CN103020733A - Method and system for predicting single flight noise of airport based on weight - Google Patents

Abstract

The invention discloses a method for predicting the single flight noise of an airport, which comprises the following steps of: giving corresponding weight to single flight noise data of the airport at first; discriminatively establishing the learning ability of a sample through the weight; training a plurality of different learning models; selecting better learning models therein to be integrated; and abstracting the noise prediction problem of the airport as the multiple regression analysis problem in mathematics. A system for realizing the prediction method disclosed by the invention comprises an airport single flight noise data acquisition module, an airport noise data storage module and an airport single flight noise prediction module. According to the method disclosed by the invention, the instability of the current prediction method based on the single learning algorithm is overcome; the prediction precision and the prediction stability are effectively increased; compared with the general optimization and all integrations of the single learning algorithm, the prediction speed is increased; and the airport noise prediction practicability is increased.

Description

The single flight noise prediction method in a kind of airport based on weight and system thereof

Technical field

The present invention relates to a kind of method of predicting the single flight noise in airport, relate in particular to a kind of single flight noise prediction in airport method based on weight, relate to simultaneously a kind of system for realizing the method.

Background technology

Along with the rapid increase of China civil aviaton passenger and freight transportation task, the Development of China's Urbanization accelerated development of surrounding area, airport, because airport space and Urban Land are more and more nearer, so the caused contradiction of the noise pollution on airport, dispute get more and more.The airport noise problem is a serious social concern, and it has not only limited the development on airport itself, brings very large impact also for life, the study and work of airport surrounding population.

Because domestic prediction theory to airport noise, Forecasting Methodology lack research, China all is to rely on external experience and method (such as the INM of US Federal Aviation Administration (FAA) proposition) to the prediction of airport noise at present, does not form unified calculating and Forecasting Methodology.And existing noise prediction method needs complicated based on the comparison hypothesis, and the environmental baseline on airport also varies, and that computation model is difficult to consider is thorough, and (as: people such as Asensio C. points out that INM can't consider the problem of aircraft taxi, has prediction deviation; The people such as Yingjie Yang point out that the model in the INM software can not provide the aircraft noise near true environment of expection to predict the outcome in some cases).The prediction of traditional airport noise mainly is to utilize real data to obtain according to noise evaluation standard and noise calculation formula, and error is large and be unfavorable for prediction to following airport noise situation.In recent years, the noise problem around China airport became increasingly conspicuous, and therefore was necessary to study as early as possible algorithm, and exploitation is applicable to the airport noise forecasting software of China.

The people such as F.Van Den Berg studies show that, utilize the rule, pattern and the acoustic theory computation model that excavate from measured data to carry out in conjunction with the predictablity rate that can effectively improve noise.Machine learning method is one of main method of setting up by measured data noise prediction model, we are attempting machine learning method is applied to airport noise prediction aspect at present, existing trial research is such as with Fuzzy Support Vector Regression machine, cluster, time series, neural networks etc. apply to the airport noise data prediction, but predicting the outcome of single learning algorithm is bad during fashion, under different conditions and data collection, do not have stability, can not keep high-precision estimated performance always.And a large amount of learning algorithm models obviously can reduce predetermined speed, increases storage space.

Summary of the invention

Technical matters

The technical problem to be solved in the present invention provides a kind of method that the noise of the single flight in airport is predicted, the method is given corresponding weight for the single flight noise data in airport first, distinguish the learning ability of formulating sample by weight, and then train a plurality of different learning models, selection wherein reasonable learning model is integrated, and a kind of system for realizing the method is provided simultaneously.

