An airport single-point noise prediction method
Patent Information
- Application Number
- CN202611217570.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-12
- Publication Date
- 2026-10-30
AI Technical Summary
[0017]1、构建动态推力系数分布模型，从历史数据逆向求解实际推力系数，按机型、飞行阶段分类建模，精准刻画其动态变化与概率分布，通过物理引导神经网络，融合物理规律与数据驱动优势，以物理约束保障预测合规性，显著提升复杂场景下的模型外推能力与预测可靠性。
Smart Images
Figure CN123014313A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of noise prediction, and more particularly to a method for predicting noise at a single point in an airport. Background Technology
[0002] Airport noise is a significant issue affecting the surrounding environment and community life. Accurate noise prediction is crucial for airport planning, operation management, noise mitigation, and community relations coordination. With the growth of air traffic, the diversification of aircraft performance, and the increasing complexity of operating modes, existing prediction methods have revealed obvious limitations in practical applications.
[0003] On the one hand, existing noise prediction methods generally use a fixed typical thrust coefficient to represent the noise emissions of a specific aircraft type at a specific stage. However, the actual thrust of an aircraft is significantly affected by real-time weather conditions, specific payload, pilot operating habits, and air traffic control instructions. It is a dynamically changing parameter, and existing methods cannot effectively characterize the dynamics and statistical distribution characteristics of this thrust coefficient, resulting in a significant decrease in prediction accuracy when deviating from standard operating conditions.
[0004] On the other hand, existing prediction methods mostly adopt offline training and fixed deployment models. Airport operating environments, fleet composition, air traffic control strategies, and seasonal climates are constantly changing, causing the predictive performance of models to gradually degrade over time. Existing technologies lack effective mechanisms to automatically and promptly assess prediction errors based on real-time monitoring data and trigger model updates, making it difficult to guarantee the accuracy and adaptability of prediction systems during long-term service.
[0005] Application content
[0006] This application aims to address, at least to some extent, the technical problems in the related art.
[0007] To achieve the above objectives, this application proposes a single-point noise prediction method for airports, comprising the following steps: S1: collecting historical airport operation data, including flight trajectory data, meteorological record data and noise monitoring data; based on the historical airport operation data, constructing a dynamic thrust coefficient distribution model and training it using a physical guided neural network to obtain a trained noise prediction model.
[0008] S2: Receive real-time flight trajectory data and real-time meteorological data, extract prediction features based on the real-time flight trajectory data, and input the prediction features and the real-time meteorological data into the trained noise prediction model to obtain noise prediction results;
[0009] S3: Continuously compare the noise prediction results with the actual monitored noise data, calculate the prediction error, and trigger a model update warning when the prediction error exceeds a preset threshold. Periodically or based on the warning, incrementally train the physical guidance neural network using newly collected airport operation data and update the dynamic thrust coefficient distribution model.
[0010] Specifically, the construction of the dynamic thrust coefficient distribution model includes: calculating the actual thrust coefficient for historical flight events based on their flight trajectory data and corresponding noise monitoring data using an inverse solution method; classifying the actual thrust coefficients according to aircraft type and flight phase, wherein the flight phase includes at least takeoff, climb, approach, and landing; for each classification, fitting the marginal probability distribution of the thrust coefficients using a kernel density estimation method, and constructing a joint probability distribution model among different thrust coefficients based on a function to form a dynamic thrust coefficient library.
[0011] Specifically, the training using a physics-guided neural network includes: constructing a dual-path neural network, where one path is used to learn noise influencing factors not covered by the physical model; training the dual-path neural network using a training dataset containing features and actual measured noise values; introducing a physical consistency constraint term into the training loss function to form a physical consistency loss function, ensuring that the network prediction results conform to basic acoustic physics laws; and using cross-validation to select the network structure and hyperparameters of the dual-path neural network and saving the trained model parameters.
[0012] Specifically, in the online prediction stage, the extraction of prediction features based on real-time flight trajectory data includes: triggering the prediction process when the target aircraft is detected to enter a preset geographical area centered on the airport; dividing the real-time flight trajectory of the target aircraft into continuous three-dimensional trajectory segments; calculating key geometric and motion features, including position, altitude, speed, and rate of change of heading, for each three-dimensional trajectory segment; and extracting the location information of the prediction point and the current meteorological conditions as features.
