest objective function value (i.e., the lowest MSE). *

[0164] Implementation details

[0165] Data preprocessing

[0166] Normalize the weights: Ensure that the weights

[0167] W satisfy the constraint conditions and

[0168] Feature engineering: Combine the flight phases (takeoff, cruise, landing) and the real-time reliability of the data sources to construct a feature vector for weight adjustment.

[0169] Surrogate model parameter settings

[0170] Gaussian process kernel function: Select an appropriate kernel function (such as the RBF kernel) to capture the non-linear relationship between the weights and the objective function.

[0171] Hyperparameter optimization: Optimize the hyperparameters of the Gaussian process by maximizing the marginal likelihood estimate to improve the fitting ability of the surrogate model.

[0172] Acquisition function optimization

[0173] Optimize the acquisition function: In each iteration, select the next weight combination W by optimizing the acquisition function (such as EI) next , ensuring efficient exploration of the weight space within a limited number of evaluations.

[0174] Real-time dynamic adjustment

[0175] Online learning: Implement an online version of Bayesian optimization to support real-time adjustment of weights during flight and respond to dynamic changes in flight status.

[0176] Computational efficiency: Optimize the algorithm implementation to ensure sufficient computational speed in a real-time system to meet the real-time requirements of air traffic management.

[0177] Steps S2 to S5:

[0178] Since this embodiment mainly focuses on the multi-source data fusion and dynamic weight adjustment in step S1, the following steps S2 to S5 briefly describe their applications in this embodiment:

[0179] Step S2: Dynamically adjust the time synchronization interval step

[0180] Adjust the time synchronization interval according to the flight phase.

[0181] Set the time synchronization interval to A seconds (e.g., 5 seconds) during takeoff and landing phases, and set the time synchronization interval to B seconds (e.g., 30 seconds) during the cruise phase.

[0182] In step S2 of dynamically adjusting the time synchronization interval, the intelligent adjustment mechanism performs dynamic management based on real-time flight data and flight status. When the system detects changes in airflows or sudden air traffic control instructions, it automatically shortens the time synchronization interval to improve the real-time performance and accuracy of predictions. To effectively manage the length of the time synchronization interval, the present invention adopts a sliding window technique, and its specific implementation process is as follows:

[0183] Window initialization:

[0184] Set the initial window size: Set the initial length of the time synchronization interval according to the flight phase (takeoff, cruise, landing) of the flight. For example, set a shorter time synchronization interval (such as 5 seconds) during takeoff and landing phases, and set a longer time synchronization interval (such as 30 seconds) during the cruise phase.

[0185] Real-time data monitoring and event detection:

[0186] Continuously monitor the flight status: The system continuously monitors data from ADS-B, meteorological satellites, and other sensors, and analyzes flight parameters in real time.

[0187] Event triggering: Once a change in airflows (such as sudden turbulence) is detected or a sudden air traffic control instruction (such as runway change, landing queue adjustment) is received, the system immediately triggers an adjustment of the time synchronization interval.

[0188] Dynamic window adjustment:

[0189] Shorten the time synchronization interval: After detecting the above events, the system automatically shortens the current time synchronization interval from the longer B seconds (such as 30 seconds) to the shorter A seconds (such as 10 seconds) to respond more quickly to rapid changes in flight status.

[0190] Sliding window update: Adopting the sliding window technique, the system adds new data points to the window at each time step while removing the oldest data points, ensuring that the window always contains the latest flight status information. In this way, the window can dynamically adapt to changes in flight status and reflect the current flight environment in real time.

[0191] Window recovery and optimization:

[0192] Recovery after the event: When the change in airflows or the sudden instruction is processed and the flight status returns to stability, the system gradually restores the time synchronization interval to the longer B seconds to optimize the use of computing resources.

[0193] Sliding window smooth transition: Through the smooth transition mechanism, it avoids the interference of frequent switching of window sizes on the prediction model, ensuring data continuity and prediction stability.

[0194] Improvement of real-time performance and accuracy:

[0195] Quick response: The sliding window technology enables the system to capture the latest changes in flight status within a short time, quickly adjust the input data range of the prediction model, and improve the real-time performance of the prediction.

[0196] Data continuity: The sliding window ensures data continuity and the integrity of the time series, enabling the prediction model to make accurate predictions based on the latest and continuous data.

