CN107392226A - The modeling method of pilot's working condition identification model - Google Patents
The modeling method of pilot's working condition identification model Download PDFInfo
- Publication number
- CN107392226A CN107392226A CN201710445742.9A CN201710445742A CN107392226A CN 107392226 A CN107392226 A CN 107392226A CN 201710445742 A CN201710445742 A CN 201710445742A CN 107392226 A CN107392226 A CN 107392226A
- Authority
- CN
- China
- Prior art keywords
- bird
- nest
- pilot
- working condition
- flying quality
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a kind of modeling method of pilot's working condition identification model, comprise the following steps：Step S1：Flying quality is pre-processed；Step S2：Feature extraction is carried out to pretreated flying quality；Step S3：Data Dimensionality Reduction is carried out to the feature extracted using Treelets algorithms；Step S4：Modeling is trained to the data after dimensionality reduction using the Gaussian process grader based on cuckoo algorithm optimization.The present invention has good time domain localization ability, also has to noise jamming immune so that characteristic value is more rigorous.Initial data is dropped to lower dimension, so that the data complexity after dimensionality reduction reduces, laid the first stone for more Accurate classification in next step.Parameter optimization is relatively easy, and is more easy to restrain.Global or approximate global optimization result is found for function as far as possible, avoids the suboptimization of model.
Description
Technical field
The present invention relates to flyer's fatigue detection and supervision early warning technology field, and in particular to pilot's working condition is known
The modeling method of other model, flight parameter when this method works according to aircraft, obtaining one using machine learning scheduling algorithm can
To assess the model of pilot work load's state.
Background technology
It is main similar to classical mode identification process for pilot's state-detection model in traditional aviation mission, its process
It is divided into：Data prediction, feature extraction, Data Dimensionality Reduction processing, machine learning algorithm classification processing.
In terms of Feature Extraction Technology, time domain signal characteristics average, variance and root mean square, be usually used to characterize signal width
It is worth characteristic and the otherness of original flight parameter.But these temporal signatures are excessively simple for the higher data of complexity,
It is difficult to the data set more complicated to information content carries out comprehensive feature description.
In terms of Data Dimensionality Reduction, principal component analytical method (PCA) is widely used in industry and scientific domain.But PCA has
The shortcomings that it can not overcome, PCA treats using all samples as an entirety, and the less direction of data variance can also wrap
Containing important information.So the algorithm may cause when dropping to low dimensional, original signal energy loss is excessive to cause distortion so that
Classification results are affected.
In terms of grader, it is SVMs (SVM) model that application at present is relatively more extensive.But SVM still has many lack
Fall into, the problem of selection of the kernel function and its parameter and penalty of SVMs is relatively difficult.In related parameter choosing
Aspect, more experiences for relying on traditional experiment, so can not generate one according to the actual conditions of input sample and most adapt to sample
Disaggregated model.It could even be possible to because parameter is improper, result over-fitting problem is produced so that classification results are affected.
Found by retrieval：
(1)
Qiu Zongjiang, Liu Huixia, Xi Qingbiao, wait《Computer engineering and application》, 2013,49 (4)：Delivered in 262-266
's《Unmanned plane PCA fault detection and diagnosis technical research [J]》In, Principal Component Analysis Algorithm (PCA) is combined variance by the technology
The detection of sensitive adaptive threshold, and applied to the fault detection and diagnosis of system for flight control computer sensor, overcome
The empty early warning and the shortcomings that wrong diagnosis in transient process of traditional pca model.But the technology point still can not overcome PCA reduction process
In the shortcomings that ignoring data variance less direction so that the algorithm can not drop data in the case where ensureing the basis of signal energy
To alap dimension.
(2)
Aircraft fault-findings [C] of Heng H, Zhang J, the Xin C. based on SVMs // international consumer electricity
Son, exchange and Web conference .IEEE, 2012：The 496-499. technologies are based on flying quality, establish SVM engine failure detection
Model, but SVM still has many defects, and the selection of the kernel function and its parameter and penalty of SVMs is relatively difficult
Problem.
