Method for identifying power quality disturbance of solar aircraft

Abstract

The invention relates to a solar aircraft electric energy quality disturbance identification method, and the method comprises the steps: obtaining an electric energy quality disturbance signal, obtaining a characteristic component color image and a wavelet time-frequency image according to the obtained electric energy quality disturbance signal, splicing the characteristic component color image and the wavelet time-frequency image to obtain a six-channel composite image, and carrying out the recognition of the electric energy quality disturbance of a solar aircraft. Performing feature fusion on the six-channel composite image through convolution to obtain a feature fusion image, and training a power quality disturbance recognition model according to the feature fusion image to obtain a trained power quality disturbance recognition model; the trained electric energy quality disturbance identification model can realize automatic identification of the electric energy quality disturbance problem of the solar aircraft, and the identification accuracy is high.

Description

Technical Field

[0001] The present application belongs to the field of power quality disturbance identification, and specifically relates to a method for identifying power quality disturbance of a solar aircraft. Background Art

[0002] Solar aircraft are equipped with solar panels that can absorb solar energy during the day to charge, thereby completing long-duration cruise and monitoring tasks. However, the solar charging process is related to the aircraft's flight attitude, atmospheric environment, meteorological conditions, and lighting conditions. The charging process inevitably faces the problem of power quality disturbance (PQD), which not only affects the charging efficiency of solar panels, but also affects the output voltage or power of solar panels, and even affects the stability of aircraft power output.

[0003] In the prior art, power quality disturbance identification can be accomplished by machine learning methods, such as random forest, support vector machine, and ResNet, etc. However, these methods require manual feature selection, which affects the recognition accuracy to a certain extent, and do not adaptively modify the network structure according to the specific dimensions of the input parameters. Under large input disturbances, the recognition network training effect is general, and it is impossible to accurately identify the power quality disturbance of solar aircraft, and the recognition accuracy is low. Summary of the invention

[0004] In order to solve the problem of low accuracy in identifying power quality disturbances of solar aircraft in the prior art, a method for identifying power quality disturbances of solar aircraft is proposed.

[0005] A method for identifying power quality disturbances of a solar aircraft, comprising:

[0006] Step 1: Obtain N power quality disturbance signals of the power supply system according to a fixed step length, each power quality disturbance signal includes an original disturbance signal and a normal signal, and obtain the noise component of the power quality disturbance signal by subtracting the original disturbance signal and the normal signal of each power quality disturbance signal;

[0007] Step 2: Perform variational modal decomposition on N power quality disturbance signals to obtain modal functions and N residual components, and each power quality disturbance signal corresponds to M modal functions and one residual component; obtain a power quality disturbance signal component matrix according to the noise components of the N power quality disturbance signals, the N original disturbance signals, the modal functions and the N residual components;

[0008] Step 3: Convert the component matrix into a pixel matrix to obtain a characteristic component color image; perform continuous wavelet transform on the original disturbance signals of N power quality disturbance signals to obtain a wavelet time-frequency image;

[0009] Step 4: splice the feature component color image and the wavelet time-frequency image to obtain a six-channel composite image, and perform feature fusion on the six-channel composite image through six-channel convolution to obtain a feature fusion image;

[0010] Step 5: training the power quality disturbance recognition model according to the feature fusion graph to obtain a trained power quality disturbance recognition model;

[0011] Step 6: Input the feature fusion graph of the power quality disturbance signal to be identified into the trained power quality disturbance identification model and output the identification result.

[0012] Beneficial Effects

[0013] The present application relates to a method for identifying power quality disturbances of solar aircraft. The method acquires a power quality disturbance signal, obtains a characteristic component color image and a wavelet time-frequency image according to the acquired power quality disturbance signal, splices the characteristic component color image and the wavelet time-frequency image to obtain a six-channel composite image, performs feature fusion on the six-channel composite image through convolution to obtain a feature fusion graph, trains a power quality disturbance identification model according to the feature fusion graph, and obtains a trained power quality disturbance identification model; the trained power quality disturbance identification model can realize automatic identification of power quality disturbance problems of solar aircraft with high identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flow chart of a method for identifying power quality disturbances of a solar aircraft according to a specific implementation of the present application;