[0013] Specifically, when updating the dynamic thrust coefficient distribution model, it is also corrected by combining real-time air traffic control instruction data. The air traffic control instructions include speed adjustment instructions, altitude restriction instructions, and route adjustment instructions. The probability distribution of the thrust coefficient is dynamically adjusted according to the instruction data to reflect the impact of control strategies on aircraft thrust usage.
[0014] Specifically, based on the noise prediction results, noise contour lines are rendered on the electronic map in real time, and visual and audible alarms are generated for areas where the predicted noise exceeds the specified threshold. At the same time, the prediction results are pushed to the airport dispatch system and the surrounding community noise monitoring platform.
[0015] Specifically, when constructing the dynamic thrust coefficient distribution model, time dimension modeling is also introduced to distinguish the thrust coefficient distribution under different seasons and time periods. Different thrust coefficient probability models are established for nighttime, daytime and peak periods to reflect the impact of time-related operating modes on noise.
[0016] The beneficial effects of the single-point noise prediction method for airports proposed in this application are as follows:
[0017] 1. Construct a dynamic thrust coefficient distribution model, reverse-engineer the actual thrust coefficient from historical data, classify and model according to aircraft type and flight stage, accurately characterize its dynamic changes and probability distribution, and integrate the advantages of physical laws and data-driven approaches through a physical-guided neural network. Physical constraints ensure the compliance of predictions and significantly improve the model extrapolation capability and prediction reliability in complex scenarios.
[0018] 2. Establish an online closed loop of prediction-monitoring-updating, automatically calculate errors and trigger early warnings, update the model based on new data increments, adapt to dynamic changes in airport environment and fleet, avoid model performance degradation, process flight trajectory and meteorological data in real time, dynamically correct thrust distribution in combination with air traffic control instructions, introduce time dimension modeling, accurately match different operating scenarios, and significantly improve the scenario awareness and timeliness of prediction. Attached Figure Description
[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0020] Figure 1 This is a flowchart of a single-point noise prediction method for airports according to this application;
[0021] Figure 2 This is a flowchart of the physical-guided neural network training process for a single-point noise prediction method for airports according to this application. Detailed Implementation
[0022] To make the technical means, inventive features, objectives, and effects of this application easier to understand, the application is further described below with reference to specific illustrations. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.
[0023] The present application will now be described in further detail with reference to the accompanying drawings.
[0024] Example: S1: Collect historical airport operation data, including flight trajectory data, meteorological record data and noise monitoring data. Based on the historical airport operation data, construct a dynamic thrust coefficient distribution model and train it using a physical guided neural network to obtain a trained noise prediction model.
[0025] This step involves building and training a noise prediction model that integrates physical laws and data-driven approaches, based on historical airport operation data. The specific process is as follows:
[0026] Collect historical operational data for the target airport over the past three years, ensuring data coverage across different seasons, time periods, and flight traffic scenarios. The data types and specific content are as follows:
[0027] Flight trajectory data: sourced from airport air traffic control automation system and aircraft ADS-B (Automatic Dependent Surveillance-Broadcast) equipment, including parameters such as aircraft type, flight number, real-time latitude and longitude, altitude, speed, heading angle, rate of climb / rate of descent, takeoff / landing timestamps, etc., with sampling frequency set to 1Hz, and data jumps and invalid missing data caused by equipment failure are removed;
[0028] Meteorological data: taken from the airport meteorological observation station and surrounding meteorological monitoring points, covering parameters such as wind speed, wind direction, atmospheric temperature, humidity, and air pressure. The temporal resolution is consistent with the flight trajectory data. A small number of missing values are supplemented by linear interpolation. At the same time, abnormal meteorological data (such as data during extreme rainstorms and strong typhoons) are marked and used separately for model validation.
[0029] Noise monitoring data: collected by noise monitoring terminals deployed at key sensitive points around the airport (such as residential areas, schools, and hospitals) and around the runway. The monitoring index is the equivalent continuous A-weighted sound level, the sampling frequency is set to 10Hz, and the location coordinates of the monitoring points and the corresponding flight event numbers are recorded simultaneously to ensure that the data can be correlated with flight trajectory data and meteorological data through time and space coordinates.
[0030] Construction of dynamic thrust coefficient distribution model
[0031] Thrust coefficient is a core parameter affecting the intensity of aircraft noise radiation. Its distribution is significantly influenced by aircraft type, flight phase, time, and air traffic control instructions. This embodiment constructs a dynamic thrust coefficient distribution model and forms a dynamic thrust coefficient library through the following steps:
[0032] Actual thrust coefficient calculation: For each historical flight event, based on its complete flight trajectory data and noise monitoring data of the corresponding time period and location, combined with the aircraft dynamics equations and acoustic radiation model, the actual thrust coefficient is calculated by inverse solution method.