[0197] Intelligently adjust the length of the time synchronization interval according to the changes in real-time flight status.

[0198] In step S2 of the present invention, the dynamic adjustment of the time synchronization interval specifically depends on the following real-time flight status factors:

[0199] Airflow changes:

[0200] Sudden changes in wind speed and direction: By real-time monitoring of meteorological satellite data, when a sharp change in wind speed or direction is detected, it indicates that there may be airflow disturbances or turbulence, and the time synchronization interval needs to be shortened for a quick response.

[0201] Air traffic control instructions:

[0202] Sudden instructions: The issuance of sudden air traffic control instructions such as runway changes and landing queue instructions requires the flight to quickly adjust the flight path or landing sequence, and the time synchronization interval needs to be shortened to update the prediction in real time.

[0203] Rapid changes in flight status:

[0204] Sharp changes in speed and altitude: Flights usually experience large speed and altitude adjustments during takeoff and landing. When rapid changes in these parameters are detected, the time synchronization interval needs to be shortened to accurately capture the dynamic changes in flight status.

[0205] Emergency events:

[0206] Emergencies: The occurrence of emergencies such as mechanical failures and passenger health problems requires the flight to respond quickly, and the time synchronization interval needs to be shortened to ensure that the prediction model can update the estimated arrival time in a timely manner.

[0207] Through the monitoring and analysis of the above real-time flight status factors, the system can intelligently adjust the length of the time synchronization interval, ensure high-precision estimated arrival time prediction during critical or abnormal flight phases, and optimize the use of computing resources when the flight status is stable.

[0208] Step S3: Deep learning optimization step

[0209] Construct an input sequence, including the fused multi-source data, the dynamically adjusted data source weights, and the dynamically adjusted time synchronization interval.

[0210] Adopt a model based on the Transformer architecture, and process the input sequence through the self-attention mechanism to capture long-range dependencies.

[0211] Use a feed-forward neural network to perform non-linear transformation on the features.

[0212] Apply positional encoding to add temporal information to each element in the input data sequence.

[0213] Adopt a sparse activation strategy, only activate some neurons for calculation, and activate all neurons when an anomaly is detected.

[0214] Step S4: Anomaly monitoring and traceability step

[0215] Set up a real-time monitoring layer to record the input, processing process, and output results of the model.

[0216] Reduce the black-box problem of the model through visual monitoring and logging.

[0217] When an anomaly occurs, implement the traceability of the calculation process to help identify the source of the problem.

[0218] Step S5: Flight estimated time of arrival prediction step

[0219] Predict the estimated time of arrival of civil aviation flights based on the optimized data obtained from the above steps S1 to S4.

[0220] Working process

[0221] Data collection and integration:

[0222] Through the real-time data collection interfaces of each data source (ADS-B, meteorological satellite, runway status monitoring, air traffic control), obtain the latest flight status information.

[0223] Adopt data cleaning and preprocessing techniques to ensure the data quality and consistency of each data source. This includes removing noise data, filling in missing values, and standardizing the data format.

[0224] Dynamic weight allocation:

[0225] According to the flight phase of the current flight, use the Bayesian optimization algorithm to dynamically adjust the weights of each data source.

[0226] The weight adjustment process responds in real time to changes in flight status, ensuring that the most relevant data sources contribute maximally to the prediction model at different flight stages.

[0227] Data fusion and feature construction:

[0228] The data from each data source is fused according to the weights through weighted average or weighted aggregation algorithms to construct comprehensive flight status features.

[0229] An input sequence containing dynamic weight adjustment information and time synchronization interval adjustment information is constructed to provide comprehensive inputs for the deep learning model.

[0230] Deep learning model processing:

[0231] The constructed input sequence is input into a deep learning model based on the Transformer architecture.

[0232] The model captures long-range dependencies in the input data through the self-attention mechanism, performs non-linear feature transformation using a feed-forward neural network, and enhances temporal information through positional encoding.

[0233] Anomaly monitoring and response:

[0234] The input, processing process, and output results of the model are monitored in real time, and detailed logs are recorded.

[0235] Through a visualization monitoring tool, operators can view the running status of the model in real time and identify potential anomalies.

[0236] When an abnormal situation is detected, the system can trace the calculation process, locate the source of the problem, and ensure the reliability of the prediction.

