CN107229916A - A kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding - Google Patents
A kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding Download PDFInfo
- Publication number
- CN107229916A CN107229916A CN201710388128.3A CN201710388128A CN107229916A CN 107229916 A CN107229916 A CN 107229916A CN 201710388128 A CN201710388128 A CN 201710388128A CN 107229916 A CN107229916 A CN 107229916A
- Authority
- CN
- China
- Prior art keywords
- noise
- airport
- data
- model
- depth
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Abstract
The invention discloses a kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding, belong to the abnormal recovery technique field of airport noise monitoring.This method obtains noise data first with the noise monitoring device on airport;Then, noise data pre-process obtaining sample set;Then, the candidate's depth noise reduction own coding model for extracting sample set feature is set, network weight initialization is carried out to each model;Then, the parameter of each model is successively trained using greedy algorithm, by back-propagation algorithm adjusting parameter, the parameter value of each model is obtained;Then, the data reconstruction error of each model is calculated, the minimum model of error is chosen, extracts the implicit depth characteristic of sample set in a model to train support vector regression model;Finally, noise monitoring data to be repaired are predicted by the model obtained using training.This method degree of intelligence is high, can accurately and efficiently repair abnormal data, effectively increases the promptness and validity of the reparation of airport noise Monitoring Data.
Description
Technical field
The invention discloses a kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding, belong to airport
Noise monitoring exception recovery technique field.
Background technology
With the development of the social economy, China increases the input in the fields such as Aeronautics and Astronautics year by year, airport quantity is also more next
It is more.But airport for passenger and cargo transport while rapid and convenient is provided, the various problems related to airport noise
Thereupon, airport noise problem has turned into one of obstacle of influence Civil Aviation Industry sustainable development.
In the arrangement of noise monitoring point, latticed monitoring point layout can be the noise situations on whole airport and periphery
Truly, in time, at large collect.This layout type needs to arrange considerable monitoring point, and node is redundancy sometimes
, not only equipment funds input is big, and post facility, which is safeguarded, also has sizable difficulty.In view of this, currently the majority airport
The mode that monitoring point is set in key area is used, to complete the collection of real time data and transfer data to terminal.
In the exploitation of airport noise monitoring system, noise and flight path the monitoring system that Australia sets up on the airport of part
Unite (NFPMS), the system does not monitor sensor-based system, also supporting management software only;Chicago airport in the U.S. also has
Airport noise actual measurement system, the system includes steady noise monitoring network subsystem and connection US Federal Aviation Administration (Federal
Aviation Administration, FAA) ATC radar aircraft flight route subsystem, daily record and
Store over 50,000 data points.The noise monitoring system of Beijing Capital Airport, is the product from B&K companies of Denmark, the system
Identification, the flight path of monitoring single incident aircraft and the height of single aircraft noise event can be achieved, noise contours is automatically generated
Whether figure, monitoring aircraft operation meet the functions such as noise abatement mission program, for scientifically grasping and controlling aircraft noise to machine
The influence of surrounding area, formulating relevant policies has important research and demonstration meaning.
But hardware device can be inevitably present the problems such as damage or aging, in view of the geographical distribution of monitoring point is extensive
With the complexity of equipment, the staff on airport usually can not rush to monitoring field in time and warping apparatus is safeguarded, lead
Cause monitoring point can not region residing for accurate acquisition noise data.Therefore, during monitoring point is failed, how the side of software is passed through
Method carries out timely repairing the problem of meriting attention as one to Monitoring Data.
The content of the invention
A kind of defect that the present invention exists for prior art, it is proposed that airport noise prison based on depth noise reduction own coding
Data recovery method is surveyed, fast and effeciently to repair extraordinary noise data in complicated airport noise environment.
The present invention adopts the following technical scheme that to solve its technical problem:
A kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding, comprises the following steps:
Step 1:Noise monitoring device using layout on airport periphery obtains the noise data of each monitoring point;
Step 2:The airport noise Monitoring Data got pre-process obtaining sample set;
Step 3:It is provided for extracting candidate's depth noise reduction autoencoder network of airport noise data feature, to therein every
Individual model carries out the initialization of network weight;
Step 4:The network parameter of each model is successively trained using greedy algorithm, network is adjusted by back-propagation algorithm
Parameter, obtains the parameter value after each model learning;
Step 5:The data reconstruction error of each model is calculated, the minimum model of reconstructed error is chosen, extracts airport noise
The implicit depth characteristic of Monitoring Data sample set in a model trains support vector regression model;
Step 6:The support vector regression model obtained using training, is carried out pre- to airport noise Monitoring Data to be repaired
Survey.