[0033] Specifically, based on the aircraft force balance equation (lift = gravity + inertial force component, thrust = drag + inertial force component), the thrust value is initially calculated by substituting data such as speed, climb rate and descent rate in the trajectory parameters. Then, it is corrected by combining noise monitoring data and noise radiation model to obtain an accurate sequence of actual thrust coefficients.
[0034] Thrust coefficient classification: A two-dimensional classification based on aircraft type and flight phase. Aircraft type classification is based on aircraft tonnage and engine type, divided into four main categories: narrow-body passenger aircraft, wide-body passenger aircraft, cargo aircraft, and general aviation aircraft. Flight phase is divided into takeoff phase, climb phase, approach phase, and landing phase according to the aircraft's operating status, while also marking special transition states (such as takeoff roll and go-around).
[0035] Probability distribution fitting: For each type of aircraft-flight phase combination, the kernel density estimation (KDE) method is used to fit the marginal probability distribution of the thrust coefficient. The kernel function is selected as a Gaussian kernel, and the bandwidth is determined by cross-validation (within the range of 0.01-0.05) to accurately characterize the distribution characteristics of a single thrust coefficient. Based on the function, a joint probability distribution model is constructed between different thrust coefficients (such as instantaneous thrust coefficient, average thrust coefficient, and thrust change rate) to capture the correlation between various thrust parameters.
[0036] Time-dimensional modeling: A time-dimensional optimization model is introduced to distinguish the differences in thrust coefficient distribution in different seasons and time periods. The time periods are divided into daytime (6:00-22:00), nighttime (22:00-6:00 the next day), and peak periods (7:00-9:00, 17:00-19:00). Independent thrust coefficient probability models are established for each time zone to reflect the impact of nighttime noise reduction flight procedures and intensive operation during peak periods on thrust usage patterns. Finally, these models are integrated to form a dynamic thrust coefficient library covering the entire scenario.
[0037] A dual-path, physically guided neural network is constructed to balance physical constraints with data-driven generalization capabilities. The specific training process is as follows:
[0038] Network architecture design: The dual-path neural network includes a physical path and a data path. The physical path takes into account physically interpretable features such as dynamic thrust coefficient, flight speed, altitude, and meteorological parameters, and learns a noise generation mechanism based on acoustic physics through two fully connected layers. The data path takes into account factors that are difficult for the physical model to cover, such as heading change rate, trajectory curvature, and historical noise correlation features, and extracts latent features through one convolutional layer and two fully connected layers. The outputs of the two paths are fused through a fusion layer, and finally output a single-point noise prediction value through an output layer.
[0039] Training dataset construction: From the preprocessed historical data, the training set, validation set and test set are divided in a ratio of 8:1:1. Each sample contains predicted features (dynamic thrust coefficient, flight trajectory parameters, meteorological parameters, time features) and the actual measurement noise value of the corresponding monitoring point. The feature data are standardized (mean is zeroed and variance is normalized) to eliminate the influence of units.
[0040] Construction of the Physical Consistency Loss Function: Based on the traditional mean square error loss function, a physical consistency constraint term is introduced to form a composite loss function. The physical constraint term includes a noise attenuation constraint with distance (based on the spherical wave radiation law, noise level is inversely proportional to distance), a noise correction constraint with wind speed and direction (noise attenuation decreases with the windward direction and increases with the windward direction), and a positive correlation constraint between thrust coefficient and noise level. Each constraint term's contribution to the total loss is adjusted by a weighting coefficient (range 0.1-0.3) to ensure that the network prediction results conform to basic acoustic physics laws.
[0041] Hyperparameter selection and training: The network hyperparameters are selected using the 5-fold cross-validation method. During training, the validation set loss is monitored in real time. When the validation set loss does not decrease for 10 consecutive iterations, training is stopped, the optimal model parameters are saved, and the trained noise prediction model is obtained.
[0042] S2: Receive real-time flight trajectory data and real-time meteorological data, extract prediction features based on the real-time flight trajectory data, and input the prediction features and the real-time meteorological data into the trained noise prediction model to obtain noise prediction results.
