CN115752708A - Airport single-point noise prediction method based on deep time convolution network - Google Patents
Airport single-point noise prediction method based on deep time convolution network Download PDFInfo
- Publication number
- CN115752708A CN115752708A CN202211344106.4A CN202211344106A CN115752708A CN 115752708 A CN115752708 A CN 115752708A CN 202211344106 A CN202211344106 A CN 202211344106A CN 115752708 A CN115752708 A CN 115752708A
- Authority
- CN
- China
- Prior art keywords
- noise
- data
- aircraft
- time
- airport
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an airport single-point noise prediction method based on a depth time convolution network, which comprises the steps of obtaining aircraft trajectory data, meteorological data and noise monitoring data around an airport; preprocessing the acquired data; matching the aircraft trajectory data set, the meteorological data set and the noise monitoring data set according to the recording time attribute, and integrating and associating the three data sets into one data set; noise data screening is carried out on the merged data set, aircraft noise is separated from background noise, and noise generated by aircraft operation is screened out; constructing a time convolution network noise prediction model comprising an encoder module and a decoder module; and training the constructed model to obtain the optimal model parameters, thereby realizing the prediction of the airport single-point noise. The method and the device correlate and match the historical flight trajectory with aircraft performance data, meteorological data, noise monitoring data and the like, and realize aviation noise time series prediction of the ground position.
Description
Technical Field
The invention belongs to the technical field of civil aviation, and particularly relates to an airport single-point noise prediction method based on a deep time convolution network.
Background
In the last decade, the civil aviation business of China is rapidly growing in all aspects, and the increase of air traffic in the future is expected, and meanwhile, the population density around airports and along air traffic routes is continuously increased, so that great challenges are brought to the control of noise around airports. The noise hazard is considered second only to air pollution as a second environmental risk factor. There are studies that have shown that airborne noise is associated with increased mortality from hypertension and cardiovascular disease, in addition to being directly associated with worsening sleep quality. In addition, there is ample evidence that children's cognitive abilities may be greatly affected and reading and memory abilities may be degraded under aviation noise exposure. In order to realize green informatization and healthy development of the civil aviation industry, an advanced technical means is urgently needed to have further scientific prediction on airport noise problems, and scientific data support is provided for relevant departments of civil aviation.
The noise prediction and monitoring of the aircraft are key technologies in the new generation of air intelligent traffic, and are the premise and the basis for realizing the green and healthy development of the civil aviation industry in the future. The accurate prediction and monitoring of the airport noise is the core basis of the airport noise control and management, because only the detailed space-time distribution situation of the noise around the airport is accurately mastered, effective measures can be taken to prevent and treat the noise pollution of the airplane, implement decision management of the noise environment and develop and utilize land resources in the peripheral area of the airport in a planned way. The accuracy of the prediction of aircraft noise will directly affect the formulation of airport noise control measures. The current airport noise model aims at evaluating the integrated noise of the whole airport fleet and is used for researching the integrated noise influence of the flight procedure. The current achievement mainly takes the comprehensive influence value of the aviation noise as a main result, and the researches on the time sequence prediction of the airport noise and the space-time evolution of the noise are not deep enough. The invention provides a prediction method based on big data and machine learning, which is different from other airport noise models calculated by using a basic physical equation and experience.
Disclosure of Invention
The purpose of the invention is as follows: the invention provides an airport single-point noise prediction method based on a depth time convolution network, which is used for carrying out correlation matching on historical flight trajectories and aircraft performance data, meteorological data, noise monitoring data and the like so as to realize aviation noise time series prediction of ground positions.
The technical scheme is as follows: the invention provides an airport single-point noise prediction method based on a deep time convolution network, which specifically comprises the following steps:
step 1: acquiring data, including aircraft trajectory data, meteorological data and noise monitoring data around an airport;
and 2, step: preprocessing the data acquired in the step (1);
and 3, step 3: matching the aircraft trajectory data set, the meteorological data set and the noise monitoring data set according to the recording time attribute, and integrating and associating the three data sets into one data set;
and 4, step 4: carrying out noise data screening on the data set merged in the step 3, separating the aircraft noise from the background noise, and screening out the noise generated by the operation of the aircraft;
and 5: constructing a time convolution network noise prediction model comprising an encoder module and a decoder module;
step 6: and (5) training the noise prediction model constructed in the step (5) to obtain the optimal model parameters, so as to realize the prediction of the airport single-point noise.