[0015] Figure 2 This is a schematic diagram of component matrix x transformation according to a specific implementation of this application;

[0016] Figure 3 This is a schematic diagram of the splicing of the characteristic component color image and the wavelet time-frequency image according to a specific implementation mode of the present application;

[0017] Figure 4 This is a schematic diagram of the structure of a power quality disturbance identification model according to a specific implementation method of the present application. DETAILED DESCRIPTION

[0018] Specific implementation method 1: The following will be combined with the attached embodiment of the present invention Figure 1 To Attachment Figure 4 , illustrate this implementation mode, and clearly and completely describe the technical solutions in the embodiments of the present invention:

[0019] A method for identifying power quality disturbances of a solar aircraft, comprising:

[0020] Step 1: Obtain N power quality disturbance signals of the power supply system according to a fixed step length, each power quality disturbance signal includes an original disturbance signal and a normal signal, and obtain the noise component of the power quality disturbance signal by subtracting the original disturbance signal and the normal signal of each power quality disturbance signal;

[0021] Step 2: Perform variational modal decomposition on N power quality disturbance signals to obtain M×N modal functions and N residual components, and each power quality disturbance signal corresponds to M modal functions and one residual component; obtain the power quality disturbance signal component matrix x according to the noise components of the N power quality disturbance signals, the N original disturbance signals, the M×N modal functions and the N residual components;

[0022] Step 3: Convert the component matrix x into a pixel matrix to obtain a characteristic component color image; perform continuous wavelet transform on the original disturbance signals of the N power quality disturbance signals to obtain a wavelet time-frequency image;

[0023] Step 4: splice the feature component color image and the wavelet time-frequency image to obtain a six-channel composite image, and perform feature fusion on the six-channel composite image through six-channel convolution to obtain a feature fusion image;

[0024] Step 5: training the power quality disturbance recognition model according to the feature fusion graph to obtain a trained power quality disturbance recognition model;

[0025] Step 6: Input the feature fusion graph of the power quality disturbance signal to be identified into the trained power quality disturbance identification model and output the identification result.

[0026] Furthermore, according to the noise components of the N power quality disturbance signals, the N original disturbance signals, the M×N modal functions and the N residual components, the power quality disturbance signal component matrix x is obtained, including:

[0027] According to the noise components of N power quality disturbance signals, an n-dimensional row matrix of noise components is established; according to N original disturbance signals, an n-dimensional row matrix of original disturbance signals is established; according to M×N modal functions, an M×N-order matrix of modal functions is established; according to N residual components, an n-dimensional row matrix of residual components is established;

[0028] Arrange the noise component n-dimensional row matrix, the original disturbance signal n-dimensional row matrix, the modal function M×N-order matrix and the residual component n-dimensional row matrix in sequence to obtain the power quality disturbance signal component matrix x;

[0029] Among them, δy(t) is the n-dimensional row matrix of the noise component, y(t) is the n-dimensional row matrix of the original disturbance signal, is the M×N order matrix of the mode function, y RES(t) is the n-dimensional row matrix of residual components.

[0030] Furthermore, the calculation formula for converting the component matrix x into a pixel matrix to obtain the characteristic component color image is:

[0031] p m,n is the feature component color image, x m,n is the value of the mth row and nth column in the component matrix x, x max and x min They represent the minimum and maximum values ​​in the component matrix x respectively, and round(·) means rounding.

[0032] Furthermore, the calculation formula of the wavelet time-frequency image obtained by performing continuous wavelet transform on the original disturbance signals of N power quality disturbance signals is:

[0033] Among them, q s,f is the wavelet time-frequency image, s is the continuous wavelet transform scale, τ is the continuous wavelet transform translation parameter, ψ is the complex conjugate of the continuous wavelet transform scale and translation parameter, f is the frequency, and j represents an imaginary number.