[0043] This step, based on real-time data, uses a trained model to dynamically predict single-point noise. The specific process is as follows:
[0044] Prediction process triggering: Real-time reception of flight trajectory data pushed by the airport air traffic control system, monitoring of the target aircraft position through geofencing technology, and automatic triggering of the noise prediction process when the aircraft enters a preset geographical area with a radius of 20 kilometers centered on the airport runway center point, while simultaneously associating the corresponding flight's aircraft type information and flight plan.
[0045] Predictive Feature Extraction: The real-time flight trajectory of the target aircraft is divided into continuous three-dimensional trajectory segments according to the time series, with each segment lasting 10 seconds (corresponding to 10 trajectory sampling points), ensuring that the trajectory segments cover the key areas above the predicted points. For each three-dimensional trajectory segment, key geometric and motion features are calculated, including the latitude and longitude, altitude, instantaneous speed, average speed, heading change rate (the ratio of the difference in heading angle between adjacent trajectory points to the time interval), and trajectory curvature of the midpoint of the trajectory segment. At the same time, the coordinate information of the predicted point (the location of the noise monitoring terminal) and the current real-time meteorological data (wind speed, wind direction, temperature) are extracted. Combined with time features (season, time period), the corresponding thrust coefficient distribution features are matched from the dynamic thrust coefficient library and integrated to form a complete predictive feature vector.
[0046] Noise prediction calculation: The predicted feature vector and real-time meteorological data are input into the trained noise prediction model. The model outputs the noise prediction results for the corresponding prediction points based on the dual-path network, including the instantaneous noise level, the equivalent continuous A-weighted sound level, and the noise duration. The output frequency of the prediction results is consistent with the sampling frequency of the trajectory data, realizing real-time dynamic prediction of noise.
[0047] S3: Continuously compare the noise prediction results with the actual monitored noise data, calculate the prediction error, and trigger a model update warning when the prediction error exceeds a preset threshold. Periodically or based on the warning, incrementally train the physical guidance neural network using newly collected airport operation data and update the dynamic thrust coefficient distribution model.
[0048] The long-term prediction accuracy of the model is ensured through real-time error feedback and incremental training. The specific process is as follows:
[0049] Continuously collect actual noise data from each noise monitoring terminal, compare the noise prediction results at the same time and the same prediction point with the actual monitoring data, calculate the prediction error, and use the root mean square error (RMSE) and mean absolute error (MAE) as the core evaluation indicators. At the same time, record the frequency of errors exceeding the allowable range.
[0050] A preset error threshold is set. When the calculated prediction error exceeds the preset threshold, or when the error of 5 consecutive sampling points exceeds the allowable range, the system will automatically trigger a model update warning and record the scenario corresponding to the error anomaly (such as specific aircraft type, special weather, or the impact of air traffic control orders).
[0051] A combination of regular and on-demand updates is adopted. The regular update cycle is set to one month, while on-demand updates are triggered by early warning signals. During an update, airport operation data generated after the early warning is triggered and within the cycle are collected. After preprocessing the data, it is added to the training dataset, and an incremental training method is used to train the physical guidance neural network. The parameters of the bottom feature extraction layer of the network are frozen, and only the parameters of the fusion layer and the output layer are updated. At the same time, the dynamic thrust coefficient distribution model is corrected by combining real-time air traffic control instruction data.
[0052] Control commands include speed adjustment commands, altitude restriction commands, and flight path adjustment commands. Correction rules are established for different command types: for example, when a speed adjustment command causes the aircraft speed to change by ±10%, the peak value of the corresponding thrust coefficient probability distribution shifts by ±5%; when an altitude restriction command causes the aircraft to climb or descend prematurely, the thrust coefficient distribution range for the corresponding flight phase is adjusted to reflect the impact of control strategies on aircraft thrust usage; after training, the dynamic thrust coefficient library and noise prediction model parameters are updated to achieve adaptive optimization of the model.
[0053] Based on noise prediction results, the system enables visualization, alerts, and data push notifications, supporting noise control in airport scheduling and surrounding communities.
[0054] The noise prediction results are overlaid on the electronic map, and noise contour lines of different colors are rendered according to the noise level (below 50dB, 50-60dB, 60-70dB, 70-80dB, above 80dB). The contour line interval is set to 5dB, and it is updated in real time to reflect the dynamic changes of the noise field, allowing managers to intuitively grasp the noise impact range.