[0237] Estimated time of arrival prediction and output:

[0238] Based on the optimized data and the processing results of the deep learning model, an estimated time of arrival prediction for civil aviation flights is generated.

[0239] The prediction results are provided to relevant aviation management systems through the user interface or interfaces for flight scheduling and resource optimization.

[0240] By integrating multi-source data and comprehensively considering various factors such as the real-time position of the flight, meteorological conditions, runway status, and air traffic instructions, the accuracy of the estimated time of arrival prediction is significantly improved.

[0241] The dynamic weight adjustment mechanism ensures that the most relevant data sources are fully utilized at different flight stages, further optimizing the prediction results.

[0242] The Bayesian optimization algorithm is used for weight allocation, which can flexibly adjust the weight allocation according to real-time data and flight status, enhancing the system's adaptability to dynamic changes.

[0243] The intelligent adjustment of the time synchronization interval mechanism enables the system to provide high-precision predictions during critical flight phases and optimize computing resources during stable flight phases.

[0244] Through weighted average or weighted aggregation algorithms, computing resources are reasonably allocated to avoid unnecessary data processing and improve the overall operating efficiency of the system.

[0245] During non-critical phases, the time synchronization interval is extended, the computing frequency is reduced, and the system burden is lowered.

[0246] High-precision positioning technology and data interpolation algorithms ensure the data quality and continuity of each data source, providing reliable data support for the prediction model.

[0247] The ground sensor network and machine learning algorithms conduct real-time monitoring and prediction of the runway status to ensure the accuracy and timeliness of runway status data.

[0248] The real-time monitoring layer records detailed model operation data. Through visualization monitoring tools, operators can intuitively understand the working process of the model and reduce the black box problem.

[0249] The anomaly monitoring and traceability mechanism can quickly locate the source of problems when issues occur, enhancing the reliability of the system and users' trust.

[0250] Accurate prediction of the estimated time of arrival provides reliable data support for aviation management departments, optimizes flight scheduling and resource allocation, and reduces flight delays and resource waste.

[0251] The priority sorting algorithm optimizes the landing sequence, improves the operating efficiency of the airport, and reduces the waiting time of flights on the ground and in the air.

[0252] By implementing the above specific methods, this embodiment demonstrates an efficient multi-source data fusion and dynamic weight adjustment mechanism, significantly improving the accuracy of civil aviation flight estimated time of arrival prediction based on ADS-B and the adaptability of the system. This method integrates multiple data sources and uses advanced optimization algorithms to dynamically adjust data weights to ensure that during different flight phases, the prediction model can make full use of the most relevant data sources to provide high-precision prediction results. At the same time, the anomaly monitoring and traceability design of the system enhances the reliability of the system and users' trust, providing strong data support and decision-making basis for aviation management departments.

[0253] Embodiment 2:

[0254] As Figures 1 to 2As shown, this embodiment details a calculation method for civil aviation flight pre-arrival data based on ADS-B, with a particular focus on the Transformer model, its multi-head attention mechanism, position encoding, and sparse activation strategy in the deep learning optimization step (step S3); meanwhile, the real-time monitoring layer, logging module, visualization analysis tool, and distributed storage technology in the anomaly monitoring and traceability steps are introduced. This method is further improved based on Embodiment 1, with a focus on model optimization and anomaly monitoring mechanisms to improve prediction accuracy and system reliability.

[0255] Step S3: Deep learning optimization step

[0256] Input sequence construction:

[0257] Fusion of data and dynamic adjustment information:

[0258] Construct an input sequence by combining the comprehensive data obtained through multi-source data fusion and dynamic weight adjustment in step S1 of Embodiment 1 with the time synchronization interval information dynamically adjusted in step S2.

[0259] The input sequence includes the weighted aggregation result of multi-source data, the dynamic weights of each data source, and the length of the current time synchronization interval.

[0260] Transformer model architecture:

[0261] Multi-layer self-attention mechanism:

[0262] The Transformer model consists of multi-layer self-attention mechanisms and fully connected feed-forward neural networks, and each layer includes a multi-head attention sub-layer and a feed-forward network sub-layer.

[0263] Multi-Head Attention:

[0264] Through multiple parallel attention heads (e.g., 8 heads), enhance the model's ability to capture dependencies between different features and different positions. Each attention head independently calculates attention weights, and finally, the outputs of each head are concatenated and linearly transformed to obtain rich feature representations.