(3)
B. [J] of the Gaussian process in machine learning worlds nervous system periodical, 2008,14 (6)：
The 3011-3015. technologies describe principle and the grader application of Gaussian process in detail, and analyze and show that Gaussian process has
There is the non-linear input of processing, independent of characteristics such as parameters.But also mentioned in article ending, in the equal of prediction Gaussian process model
, it is necessary to obtain the result of global optimization when value and covariance.This process can by carry out negative logarithm marginal probability on
The minimum of hyper parameter obtains, general to determine that hyper parameter has gradient descent method and Newton method, but such method becomes with initial
The number of dimensions quantitative change of amount is big, and algorithm is easily trapped into local extremum, and the selection of initial value has a great influence to algorithm.
The content of the invention
For above shortcomings in the prior art, the invention provides a kind of pilot's working condition identification model
Modeling method, it this method solve following technical problem：
(1) characteristic is excessively simple, and noise jamming is difficult to avoid that.
(2) more preferable dimension-reduction algorithm is needed so that can be in the case where ensureing original signal energy abundance, by data dimension
Reduce as far as possible.
(3) rigorous sorting algorithm is needed, adapts it to new stronger, is not to be completely dependent on parameter.And sorting algorithm should
This can handle nonlinear data.
The present invention is achieved by the following technical solutions.
A kind of modeling method of pilot's working condition identification model, comprises the following steps：
Step S1：Flying quality is pre-processed；
Step S2：Feature extraction is carried out to pretreated flying quality；
Step S3：Data Dimensionality Reduction is carried out to the feature extracted using Treelets algorithms；
Step S4：The characteristic after dimensionality reduction is instructed using the Gaussian process grader based on cuckoo algorithm optimization
Practice modeling.
Preferably, in step S1, longitudinal acceleration, side acceleration, normal acceleration, pitching in flying quality are chosen
Totally 11 flights are joined for angle, yaw angle, roll angle, ground velocity, rate of pitch, yaw rate, angular velocity in roll and angle of attack
Number, and flight parameter is normalized, obtain flight parameter data set；Flight parameter data set is carried out using 2 classes
Demarcation, i.e., {+1, -1 }, wherein ,+1 represents normal workload, and -1 represents non-normal working load, obtains pretreated fly
Row data.
Preferably, the step S2 comprises the following steps：
Step S2.1：Time domain signal characteristics are introduced, extract the average of pretreated flying quality, variance and square
Root；
Step S2.2：Calculate wavelet singular entropy：
Step S2.2.1：Obtain pretreated flying quality sequence；
Step S2.2.2：Wavelet analysis is carried out to pretreated flying quality sequence, obtains matrix of wavelet coefficients A；
To singular value features value λi(i=1,2 ... ..., r), wherein r ∈ R, represent singular eigenvalue problem sum, and all singular value features
Value λ1≥λ2≥…≥λr≥0；
Step S2.2.4：λk/λ1＞ 0.01%, k rank wavelet singular entropies are calculated, wherein, λkIt is big for kth in step S2.2.3
Singular value features value, λ1For singular value features value maximum in step S2.2.3.
Preferably, the k=5.
Preferably, step S3 comprises the following steps：
Step S3.1：At the 0th layer of the feature extracted, pretreated flying quality covariance matrix and similar is calculated
Property measurement；
Step S3.2：Since the 1st layer of the feature extracted, step S3.1 is repeated in, finds similarity measurement most
High covariance matrix, and the two-dimensional vector to finding carries out PCA conversion, obtains Jacobi's transformation matrix；
Step S3.3：To the Jacobi's transformation matrix obtained in step S3.2, approximating function and Detailfunction are defined, and
Determine that Trelets is represented according to the parameter of Jacobi's transformation, dimensionality reduction is carried out to given characteristic afterwards.