[0055] The system has preset noise thresholds (70dB during the day and 55dB at night, which can be adjusted according to the area's functions). When the predicted noise exceeds the corresponding threshold, the system automatically generates a visual alarm (marked as a red warning area on the map) and an audible alarm (terminal prompt tone). At the same time, it pushes the warning information, noise prediction data, and corresponding flight information to the airport dispatch system to provide a basis for dispatchers to adjust flight procedures (such as optimizing routes and adjusting takeoff times).
[0056] The noise prediction results (daily average, hourly average, and peak value) will be simultaneously pushed to the noise monitoring platform and environmental protection supervision platform of the surrounding communities, and made public through web and mobile terminals to ensure the public's right to know. At the same time, it will provide data support for environmental protection departments to carry out noise pollution control and airport operation optimization plans.
[0057] The present application and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present application. The actual structure is not limited to this. In conclusion, if a person skilled in the art is inspired by this description and designs a similar structure and embodiment without departing from the spirit of the present application, such design should fall within the protection scope of the present application.
Claims
1. A method for predicting single-point noise at an airport, characterized in that, Includes the following steps: S1: Collect historical airport operation data, including flight trajectory data, meteorological record data and noise monitoring data. Based on the historical airport operation data, construct a dynamic thrust coefficient distribution model and train it using a physical guided neural network to obtain a trained noise prediction model. S2: Receive real-time flight trajectory data and real-time meteorological data, extract prediction features based on the real-time flight trajectory data, and input the prediction features and the real-time meteorological data into the trained noise prediction model to obtain noise prediction results; S3: Continuously compare the noise prediction results with the actual monitored noise data, calculate the prediction error, and trigger a model update warning when the prediction error exceeds a preset threshold. Periodically or based on the warning, incrementally train the physical guidance neural network using newly collected airport operation data and update the dynamic thrust coefficient distribution model.
2. The airport single-point noise prediction method according to claim 1, characterized in that, The construction of the dynamic thrust coefficient distribution model includes: calculating the actual thrust coefficient for historical flight events based on their flight trajectory data and corresponding noise monitoring data using an inverse solution method; classifying the actual thrust coefficients according to aircraft type and flight phase, wherein the flight phase includes at least takeoff, climb, approach, and landing; for each classification, fitting the marginal probability distribution of the thrust coefficient using a kernel density estimation method, and constructing a joint probability distribution model among different thrust coefficients based on a function to form a dynamic thrust coefficient library.
3. The airport single-point noise prediction method according to claim 1, characterized in that, The training of the physical-guided neural network includes: constructing a dual-path neural network, where one path is used to learn noise influencing factors not covered by the physical model; training the dual-path neural network using a training dataset containing features and actual measured noise values; introducing a physical consistency constraint term into the training loss function to form a physical consistency loss function, so as to ensure that the network prediction results conform to the basic acoustic physics laws; and using a cross-validation method to select the network structure and hyperparameters of the dual-path neural network and saving the trained model parameters.
4. The airport single-point noise prediction method according to claim 1, characterized in that, In the online prediction phase, the extraction of prediction features based on real-time flight trajectory data includes: triggering the prediction process when the target aircraft is detected to enter a preset geographical area centered on the airport; dividing the real-time flight trajectory of the target aircraft into continuous three-dimensional trajectory segments; calculating key geometric and motion features, including position, altitude, speed, and heading change rate, for each three-dimensional trajectory segment; and extracting the location information of the prediction point and the current meteorological conditions as features.
5. The airport single-point noise prediction method according to claim 1, characterized in that, When updating the dynamic thrust coefficient distribution model, it is also corrected by combining real-time air traffic control instruction data, including speed adjustment instructions, altitude restriction instructions, and route adjustment instructions. The probability distribution of the thrust coefficient is dynamically adjusted according to the instruction data to reflect the impact of control strategies on aircraft thrust usage.
6. The airport single-point noise prediction method according to claim 1, characterized in that, Based on the noise prediction results, noise contour lines are rendered on the electronic map in real time, and visual and audible alarms are generated for areas where the predicted noise exceeds the specified threshold. At the same time, the prediction results are pushed to the airport dispatch system and the surrounding community noise monitoring platform.
7. The airport single-point noise prediction method according to claim 2, characterized in that, When constructing the dynamic thrust coefficient distribution model, time dimension modeling is also introduced to distinguish the thrust coefficient distribution under different seasons and time periods. Different thrust coefficient probability models are established for nighttime, daytime and peak periods to reflect the impact of time-related operating modes on noise.