Further, the aircraft trajectory data in the step (1) comprises recording time, longitude of the aircraft, latitude of the aircraft, altitude of the aircraft, speed of the aircraft, model of the aircraft, heading of the aircraft, and type of the approach and departure of the aircraft; the meteorological data comprise recording time, local atmospheric pressure, temperature, relative humidity, wind speed and wind direction of an airport; the noise monitoring data of the airport periphery comprises recording time and noise values.
Further, the step (2) is realized as follows:
step 2.1: and (3) carrying out data cleaning on the air device track data, the meteorological data and the noise monitoring data around the airport: if the repeated values exist in the data set, only the first repeated track point is reserved, and redundant repeated track points are deleted; deleting or interpolating information data missing in the data set; replacing the abnormal data values by adopting a regularization algorithm to smooth the track data;
step 2.2: and (3) carrying out coordinate transformation on the aircraft trajectory data: the longitude and latitude coordinates of the aircraft are converted into a new coordinate system with an airport as a center, the longitude and latitude coordinates in a track data set are converted into a geocentric geocoded rectangular coordinate system, and the calculation processes of geocentric geocoded coordinate values X, Y and Z are as follows:
in the formula: alpha represents the semimajor axis radius of the earth, e represents the eccentricity of the earth, r represents the radius of curvature of the earth,
radian corresponding to latitude is represented, lambda represents radian corresponding to longitude, and h represents aircraft height;
secondly, the earth-center-earth-fixed rectangular coordinate system is converted into a northeast rectangular coordinate system with the airport reference point as the coordinate origin, and the coordinate values xaast, ynorth, zup under the northeast rectangular coordinate system are calculated as follows:
Xeast＝-(X-x o )sinλ o -(Y-y o )cosλ o (5)
in the formula: (x) 0 ,y 0 ,z 0 ) Representing airport reference point coordinates under a geocentric-geostationary rectangular coordinate system,
denotes the radian, lambda, corresponding to the latitude of the origin 0 Denotes the radian, h, corresponding to the longitude of the origin 0 Representing the height of the origin;
step 2.3: and (3) performing derivative attribute construction on the aircraft trajectory data set:
constructing derivative attributes of the relative position of the aircraft relative to the ground monitoring point, wherein the derivative attributes comprise the relative distance S from the aircraft to the monitoring point, the lateral distance L and the elevation angle beta from the monitoring point to the aircraft position; the smaller the relative position of the aircraft and the ground monitoring point is, the greater the noise influence is; the smaller the lateral distance, the smaller the attenuation of the energy of the noise in the air propagation;
constructing track attributes: finding out ground projection points of track points with the height being not 0 and closest to the airport, and calculating the angle between the connection line of the projection points and the center point of each runway and the straight line where the runway is located; in all angles, the runway corresponding to the minimum angle is the runway matched with the flight event;
step 2.4: data characteristic standardization treatment: the minimum and maximum value standardization processing is carried out on the numerical attributes in the three cleaned data sets, and the data are mapped into the range of 0-1.
Further, the step (3) is realized as follows:
step 3.1: data set time scale processing: the aircraft trajectory data is defined according to the current position of the aircraft and the information received last every second in ADS-B information transmitted by the aircraft; the sampling frequency of the meteorological data and the noise monitoring data is fixed, and is respectively once per minute and once per second;
step 3.2: data merging: and merging the time attributes of the flight path data set and the noise monitoring data, merging the meteorological data according to the number of minutes of the recording time of the two merged data sets, and finally merging the data sets to obtain a data set which comprises a flight path data set per second, a noise monitoring data set per second and the meteorological data of the current minute time.
Further, the step (4) is realized as follows:
step 4.1: background noise filtering, namely setting a background noise threshold, and filtering all data lower than the background noise threshold to obtain new noise data under a time sequence;
step 4.2: time series fragment length filtering: setting a time length threshold, checking all new time sequence segments, and removing all time sequence segments with the time length smaller than the time length threshold;
step 4.3: ADS-B based aircraft noise event identification: checking the time sequence segment filtered in the step 4.2, and detecting whether the aircraft within 3km away from the monitoring point exists in the current time step; if not, the current time step is the non-aviation noise event, the time step is removed, and the non-aviation noise event is separated from the data set.
