A method and device for detecting attachments on ocean current generator blades under variable operating conditions

Abstract

The present invention relates to a method and device for detecting attachments on ocean current generator blades under variable operating conditions. The method comprises: performing Hilbert transform and variational modal decomposition on multiple sets of collected ocean current generator electrical signals to obtain corresponding instantaneous frequency signals and their characteristic components; processing the characteristic components of the instantaneous frequency signals using a K-means clustering method by randomly initializing cluster centers and numbers to determine pre-division information of the ocean current generator operating conditions; selecting samples with the smallest mutual information entropy within each pre-divided operating condition as initialization cluster centers and performing secondary clustering; determining the number of operating conditions using a clustering evaluation index to obtain the optimal number of divided operating conditions; and establishing a PCA detection model for blade attachments within each determined optimal divided operating condition, outputting the detection results. Compared with the prior art, the present invention improves the flexibility and accuracy of detecting attachments on ocean current generator blades under variable operating conditions by dividing the operating conditions.

Description

Technical Field

[0001] The present invention relates to the field of new energy power generation status monitoring, and in particular to a method and device for detecting attachments on ocean current generator blades under variable working conditions based on mutual information clustering. Background Art

[0002] In recent years, ocean current energy has attracted widespread attention as a renewable energy source, offering advantages such as cleanliness, reliability, high energy density, and predictability. However, ocean current generators operate in complex environments over long periods of time, and their blades are susceptible to buildup, which can reduce power generation quality. If not promptly addressed, this can accelerate wear and corrosion, leading to irreversible failures such as blade breakage. Therefore, it is crucial to monitor the machine's status and perform repairs or promptly remove buildup to prevent system damage.

[0003] In addition, affected by factors such as tidal flow velocity, salinity, and temperature, ocean current generators usually operate under variable flow rates. Changes in flow velocity and blade attachments will affect the amplitude and frequency of the stator current, reducing the accuracy of ocean current generator blade attachment detection under variable operating conditions. Existing ocean current generator blade attachment detection methods based on stator electrical signals mostly perform detection under a single operating condition, ignoring the impact of changes in operating conditions on the stator electrical signals of ocean current generators. The accuracy of blade attachment detection is poor, with high missed reporting and false detection rates. Summary of the Invention

[0004] The present invention aims to provide a method and device for detecting debris on ocean current generator blades under variable operating conditions based on mutual information clustering. This method overcomes the problem of low accuracy in detecting debris on ocean current generator blades under variable operating conditions. The most important aspect of the present invention is that it utilizes operating condition classification after preprocessing the stator current to address the problem of adaptive adaptation to the current generator's current operating conditions, thereby improving the accuracy of detecting debris on ocean current generator blades under variable operating conditions.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A method for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering comprises the following steps:

[0007] Step 1) Collect multiple sets of electrical signals from ocean current generators Perform Hilbert transform and variational mode decomposition to obtain the corresponding instantaneous frequency signal and its characteristic components

[0008] Step 2) Use K-means clustering method to process the characteristic components of instantaneous frequency signal by randomly initializing the cluster center and number Determine the pre-division information of the ocean current generator working conditions

[0009] Step 3) Select mutual information entropy in each pre-divided working condition The smallest sample is used as the initial cluster center for secondary clustering;

[0010] Step 4) Use the clustering evaluation index S_Dbw to determine the number of working conditions and obtain the optimal number of divided working conditions

[0011]

[0012] Step 5) Establish a PCA detection model for blade attachments in each determined optimal division condition and output the detection results.

[0013] In step 1), in order to enhance the blade attachment feature, the stator electrical signal of the ocean current generator is pre-processed. The principle of enhancing the attachment feature is: the Hilbert transform method is used to pre-process the single ocean current generator electrical signal C to obtain the instantaneous frequency signal f e , further use the variational mode decomposition VMD method to decompose the instantaneous frequency to obtain m signal components u 1,i (i=1,2,...,m), and extract the signal components containing the characteristics of leaf attachments. On the basis of discarding other irrelevant components, the characteristics of leaf attachments are enhanced.