[0034] Furthermore, the characteristic component color image and the wavelet time-frequency image are spliced ​​to obtain a six-channel composite image, including:

[0035] The sizes of the characteristic component color image and wavelet time-frequency image are normalized to 224×224×3 respectively; the normalized characteristic component color image is decomposed into three 224×224×1 grayscale images through the R channel, G channel and B channel respectively; the normalized wavelet time-frequency image is decomposed into three 224×224×1 grayscale images through the R channel, G channel and B channel respectively; each 224×224×1 grayscale image corresponds to a 224×224×1 pixel matrix; all 224×224×1 pixel matrices are spliced ​​in the order of image decomposition to obtain a 224×224×6 six-channel composite image.

[0036] Furthermore, the power quality disturbance identification model is an improved multi-channel ResNet network.

[0037] The number of convolution kernels in the improved multi-channel ResNet network is modified accordingly according to the size of the input feature fusion map, thus ensuring that the number of channels is adapted to the image size;

[0038]

[0039] Among them, l in and l out are the dimensions of the input feature matrix and the output feature matrix respectively, A u,v (lout ) is the l-dimensional feature matrix output after the input feature matrix is ​​calculated by l convolution kernels, x i,j is the element of the input matrix, k u-i,v-j is the internal element of l convolution kernels, B is the layer convolution bias, u, v represent the rows and columns of the output l-dimensional feature matrix.

[0040] Furthermore, the power quality disturbance identification model includes a feature extraction module, a feature enhancement module and an identification output module;

[0041] The feature extraction module is used to make the input power quality disturbance signal feature fusion map pass through the No. 1 convolution layer and the maximum pooling layer in sequence to obtain the feature map of the power quality disturbance signal, and input the power quality disturbance signal feature map into the feature enhancement module;

[0042] The feature enhancement module is used to make the input power quality disturbance signal feature map pass through the No. 1 residual module, the No. 2 residual module, the No. 3 residual module and the No. 4 residual module in sequence to obtain the power quality disturbance signal feature enhancement map, and input the power quality disturbance signal feature enhancement map to the output module;

[0043] The recognition output module is used to make the input power quality disturbance signal feature enhancement map pass through the average pooling layer, the first fully connected layer, the second fully connected layer, the third fully connected layer and the Softmax output in sequence, and output the recognition result; the recognition result includes the category and the probability value corresponding to the category.

[0044] Further, the No. 1 residual module is used to convolve and activate the input power quality disturbance signal feature map to obtain a first power quality disturbance signal feature enhancement map, and send the first power quality disturbance signal feature enhancement map to the No. 2 residual module;

[0045] The No. 1 residual module includes a No. 2 convolution layer, a No. 1 ReLU activation function, a No. 3 convolution layer, and a No. 2 ReLU activation function connected in sequence, and the No. 1 residual module also includes a No. 1 residual connection; the No. 1 residual connection includes a 1×1 convolution layer, and the No. 1 residual connection is used to convolve the input power quality disturbance signal feature map and add the convolved power quality disturbance signal feature map with the activated power quality disturbance signal feature map output by the No. 2 ReLU activation function to obtain a first power quality disturbance signal feature enhancement map;

[0046] The No. 2 residual module is used to convolve and activate the input power quality disturbance signal feature map to obtain a second power quality disturbance signal feature enhancement map, and send the second power quality disturbance signal feature enhancement map to the No. 3 residual module; the No. 2 residual module includes a No. 4 convolution layer, a No. 3 ReLU activation function, a No. 5 convolution layer, and a No. 6 ReLU activation function connected in sequence;

[0047] Residual module No. 3 is used to convolve and activate the input power quality disturbance signal feature map to obtain a third power quality disturbance signal feature enhancement map, and send the third power quality disturbance signal feature enhancement map to residual module No. 4; residual module No. 3 includes convolution layer No. 6, ReLU activation function No. 5, convolution layer No. 7, and ReLU activation function No. 6 connected in sequence, and residual module No. 3 also includes residual connection No. 2; residual connection No. 2 includes a 1×1 convolution layer, and residual connection No. 2 is used to convolve the input second power quality disturbance signal feature map and add the convolved second power quality disturbance signal feature map with the activated second power quality disturbance signal feature map output by ReLU activation function No. 6 to obtain the third power quality disturbance signal feature enhancement map;

[0048] Residual module No. 4 is used to convolve and activate the input power quality disturbance signal feature map to obtain the power quality disturbance signal feature enhancement map, and send the power quality disturbance signal feature enhancement map to the recognition output module; residual module No. 4 includes convolution layer No. 8, ReLU activation function No. 7, convolution layer No. 9, and ReLU activation function No. 8 connected in sequence.