Further, the encoder module in step (5) adopts an inflation causal convolution, each time block in the encoder is composed of two inflation causal convolutions and residual errors, and the inflation coefficient d exponential level of the inflation causal convolution in each residual error block increases by d =2 n The coding structure with a plurality of time blocks stacked can quickly complete the coding of long-time sequence data.
Further, the decoder module in step (5) includes a variant of the residual neural network and a full-connection layer, and the prediction is made on the aircraft noise value at the future time by constructing a covariate of the residual grid variant module capturing the historical information and the prediction time, and the calculation formula is as follows:
δ t+k ＝R(X t+k )+h t (8)
in the formula: delta t+k Representing a residual neural network output; r (-) is a residual calculation function formed by a residual network variant structure; x t+k A factor variable representing noise at time t + k; h is t Representing the output of the encoder.
Further, the step (6) is realized as follows:
step 6.1: data set partitioning: dividing the data processed in the step 4 into a training set, a testing set and a verification set for training a noise prediction model;
step 6.2: initializing and setting parameters of a noise prediction model: setting the structural parameters and the internal parameters of the constructed model, wherein the structural parameters and the internal parameters comprise the length of a coding and decoding sequence, the number of hidden layers, the size of a convolution kernel, the number of training samples in each batch, the maximum iteration times, the initial learning rate, weight parameters and the like; selecting a parameter value with the best performance of a noise prediction model;
step 6.3: updating the parameters of the noise prediction model, and performing iterative updating on the established model parameters by using a training set: selecting an L-1 loss function to carry out model training, randomly selecting a certain number of samples from a training set in each training process, segmenting the samples into a coding and decoding sequence according to the sequence length, taking the coding sequence as the input of the model, obtaining a noise prediction output value at each time step through a coding and decoding stage, and then calculating the loss function by combining a real value corresponding to the time step until the loss function reaches a set threshold value, or adjusting a weight and repeatedly training;
step 6.4: testing a noise prediction model: testing the trained model by using a test set, testing the error of the predicted value of the noise prediction model, and finishing the model training if the error is in accordance with the expectation; otherwise, the training process is repeated until the noise prediction model test reaches a desired level.
Has the advantages that: compared with the prior art, the invention has the beneficial effects that:
1. the method comprises the steps of building an aviation noise prediction model based on big data and a machine learning technology, rapidly processing an ultra-large-scale data set by using deep time convolution, rapidly executing the model in a short time, providing accurate noise prediction according to time granularity per second, and evaluating thousands of flight trajectories in a few seconds when a new flight program is designed;
2. the method can predict the transient noise generated by the aircraft with high precision: the current airport noise model aims at evaluating the comprehensive noise of the whole airport fleet, and researches the comprehensive noise influence of a single aircraft and the noise influence of the comprehensive flight path of the whole fleet, namely the current model researches a single noise event and an accumulated noise event and cannot accurately predict the instantaneous noise of the aircraft; the noise model established by the invention is a noise time series research composed of instantaneous noise generated by each aircraft, and can predict the instantaneous noise of the aircraft with high precision;
3. the model can be established without setting a complex simulation environment: the model established by the invention can output the noise influence generated by the aircraft only by inputting a small number of parameters, and can be used as a daily noise prediction and monitoring tool for an airport after the model is established;
4. the system has a data driving characteristic, can create a generation model for any airport from a historical data generation model, can acquire noise related statistical data of the area around the airport, provides noise environment analysis of the airport for related management departments so as to contribute to land planning around the airport, airport noise legislation, noise reduction flight program development and the like, and provides better data support for airport noise management;
5. the airport surrounding noise monitoring with low cost and large range can be realized.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic representation of the derivative attributes of the relative position of the aircraft with respect to the ground monitoring points;
fig. 3 is a diagram of a noise prediction model architecture based on a deep time convolution network.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
The invention provides an airport single-point noise prediction method based on a deep time convolution network, which specifically comprises the following steps as shown in figure 1:
step 1: preparing data: and reading aircraft track data, meteorological data and noise monitoring data. The data attributes of the aircraft trajectory data comprise recording time, longitude of the aircraft, latitude of the aircraft, altitude of the aircraft, speed of the aircraft, model of the aircraft, heading of the aircraft and type of the approach and departure of the aircraft; the meteorological data attributes comprise recording time, local atmospheric pressure, temperature, relative humidity, wind speed and wind direction of an airport; the noise monitoring data around the airport is data for monitoring and measuring the noise of the aircraft according to the requirements of 'airport surrounding aircraft noise environmental standard', and the data attribute comprises recording time and a noise value.