[0014] The step 2) is specifically as follows: using multiple sets of instantaneous frequency signals The characteristic component containing the characteristics of leaf attachment Randomly initialize the number of clusters b and cluster centers a b , perform a preliminary division of the current working condition and divide each sample into the corresponding initial working condition:

[0015]

[0016] Among them, γ (b) is the sample matrix divided into the bth working condition, u 1i is the i-th independent sample in the first signal component.

[0017] The step 3) comprises the following steps:

[0018] Step 3-1) Determine each pre-divided working condition γ obtained in step 2) (b) , where γ (b) Represents the sample matrix of the bth working condition;

[0019] Step 3-2) Determine the mutual information entropy between samples in each pre-divided working condition:

[0020]

[0021] in, and are independent samples in the bth pre-divided working condition, K b is the number of samples in the bth operating condition, and Represents samples and The probability of Representation sample and The joint probability distribution of

[0022] Step 3-3) Use the strategy of minimizing entropy to select the central sample, and obtain the sample with the minimum mutual information entropy in each pre-divided working condition as follows:

[0023]

[0024] Among them, v b is the central sample of the bth working condition;

[0025] Step 3-4) The sample with the smallest mutual information entropy in each working condition is used as the initial working condition center, and secondary clustering is performed to obtain clustering results under different initialization cluster numbers.

[0026] The clustering evaluation index is:

[0027] S_Dbw=Scat+Dens_bw

[0028] Among them, Scat and Dens_bw are the intra-class density and inter-class density of each cluster sample respectively.

[0029] The optimal number of divided working conditions is the number of divided working conditions corresponding to the minimum clustering evaluation index.

[0030] That is, the present invention utilizes the proposed mutual information clustering method and multiple groups of signal instantaneous frequencies The amount produced The mutual information clustering method proposed in this invention helps the traditional K-means clustering method to select reliable working condition cluster centers. The principle is: the current working condition of the ocean current generator is pre-divided by using K-means clustering, and the number of clusters K1 and the cluster center a are randomly initialized. b(b=1,2,...,K1) pre-divide the current working condition and calculate the initial category of each sample; use the mutual information entropy to initialize the working condition center, and obtain the mutual information entropy between its samples and the sample with the smallest mutual information entropy in each pre-divided working condition; use the sample with the smallest mutual information entropy in each working condition as the initialized working condition center, and use the clustering evaluation index to determine the number of working conditions. The clustering evaluation index S_Dbw is determined by the intra-class compactness and inter-class density. The higher the intra-class compactness, the better, and the lower the inter-class density, the better. Therefore, the smaller the S_Dbw index, the better the working condition division effect. The clustering result with the smallest S_Dbw index is selected as the optimal working condition division result.

[0031] The present invention can provide a reliable working condition initialization center and working condition number for the division of the current working condition of the ocean current generator, thereby improving the accuracy of the division of the current working condition of the ocean current generator. Further, a principal component analysis model is established in each working condition interval to detect blade attachments. The statistics and control limits of each principal component analysis model are calculated respectively. If the statistics exceed its control limit, it indicates that the blades of the ocean current generator are attached by attachments. By dividing the current working condition of the ocean current generator instead of detecting under multiple working conditions together, the influence of working condition changes on the stator electrical signal of the ocean current generator is avoided, the problem of low accuracy in detecting blade attachments of the ocean current generator under variable working conditions is solved, and the flexibility and accuracy of detecting blade attachments of the ocean current generator under variable working conditions is improved.