[0049] Furthermore, the number of full connections of the No. 1 fully connected layer is 2000, the number of full connections of the No. 2 fully connected layer is 1000, and the number of full connections of the No. 3 fully connected layer is 1000. Specifically, the fully connected layer can ensure that the feature quantity is adapted to the number of channels.

[0050] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. A method for identifying power quality disturbances of solar aircraft, characterized in that: include: Step 1: Obtain N power quality disturbance signals of the power supply system according to a fixed step length, each power quality disturbance signal includes an original disturbance signal and a normal signal, and obtain the noise component of the power quality disturbance signal by subtracting the original disturbance signal and the normal signal of each power quality disturbance signal; Step 2: Perform variational modal decomposition on N power quality disturbance signals to obtain M×N modal functions and N residual components. Each power quality disturbance signal corresponds to M modal functions and one residual component. A power quality disturbance signal component matrix x is obtained according to the noise components of the N power quality disturbance signals, the N original disturbance signals, the M×N modal functions and the N residual components; Step 3: Convert the component matrix x into a pixel matrix to obtain a characteristic component color image; perform continuous wavelet transform on the original disturbance signals of the N power quality disturbance signals to obtain a wavelet time-frequency image; Step 4: splice the feature component color image and the wavelet time-frequency image to obtain a six-channel composite image, and perform feature fusion on the six-channel composite image through six-channel convolution to obtain a feature fusion image; Step 5: training the power quality disturbance recognition model according to the feature fusion graph to obtain a trained power quality disturbance recognition model; Step 6: Input the feature fusion graph of the power quality disturbance signal to be identified into the trained power quality disturbance identification model and output the identification result.

2. A method for identifying power quality disturbances of a solar powered aircraft according to claim 1, characterized in that: According to the noise components of N power quality disturbance signals, N original disturbance signals, M×N modal functions and N residual components, the power quality disturbance signal component matrix x is obtained, including: According to the noise components of N power quality disturbance signals, an n-dimensional row matrix of noise components is established; according to N original disturbance signals, an n-dimensional row matrix of original disturbance signals is established; according to M×N modal functions, an M×N-order matrix of modal functions is established; according to N residual components, an n-dimensional row matrix of residual components is established; Arrange the noise component n-dimensional row matrix, the original disturbance signal n-dimensional row matrix, the modal function M×N-order matrix and the residual component n-dimensional row matrix in sequence to obtain the power quality disturbance signal component matrix x; Among them, δy(t) is the n-dimensional row matrix of the noise component, y(t) is the n-dimensional row matrix of the original disturbance signal, is the M×N order matrix of the mode function, y RES (t) is the n-dimensional row matrix of residual components.

3. A method for identifying power quality disturbances of solar powered aircraft according to claim 1, characterized in that: The calculation formula for converting the component matrix x into a pixel matrix to obtain the characteristic component color image is: p m,n is the feature component color image, x m,n is the value of the mth row and nth column in the component matrix x, x max and x min They represent the minimum and maximum values ​​in the component matrix x respectively, and round(·) means rounding.

4. A method for identifying power quality disturbances of a solar powered aircraft according to claim 3, characterized in that: The calculation formula of the wavelet time-frequency image obtained by continuous wavelet transform of the original disturbance signals of N power quality disturbance signals is: Among them, q s,f is the wavelet time-frequency image, s is the continuous wavelet transform scale, τ is the continuous wavelet transform translation parameter, ψ is the complex conjugate of the continuous wavelet transform scale and translation parameter, f is the frequency, and j represents an imaginary number.

5. A method for identifying power quality disturbances of solar powered aircraft according to claim 4, characterized in that: The characteristic component color image and the wavelet time-frequency image are spliced ​​to obtain a six-channel composite image, including: The sizes of the characteristic component color image and wavelet time-frequency image are normalized to 224×224×3 respectively; the normalized characteristic component color image is decomposed into three 224×224×1 grayscale images through the R channel, G channel and B channel respectively; the normalized wavelet time-frequency image is decomposed into three 224×224×1 grayscale images through the R channel, G channel and B channel respectively; each 224×224×1 grayscale image corresponds to a 224×224×1 pixel matrix; all 224×224×1 pixel matrices are spliced ​​in the order of image decomposition to obtain a 224×224×6 six-channel composite image.