Step 2: data preprocessing, in order to make noise prediction better at a later stage, must ensure high quality of data. Therefore, the problems of data missing, data duplication, data abnormality and the like which may exist need to be solved through data preprocessing, and original data of the data is converted into a form required by subsequent research. The method comprises the following specific steps:
step 2.1: and carrying out data cleaning on the aircraft trajectory data, the meteorological data and the noise monitoring data. If the repeated values exist in the data set, only the first repeated track point is reserved, and redundant repeated track points are deleted; if the missing information data in the data set exceeds 85%, deleting the flight path data; if the data is less than 85% missing, interpolating the missing value according to other data; and replacing the abnormal data values by adopting a regularization algorithm, and enabling the track data to be smoother.
Step 2.2: and carrying out coordinate transformation on the aircraft trajectory information data set. In order to calculate the position of the aircraft conveniently, the longitude and latitude coordinates of the aircraft need to be converted into a new coordinate system taking the airport as the center.
Firstly, longitude and latitude coordinates in a track data set are converted into a geocentric geocoded rectangular coordinate system, and the calculation processes of geocentric geocoded coordinate values X, Y and Z are as follows:
in the formula: alpha represents the semimajor axis radius of the earth, e represents the eccentricity of the earth, r represents the radius of curvature of the earth,
denotes radians for latitude, λ denotes radians for longitude, and h denotes aircraft altitude.
And converting the earth center earth fixed rectangular coordinate system into a northeast rectangular coordinate system with the airport reference point as the origin of coordinates. The X-axis direction of the northeast rectangular coordinate system is the true east direction, the Y-axis direction is the true north direction, and the Z-axis direction is the direction perpendicular to the earth surface where the airport reference point is located. The coordinate values Xeast, ynorth, zup in the northeast coordinate system are calculated as follows:
Xeast＝-(X-x o )sinλ o -(Y-y o )cosλ o (5)
in the formula: (x) 0 ,y 0 ,z 0 ) Representing airport reference point coordinates under a geocentric-geostationary rectangular coordinate system,
denotes the radian, lambda, corresponding to the latitude of the origin 0 Denotes the radian, h, corresponding to the longitude of the origin 0 Representing the height of the origin.
Step 2.3: and performing derivative attribute construction on the aircraft trajectory data set.
As shown in fig. 2, derivative attributes of the relative position of the aircraft with respect to the ground monitoring points are constructed, including the relative distance S of the aircraft to the monitoring points, the lateral distance L, and the elevation angle β of the monitoring points to the aircraft position. In the noise model, the aircraft position attribute is an important factor that affects the magnitude of the noise. In the basic physical equation and the empirical calculation model, the smaller the relative position of the aircraft and the ground monitoring point is, the larger the noise influence is; the smaller the lateral distance, the less the energy of the noise is attenuated in the air propagation.
Constructing track attributes: firstly, finding a ground projection point of a track point which is closest to an airport and has a height different from 0, and then calculating an angle between a connecting line of the projection point and the central point of each runway and a straight line where the runway is located; in all angles, the runway corresponding to the smallest angle is the runway matched with the flight event.
Step 2.4: data characterization and (6) carrying out standardization processing. And carrying out minimum-maximum (Min-max) standardization on the cleaned data set, and mapping the data into a range of 0-1, thereby eliminating errors caused by different dimensions and enabling different indexes to have comparability.
And step 3: and matching the data sets. And matching the aircraft trajectory data set, the meteorological data set and the noise monitoring data set according to the recording time attribute, and integrating and associating the three data sets into one data set for research. The method comprises the following specific steps:
step 3.1: and (5) processing the time scale of the data set. Due to the inconsistent sampling frequency of the recording time of the three data sets, the data sets need to be time-scaled. The data source of the aircraft trajectory data set is broadcast Automatic Dependent Surveillance-broadcast (ADS-B) data, which has a very high and non-fixed sampling frequency of the aircraft during flight. In order to change its sampling frequency to each second, the current position of the aircraft is defined according to the last received information per second of the ADS-B information transmitted by it. The sampling frequencies of the meteorological data and the noise monitoring data are fixed, namely once per minute and once per second respectively, and do not need to be processed.