The particular content of the step 2 is:
Airport noise monitoring point measured data collection is normalized, the airport noise sample after being pre-processed
Collection.
The particular content of the step 3 is:
The depth noise reduction autoencoder network weight is initialized, makes the weight of connection adjacent two layers and the bias term of each layer be
Obey the random value of standardized normal distribution.
Particular content in the step 4 is:
Instructed successively come successively training network, the i.e. mode using the output of front layer as next layer of input using greedy algorithm
Practice, when training each layer parameter, fixed other each layer parameters keep constant;Each layer of input is required for adding when successively training
Enter noise, i.e., the value of neuron is reset to 0 with 0.1 probability;Then by back-propagation algorithm come percentage regulation noise reduction
Autoencoder network parameter, i.e., update network weight using the continuous iteration of gradient descent method, object function is progressivelyed reach fixed threshold
Value, completes the process of adjustment network parameter, the parameter value after being learnt.
The particular content of the step 6 is:
One is chosen from multiple airport noise monitoring points as abnormity point, other monitoring points utilize depth as feature set
The implicit depth characteristic that noise reduction autoencoder network is extracted is spent, support vector regression model is trained, to the noise of monitoring to be repaired
Value is predicted, and is predicted the outcome using this as data reparation result, completes the reparation of abnormal data.
Beneficial effects of the present invention are as follows:
1st, this method has intelligence learning ability, and the airport noise sample set needed for repairing is easily obtained.
2nd, using depth noise reduction autoencoder network learning data feature, by reconstructing input data and learning input data
Compression expression, obtains the new expression-form that airport noise Monitoring Data sample set implies depth characteristic, not only fully reflects defeated
Enter the characteristic of data, also considerably reduce the redundancy existed between relevant information.
3rd, Monitoring Data to be repaired is predicted using support vector regression algorithm, can preferably solves to deposit in data
Nonlinear problem, improve Generalization Capability.
4th, this method can effectively improve the reparation of airport noise Monitoring Data in complicated airport noise monitoring of environmental
Promptness and validity.
Brief description of the drawings
Fig. 1 is flow chart of the method for the present invention.
Embodiment
The invention is described in further detail below in conjunction with the accompanying drawings.
The flow of the airport noise data of monitoring point restorative procedure based on depth noise reduction own coding of the invention as shown in figure 1,
Specifically include following steps:
Step 1:Noise monitoring device using layout on airport periphery obtains the noise data of each monitoring point;
Using the noise monitoring device being laid out on airport periphery, these monitoring devices can monitor the airport of position in real time
Noise data.The real-time noise data of each monitoring point are obtained by these monitoring devices.
Step 2:The airport noise Monitoring Data got pre-process obtaining sample set.
The airport noise Monitoring Data got is normalized by formula below:
Wherein, diIt is an airport noise Monitoring Data, Max and Min are represented in airport noise Monitoring Data most respectively
Big value and minimum value, xiIt is diNormalization result.Thus number of samples is obtained for trnTraining sample set
It is te with number of samplesnTest sample collectionWherein:For the 1st training data,For the 2nd instruction
Practice data,For trnIndividual training data,For the 1st test data,For the 2nd test data,For tenIt is individual to survey
Try data.
Step 3:It is provided for extracting candidate's depth noise reduction autoencoder network of airport noise data feature, to therein every
Individual model carries out the initialization of network weight.
It is provided for extracting candidate's depth noise reduction autoencoder network M=that noise monitoring data in airport imply depth characteristic
{m1,m2,...,mk, wherein:m1For the 1st candidate family, m2For the 2nd candidate family, mkFor k-th of candidate family.To in M
Each model mi, initialize network weightWherein W is the weight of all connection adjacent two layers,
B is the bias term of each layer, and N (0,1) is standardized normal distribution.
Step 4:The network parameter of each model is successively trained using greedy algorithm, network is adjusted by back-propagation algorithm
Parameter, obtains the parameter value after each model learning.