Preferably, step S4 comprises the following steps：
Step S4.1：The classification based training of Gauss two is carried out to the characteristic after dimensionality reduction；
Step S4.2：Cuckoo algorithm optimization training is carried out to the characteristic after the classification based training of Gauss two：
Step S4.2.1：Select majorized function：Log-likelihood function；
Step S4.2.2：Initiation parameter n, P are seta, MaxGeneration, wherein, n be Bird's Nest population quantity, Pa
The probability being found for parasitic Bird's Nest, i.e., the Bird's Nest ratio abandoned at random, MaxGeneration is maximum iteration；
Step S4.2.3：Flown by cuckoo Levy, produce a new Bird's Nest i, and evaluate Bird's Nest i fitness fi；
Step S4.2.4：Randomly choose a Bird's Nest j, evaluation Bird's Nest j fitness fj, and with fitness fiCompare；
Step S4.2.5：If fj＞ fi, substitute Bird's Nest i to turn into current the most new position with Bird's Nest j, otherwise Bird's Nest i is still
For latest position；
Step S4.2.6：Abandon PaThe Bird's Nest being found under probability, and sort and compare in remaining Bird's Nest, generate this
Take turns the current optimal solution after iteration；
Step S4.2.7：Judge whether current iteration number has been maxed out, if not up to, jumping to step
S4.2.3 carries out next round iteration, if it has, carrying out step S4.2.8；
Step S4.2.8：Optimal solution is exported, obtains pilot's working condition identification model.
Preferably, in the step S4.2.2, initiation parameter is arranged to：N=25, Pa=0.25, MaxGeneration
=1000；Wherein, n be Bird's Nest population quantity, PaThe probability being found for parasitic Bird's Nest, i.e., the Bird's Nest ratio abandoned at random,
MaxGeneration is maximum iteration.
Preferably, also comprise the following steps：
Step S5, for training remaining pretreated flying quality to know as test data to pilot's working condition
Other model is verified, obtains the precision of model, so as to the efficiency of evaluation model；Contrast Treelets dimensionality reductions and PCA simultaneously
Influence of the dimensionality reduction to model efficiency, obtains best identified model.
By adopting the above-described technical solution, the beneficial effects of the invention are as follows：
1) wavelet singular entropy feature is added in feature extraction, due to its good time domain localization ability, it is carried
Signal more complicated feature.In addition to a certain extent, also have to noise jamming immune so that characteristic value is more rigorous.
2) compared to PCA, Treelets is a kind of Multi-scale model, and this method is adaptive algorithm, as a result for can be anti-
Reflect the hierarchical tree and orthogonal basis of data structure.It, which can drop to initial data less dimension, can but retain more original letters
Cease energy.And Treelets algorithms used herein can drop to initial data more as multi-scale self-adaptive algorithm
Add low dimension, so that the data complexity after dimensionality reduction reduces, laid the first stone for more Accurate classification in next step.
3) compared to SVM, Gaussian process is a kind of nonparametric model, can easily processing system uncertainty with making an uproar
Sound observes modeling problem, and has simplification and flexible nonparametric inference ability, and it is showed often calculates better than similar classification
Method.Gaussian process can select hyper parameter and kernel function according to training data, and parameter optimization is relatively easy, and is more easy to restrain.
4) in addition in order to avoid model produces local minimum problem, the application introduces cuckoo optimized algorithm (CS).
Due to the random selection of the algorithm, local optimum can be evaded to a certain extent, found as far as possible for function global or near
Like global optimization result, the suboptimization of model is avoided.
Brief description of the drawings
The detailed description made by reading with reference to the following drawings to non-limiting example, further feature of the invention,
Objects and advantages will become more apparent upon：
Fig. 1 is Treelets and the nicety of grading comparison diagram after PCA dimensionality reductions；
Fig. 2 is Treelets-CS-GP typical classification result figures；
Fig. 3 is TPCA-CS-GP typical classification result figures.
Embodiment
Embodiments of the invention are elaborated below：The present embodiment is carried out lower premised on technical solution of the present invention
Implement, give detailed embodiment and specific operating process.It should be pointed out that to one of ordinary skill in the art
For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the guarantor of the present invention
Protect scope.