Step 3.2: and (6) merging the data. The method comprises the steps of firstly merging the time attributes of a flight path data set and a noise data set, and then merging meteorological data according to the minutes of the recording time of the two merged data sets, namely, a finally-synthesized data set comprises a flight path data set per second, a noise monitoring data set per second and a meteorological data set at the current minute time.
And 4, step 4: aircraft noise event identification. The airborne noise monitoring equipment located near the airport may be disturbed by the environmental background noise, and therefore noise data screening is required on the data set combined in step 3 to separate the aircraft noise from the background noise and screen out the noise generated by the aircraft operation.
Step 4.1: and filtering background noise. Setting a background noise threshold (the background noise threshold is selected to be related to the environmental noise near the ground monitoring point), and filtering all data lower than the background noise threshold to obtain the noise data in a new time series.
And 4.2: time series fragment length filtering. Setting a time length threshold (the selection of the time length threshold is related to the duration of a single noise event received by the ground monitoring point), checking all new time sequence segments, and rejecting all time sequence segments with the time length smaller than the time length threshold.
Step 4.3: ADS-B based aircraft noise event identification. And (4) checking the time sequence segment filtered in the step 4.2, and detecting whether the aircraft within 3km away from the monitoring point exists in the current time step. If not, the current time step is a non-aviation noise event, and the time step is eliminated. And finally, separating the non-aviation noise events from the data set.
And 5: constructing a time-convolutional network noise prediction model comprising an encoder block and a decoder block, as shown in fig. 3, wherein the encoder uses a dilation causal convolution, there is a time delay phenomenon in the noise modeling problem, and it is necessary to consider a priori data heavily. The noise generated by the aircraft takes time to propagate to the noise monitoring station, and therefore, the noise events generated by the previous time series in the aircraft trajectory are taken into account when predicting the noise at the next time step. Each time block in the encoder is composed of two dilation causal volumes and a residual. The expansion coefficient d of the dilation causal convolution in each residual block grows exponentially d =2 n The coding structure with a plurality of time blocks stacked can quickly complete the coding of long-time sequence data.
The decoder module comprises a variant of a residual error neural network and a full connection layer, and predicts the aircraft noise value at the future time by constructing a covariate of the residual error grid variant module capturing historical information and the prediction time, wherein the calculation formula is as follows:
δ t+k ＝R(X t+k )+h t (8)
in the formula: delta t+k Representing a residual neural network output; r (-) is a residual calculation function formed by a residual network variant structure; x t+k A factor variable representing noise at time t + k; h is t Representing the output of the encoder.
The efficient residual error expansion causal convolution operation unit in the deep time convolution network also fully fits the time delay characteristic of the aircraft noise, so that the future information is prevented from being leaked, and the prediction accuracy is improved.
Step 6: and training a noise prediction model, obtaining optimal model parameters and realizing the prediction of the airport single-point noise.
Step 6.1: and (4) dividing the data set. Using the data set processed in the step 4 for training a data set of a noise model; dividing the processed data set into three different data sets, selecting 60% of the data sets as a training set, 20% of the data sets as a testing set, and finally 20% of the data sets as a verification set to carry out model evaluation.
Step 6.2: and initializing and setting model parameters. And setting the structural parameters and the internal parameters of the constructed model, wherein the structural parameters and the internal parameters comprise the length of a coding and decoding sequence, the number of hidden layers, the size of a convolution kernel, the number of training samples in each batch, the maximum iteration number, the initial learning rate, weight parameters and the like. And (4) setting value ranges for each parameter to carry out experimental comparison, and selecting the parameter value with the best model performance.
Step 6.3: and updating the model parameters. And carrying out iterative updating on the established model parameters by using the training set. In order to reduce the sensitivity of non-aviation environmental noise in the data set, an L-1 loss function is selected for model training. During each epoch training procedure, a certain number of samples are randomly selected from the training set, and are split into coding and decoding sequences according to the sequence length. And taking the coding sequence as the input of the model, obtaining a noise prediction output value at each time step through a coding and decoding stage, and then calculating a loss function by combining a real value corresponding to the time step until the loss function reaches a set threshold, otherwise, adjusting the weight and repeating the training.