[0265] Self-attention mechanism formula: As Figure 5 shown;

[0266] Multi-Head Attention formula: As Figure 6 shown

[0267] Position Encoding:

[0268] Since the Transformer model itself does not have the ability to handle the order of sequence data, positional encoding is introduced to provide temporal information.

[0269] Positional encoding generated by sine and cosine functions: As Figure 7 shown

[0270] Feed-forward neural network (feed-forward neural network part):

[0271] Nonlinear feature transformation:

[0272] In the feed-forward neural network sublayer of each Transformer layer, a fully connected layer is applied for nonlinear transformation. It usually includes two linear transformation layers and a ReLU activation function to increase the model's expressive power and capture complex nonlinear relationships.

[0273] Feed-forward neural network formula: As Figure 8 shown

[0274] Sparse activation strategy:

[0275] Dynamic gating mechanism:

[0276] The dynamic gating mechanism is adopted to implement the sparse activation strategy, and only part of the neurons are activated in the normal state (for example, the activation ratio is 30%) to reduce the computational amount and improve the operation efficiency.

[0277] Sparse activation strategy formula: As Figure 9 shown

[0278] Anomaly detection and full neuron activation:

[0279] When the system detects an anomaly (such as a drastic change in flight state, an emergency event, etc.), all neurons are activated through a global signal to ensure that the model can respond comprehensively in case of an emergency and improve the accuracy of prediction and the reliability of the system.

[0280] Model training and optimization:

[0281] Loss function and optimization algorithm:

[0282] The mean squared error (MSE) is used as the loss function to measure the gap between the predicted value and the true predicted arrival time.

[0283] Loss function formula: As Figure 10 shown;

[0284] The Adam optimization algorithm is adopted to update the model parameters to accelerate convergence and improve the model performance.

[0285] Adam optimization algorithm update formula: As Figure 11 shown

[0286] Step S4: Abnormal Monitoring and Traceability Step

[0287] Construction of the Real-time Monitoring Layer (Part of Step S4, Real-time Monitoring Layer):

[0288] Integrated Logging Module:

[0289] Set up a logging module to record the input data of the model, weight adjustment, time synchronization process, and output results in real time.

[0290] Distributed Storage Technology Formula: As Figure 12 shown;

[0291] Visualization Analysis Tool:

[0292] Integrate visualization analysis tools (such as Grafana, Kibana, etc.) to display the running status of the model at each time step in real time.

[0293] Multi-dimensional Data Dynamic Query and Backtracking Formula: As Figure 13 shown

[0294] Reduction of Black Box Problem (Part of Step S4, Transparency Enhancement):

[0295] Enhancement of Transparency:

[0296] Through real-time recording and visual monitoring, operators can intuitively understand the calculation process and decision-making basis of the model, reducing the black box problem.

[0297] Detailed Log and Report Generation Formula: As Figure 14 shown

[0298] Abnormal Handling and Traceability (Part of Step S4, Abnormal Detection and Response):

[0299] Abnormal Detection and Response:

[0300] When the system detects an abnormality during the model operation (such as excessive prediction error, abnormal input data, etc.), it automatically triggers an alarm mechanism to notify relevant personnel to take measures.

[0301] Abnormal Detection Formula: As Figure 15 shown

[0302] Calculation Process Backtracking:

[0303] When an abnormality occurs, the real-time monitoring layer can trace back to the specific calculation process, identify possible problems in the data source, algorithm adjustment, or model inference process, and help quickly locate and solve the problem.

[0304] Backtracking Step Formula: As Figure 16 shown;

[0305] Working process

[0306] Input sequence preparation and model processing:

[0307] Integrate the multi-source data, dynamic weights, and time synchronization interval information processed in steps S1 and S2 to construct an input sequence.

[0308] Input the input sequence into a deep learning model based on the Transformer architecture for predicted arrival time prediction.

[0309] Model operation and sparse activation:

[0310] In the normal state, the model processes data through the multi-head self-attention mechanism and the feed-forward neural network, and only activates some neurons to reduce the computational burden.

[0311] When an anomaly is detected, the dynamic gating mechanism triggers the full neuron activation mode to ensure that the model can comprehensively respond to complex flight state changes.