Embodiment
The modeling method for pilot's working condition identification model that the present embodiment provides, comprises the following steps：
Step S1：Flying quality is pre-processed
Step S2：Feature extraction is carried out to pretreated flying quality；
Step S3：Data Dimensionality Reduction is carried out to the feature extracted using Treelets algorithms；
Step S4：The characteristic after dimensionality reduction is divided using the Gaussian process grader based on cuckoo algorithm optimization
Class (training modeling)；
Further, in step S1, longitudinal acceleration, side acceleration, normal acceleration, pitching in flying quality are chosen
Totally 11 flights are joined for angle, yaw angle, roll angle, ground velocity, rate of pitch, yaw rate, angular velocity in roll and angle of attack
Number, and flight parameter is normalized, obtain flight parameter data set；Flight parameter data set is carried out using 2 classes
Demarcation, i.e., {+1, -1 }, wherein ,+1 represents normal workload, and -1 represents non-normal working load, obtains pretreated fly
Row data.
Further, step 2 comprises the following steps：
Step S2.1：Time domain signal characteristics are introduced, extract the average, variance and root mean square of pretreated flying quality；
Step S2.2：Calculate the wavelet singular entropy of pretreated flying quality；
Step S2.2.1：Obtain pretreated flying quality sequence；
Step S2.2.2：Wavelet analysis is carried out to pretreated flying quality s (t) sequence；According to wavelet mother function
Wavelet transformation：Wherein, m is scale factor, and τ is shift factor；Wavelet transformation is
Number can be defined by following inner product formula：Flying quality obtains small echo after wavelet transformation
Coefficient matrices A；
Step S2.2.3：Singular value features value decomposition is carried out to matrix of wavelet coefficients A；A is decomposed into orthogonal matrix U, just
Hand over matrix VTWith diagonal matrix Λ, i.e. A=UAVT；Diagonal matrix Λ eigenvalue λi(i=1,2 ..., r) is the unusual of matrix A
Value tag value, and all singular value features value λ1≥λ2≥…≥λr＞ 0；
Step S2.2.4：λk/λ1＞ 0.01%, wherein λkFor the big characteristic value of the kth obtained according to step S2.2.3, λ1
For characteristic value maximum in step S2.2.3.According to information entropy principle, using the singular value after wavelet transformation as object, small echo is obtained
Singular entropy：
Further, the k=5.
Further, step S3 comprises the following steps：
Step S3.1：In l=0, i.e., the 0th layer, each observation signal x represents flying quality parameter x(0)=[s0,1...,
s0, p]T, s0, k=xk, calculate the covariance matrix of flying qualityWith similarity measurement
∑ij=E [(si-Esi)(sj-Esj)], Mij=| ρij|+λ|∑ij|；
Wherein：x(0)Represent the flying quality parameter matrix set that observation signal x is represented, s0, pRepresent in all flight parameters
P-th of flying quality, s0, kRepresent k-th of flying quality in all flight parameters.Now for x(0)I-th group in matrix and the
J group data, ∑ijRepresent x(0)Covariance matrix in set between i-th group of data and jth group data, ρijRepresent ∑ijNormalizing
The intermediate parameters obtained after change, EsiRepresent the desired value of i-th group of data, EsjRepresent the desired value of jth group data, MijRepresent the
Similarity measurement between i groups data and jth group data, λ represent weight parameter, can carry out related adjustment as needed.