Step 6.4: and (4) testing the model. And testing the trained model by using the divided test set, testing the error of the predicted value of the model, and finishing the model training if the error is in accordance with the expectation. Otherwise, the training process is repeated until the model test reaches the desired expected level.
The method takes a fertile new bridge airport as a research object, and selects 5 noise monitoring points in the area around the 15/33 runway of the new bridge airport to carry out noise prediction research. The airport single-point noise prediction method based on the depth time convolution network established by the patent can accurately predict the noise of an aircraft with the average absolute error of 1.7dB above 55dB in airport measured data, and can realize the noise monitoring of the aircraft near a noise monitoring point.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Furthermore, it should be understood that although the specification has been described in terms of embodiments, not every embodiment includes every single embodiment, and such description is for clarity purposes only, and it will be appreciated by those skilled in the art that the specification as a whole can be combined as appropriate to form other embodiments as will be apparent to those skilled in the art.
Claims (8)
1. An airport single-point noise prediction method based on a deep time convolution network is characterized by comprising the following steps:
step 1: preparing data, including aircraft trajectory data, meteorological data and noise monitoring data around an airport;
step 2: preprocessing the data acquired in the step (1);
and 3, step 3: matching the aircraft trajectory data set, the meteorological data set and the noise monitoring data set according to the recording time attribute, and integrating and associating the three data sets into one data set;
and 4, step 4: carrying out noise data screening on the data set merged in the step 3, separating the aircraft noise from the background noise, and screening out the noise generated by the operation of the aircraft;
and 5: constructing a time convolution network noise prediction model comprising an encoder module and a decoder module;
step 6: and (5) training the noise prediction model constructed in the step (5) to obtain the optimal model parameters, so as to realize the prediction of the airport single-point noise.
2. The method for predicting airport single-point noise based on the deep time convolution network of claim 1, wherein the aircraft trajectory data in the step (1) comprises recording time, longitude of the aircraft, latitude of the aircraft, altitude of the aircraft, speed of the aircraft, model of the aircraft, heading of the aircraft, type of the aircraft entering and leaving a field; the meteorological data comprise recording time, local atmospheric pressure, temperature, relative humidity, wind speed and wind direction of an airport; the noise monitoring data of the airport periphery comprises recording time and noise values.
3. The method for airport single-point noise prediction based on the deep time convolution network as claimed in claim 1, wherein the step (2) is implemented as follows:
step 2.1: and (3) carrying out data cleaning on the air vehicle track data, meteorological data and noise monitoring data around the airport: if the repeated values exist in the data set, only the first repeated track point is reserved, and redundant repeated track points are deleted; deleting or interpolating information data missing in the data set; replacing the abnormal data values by adopting a regularization algorithm to smooth the track data;
step 2.2: and (3) carrying out coordinate transformation on the aircraft trajectory data: the longitude and latitude coordinates of the aircraft are converted into a new coordinate system with an airport as a center, the longitude and latitude coordinates in a track data set are converted into a geocentric geocoded rectangular coordinate system, and the calculation processes of geocentric geocoded coordinate values X, Y and Z are as follows:
in the formula: alpha represents the semimajor axis radius of the earth, e represents the eccentricity of the earth, r represents the radius of curvature of the earth,
radian corresponding to latitude, lambda represents radian corresponding to longitude, and h represents aircraft height;
secondly, the earth-center-earth-fixed rectangular coordinate system is converted into a northeast rectangular coordinate system with the airport reference point as the coordinate origin, and the coordinate values xaast, ynorth, zup under the northeast rectangular coordinate system are calculated as follows:
Xeast＝-(X-x o )sinλ o -(Y-y o )cosλ o (5)
in the formula: (x) 0 ,y 0 ,z 0 ) Representing airport reference point coordinates under a geocentric-geostationary rectangular coordinate system,
radian, λ, corresponding to latitude of origin 0 Denotes the radian, h, corresponding to the longitude of the origin 0 Representing the height of the origin;
step 2.3: and (3) performing derivative attribute construction on the aircraft trajectory data set:
constructing derivative attributes of the relative position of the aircraft relative to the ground monitoring point, wherein the derivative attributes comprise the relative distance S from the aircraft to the monitoring point, the lateral distance L and the elevation angle beta from the monitoring point to the aircraft position; the smaller the relative position of the aircraft and the ground monitoring point is, the greater the noise influence is; the smaller the lateral distance, the smaller the attenuation of the energy of the noise in the air propagation;
constructing the track attribute: finding out ground projection points of track points with the height being not 0 and closest to the airport, and calculating the angle between the connection line of the projection points and the center point of each runway and the straight line where the runway is located; in all angles, the runway corresponding to the minimum angle is the runway matched with the flight event;
step 2.4: data characteristic standardization treatment: the minimum and maximum value standardization processing is carried out on the numerical attributes in the three cleaned data sets, and the data are mapped into the range of 0-1.