[0312] Anomaly monitoring and logging:

[0313] The real-time monitoring layer records every step of the model's calculation details, including input data, weight adjustment, time synchronization process, and output results.

[0314] The visualization analysis tool displays the model operation status in real time, supporting operators for dynamic query and retrospective analysis.

[0315] Predicted arrival time prediction and feedback:

[0316] Based on the output of the model, generate accurate predicted arrival time prediction results for civil aviation flights.

[0317] The prediction results are provided to the relevant aviation management system through the user interface or interface for flight scheduling and resource optimization.

[0318] The multi-head self-attention mechanism enhances the model's ability to capture different features and can better understand and process complex flight data relationships.

[0319] The positional encoding provides rich temporal information to ensure that the model can accurately understand the time sequence and dependency relationship of the data.

[0320] The sparse activation strategy only activates some neurons in the normal state through the dynamic gating mechanism, significantly reducing the computational amount and energy consumption.

[0321] In case of an anomaly, the model can quickly switch to the full neuron activation mode to ensure the accuracy of the prediction and the response ability of the system.

[0322] The real-time monitoring layer enhances the interpretability of the system by integrating a logging module and a visualization analysis tool, helping operators understand and verify the calculation process of the model.

[0323] Distributed storage technology ensures the high availability and integrity of log data, supporting the efficient management and query of large-scale data.

[0324] Through detailed logging and visual monitoring, the system can quickly identify and respond to anomalies during operation, enhancing the overall reliability and stability of the system.

[0325] The anomaly handling and traceability mechanism ensures that the system can still operate efficiently and make accurate predictions in complex flight environments and emergencies.

[0326] The highly accurate and real-time prediction of the estimated time of arrival provides reliable data support for aviation management departments, optimizing flight scheduling and resource allocation, and reducing flight delays and resource waste.

[0327] The traceability and transparency of the system enhance users' trust in the prediction results and strengthen the decision-making support ability of aviation management.

[0328] By implementing the above specific methods, this embodiment demonstrates a deep learning optimization and anomaly monitoring method based on Transformer, significantly improving the accuracy of civil aviation flight estimated time of arrival prediction based on ADS-B and the reliability of the system. Specifically, it is manifested in:

[0329] The application of the multi-head attention mechanism and positional encoding enables the model to better understand and process the complex relationships and temporal characteristics in multi-source data, improving the prediction accuracy.

[0330] The sparse activation strategy, through a dynamic gating mechanism, only activates some neurons when necessary, reducing the computational burden and energy consumption, and improving the system operation efficiency.

[0331] The intelligent anomaly monitoring and traceability mechanism enables the system to quickly respond to emergencies, maintain efficient and stable operation, and ensure the reliability of the prediction results.

[0332] The real-time monitoring and visualization analysis tool provides comprehensive model operation information, helping operators timely understand the system status and make accurate management decisions.

[0333] Through detailed logging and visual monitoring, the black box problem of the model is reduced, the transparency of the system and users' trust are enhanced, providing more reliable technical support for aviation management departments.

[0334] In summary, through the introduction of an advanced Transformer model architecture, a sparse activation strategy, and a comprehensive anomaly monitoring and traceability mechanism, this embodiment significantly improves the performance and reliability of the civil aviation flight pre-arrival time prediction method based on ADS-B, providing strong technical support and decision-making basis for air traffic management.

[0335] Embodiment 3:

[0336] As Figures 1 to 2 shown, this embodiment details a calculation method for civil aviation flight pre-arrival data based on ADS-B, with a particular focus on the anomaly monitoring and traceability step (Step S4) to ensure the interpretability of the model and the reliability of the system. This method further improves on the basis of Embodiment 1 and Embodiment 2, and mainly introduces the construction of the real-time monitoring layer, the integration of the log recording module, and the application of the visualization analysis tool. The specific steps are as follows:

[0337] Step S4: Anomaly monitoring and traceability step

[0338] Construction of the real-time monitoring layer

[0339] Integration of the log recording module:

[0340] Set up a log recording module to record the input data of the model, weight adjustment, time synchronization process, and output results in real time.

[0341] Distributed storage technology formula: As Figure 17 shown

[0342] Visualization analysis tool:

[0343] Integrate visualization analysis tools (such as Grafana, Kibana, etc.) to display the running status of the model at each time step in real time.