Step S3.2：Since the first layer of the feature extracted, step S3.1 is repeated in, finds similarity measurement most
High covariance matrix, and the two-dimensional vector to finding carries out PCA conversion, obtains Jacobi's transformation matrix；
1st, similarity measurement highest matrix is found Wherein α and β representing matrixsThe position of most like two variables in matrix, maximizing in argmax representing matrixs,It is l-1 layer similarities
Matrix, coordinate position in i and j representing matrixs；
2nd, PCA conversion is carried out to the two-dimensional vector found, finds Jacobi's transformation matrix J (α, β, θl)；Wherein θiRepresent the
The angle of i variable rotation；
Wherein：
Wherein c=cos (θl), s=sin (θl), to xα, xβPCA dimensionality reductions are carried out, that is, find an anglec of rotation θl, | θl|
≤ π/4,The new base B in spacel=JTBl-1, new coordinate is x(l)=JTx(l-1), renewal
Similarity measurement
Step S3.3：After Jacobi's transformation, it will be assumed thatI.e. α, β are respectively first principal component and the second master
Composition, it is defined on l tree constructions, l layers are with variableThe poor variable of l layers isMeanwhile definition is forced
Nearly function and Detailfunction are respectively φl, ψl, as basic function Blα and β arrange.At this moment, on l layer tree constructions, we obtain
Treelets is represented：
For given base B=(w1..., wp), wherein p represents that the vectorial total x data of base are reduced to k and retain original letter
Number energy is：
Wherein：slRepresent the 1st layer and variable,Represent in l layer flight parameters, position be α data, dlRepresent the
The poor variable of l layers,Represent in l layer flight parameters, position is α data；
Further, step S4 comprises the following steps：
Step S4.1：The classification based training of Gauss two is carried out to the characteristic after dimensionality reduction；
Step S4.2：Cuckoo algorithm optimization training is carried out to the characteristic after the classification based training of Gauss two：
Step S4.2.1：Select majorized function：Log-likelihood function；
Step S4.2.2：Initiation parameter n, P are seta, MaxGeneration, wherein, n be Bird's Nest population quantity, Pa
The probability being found for parasitic Bird's Nest, i.e., the Bird's Nest ratio abandoned at random, MaxGeneration is maximum iteration；
Step S4.2.3：Flown by cuckoo Levy, produce a new Bird's Nest i, and evaluate Bird's Nest i fitness fi；
Step S4.2.4：Randomly choose a Bird's Nest j, evaluation Bird's Nest j fitness fj, and with fitness fiCompare；
Step S4.2.5：If fj＞ fi, substitute Bird's Nest i to turn into current the most new position with Bird's Nest j, otherwise Bird's Nest i is still
For latest position；
Step S4.2.6：Abandon PaThe Bird's Nest being found under probability, and sort and compare in remaining Bird's Nest, generate this
Take turns the current optimal solution after iteration；
Step S4.2.7：Judge whether current iteration number has been maxed out, if not up to, jumping to step
S4.2.3 carries out next round iteration, if it has, carrying out step S4.2.8；
Step S4.2.8：Optimal solution is exported, obtains pilot's working condition identification model.
Further, in the step S4.2.2, initiation parameter is arranged to：N=25, Pa=0.25,
MaxGeneration=1000.
Further, also comprise the following steps：
Step S5, for training remaining pretreated flying quality to know as test data to pilot's working condition
Other model is verified, obtains the precision of model, so as to the efficiency of evaluation model；Contrast Treelets dimensionality reductions and PCA simultaneously
Influence of the dimensionality reduction to model efficiency, obtains best identified model.
In flying quality pretreatment stage：
From flying quality storage device choose longitudinal acceleration, side acceleration, normal acceleration, the angle of pitch, yaw angle,
Roll angle, ground velocity, rate of pitch, yaw rate, angular velocity in roll and 11 flight parameters of angle of attack.In order to avoid difference
Flight parameter difference dimension is had an impact to result, and flight parameter is normalized.The contrast flight of corresponding data collection is appointed
Business terminates the NASA-TLX scale nominal data collection of record, is simply demarcated using 2 classes, i.e., {+1, -1 }, represents normal work
Load and non-normal working load.
In flying quality feature extraction phases：
Average, variance, root mean square and the wavelet singular entropy feature of flying quality after extraction pretreatment.
In the Feature Dimension Reduction stage：
Arbor dimensionality reduction (Treelets) algorithm is respectively adopted to drop the feature of extraction with principal component analysis (PCA) algorithm
Dimension processing.
In the Gaussian process modelling phase optimized based on cuckoo：
K folding crosss over model pair are carried out with demarcation collection to the feature set using Treelets algorithms and PCA algorithm dimensionality reductions respectively
Data set is trained, k=5, and cuckoo algorithm initialization parameter is arranged to, n=25, Pa=0.25, MaxGeneration=
1000。
In model Qualify Phase：
For training remaining data to be verified as test data, the precision of model is obtained, so as to evaluation model
Efficiency.Contrasting the influence of Treelets dimensionality reductions and PCA dimensionality reductions to model efficiency obtains best identified model simultaneously.