In order to learn the feature to more robust, each layer of input is required for adding noise, i.e. input data x is by adopting
The mode for taking random selection neuron to reset to 0 adds noise and obtainedPlus data of making an uproar obtain hidden layer by coding function and represented
h.Coding function is defined as:
WhereinW(1)Represent the weight of connection input layer and hidden layer;b(1)Represent the bias term of input layer;To add data of making an uproar.In the same manner, decoding functions are defined as:
Wherein W(2)Represent the connection weight of hidden layer and output layer;b(2)Represent the bias term of hidden layer.By above-mentioned formula
Understand, outputIt is represented by:
Wherein hW,b() is the reconstruction of function of input data.
To training dataUsing greedy algorithm successively training network parameter θ(j), j=1,2 ...,
mi, i.e., the first layer of depth noise reduction autoencoder network is trained first with input data, the parameter W of the first layer network is generated(1)And b(1);Then the output of first layer is continued to train the parameter W for obtaining the second layer as the input of the second layer(2)And b(2);It is finally right
Each layer uses same strategy, the i.e. mode using the output of front layer as next layer of input to train successively below.For above-mentioned instruction
The mode of white silk, when each layer parameter is trained, can fix other each layer parameters and keep constant.
By the parameter of back-propagation algorithm percentage regulation noise reduction autoencoder network, i.e., updated using gradient descent method iteration
Weight, renewal process is represented by:
Wherein parameter alpha is learning rate,For the connection weight of i-th of neuron in l layer networks to jth neuron,
b(l)For the bias term in l layer networks.When object function reaches that certain threshold value completes the whole network development process of fine setting, learned
Parameter value θ after habitmi.For a sample set { x containing m sample1,x2,...,xm, wherein:x1For the 1st input number
According to x2For second input data, xmFor m-th of input data, object function is represented by:
Wherein:xiFor original input data,To add data of making an uproar.
To further reduce over the risk of fitting and improving the generalization of network, to learning parameter there is provided L2 regularizations about
Beam, object function can be further rewritten as:
Wherein θ={ W, b };W is the weight of all connection adjacent two layers;B is the bias term of each layer;λ is then used to measure number
According to the weight between reconstruct degree and regularization constraint.
Step 5:The data reconstruction error of each model is calculated, the minimum model of reconstructed error is chosen, extracts airport noise
The implicit depth characteristic of Monitoring Data sample set in a model trains support vector regression model.
To test dataM is calculated according to above-mentioned objective function EquationiThe data reconstruction error of modelMake M=M { mi, aforesaid operations are carried out to each model in candidate's depth noise reduction autoencoder network M.FromThe minimum model m of middle selected valuemin, x is extracted respectivelytrAnd xteIn mminImplicit depth characteristic d in modeltrWith
dte.Using ε-SVR regression algorithms with dtrIt is trained on the data set being characterized, obtains regression model.
Step 6:The support vector regression model obtained using training, is carried out pre- to airport noise Monitoring Data to be repaired
Survey.
One is chosen from multiple airport noise monitoring points as abnormity point, other monitoring points are as feature set, according to
The regression model practised, utilizes the implicit depth characteristic d of test datatePredicted value is calculated, to the noise of monitoring point to be repaired
Value is predicted, and is predicted the outcome using this as the reparation result of abnormal data, completes the reparation of abnormal data.
Claims (5)
1. a kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding, it is characterised in that including following step
Suddenly:
Step 1:Noise monitoring device using layout on airport periphery obtains the noise data of each monitoring point;
Step 2:The airport noise Monitoring Data got pre-process obtaining sample set;
Step 3:It is provided for extracting candidate's depth noise reduction autoencoder network of airport noise data feature, to each mould therein
Type carries out the initialization of network weight;
Step 4:The network parameter of each model is successively trained using greedy algorithm, adjusting network by back-propagation algorithm joins
Number, obtains the parameter value after each model learning;
Step 5:The data reconstruction error of each model is calculated, the minimum model of reconstructed error is chosen, extracts airport noise monitoring
The implicit depth characteristic of set of data samples in a model trains support vector regression model;
Step 6:The support vector regression model obtained using training, is predicted to airport noise Monitoring Data to be repaired.