[0344] Multi-dimensional data dynamic query and backtracking formula: As Figure 18 shown

[0345] Reduction of the black box problem

[0346] Improvement of transparency:

[0347] Through real-time recording and visual monitoring, operators can intuitively understand the calculation process and decision-making basis of the model, reducing the black box problem.

[0348] Detailed log and report generation formula: As Figure 19 shown

[0349] Anomaly handling and traceability

[0350] Anomaly detection and response:

[0351] When the system detects an anomaly during model operation (such as excessive prediction error, abnormal input data, etc.), it automatically triggers an alarm mechanism to notify relevant personnel to take measures.

[0352] Anomaly detection formula: as Figure 20 shown

[0353] Calculation process backtracking:

[0354] When an anomaly occurs, the real-time monitoring layer can trace back to the specific calculation process, identify potential problems in the data source, algorithm adjustment, or model inference process, and help quickly locate and solve the problem.

[0355] Backtracking step formula: as Figure 21 shown

[0356] Working process

[0357] Model operation and anomaly monitoring:

[0358] During the normal operation of the model, the input data, weight adjustment, time synchronization process, and output results are monitored in real time, and detailed logs are recorded.

[0359] The visualization analysis tool shows the running status of the model in real time. Operators can dynamically view and analyze the running situation of the model through the multi-dimensional query function.

[0360] Anomaly detection and alarm:

[0361] The system monitors the difference between the prediction result and the actual result in real time through a preset anomaly detection formula. When the detected prediction error exceeds the set threshold, the system automatically triggers an alarm mechanism to notify relevant personnel to intervene.

[0362] Calculation process backtracking and problem location:

[0363] When an anomaly occurs, the system traces back the calculation process before and after the anomaly event through the log records and visualization analysis tool of the real-time monitoring layer.

[0364] By analyzing the input data, weight adjustment, time synchronization process, and output results, quickly identify the source of the problem, such as abnormal data source, incorrect weight adjustment, or model inference deviation.

[0365] Problem solving and system optimization:

[0366] Relevant personnel take corresponding measures to solve the problem according to the backtracking analysis results, such as data source repair, algorithm adjustment, or model retraining.

[0367] The system optimizes the log record and anomaly detection mechanism according to the processing results, improving the overall reliability and prediction accuracy of the system.

[0368] The real-time monitoring layer records and displays the running status of the model at each time step by integrating a logging module and a visualization analysis tool, helping operators intuitively understand the calculation process of the model and reducing the black-box problem.

[0369] The multi-dimensional data dynamic query and backtracking function supports operators to deeply analyze the model behavior and performance, quickly locate and solve abnormal problems.

[0370] Through the real-time anomaly detection and response mechanism, the system can quickly identify and respond to abnormal situations during operation, ensuring the accuracy of prediction results and the stable operation of the system.

[0371] The calculation process backtracking mechanism helps quickly locate the source of problems, reduces system downtime, and improves overall reliability.

[0372] Detailed logging and visualization analysis tools provide comprehensive model operation information for operators, supporting more accurate and timely aviation management decisions.

[0373] Accurate predicted arrival time prediction and efficient anomaly handling mechanism optimize flight scheduling and resource allocation, reducing flight delays and resource waste.

[0374] Through the transparent model operation process and detailed logging, the trust of users in the prediction results is enhanced, and the usability and user satisfaction of the system are improved.

[0375] The visualization monitoring tool provides an intuitive display of the system running status, helping operators promptly discover and handle potential problems, ensuring the efficient operation of the system.

[0376] By implementing the above specific methods, this embodiment demonstrates an anomaly monitoring and traceability mechanism, significantly improving the system transparency and reliability of the civil aviation flight predicted arrival time prediction method based on ADS-B. The real-time monitoring layer provides comprehensive model operation information by integrating a logging module and a visualization analysis tool, helping operators timely understand and verify the calculation process of the model and reducing the black-box problem. At the same time, the anomaly detection and calculation process backtracking mechanism ensures that the system can quickly locate and solve problems when anomalies occur, improving the overall reliability of the system and user trust. This mechanism provides strong technical support and decision-making basis for aviation management departments, optimizes flight scheduling and resource allocation, and improves the efficiency and safety of air traffic management.