[0032] A device for detecting attachments on blades of ocean current generators under variable operating conditions based on mutual information clustering, comprising:

[0033] An acquisition module is used to collect multiple sets of electrical signals from ocean current generators;

[0034] The signal preprocessing module is used to perform Hilbert transform and variational mode decomposition on multiple sets of electrical signals collected from the ocean current generator to obtain the corresponding instantaneous frequency signals and their characteristic components;

[0035] The preliminary clustering module uses the K-means clustering method to process the characteristic components of the instantaneous frequency signal by randomly initializing the cluster center and number, and determines the pre-division information of the ocean current generator working conditions;

[0036] The secondary clustering module selects the sample with the smallest mutual information entropy in each pre-divided working condition as the initial cluster center for secondary clustering;

[0037] The module for determining the number of optimal divided working conditions is used to determine the number of working conditions by using clustering evaluation indicators to obtain the optimal number of divided working conditions;

[0038] The detection module is used to establish a PCA detection model for blade attachments in each determined optimal division working condition and output the detection results.

[0039] The initial clustering module performs the following steps: using the characteristic component containing the leaf attachment feature in the instantaneous frequency signal to randomly initialize the cluster number b and cluster center a b , perform a preliminary division of the current working condition and divide each sample into the corresponding initial working condition:

[0040]

[0041] Among them, γ (b) is the sample matrix divided into the bth working condition, u 1i is the i-th independent sample in the first signal component.

[0042] The secondary clustering module performs the following steps:

[0043] Step 3-1) Determine the pre-divided working conditions γ obtained by the preliminary clustering module (b) , where γ (b) Represents the sample matrix of the bth working condition;

[0044] Step 3-2) Determine the mutual information entropy between samples in each pre-divided working condition:

[0045]

[0046] in, and are independent samples in the bth pre-divided working condition, K b is the number of samples in the bth operating condition, and Represents samples and The probability of Representation sample and The joint probability distribution of

[0047] Step 3-3) Use the strategy of minimizing entropy to select the central sample, and obtain the sample with the minimum mutual information entropy in each pre-divided working condition as follows:

[0048]

[0049] Among them, v b is the central sample of the bth working condition;

[0050] Step 3-4) The sample with the smallest mutual information entropy in each working condition is used as the initial working condition center, and secondary clustering is performed to obtain clustering results under different initialization cluster numbers.

[0051] The clustering evaluation index is:

[0052] S_Dbw=Scat+Dens_bw

[0053] Among them, Scat and Dens_bw are the intra-class density and inter-class density of each cluster sample respectively.

[0054] The optimal number of divided working conditions is the number of divided working conditions corresponding to the minimum clustering evaluation index.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] (1) The present invention takes into account the characteristics of the current working conditions of the ocean current generator and uses the mutual information clustering method to divide the current working conditions. This method improves the traditional K-means clustering method, selects a reliable initial working condition center for the working condition division, and improves the accuracy of the ocean current generator working condition division.

[0057] (2) The present invention divides the working conditions based on mutual information clustering and detects blade attachments in each working condition of the ocean current generator, thereby avoiding the impact of working condition changes on the stator electrical signal of the ocean current generator, solving the problem of variable working conditions of the ocean current generator under variable flow rates, and improving the flexibility and accuracy of detecting blade attachments of the ocean current generator under variable working conditions.

[0058] (3) The present invention is also applicable to the field of blade fault detection in renewable energy power generation under complex environments and has a wide range of applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flow chart of the method of the present invention;

[0060] Figure 2 Schematic diagram of central samples of various working conditions obtained by minimizing mutual information entropy in one embodiment;

[0061] Figure 3 This is a visualization diagram of the working condition division results based on mutual information clustering in one embodiment. DETAILED DESCRIPTION

[0062] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0063] The present invention proposes a method for detecting attachments on ocean current generator blades under variable working conditions based on mutual information clustering. Figure 1The following is a flow chart of the present invention. The following describes the detection of blade attachments on ocean current generators using the mutual information clustering method proposed in the present invention. This method primarily involves three steps: signal acquisition and preprocessing, operating condition classification using mutual information clustering, and blade attachment detection. The specific steps of the present invention are illustrated using faulty electrical signals from ocean current generator blades collected from an ocean current power generation system experimental platform. The following is a sample:

[0064] (1) Signal acquisition and preprocessing

[0065] This section corresponds to Figure 1 Specifically, this example collected stator electrical signals from a tidal power generation experimental platform, both in a healthy state and in a blade attachment state. The experimental platform primarily consists of a turbine (with a blade radius of 0.3 m), a direct-drive permanent magnet synchronous ocean current generator with an 8-pole pair count, a data acquisition and status monitoring system, and a flow system simulating submarine tidal currents. The sampling frequency was 1 kHz. In this experimental system, ropes with mass attached to the ocean current generator blades were used to simulate blade attachments, and their impact on the stator electrical signals was observed.