2. a kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding according to claim 1, its
It is characterised by:The particular content of the step 2 is:
Airport noise monitoring point measured data collection is normalized, the airport noise sample set after being pre-processed.
3. a kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding according to claim 1, its
It is characterised by:The particular content of the step 3 is:
The depth noise reduction autoencoder network weight is initialized, makes the weight of connection adjacent two layers and the bias term of each layer to obey
The random value of standardized normal distribution.
4. a kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding according to claim 1, its
It is characterised by:Particular content in the step 4 is:
Trained successively come successively training network, the i.e. mode using the output of front layer as next layer of input using greedy algorithm,
When training each layer parameter, fixed other each layer parameters keep constant;Each layer of input is required for addition to make an uproar when successively training
Sound, i.e., reset to 0 with 0.1 probability by the value of neuron;Then it is self-editing come percentage regulation noise reduction by back-propagation algorithm
Code network parameter, i.e., update network weight using the continuous iteration of gradient descent method, object function is progressivelyed reach fixed threshold, complete
Into the process of adjustment network parameter, the parameter value after being learnt.
5. a kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding according to claim 1, its
It is characterised by:The particular content of the step 6 is:
One is chosen from multiple airport noise monitoring points as abnormity point, other monitoring points are dropped as feature set using depth
The implicit depth characteristic that autoencoder network of making an uproar is extracted, trains support vector regression model, the noise figure of monitoring to be repaired is entered
Row prediction, and predicted the outcome using this as data reparation result, complete the reparation of abnormal data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710388128.3A CN107229916A (en) | 2017-05-27 | 2017-05-27 | A kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710388128.3A CN107229916A (en) | 2017-05-27 | 2017-05-27 | A kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107229916A true CN107229916A (en) | 2017-10-03 |
Family
ID=59934108
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710388128.3A Pending CN107229916A (en) | 2017-05-27 | 2017-05-27 | A kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107229916A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108629675A (en) * | 2018-04-27 | 2018-10-09 | 信阳师范学院 | A kind of Multiple trees financial alert method |
CN111144303A (en) * | 2019-12-26 | 2020-05-12 | 华北电力大学(保定) | Power line channel transmission characteristic identification method based on improved denoising autoencoder |
CN112308104A (en) * | 2019-08-02 | 2021-02-02 | 杭州海康威视数字技术股份有限公司 | Abnormity identification method and device and computer storage medium |
CN113792754A (en) * | 2021-08-12 | 2021-12-14 | 国网江西省电力有限公司电力科学研究院 | Method for processing DGA (differential global alignment) online monitoring data of converter transformer by removing different elements and then repairing |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103810392A (en) * | 2013-12-13 | 2014-05-21 | 北京航空航天大学 | Degradation data missing interpolation method based on support vector machine and RBF neural network |
CN104537684A (en) * | 2014-06-17 | 2015-04-22 | 浙江立元通信技术股份有限公司 | Real-time moving object extraction method in static scene |
CN105469123A (en) * | 2015-12-30 | 2016-04-06 | 华东理工大学 | Missing data completion method based on k plane regression |
CN105868786A (en) * | 2016-04-01 | 2016-08-17 | 山东正晨科技股份有限公司 | Car logo identifying method based on self-coding pre-training deep neural network |
CN106485353A (en) * | 2016-09-30 | 2017-03-08 | 中国科学院遥感与数字地球研究所 | Air pollutant concentration forecasting procedure and system |
CN106709640A (en) * | 2016-12-15 | 2017-05-24 | 华南理工大学 | Method for predicting ship energy consumption based on deep learning and support vector regression |
-
2017
- 2017-05-27 CN CN201710388128.3A patent/CN107229916A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103810392A (en) * | 2013-12-13 | 2014-05-21 | 北京航空航天大学 | Degradation data missing interpolation method based on support vector machine and RBF neural network |
CN104537684A (en) * | 2014-06-17 | 2015-04-22 | 浙江立元通信技术股份有限公司 | Real-time moving object extraction method in static scene |