4. The method for airport single-point noise prediction based on the deep time convolution network as claimed in claim 1, wherein the step (3) is implemented as follows:
step 3.1: data set time scale processing: the aircraft trajectory data is defined according to the current position of the aircraft and the information received last every second in ADS-B information transmitted by the aircraft; the sampling frequency of the meteorological data and the noise monitoring data is fixed, and is respectively once per minute and once per second;
step 3.2: data merging: and merging the time attributes of the flight path data set and the noise monitoring data, merging the meteorological data according to the minutes of the recording time of the two merged data sets, and finally merging the data sets, wherein the finally synthesized data set comprises the flight path data set per second, the noise monitoring data set per second and the meteorological data of the current minute time.
5. The airport single-point noise prediction method based on the deep time convolution network as claimed in claim 1, wherein the step (4) is implemented as follows:
step 4.1: background noise filtering, namely setting a background noise threshold, and filtering all data lower than the background noise threshold to obtain new noise data under a time sequence;
step 4.2: time series fragment length filtering: setting a time length threshold, checking all new time sequence segments, and removing all time sequence segments with the time length smaller than the time length threshold;
step 4.3: ADS-B based aircraft noise event identification: checking the time sequence segment filtered in the step 4.2, and detecting whether the aircraft within 3km away from the monitoring point exists in the current time step; if not, the current time step is the non-aviation noise event, the time step is removed, and the non-aviation noise event is separated from the data set.
6. The airport single-point noise prediction method based on the deep time convolution network as claimed in claim 1, wherein the encoder module in step (5) adopts dilation causal convolution, each time block in the encoder is composed of two dilation causal convolutions and residuals, and the dilation coefficient d of the dilation causal convolution in each residual block grows exponentially d =2 n The coding structure with a plurality of time blocks stacked can quickly complete the coding of long-time sequence data.
7. The airport single-point noise prediction method based on the deep time convolutional network as claimed in claim 1, wherein the decoder module in step (5) comprises a variant of a residual neural network and a fully connected layer, and the prediction of the aircraft noise value at the future time is made by constructing a residual grid variant module to capture historical information and covariates at the prediction time, and the calculation formula is as follows:
δ t+k ＝R(X t+k )+h t (8)
in the formula: delta t+k Representing a residual neural network output; r (-) is a residual calculation function formed by a residual network variant structure; x t+k A variable representing an influence factor of noise at time t + k; h is t Representing the output of the encoder.