[0066] The above experimental system is used to simulate the running state of the ocean current generator blade when it is attached. When an 80g attachment is attached to a single blade, the stator electrical signals of the four current working conditions, the healthy state and the blade attachment state, are collected respectively. In order to expand the blade attachment characteristics and facilitate the subsequent detection of attachments on the ocean current generator blade, the multiple sets of electrical signals collected by the experimental system are Processing, using Hilbert and VMD to pre-process the collected ocean current electromechanical signals to obtain instantaneous frequency signals Select the signal component containing the characteristics of leaf attachment It will also be used for the subsequent classification of the current operating conditions of ocean current generators and the detection of blade attachments.

[0067] (2) Working condition division using mutual information clustering

[0068] The mutual information clustering method proposed in this invention is used to classify the current working conditions of the ocean current generator. The specific steps are as follows:

[0069] Step 2) Use K-means clustering to pre-divide the current working condition of the ocean current generator. Use the selected signal component u1 to randomly initialize the number of clusters b = 2, 3, 4, 5, 6 (6 is the maximum number of working conditions that the ocean current generator can operate in this example experimental platform, and can be set as needed in other application fields) and the cluster center a b (b=2,3,4,5,6) Perform a preliminary division of the current working condition and divide each sample into the corresponding initial working condition:

[0070]

[0071] Among them, γ (b) is the sample matrix divided into the bth working condition, u 1i is the i-th independent sample in the first signal component.

[0072] Step 3) Initialize the working condition center using the strategy of minimizing mutual information entropy and further cluster it.

[0073] First, calculate the mutual information entropy between independent samples in each pre-divided working condition:

[0074]

[0075] in, and are independent samples in the bth pre-divided working condition, and Represents samples and The probability of Representation sample and The mutual information entropy matrix of each pre-divided working condition can be obtained by the joint probability distribution of .

[0076] Then, the sample with the smallest mutual information entropy in each pre-divided working condition is selected by minimizing the mutual information entropy:

[0077]

[0078] Among them, v b is the central sample of the bth working condition, and the sample with the smallest mutual information entropy in each of the above working conditions is used as the initial working condition center. Figure 2 The center samples of each working condition are obtained by minimizing the mutual information entropy in this embodiment. It can be seen that each center sample can well represent each working condition. Finally, the above center samples are used as the initial working condition centers, and further iterative clustering is performed to obtain a reliable working condition classification result.

[0079] Step 4) Determine the number of operating conditions based on the S_Dbw index. The S_Dbw index is:

[0080] S_Dbw=Scat+Dens_bw

[0081] Among them, Scat and Dens_bw are the density within the sample class and the density between classes.

[0082] The formula for determining the optimal number of operating conditions is:

[0083] S_Dbw b* =min(S_Dbw b ),b=1,2,...,K1

[0084] Among them, b* is the optimal number of working conditions, K1 is the maximum number of initial clusters determined in step 2), that is, the working condition division number with the minimum S_Dbw index is selected. * As the final working condition division result of the ocean current generator.

[0085] In this embodiment, when the number of working conditions is 4, the minimum S_Dbw index is 1.1103. When the number of working conditions is 5 or 6, the S_Dbw index becomes null through loop iteration. Therefore, the result when the number of working conditions is 4 is selected as the optimal result to complete the classification of the current working conditions of the marine turbine blade attachment. In this embodiment, the accuracy reaches 96.5%. Figure 3 It can be seen that the working condition classification method based on mutual information clustering proposed in the present invention can clearly identify different working conditions of the ocean current generator under variable flow rate, which will be beneficial to the accurate detection of blade faults.