The specific embodiment of the present invention is described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make various deformations or amendments within the scope of the claims, this not shadow
Ring the substantive content of the present invention.
Claims (8)
1. a kind of modeling method of pilot's working condition identification model, it is characterised in that comprise the following steps：
Step S1：Flying quality is pre-processed；
Step S2：Feature extraction is carried out to pretreated flying quality；
Step S3：Data Dimensionality Reduction is carried out to the feature extracted using Treelets algorithms；
Step S4：The characteristic after dimensionality reduction is trained using the Gaussian process grader based on cuckoo algorithm optimization and built
Mould.
2. the modeling method of pilot's working condition identification model according to claim 1, it is characterised in that step S1
In, choose flying quality in longitudinal acceleration, side acceleration, normal acceleration, the angle of pitch, yaw angle, roll angle, ground velocity,
Rate of pitch, yaw rate, angular velocity in roll and angle of attack totally 11 flight parameters, and normalizing is carried out to flight parameter
Change is handled, and obtains flight parameter data set；Flight parameter data set is demarcated using 2 classes, i.e., {+1, -1 }, wherein ,+1 generation
Table normal workload, -1 represents non-normal working load, obtains pretreated flying quality.
3. the modeling method of pilot's working condition identification model according to claim 1, it is characterised in that the step
S2 comprises the following steps：
Step S2.1：Time domain signal characteristics are introduced, extract the average, variance and root mean square of pretreated flying quality；
Step S2.2：Calculate wavelet singular entropy：
Step S2.2.1：Obtain pretreated flying quality sequence；
Step S2.2.2：Wavelet analysis is carried out to pretreated flying quality sequence, obtains matrix of wavelet coefficients A；
Step S2.2.3：Singular value features value decomposition is carried out to matrix of wavelet coefficients A, obtains singular value features value λi(i=1,
2 ... ..., r), wherein r ∈ R, represent singular eigenvalue problem sum, and all singular value features value λ1≥λ2≥…≥λr≥0；
Step S2.2.4：λk/λ1＞ 0.01%, k rank wavelet singular entropies are calculated, wherein, λkFor big strange of kth in step S2.2.3
Different value tag value, λ1For singular value features value maximum in step S2.2.3.
4. the modeling method of pilot's working condition identification model according to claim 3, it is characterised in that the k=
5。
5. the modeling method of pilot's working condition identification model according to claim 1, it is characterised in that step S3 bags
Include following steps：
Step S3.1：At the 0th layer of the feature extracted, pretreated flying quality covariance matrix and similarity measurements are calculated
Amount；
Step S3.2：Since the 1st layer of the feature extracted, step S3.1 is repeated in, finds similarity measurement highest
Covariance matrix, and the two-dimensional vector to finding carries out PCA conversion, obtains Jacobi's transformation matrix；
Step S3.3：To the Jacobi's transformation matrix obtained in step S3.2, definition approximating function and Detailfunction, and according to
The parameter of Jacobi's transformation determines that Trelets is represented, carries out dimensionality reduction to given characteristic afterwards.
6. the modeling method of pilot's working condition identification model according to claim 1, it is characterised in that step S4 bags
Include following steps：
Step S4.1：The classification based training of Gauss two is carried out to the characteristic after dimensionality reduction；
Step S4.2：Cuckoo algorithm optimization training is carried out to the characteristic after the classification based training of Gauss two：
Step S4.2.1：Select majorized function：Log-likelihood function；
Step S4.2.2：Initiation parameter n, P are seta, MaxGeneration, wherein, n be Bird's Nest population quantity, PaTo post
The probability that raw Bird's Nest is found, i.e., the Bird's Nest ratio abandoned at random, MaxGeneration is maximum iteration；
Step S4.2.3：Flown by cuckoo Levy, produce a new Bird's Nest i, and evaluate Bird's Nest i fitness fi；
Step S4.2.4：Randomly choose a Bird's Nest j, evaluation Bird's Nest j fitness fj, and with fitness fiCompare；
Step S4.2.5：If fj＞ fi, substitute Bird's Nest i to turn into current the most new position with Bird's Nest j, otherwise Bird's Nest i is remained as most
New position；
Step S4.2.6：Abandon PaThe Bird's Nest being found under probability, and sort and compare in remaining Bird's Nest, generate this wheel and change
Current optimal solution after generation；
Step S4.2.7：Judge whether current iteration number has been maxed out, if not up to, jumping to step S4.2.3 and entering
Row next round iteration, if it has, carrying out step S4.2.8；
Step S4.2.8：Optimal solution is exported, obtains pilot's working condition identification model.