8. The airport single-point noise prediction method based on the deep time convolution network as claimed in claim 1, wherein the step (6) is implemented as follows:
step 6.1: data set partitioning: dividing the data processed in the step 4 into a training set, a testing set and a verification set for training a noise prediction model;
step 6.2: initializing and setting parameters of a noise prediction model: setting the structural parameters and the internal parameters of the constructed model, wherein the structural parameters and the internal parameters comprise the length of a coding and decoding sequence, the number of hidden layers, the size of a convolution kernel, the number of training samples in each batch, the maximum iteration times, the initial learning rate, weight parameters and the like; selecting a parameter value with the best performance of a noise prediction model;
step 6.3: updating the parameters of the noise prediction model, and performing iterative updating on the established model parameters by using a training set: selecting an L-1 loss function to carry out model training, randomly selecting a certain number of samples from a training set in each training process, segmenting the samples into a coding and decoding sequence according to the sequence length, taking the coding sequence as the input of the model, obtaining a noise prediction output value at each time step through a coding and decoding stage, and then calculating the loss function by combining a real value corresponding to the time step until the loss function reaches a set threshold value, or adjusting a weight and repeatedly training;
step 6.4: testing a noise prediction model: testing the trained model by using a test set, testing the error of the predicted value of the noise prediction model, and finishing the model training if the error is in accordance with the expectation; otherwise, the training process is repeated until the noise prediction model test reaches a desired level.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211344106.4A CN115752708A (en) | 2022-10-31 | 2022-10-31 | Airport single-point noise prediction method based on deep time convolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211344106.4A CN115752708A (en) | 2022-10-31 | 2022-10-31 | Airport single-point noise prediction method based on deep time convolution network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115752708A true CN115752708A (en) | 2023-03-07 |
Family
ID=85354438
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211344106.4A Pending CN115752708A (en) | 2022-10-31 | 2022-10-31 | Airport single-point noise prediction method based on deep time convolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115752708A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116086599A (en) * | 2023-04-10 | 2023-05-09 | 中国飞行试验研究院 | Method, device, equipment and storage medium for acquiring low-altitude radiation sound field of aircraft |
| CN117076839A (en) * | 2023-10-17 | 2023-11-17 | 中国民用航空总局第二研究所 | Airport aircraft track dynamic prediction method based on dual incremental neural network |
-
2022
- 2022-10-31 CN CN202211344106.4A patent/CN115752708A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116086599A (en) * | 2023-04-10 | 2023-05-09 | 中国飞行试验研究院 | Method, device, equipment and storage medium for acquiring low-altitude radiation sound field of aircraft |
| CN116086599B (en) * | 2023-04-10 | 2023-10-20 | 中国飞行试验研究院 | Method, device, equipment and storage medium for acquiring low-altitude radiation sound field of aircraft |
| CN117076839A (en) * | 2023-10-17 | 2023-11-17 | 中国民用航空总局第二研究所 | Airport aircraft track dynamic prediction method based on dual incremental neural network |
| CN117076839B (en) * | 2023-10-17 | 2023-12-26 | 中国民用航空总局第二研究所 | Airport aircraft track dynamic prediction method based on dual incremental neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110264709B (en) | Method for predicting traffic flow of road based on graph convolution network | |
| CN115752708A (en) | Airport single-point noise prediction method based on deep time convolution network | |
| CN112819207B (en) | Geological disaster space prediction method, system and storage medium based on similarity measurement | |
| CN108805350B (en) | Search and rescue range prediction method based on multi-dimensional Monte Carlo theory | |
| CN111680870B (en) | Comprehensive evaluation method for quality of target motion trail | |
| CN110826788A (en) | Airport scene variable slide-out time prediction method based on big data deep learning | |
| WO2021082394A1 (en) | Layout-variable taxiing-out time prediction system based on big data deep learning | |
| CN115564114B (en) | Airspace carbon emission short-term prediction method and system based on graph neural network | |
| CN110570693A (en) | Flight operation time prediction method based on reliability | |
| CN109062245B (en) | Reliability intelligent distribution method for unmanned aerial vehicle ground station system software | |
| CN114842681A (en) | Airport scene flight path prediction method based on multi-head attention mechanism | |
| CN111179592A (en) | Urban traffic prediction method and system based on spatio-temporal data flow fusion analysis | |
| CN116681153A (en) | TEC forecasting method based on LSTM neural network and GNSS historical observation data | |
| CN113284369B (en) | Prediction method for actually measured airway data based on ADS-B | |
| CN116384565A (en) | Hierarchical atmospheric ozone concentration prediction method based on missing data filling | |
| CN116910919B (en) | Filling method and device under Gao Queshi rate of aircraft track | |
| CN116543603B (en) | Flight path completion prediction method and device considering airspace situation and local optimization | |
| CN116976227A (en) | Storm water increasing forecasting method and system based on LSTM machine learning | |
| CN115438453B (en) | Method for constructing road network facility topological structure by using observation data | |
| CN115907079B (en) | Airspace traffic flow prediction method based on attention space-time diagram convolutional network | |
| CN116796805A (en) | PM2.5 concentration prediction method based on Gaussian process regression and deep learning | |
| CN116384814A (en) | Airport runway capacity assessment method based on machine learning under multi-factor influence | |
| CN113744888B (en) | Regional epidemic trend prediction and early warning method and system | |
| CN110533241B (en) | Terminal area take-off and landing capacity prediction system | |
| Ma et al. | A Method for Establishing Tropospheric Atmospheric Refractivity Profile Model Based on Multiquadric RBF and k-means Clustering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |