Feature extraction method for flight parameter data based on BPCA-SPCA

Abstract

The application discloses a flight parameter data feature extraction method based on BPCA-SPCA, and specifically comprises the following steps: obtaining a required data sample set, and formulating a standardized data set; based on the standardized data set, estimating load coefficients and variances according to an expectation maximization algorithm until the load coefficients and variances converge; initializing A=[α1,...,α k ] wherein the load matrix obtained in step two is taken as an initialization matrix of the A matrix in SPCA; the number of principal components automatically learned in BPCA is assigned to the dimension of the A matrix array vector, namely k=d; based on the given A, the elastic net model is calculated about the estimation value; based on the calculation result of the estimation value, the matrix (X T C)B is given, the matrix (X T C)B is singular value decomposed until the model converges; and the load vector is returned to realize sparse extraction of data feature vectors. The principal components of the flight parameter data extracted based on the BPCA-SPCA can effectively express the original data features, the feature sparse vectors can provide effective input information for subsequent fault diagnosis, and the efficiency of the fault diagnosis is indirectly improved.

Description

Technical Field

[0001] This invention relates to the field of data extraction technology, specifically to a method for extracting flight parameter data features based on BPCA-SPCA. Background Technology

[0002] With the increasing integration of aircraft systems and the rapid development of sensor monitoring technology, flight parameter data during flight testing and operation are becoming more diverse and the sampling volume is much larger. The large volume and high complexity of flight parameter data lead to low efficiency in system fault diagnosis and prediction based on flight parameter data.

[0003] Bayesian Principal Component Analysis (BPCA) projects high-dimensional data onto several directions with the largest covariance matrix, decomposing the high-dimensional data into several principal directions with the largest cocumulative variance. Using information from a few principal directions to approximate the high-dimensional spatial data information can only reduce the dimensionality of the data.

[0004] Sparse Principal Component Analysis (SPC) combines principal component analysis with a linear regression model to obtain principal component vectors and sparse data. However, using either of these methods alone is inefficient and can negatively impact subsequent tasks such as aircraft system fault diagnosis.

[0005] Therefore, it is urgent to study a flight parameter data feature extraction method based on BPCA-SPCA, which is of great significance for improving the efficiency of operation and maintenance work. Summary of the Invention

[0006] The purpose of this invention is to provide a flight parameter data feature extraction method based on BPCA-SPCA, which solves the problem that the efficiency of using multiple methods individually in the prior art is low and can easily have an adverse effect on subsequent aircraft system fault diagnosis and other work.

[0007] To achieve the above objectives, this invention provides the following technical solution: a flight parameter data feature extraction method based on BPCA-SPCA, specifically including the following steps: using the Laida criterion to reduce data noise and smooth out outliers between different groups; then, BPCA automatically determines the data dimensionality reduction dimension; and finally, SPCA performs sparsity processing on the BPCA-reduced data to achieve data feature extraction, specifically including:

[0008] Step 1: Obtain the required data sample set and develop a standardized dataset;

[0009] Step 2: Based on the standardized dataset, estimate the loading coefficients and variances using the expectation-maximization algorithm until the loading coefficients and variances converge.

[0010] Step 3: Initialize A = [α1, ..., α] k ], where the load matrix obtained in step two is... As the initialization matrix of matrix A in SPCA; the number of principal components d automatically learned in BPCA is assigned to the dimension of the column vector of matrix A, i.e., k = d;

[0011] Step 4: Calculate the elastic network model based on the given A. about The estimated value;

[0012] Step 5, based on step 4 The calculation results, given the matrix (X) T X)B, array(X)B T X)B singular value decomposition;

[0013] Step 6: Repeat steps 4 and 5 until the model converges;

[0014] Step 7: Load vector normalization. Achieve sparse extraction of data feature vectors.

[0015] Preferably, the data sample set in step 1 is specifically: The standardized data specifically refers to,

[0016]

[0017] Preferably, the specific formula for calculating the load factor in step 2 is as follows: The specific formula for calculating the variance is as follows: in

[0018] Preferably, in step 4, when estimating λ and λ 1，j At that time, using a relatively small fixed value λ, the regularization coefficient λ of the L1 penalty is determined by cross-validation. 1，j .

[0019] Preferably, in step 5, the singular value decomposition of matrix B is specifically...

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. This invention first uses the Laida criterion to reduce noise in flight parameter data and smooth out outliers between different groups. Then, BPCA automatically determines the dimensionality reduction dimension of the flight parameter data. Based on SPCA, the data after BPCA dimensionality reduction is sparsely processed to achieve BPCA-SPCA flight parameter data feature extraction. The principal components extracted based on BPCA-SPCA flight parameter data can effectively express the original data features. The sparse feature vector can provide effective input information for subsequent fault diagnosis, indirectly improving the efficiency of fault diagnosis.

[0022] 2. In this invention, the landing gear system, which has a high failure rate in aircraft systems, is selected as the analysis object. The flight parameter data features of the system are extracted, and the data sparsity is verified by the method proposed in this invention. It can effectively express the original data features, and the feature sparse vector can provide effective input information for subsequent fault diagnosis, thereby indirectly improving the efficiency of fault diagnosis. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the model of the flight parameter data feature extraction method based on BPCA-SPCA proposed in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1: This invention proposes a flight parameter data feature extraction method based on BPCA-SPCA. It employs the Laida criterion for data denoising and smooths outliers between different groups. Then, BPCA automatically determines the dimensionality reduction dimension, and SPCA performs sparsity processing on the BPCA-reduced data to achieve feature extraction. Specifically, this includes:

[0026] Step 1: Obtain the required data sample set, specifically... And a standardized dataset was developed, specifically,

[0027] Step 2: Based on the standardized dataset, estimate the loading coefficients and variance using the expectation-maximization algorithm until the loading coefficients and variance converge. The specific formula for calculating the loading coefficients is as follows: The specific formula for calculating variance is as follows:

[0028] Step 3: Initialize A = [α1, ..., α] k ], where the load matrix obtained in step two is... As the initialization matrix of matrix A in SPCA; the number of principal components d automatically learned in BPCA is assigned to the dimension of the column vector of matrix A, i.e., k = d;

[0029] Step 4: Calculate the elastic network model based on the given A. about The estimated values, in estimating λ and λ 1，j At that time, using a relatively small fixed value λ, the regularization coefficient λ of the L1 penalty is determined by cross-validation. 1，j ;

[0030] Step 5, based on step 4 The calculation results give the matrix (X) T X)B,) matrix(X)B,) T X)B singular value decomposition, the specific decomposition formula is as follows:

[0031] Step 6: Repeat steps 4 and 5 until the model converges;

[0032] Step 7: Load vector normalization. Achieve sparse extraction of data feature vectors.

[0033] This study proposes a flight parameter data feature extraction method based on BPCA-SPCA. Before implementing this method, the Laida criterion is used to reduce noise in the flight parameter data and smooth out outliers between different groups. The dimensionality reduction dimension of the flight parameter data is automatically determined based on BPCA, and the data after dimensionality reduction by BPCA is sparsely processed based on SPCA to achieve BPCA-SPCA flight parameter data feature extraction.

[0034] Example 2: The landing gear system, which has a high failure rate in aircraft systems, is selected as the analysis object. Flight parameter data including system (force (P), (degree (V), left / right brake command (I), (pressure) U) and other monitoring parameters are selected from the flight parameter data of the landing gear braking system (as shown in Table 1). The noise reduction model of typical aircraft system flight parameter data based on the Laida criterion is verified, and the feature extraction process of typical aircraft system flight parameter data based on the BPCA-SPCA model is verified.

[0035] Table 1. Sample flight parameter data before noise reduction (partial)

[0036]

[0037]

[0038] Each column of sample data first uses the Laida criterion to denoise the flight parameter data and smooth out outliers between different groups. Then, after normalization using the above method, the sample data is adjusted to obtain a new sample data table, as shown in Table 2, which contains some of the denoised flight parameter data samples.

[0039] Table 2. Sample flight parameter data after noise reduction (partial)

[0040]

[0041] According to the BPCA algorithm steps, the loading coefficient vector matrix of all the above-mentioned monitoring parameter variables is obtained. The PCs (Principal Components) under the first few largest variance percentages are extracted in the BPCA process. The number of principal components after BPCA training is used as the initial input dimension of SPCA. According to the SPCA model algorithm steps above, the loading coefficient vector matrix of all monitoring parameter variables of SPCA is obtained. See the table below for details. The table below is a comparison of PCA, BPCA and BPCA-SPCA models.

[0042] Model Cumulative total variance Percentage of non-zero loads Convergence time PCA 100％ 100％ 19.285s BPCA 97.4％ 100％ 12.936s BPCA-SPCA 89.7％ 57.4％ 17.855s

[0043] In summary, regarding the total cumulative variance, due to the sparsity of the PCs load vector coefficients, the percentage of the final cumulative total variance of the BPCA-SPCA model is smaller than that of the traditional PCA. However, in terms of sparsity representation, the BPCA-SPCA model has a significant advantage over both traditional PCA and BPCA. It can be seen that BPCA-SPCA has a better ability to dilute data and can quickly obtain principal component vectors, effectively representing the features of the original data. The sparse feature vectors can provide effective input information for subsequent fault diagnosis, indirectly improving the efficiency of fault diagnosis.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for feature extraction of flight parameter data based on BPCA-SPCA, characterized in that, The Laida criterion is used to reduce data noise and smooth out outliers between different groups. Then, BPCA automatically determines the dimensionality reduction dimension. Finally, SPCA performs sparsity processing on the BPCA-reduced data to achieve feature extraction, specifically including: Step 1: Obtain the required data sample set and develop a standardized dataset; Step 2: Based on the standardized dataset, estimate the loading coefficients and variance using the expectation-maximization algorithm until the loading coefficients and variance converge; the specific formula for calculating the loading coefficients is as follows: Step 3: Initialize A = [α1,...,α] k ], where the result obtained in step two As the initialization matrix of matrix A in SPCA; the number of principal components d automatically learned in BPCA is assigned to the dimension of the column vector of matrix A, i.e., k = d; Step 4: Calculate the elastic network model based on the given A. about The estimated value; Step 5, based on step 4 The calculation results for matrix (X) T X)B singular value decomposition; Step 6: Repeat steps 4 and 5 until the model converges; Step 7: Load vector normalization. j = 1, ..., k, to achieve sparse extraction of data feature vectors.

2. The method for extracting flight parameter data features based on BPCA-SPCA according to claim 1, characterized in that: The data sample set in step 1 is specifically as follows: The standardized data specifically refers to, 3. The method for extracting flight parameter data features based on BPCA-SPCA according to claim 1, characterized in that: The specific formula for calculating the load factor in step 2 is as follows: The specific formula for calculating the variance is as follows: in 4. The method for extracting flight parameter data features based on BPCA-SPCA according to claim 1, characterized in that: In step 4, λ and λ are estimated. 1，j At that time, using a relatively small fixed value λ, the regularization coefficient λ of the L1 penalty is determined by cross-validation. 1,j .

5. The method for extracting flight parameter data features based on BPCA-SPCA according to claim 1, characterized in that: In step 5 (X) T X)B singular value decomposition, specifically

Images