Technical scheme

In order to solve above-mentioned technical matters, the single flight noise prediction method in the airport based on weight of the present invention comprises the following steps:

Step 1: collect the raw data of the single flight noise in airport and carry out the data normalized;

Step 2: the raw data of part airport flight noise makes up the first training sample set S that obtains the single flight noise prediction in airport, S={ (x in the extraction step one _i, y _i) | i=1,2 ..., n}, wherein x _i=(x _I1, x _I2..., x _Ip) be input item, y _i∈ R is the decision-making item, and n is the number of sample data among the S, extracts residue air station flight noise raw data and makes up the second test sample book collection S that obtains the single flight noise prediction in airport ^*

Step 3: the training sample set S that step 2 is obtained trains, and is met the single flight noise prediction model in airport of accuracy requirement, and concrete substep is as follows:

3.1: select a kind of basic studies device M, and setting training pattern number is T;

3.2: give weight for each the bar sample data of the first original training sample collection S in the step 2, suppose that at first each bar sample data is of equal importance, then the weight setting of sample data is

I=1,2..., n, described n are the total number of sample among the S;

3.3: suppose t model of current training, t ∈ [1, T] is according to current sample data weight w _tProbability distribution, from the first training sample set S, have the larger sample of the weight of taking-up obtain new training set S with putting back to _t, use described basic studies device M to train, obtain forecast model M _t

3.4: with the forecast model M of gained _tThe first training sample set S is predicted, obtain predicted value f _t(x _i), i=1,2 ..., n, described n are the total number of sample among the S;

3.5: the loss value set that calculates every sample data of the first training sample set S

A=max{|f wherein _t(x _i)-y _i|, i=1,2 ..., n, n are the number of samples of the first training sample set S, according to sample weights w _tObtain forecast model M _tThe average loss value

The L_avr if satisfy condition (t) 〉=0.5 then jumps to substep 3.3, otherwise execution in step 3.6;

3.6: set a threshold alpha for sample losses value set { loss (i) }, i=1 wherein, 2 ..., n, n is the number of samples of S, if the loss value loss (i) of i bar sample data＞α in the sample set sets the υ that is labeled as of this sample _tOtherwise set the υ that is labeled as of this sample (i)=-1, _t(i)=1, then set forecast model M _tBe labeled as m (t), m (t)={ υ _t(i) w _t(i) }, i=1 wherein, 2..., n, n are the number of samples of S, and namely m (t) is n-dimensional vector, and its element is the mark of each sample among the S and the product of its weight;

3.7: by forecast model M _tAverage loss value L_avr (t) obtain its degree of confidence β _t=L_ave (t)/(1-L_ave (t)), then the weight of every sample data is updated among the first training sample set S $w_{t+1}\left(i\right)=w_t\left(i\right){\mathrm{\β}}_t^{1-L\_\mathrm{ave}\left(t\right)}/Z_t,$ Described Z _tNormalization coefficient, so that $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{w}_{t+1}\left(i\right)=1;$

3.8: set forecast model M _tWeight be

3.9: set forecast model set { M ₁..., M _TThe average integrated H that is labeled as,

Wherein m (t) is forecast model M _tMark, wh _tBe its weight, then H is n-dimensional vector, and its element is that the mark of each sample among the first training sample set S is about the weighted mean value of T forecast model;

3.10: given n-dimensional vector O, its element is 1, sets a reference vector R with the projection of O on the determined lineoid of integrated mark H, and RH=0 is then arranged, and supposes R=O+bH, then has $b=\frac{-O\·H}{{\left|H\right|}^2};$

3.11: set forecast model M _tThe angle of mark m (t) and reference vector R be θ (t), T=1,2..., T sorts to θ (t), choose less than

The corresponding forecast model M of angle _t, form the set of new forecast model M ' ₁..., M ' _s, corresponding forecast model weight sets be combined into wh ' ₁..., wh ' _s;

3.12: to the set of resulting forecast model M ' ₁..., M ' _sBe weighted to integrate and obtain the single flight noise prediction model in final airport;

Step 4: utilize the single flight noise prediction model in final airport of step 3 gained to the second test sample book collection S ^*Predict and try to achieve square error, with the measurement condition of square error as precision, be met the actual prediction model of accuracy requirement;

Step 5: utilize the actual prediction model of step 4 gained that the noise effect of the single flight of a certain specific airport in future is predicted.

The raw data of the single flight noise in described airport comprises the information such as longitude and latitude, height above sea level, flight flight path, section, speed, humiture and day night equivalent sound level.

Method of the present invention is abstracted into multiple regression analysis problem in the mathematics with the airport noise forecasting problem, with various airport noise influence factors as independent variable, noise grade is as dependent variable, simulate their funtcional relationship in conjunction with machine learning algorithm, and then only need the single flight noise effect factor in airport of a certain specific the unknown just can obtain its noise figure.On this basis, choosing multiple good airport noise predicts the outcome to be optimized to integrate and is finally predicted the outcome.

System for realizing described Forecasting Methodology of the present invention comprises for the airport of harvester sound list flight noise single flight noise data acquisition module, be used for airport noise data memory module that storage airport single flight noise data acquisition module collects and carry out the single flight noise prediction in airport module according to the data of airport noise data memory module.

Beneficial effect

The single flight noise prediction method in airport of the present invention has overcome the instability of existing Forecasting Methodology based on single learning algorithm, thereby the sample of distinguishing different learning abilities by sample weighting is trained a plurality of learners, therefrom selecting good single learning algorithm to integrate for airport noise predicts, Effective Raise the stability of precision of prediction and prediction, be compared to the single learning algorithm of general optimization and all integrate and improved predetermined speed, strengthened the practicality of airport noise prediction.

Description of drawings

Fig. 1 is training study device process flow diagram of the present invention;

Fig. 2 is the single flight noise prediction in the airport method flow diagram that the present invention proposes;

Fig. 3 is airport of the present invention single flight Noise Prediction System figure.

Embodiment

Embodiment one:

Present embodiment is a kind of single flight noise prediction in airport method based on weight, as shown in Figure 1 and Figure 2, comprises the following steps:

Step 1: collect the raw data of the single flight noise in airport and carry out the data normalized; The single flight noise data of present embodiment on certain each monitoring point, large-scale International airport in 2010 is as sample instance.The single flight noise data in airport comprises longitude and latitude, height above sea level, flight flight path, section, speed, humiture and day night equivalent sound level (DNL), for avoiding in the data every value order of magnitude training effect of impact that has big difference, raw data is carried out normalized, so that everyly all normalize to [0 according to its span, 1] interval, employed formula is

$y=\frac{x-x_{\min }}{x_{\max }-x_{\min }};$

With longitude and latitude, height above sea level, flight flight path, section, speed, humiture etc. in the single flight noise data sample of airport as input item X={x _i| i=1,2 ..., n}, ENL is as a decision-making Y={y round the clock _i| i=1,2 ..., n} forms data set S=(X, Y)={ (x _i, y _i) | i=1,2 ..., n}, for use in subsequent treatment, wherein n is the noise data number of samples;

Step 2: the raw data of part airport flight noise makes up the first training sample set S that obtains the single flight noise prediction in airport, S={ (x in the extraction step one _i, y _i) | i=1,2 ..., n}, wherein x _i=(x _I1, x _I2..., x _Ip) be input item, y _i∈ R is the decision-making item, and n is the number of sample data among the S, extracts residue air station flight noise raw data and makes up the second test sample book collection S that obtains the single flight noise prediction in airport ^*

Step 3: the training sample set S that step 2 is obtained trains, and is met the single flight noise prediction model in airport of accuracy requirement, and concrete substep is as follows:

3.1: select a kind of basic studies device M, and setting training pattern number is T;

3.2: give weight for each the bar sample data of the first original training sample collection S in the step 2, suppose that at first each bar sample data is of equal importance, then the weight setting of sample data is

Described n is the total number of sample among the S;

3.3: suppose t model of current training, t ∈ [1, T] is according to current sample data weight w _tProbability distribution, from the first training sample set S, have the larger sample of the weight of taking-up obtain new training set S with putting back to _t, use described basic studies device M to train, obtain forecast model M _t

3.4: with the forecast model M of gained _tThe first training sample set S is predicted, obtain predicted value f _t(x _i), i=1,2 ..., n, described n are the total number of sample among the S;

3.5: the loss value set that calculates every sample data of the first training sample set S A=max{|f wherein _t(x _i)-y _i|, i=1,2 ..., n, n are the number of samples of the first training sample set S, according to sample weights w _tObtain forecast model M _tThe average loss value

The L_avr if satisfy condition (t) 〉=0.5 then jumps to substep 3.3, otherwise execution in step 3.6;