7. the modeling method of pilot's working condition identification model according to claim 6, it is characterised in that the step
In S4.2.2, initiation parameter is arranged to：N=25, Pa=0.25, MaxGeneration=1000；Wherein, n is Bird's Nest population
Quantity, PaThe probability being found for parasitic Bird's Nest, i.e., the Bird's Nest ratio abandoned at random, MaxGeneration is greatest iteration
Number.
8. the modeling method of pilot's working condition identification model according to any one of claim 1 to 7, its feature exist
In also comprising the following steps：
Step S5, for training remaining pretreated flying quality to identify mould to pilot's working condition as test data
Type is verified, obtains the precision of model, so as to the efficiency of evaluation model；Contrast Treelets dimensionality reductions and PCA dimensionality reductions simultaneously
Influence to model efficiency, obtain best identified model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710445742.9A CN107392226A (en) | 2017-06-13 | 2017-06-13 | The modeling method of pilot's working condition identification model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710445742.9A CN107392226A (en) | 2017-06-13 | 2017-06-13 | The modeling method of pilot's working condition identification model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107392226A true CN107392226A (en) | 2017-11-24 |
Family
ID=60332670
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710445742.9A Pending CN107392226A (en) | 2017-06-13 | 2017-06-13 | The modeling method of pilot's working condition identification model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107392226A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110472659A (en) * | 2019-07-05 | 2019-11-19 | 中国平安人寿保险股份有限公司 | Data processing method, device, computer readable storage medium and computer equipment |
| CN112949524A (en) * | 2021-03-12 | 2021-06-11 | 中国民用航空飞行学院 | Engine fault detection method based on empirical mode decomposition and multi-core learning |
| CN113313049A (en) * | 2021-06-11 | 2021-08-27 | 北京百度网讯科技有限公司 | Method, device, equipment, storage medium and computer program product for determining hyper-parameters |
| CN113344423A (en) * | 2021-06-28 | 2021-09-03 | 北京航空航天大学 | Pilot scene applicability diagnosis method and system based on machine learning |
| CN113554072A (en) * | 2021-07-09 | 2021-10-26 | 北京军懋国兴科技股份有限公司 | Flight action dividing method, computer readable storage medium and terminal equipment |
| CN116821644A (en) * | 2023-03-23 | 2023-09-29 | 南京航空航天大学 | Flight data identification method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106197424A (en) * | 2016-06-28 | 2016-12-07 | 哈尔滨工业大学 | The unmanned plane during flying state identification method that telemetry drives |
| CN106203526A (en) * | 2016-07-19 | 2016-12-07 | 中国人民解放军海军航空工程学院 | Target behavior pattern online classification method based on multidimensional characteristic |
-
2017
- 2017-06-13 CN CN201710445742.9A patent/CN107392226A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106197424A (en) * | 2016-06-28 | 2016-12-07 | 哈尔滨工业大学 | The unmanned plane during flying state identification method that telemetry drives |
| CN106203526A (en) * | 2016-07-19 | 2016-12-07 | 中国人民解放军海军航空工程学院 | Target behavior pattern online classification method based on multidimensional characteristic |
Non-Patent Citations (4)
| Title |
|---|
| ANN B. LEE 等: "TREELETS—AN ADAPTIVE MULTI-SCALE BASIS FOR SPARSE UNORDERED DATA", 《THE ANNALS OF APPLIED STATISTICS》 * |
| WANG QING等: "Aerodynamic Modeling and Parameter Estimation from QAR Data of an Airplane Approaching a High-altitude Airport", 《CHINESE JOURNAL OF AERONAUTICS》 * |
| 王珍珍 等: "铁路扣件图像边缘检测中LoG参数优化", 《计算机工程与应用》 * |
| 蔡正祥 等: "基于FPA 优化的GP 算法的飞行员认知状态识别", 《电光与控制》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110472659A (en) * | 2019-07-05 | 2019-11-19 | 中国平安人寿保险股份有限公司 | Data processing method, device, computer readable storage medium and computer equipment |