[0377] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A method for calculating civil aviation flight arrival data based on ADS-B, characterized by: The following steps are involved: S1. Multi-source data fusion steps: Integrate multi-source data from ADS-B data, meteorological satellite data, airport runway status monitoring data and air traffic control real-time instruction data; Dynamically adjust the weights of the above data sources according to the flight phase of the flight; S2. Steps to dynamically adjust the time synchronization interval: Adjust the time synchronization interval according to the flight phase, take-off, cruising, and landing; The time synchronization interval is set to A seconds during takeoff and landing, and the time synchronization interval is set to B seconds during cruising; Intelligently adjust the time synchronization interval length according to the changes in real-time flight status; S3. Deep learning optimization steps: Construct an input sequence, including fused multi-source data, dynamically adjusted data source weights, and dynamically adjusted time synchronization intervals; A model based on the Transformer architecture is used to process input sequences through a self-attention mechanism to capture long-distance dependencies. Use feedforward neural network to perform nonlinear transformation on features; Applying positional encoding adds timing information to each element in the input data sequence; A sparse activation strategy is used to activate only some neurons for calculation, and all neurons are activated when an anomaly is detected; S4. Abnormal monitoring and traceability steps: Establish a real-time monitoring layer to record the model’s input, processing, and output results; Reduce the black box problem of models through visual monitoring and logging; When an exception occurs, the computing process can be traced back to help identify the source of the problem; S5. Flight arrival time prediction steps: Based on the optimized data obtained in the above steps S1 to S4, the expected arrival time of the civil aviation flight is predicted.

2. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S1 multi-source data fusion step, the ADS-B data includes the real-time position, speed, altitude and heading information of the flight. The real-time position of the flight adopts high-precision positioning technology to improve the spatial resolution and accuracy of the data.

3. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S1 multi-source data fusion step, the meteorological satellite data includes wind speed, wind direction, airflow and temperature information. The meteorological data is acquired through a real-time data stream interface, and a data interpolation algorithm is used to supplement missing data to ensure the continuity and integrity of meteorological information.

4. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S1 multi-source data fusion step, the airport runway status monitoring data includes runway usage status, slippery degree and snow accumulation conditions. The runway status data is collected in real time through a ground sensor network, and a machine learning algorithm is used to predict and evaluate the runway slippery degree.

5. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S1 multi-source data fusion step, the air traffic control real-time instruction data includes the relative position of the flight to other flights, landing queue and runway change information. The real-time instruction data is obtained through the air traffic management system interface, and the instructions are processed using a priority sorting algorithm to optimize the flight landing order.

6. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: The dynamic data source weight adjustment in the S1 multi-source data fusion step adopts a weighted average or weighted aggregation algorithm, wherein the weight changes dynamically according to the flight phase of the flight and the real-time reliability of the data source. The weight adjustment is based on a Bayesian optimization algorithm to achieve optimal weight distribution.

7. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S2 step of dynamically adjusting the time synchronization interval, the intelligent adjustment mechanism is based on real-time flight data and flight status. When airflow changes or sudden control instructions are detected, the time synchronization interval is automatically shortened, and the sliding window technology is used to dynamically manage the time synchronization interval to improve the real-time and accuracy of the prediction.

8. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S3 deep learning optimization step, the Transformer model includes a multi-layer self-attention mechanism and a fully connected feedforward neural network. The self-attention mechanism uses multi-head attention to enhance the model's ability to capture different features, and the position encoding is generated using sine and cosine functions to provide rich timing information.

9. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S3 deep learning optimization step, the sparse activation strategy selectively activates neurons through a dynamic gating mechanism, activating only a specific proportion of neurons under normal conditions to reduce the amount of computation, and triggers the activation of all neurons through a global signal when an abnormality is detected to ensure the comprehensive responsiveness of the model in emergency situations.

10. The method for calculating the expected arrival data of civil aviation flights based on ADS-B according to claim 1, characterized in that: In the S4 anomaly monitoring and traceability step, the real-time monitoring layer records and displays the model's input data, weight adjustment, time synchronization process and output results at each time step in real time by integrating a logging module and a visual analysis tool. The logging uses distributed storage technology to ensure high availability and integrity of the data, and the visual analysis tool supports dynamic query and backtracking of multi-dimensional data to improve the interpretability of the model and the decision support capabilities of operators.

Images