3.6: set a threshold alpha for sample losses value set { loss (i) }, i=1 wherein, 2 ..., n, n is the number of samples of S, if the loss value loss (i) of i bar sample data＞α in the sample set sets the υ that is labeled as of this sample _tOtherwise set the υ that is labeled as of this sample (i)=-1, _t(i)=1, then set forecast model M _tBe labeled as m (t), m (t)={ υ _t(i) w _t(i) }, i=1 wherein, 2..., n, n are the number of samples of S, and namely m (t) is n-dimensional vector, and its element is the mark of each sample among the S and the product of its weight;

3.7: by forecast model M _tAverage loss value L_avr (t) obtain its degree of confidence β _t=L_ave (t)/(1-L_ave (t)), then the weight of every sample data is updated among the first training sample set S $w_{t+1}\left(i\right)=w_t\left(i\right){\mathrm{\β}}_t^{1-L\_\mathrm{ave}\left(t\right)}/Z_t,$ Described Zt is normalization coefficient, so that $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{w}_{t+1}\left(i\right)=1;$

3.8: set forecast model M _tWeight be

3.9: set forecast model set { M ₁..., M _TThe average integrated H that is labeled as,

Wherein m (t) is forecast model M _tMark, wh _tBe its weight, then H is n-dimensional vector, and its element is that the mark of each sample among the first training sample set S is about the weighted mean value of T forecast model;

3.10: given n-dimensional vector O, its element is 1, sets a reference vector R with the projection of O on the determined lineoid of integrated mark H, and RH=0 is then arranged, and supposes R=O+bH, then has $b=\frac{-O\·H}{{\left|H\right|}^2};$

3.11: set forecast model M _tThe angle of mark m (t) and reference vector R be θ (t), T=1,2..., T sorts to θ (t), choose less than

The corresponding forecast model M of angle _t, form the set of new forecast model M ' ₁..., M ' _s, corresponding forecast model weight sets be combined into wh ' ₁..., wh ' _s;

3.12: to the set of resulting forecast model M ' ₁..., M ' _sBe weighted to integrate and obtain the single flight noise prediction model in final airport;

Step 4: utilize the single flight noise prediction model in final airport of step 3 gained to the second test sample book collection S ^*Predict and try to achieve square error, with the measurement condition of square error as precision, be met the actual prediction model of accuracy requirement;

Step 5: utilize the actual prediction model of step 5 gained that the noise effect of the single flight of a certain specific airport in future is predicted.

Embodiment two:

As shown in Figure 3, present embodiment is a kind of system for realizing embodiment one described method, and it comprises:

Be used for gathering airport single flight noise data acquisition module of the single flight noise in airport, the data in the single flight noise data acquisition module of described airport comprise airport noise data training set and airport noise data verification collection;

The data of airport noise data training set single flight noise prediction model in input airport after airport single flight noise prediction model training module is trained is selected integrate module;

The single flight noise prediction model in airport selects integrate module to integrate the data of the single flight noise prediction model of the concentrated data of airport noise data verification and airport training module output, obtain the single flight noise prediction model in airport, utilize this forecast model then can carry out the prediction of airport noise.

Claims (3)

1. the single flight noise in the airport based on weight is selected the consolidated forecast method, it is characterized in that, comprises the following steps:

Step 1: collect the raw data of the single flight noise in airport and carry out the data normalized;

Step 2: the raw data of part airport flight noise makes up the first training sample set S that obtains the single flight noise prediction in airport in the extraction step one, extracts residue air station flight noise raw data and makes up the second test sample book collection S that obtains the single flight noise prediction in airport ^*

Step 3: the training sample set S that step 2 is obtained trains, and is met the single flight noise prediction model in airport of accuracy requirement, and concrete substep is as follows:

3.1: select a kind of basic studies device M, and setting training pattern number is T;

3.2: give weight for each the bar sample data of the first original training sample collection S in the step 2, suppose that at first each bar sample data is of equal importance, then the weight setting of sample data is

I=1,2..., n, described n are the total number of sample among the S;