6. The method for identifying power quality disturbance of solar powered aircraft according to claim 1, characterized in that: The power quality disturbance identification model is an improved multi-channel ResNet network.

7. A method for identifying power quality disturbances of a solar powered aircraft according to claim 5, characterized in that: The power quality disturbance identification model includes a feature extraction module, a feature enhancement module and an identification output module; The feature extraction module is used to make the input power quality disturbance signal feature fusion map pass through the No. 1 convolution layer and the maximum pooling layer in sequence to obtain the feature map of the power quality disturbance signal, and input the power quality disturbance signal feature map into the feature enhancement module; The feature enhancement module is used to make the input power quality disturbance signal feature map pass through the No. 1 residual module, the No. 2 residual module, the No. 3 residual module and the No. 4 residual module in sequence to obtain the power quality disturbance signal feature enhancement map, and input the power quality disturbance signal feature enhancement map to the output module; The recognition output module is used to make the input power quality disturbance signal feature enhancement map pass through the average pooling layer, the first fully connected layer, the second fully connected layer, the third fully connected layer and the Softmax output in sequence, and output the recognition result; the recognition result includes the category and the probability value corresponding to the category.

8. A method for identifying power quality disturbances of a solar powered aircraft according to claim 7, characterized in that: The No. 1 residual module is used to convolve and activate the input power quality disturbance signal feature map to obtain a first power quality disturbance signal feature enhancement map, and send the first power quality disturbance signal feature enhancement map to the No. 2 residual module; The No. 1 residual module includes a No. 2 convolution layer, a No. 1 ReLU activation function, a No. 3 convolution layer, and a No. 2 ReLU activation function connected in sequence, and the No. 1 residual module also includes a No. 1 residual connection; the No. 1 residual connection includes a 1×1 convolution layer, and the No. 1 residual connection is used to convolve the input power quality disturbance signal feature map and add the convolved power quality disturbance signal feature map with the activated power quality disturbance signal feature map output by the No. 2 ReLU activation function to obtain a first power quality disturbance signal feature enhancement map; The No. 2 residual module is used to convolve and activate the input power quality disturbance signal feature map to obtain a second power quality disturbance signal feature enhancement map, and send the second power quality disturbance signal feature enhancement map to the No. 3 residual module; the No. 2 residual module includes a No. 4 convolution layer, a No. 3 ReLU activation function, a No. 5 convolution layer, and a No. 6 ReLU activation function connected in sequence; Residual module No. 3 is used to convolve and activate the input power quality disturbance signal feature map to obtain a third power quality disturbance signal feature enhancement map, and send the third power quality disturbance signal feature enhancement map to residual module No. 4; residual module No. 3 includes convolution layer No. 6, ReLU activation function No. 5, convolution layer No. 7, and ReLU activation function No. 6 connected in sequence, and residual module No. 3 also includes residual connection No. 2; residual connection No. 2 includes a 1×1 convolution layer, and residual connection No. 2 is used to convolve the input second power quality disturbance signal feature map and add the convolved second power quality disturbance signal feature map with the activated second power quality disturbance signal feature map output by ReLU activation function No. 6 to obtain the third power quality disturbance signal feature enhancement map; Residual module No. 4 is used to convolve and activate the input power quality disturbance signal feature map to obtain the power quality disturbance signal feature enhancement map, and send the power quality disturbance signal feature enhancement map to the recognition output module; residual module No. 4 includes convolution layer No. 8, ReLU activation function No. 7, convolution layer No. 9, and ReLU activation function No. 8 connected in sequence.

9. A method for identifying power quality disturbances of a solar powered aircraft according to claim 8, characterized in that: The number of full connections of fully connected layer 1 is 2000, the number of full connections of fully connected layer 2 is 1000, and the number of full connections of fully connected layer 3 is 1000.

Images