| CN110472659B (en) * | 2019-07-05 | 2024-03-08 | 中国平安人寿保险股份有限公司 | Data processing method, device, computer readable storage medium and computer equipment |
| CN112949524A (en) * | 2021-03-12 | 2021-06-11 | 中国民用航空飞行学院 | Engine fault detection method based on empirical mode decomposition and multi-core learning |
| CN113313049A (en) * | 2021-06-11 | 2021-08-27 | 北京百度网讯科技有限公司 | Method, device, equipment, storage medium and computer program product for determining hyper-parameters |
| CN113344423A (en) * | 2021-06-28 | 2021-09-03 | 北京航空航天大学 | Pilot scene applicability diagnosis method and system based on machine learning |
| CN113554072A (en) * | 2021-07-09 | 2021-10-26 | 北京军懋国兴科技股份有限公司 | Flight action dividing method, computer readable storage medium and terminal equipment |
| CN113554072B (en) * | 2021-07-09 | 2024-04-16 | 北京军懋国兴科技股份有限公司 | Flight action dividing method, computer readable storage medium and terminal equipment |
| CN116821644A (en) * | 2023-03-23 | 2023-09-29 | 南京航空航天大学 | Flight data identification method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107392226A (en) | The modeling method of pilot's working condition identification model | |
| CN104459668B (en) | radar target identification method based on deep learning network | |
| CN107122375B (en) | Image subject identification method based on image features | |
| CN110006649A (en) | A kind of Method for Bearing Fault Diagnosis based on improvement ant lion algorithm and support vector machines | |
| CN102915445A (en) | Method for classifying hyperspectral remote sensing images of improved neural network | |
| CN108629373B (en) | Image classification method, system, equipment and computer readable storage medium | |
| CN106628097A (en) | Ship equipment fault diagnosis method based on improved radial basis function neutral network | |
| Vatsavai et al. | Machine learning approaches for high-resolution urban land cover classification: a comparative study | |
| CN103632155A (en) | Remote-sensing image variation detecting method based on slow characteristic analysis | |
| CN103544499A (en) | Method for reducing dimensions of texture features for surface defect detection on basis of machine vision | |
| Lee et al. | Black-box classifier interpretation using decision tree and fuzzy logic-based classifier implementation | |
| Alburshaid et al. | Palm trees detection using the integration between gis and deep learning | |
| Zheng et al. | Research on feature extracted method for flutter test based on EMD and CNN | |
| Dharwadkar et al. | Floriculture classification using simple neural network and deep learning | |
| Kramer et al. | Wind power ramp event prediction with support vector machines | |
| Zhou et al. | Wavelet domain multi-view active learning for hyperspectral image analysis | |
| Al Iqbal | Eclectic rule extraction from Neural Networks using aggregated Decision Trees | |
| Marushko | Using ensembles of neural networks with different scales of input data for the analysis of telemetry data | |
| CN112101263B (en) | Active learning sample selection method and system and computer readable storage medium | |
| Liu et al. | Deep convolution neural networks for the classification of robot execution failures | |
| Wei et al. | Selection of optimal packaging methods for different food based on big data analysis | |
| CN112488146A (en) | Airplane danger identification and trend prediction method based on deep learning | |
| Zhou et al. | An Anomaly Detection Method for UAV Based on Wavelet Decomposition and Stacked Denoising Autoencoder | |
| Su et al. | Active learning based on SVM and representativity in a coal mining environment | |
| CN112001450A (en) | Ship inertial navigation system fault identification method |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171124 |
|
| RJ01 | Rejection of invention patent application after publication |