CN105469123A (en) * | 2015-12-30 | 2016-04-06 | 华东理工大学 | Missing data completion method based on k plane regression |
CN105868786A (en) * | 2016-04-01 | 2016-08-17 | 山东正晨科技股份有限公司 | Car logo identifying method based on self-coding pre-training deep neural network |
CN106485353A (en) * | 2016-09-30 | 2017-03-08 | 中国科学院遥感与数字地球研究所 | Air pollutant concentration forecasting procedure and system |
CN106709640A (en) * | 2016-12-15 | 2017-05-24 | 华南理工大学 | Method for predicting ship energy consumption based on deep learning and support vector regression |
Non-Patent Citations (4)
Title |
---|
PASCAL VINCENT 等: ""Extracting and Composing Robust Features with Denoising Autoencoders"", 《PROC.OF THE 25TH ACM INTERNATIONAL CONFERENCE ON MACHINE》 * |
PASCAL VINCENT 等: ""Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion"", 《JOURNAL OF MACHINE LEARNING RESEARCH》 * |
时培明 等: ""基于深度学习特征提取和粒子群支持向量机状态识别的齿轮智能故障诊断"", 《中国机械工程》 * |
陈曦 等: ""基于观察学习的机场噪声监测点关联预测研究"", 《计算机工程与科学》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108629675A (en) * | 2018-04-27 | 2018-10-09 | 信阳师范学院 | A kind of Multiple trees financial alert method |
CN112308104A (en) * | 2019-08-02 | 2021-02-02 | 杭州海康威视数字技术股份有限公司 | Abnormity identification method and device and computer storage medium |
CN111144303A (en) * | 2019-12-26 | 2020-05-12 | 华北电力大学(保定) | Power line channel transmission characteristic identification method based on improved denoising autoencoder |
CN113792754A (en) * | 2021-08-12 | 2021-12-14 | 国网江西省电力有限公司电力科学研究院 | Method for processing DGA (differential global alignment) online monitoring data of converter transformer by removing different elements and then repairing |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Dinda et al. | An integrated simulation approach to the assessment of urban growth pattern and loss in urban green space in Kolkata, India: A GIS-based analysis | |
CN107229916A (en) | A kind of airport noise Monitoring Data restorative procedure based on depth noise reduction own coding | |
CN111209434B (en) | Substation equipment inspection system and method based on multi-source heterogeneous data fusion | |
CN112819207B (en) | Geological disaster space prediction method, system and storage medium based on similarity measurement | |
CN113240688A (en) | Integrated flood disaster accurate monitoring and early warning method | |
CN111695731B (en) | Load prediction method, system and equipment based on multi-source data and hybrid neural network | |
CN110766212A (en) | Ultra-short-term photovoltaic power prediction method for historical data missing electric field | |
CN110009614A (en) | Method and apparatus for output information | |
KR101183105B1 (en) | Method of establishing information of cloud data and establishing system of information of cloud data | |
CN109978870A (en) | Method and apparatus for output information | |
Wu et al. | Long-term 4D trajectory prediction using generative adversarial networks | |
Lee et al. | Applying machine learning methods to detect convection using Geostationary Operational Environmental Satellite-16 (GOES-16) advanced baseline imager (ABI) data | |
CN115545334B (en) | Land utilization type prediction method and device, electronic equipment and storage medium | |
KR102454007B1 (en) | Apparatus and method for calculating solar insolation quantity based on artificial intelligence using artificial satellite observation data | |
CN103942251A (en) | Method and system for inputting high altitude meteorological data into database based on multiple quality control methods | |
CN116451514A (en) | Bridge structure state evaluation method and device and electronic equipment | |
Ren et al. | DLformer: A dynamic length transformer-based network for efficient feature representation in remaining useful life prediction | |
Shang et al. | Spatiotemporal reflectance fusion using a generative adversarial network | |
CN117556197B (en) | Typhoon vortex initialization method based on artificial intelligence | |
CN111275059B (en) | Image processing method and device and computer readable storage medium | |
CN116681962A (en) | Power equipment thermal image detection method and system based on improved YOLOv5 | |
CN114263855B (en) | Natural gas transportation pipeline leakage prediction method and application thereof | |
Sharma et al. | Deep Learning Based Prediction Of Weather Using Hybrid_stacked Bi-Long Short Term Memory | |
CN115391523A (en) | Wind power plant multi-source heterogeneous data processing method and device | |
CN114154622A (en) | Algorithm model for traffic operation system flow data acquisition missing completion |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171003 |
|
WD01 | Invention patent application deemed withdrawn after publication |