3.3: suppose t model of current training, t ∈ [1, T] is according to current sample data weight w _tProbability distribution, from the first training sample set S, have the larger sample of the weight of taking-up obtain new training set S with putting back to _t, use described basic studies device M to train, obtain forecast model M _t

3.4: with the forecast model M of gained _tThe first training sample set S is predicted, obtain predicted value f _t(x _i), i=1,2 ..., n, described n are the total number of sample among the S;

3.5: the loss value set that calculates every sample data of the first training sample set S A=max{|f wherein _t(x _i)-y _i|, i=1,2 ..., n, n are the number of samples of the first training sample set S, according to sample weights w _tObtain forecast model M _tThe average loss value

The L_avr if satisfy condition (t) 〉=0.5 then jumps to substep 3.3, otherwise execution in step 3.6;

3.6: set a threshold alpha for sample losses value set { loss (i) }, i=1 wherein, 2 ..., n, n is the number of samples of S, if the loss value loss (i) of i bar sample data＞α in the sample set sets the υ that is labeled as of this sample _tOtherwise set the υ that is labeled as of this sample (i)=-1, _t(i)=1, then set forecast model M _tBe labeled as m (t), m (t)={ υ _t(i) w _t(i) }, i=1 wherein, 2..., n, n are the number of samples of S, and namely m (t) is n-dimensional vector, and its element is the mark of each sample among the S and the product of its weight;

3.7: by forecast model M _tAverage loss value L_avr (t) obtain its degree of confidence β _t=L_ave (t)/(1-L_ave (t)), then the weight of every sample data is updated among the first training sample set S $w_{t+1}\left(i\right)=w_t\left(i\right){\mathrm{\β}}_t^{1-L\_\mathrm{ave}\left(t\right)}/Z_t,$ Described Z _tNormalization coefficient, so that $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{w}_{t+1}\left(i\right)=1;$

3.8: set forecast model M _tWeight be

3.9: set forecast model set { M ₁..., M _TThe average integrated H that is labeled as,

Wherein m (t) is forecast model M _tMark, wh _tBe its weight, then H is n-dimensional vector, and its element is that the mark of each sample among the first training sample set S is about the weighted mean value of T forecast model;

3.10: given n-dimensional vector O, its element is 1, sets a reference vector R with the projection of O on the determined lineoid of integrated mark H, and RH=0 is then arranged, and supposes R=O+bH, then has $b=\frac{-O\·H}{{\left|H\right|}^2};$

3.11: set forecast model M _tThe angle of mark m (t) and reference vector R be θ (t),

T=1,2..., T sorts to θ (t), choose less than The corresponding forecast model M of angle _t, form the set of new forecast model M ' ₁..., M ' _s, corresponding forecast model weight sets be combined into wh ' ₁..., wh ' _s;

3.12: to the set of resulting forecast model M ' ₁..., M ' _sBe weighted to integrate and obtain the single flight noise prediction model in final airport;

Step 4: utilize the single flight noise prediction model in final airport of step 3 gained to the second test sample book collection S ^*Predict and try to achieve square error, with the measurement condition of square error as precision, be met the actual prediction model of accuracy requirement;

Step 5: utilize the actual prediction model of step 4 gained that the noise effect of the single flight of a certain specific airport in future is predicted.

2. the single flight noise in the airport based on weight as claimed in claim 1 is selected the consolidated forecast method, it is characterized in that the raw data of the single flight noise in the described airport of step 1 comprises the information such as longitude and latitude, height above sea level, flight flight path, section, speed, humiture and day night equivalent sound level.

3. a system that is used for realizing the method for claim 1 is characterized in that, comprising:

Be used for gathering airport single flight noise data acquisition module of the single flight noise in airport, the data in the single flight noise data acquisition module of described airport comprise airport noise data training set and airport noise data verification collection;

The data of airport noise data training set single flight noise prediction model in input airport after airport single flight noise prediction model training module is trained is selected integrate module;

The single flight noise prediction model in airport selects integrate module to integrate the data of the single flight noise prediction model of the concentrated data of airport noise data verification and airport training module output, obtains the single flight noise prediction model in airport.

Images