[0086] (3) Detection of leaf attachments

[0087] This section corresponds to Figure 1 (See step 5 below). After the operating conditions are divided, a principal component analysis model is established within each operating condition to detect debris on the ocean current generator blades. The blade debris detection results are output. If the statistical value exceeds the control limit, it indicates that the ocean current generator blades are adhered to debris. In the experimental platform of this embodiment, the detection accuracy reached 97.5%. If excessive debris is detected on the blades, the machine should be shut down immediately for cleaning to avoid serious faults such as rotor wear and blade breakage caused by the debris.

[0088] This embodiment proposes a method for detecting debris on ocean current turbine blades under variable operating conditions based on mutual information clustering. This improves upon traditional K-means clustering, providing reliable initialization of the operating condition center and the number of operating conditions for classifying the current operating condition of the ocean current turbine, thereby improving the accuracy of the classification. Furthermore, detection of debris on the turbine blades is performed within each operating condition, enhancing the flexibility and accuracy of detection under variable operating conditions and addressing the high false negative and false positive rates of detection under variable operating conditions.

[0089] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A method for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering, characterized in that: The following steps are involved: Step 1) performing Hilbert transform and variational mode decomposition on multiple sets of electrical signals collected from ocean current generators to obtain corresponding instantaneous frequency signals and their characteristic components; Step 2) by randomly initializing the cluster centers and numbers, the characteristic components of the instantaneous frequency signal are processed using the K-means clustering method to determine the pre-division information of the ocean current generator working conditions; Step 3) Select the sample with the smallest mutual information entropy in each pre-divided working condition as the initial cluster center and perform secondary clustering; Step 4) using the clustering evaluation index to determine the number of working conditions and obtain the optimal number of divided working conditions; Step 5) Establish a PCA detection model for blade attachments in each determined optimal division condition and output the detection results.

2. The method for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 1, characterized in that: The step 2) is specifically as follows: using the characteristic component containing the leaf attachment characteristics in the instantaneous frequency signal to randomly initialize the cluster number b and cluster center a b , perform a preliminary division of the current working condition and divide each sample into the corresponding initial working condition: Among them, γ (b) is the sample matrix divided into the bth working condition, u 1i is the i-th independent sample in the first signal component.

3. The method for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 1, characterized in that: The step 3) comprises the following steps: Step 3-1) Determine each pre-divided working condition γ obtained in step 2) (b) , where γ (b) Represents the sample matrix of the bth working condition; Step 3-2) Determine the mutual information entropy between samples in each pre-divided working condition: in, and are independent samples in the bth pre-divided working condition, K b is the number of samples in the bth operating condition, and Represents samples and The probability of Representation sample and The joint probability distribution of Step 3-3) Use the strategy of minimizing entropy to select the central sample, and obtain the sample with the minimum mutual information entropy in each pre-divided working condition as follows: Among them, v b is the central sample of the bth working condition; Step 3-4) The sample with the smallest mutual information entropy in each working condition is used as the initial working condition center, and secondary clustering is performed to obtain clustering results under different initialization cluster numbers.

4. The method for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 1, characterized in that: The clustering evaluation index is: S_Dbw=Scat+Dens_bw Among them, Scat and Dens_bw are the intra-class density and inter-class density of each cluster sample respectively.

5. The method for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 1, characterized in that: The optimal number of divided working conditions is the number of divided working conditions corresponding to the minimum clustering evaluation index.

6. A device for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering, characterized in that: include: An acquisition module is used to collect multiple sets of electrical signals from ocean current generators; The signal preprocessing module is used to perform Hilbert transform and variational mode decomposition on multiple sets of electrical signals collected from the ocean current generator to obtain the corresponding instantaneous frequency signals and their characteristic components; The preliminary clustering module uses the K-means clustering method to process the characteristic components of the instantaneous frequency signal by randomly initializing the cluster center and number, and determines the pre-division information of the ocean current generator working conditions; The secondary clustering module selects the sample with the smallest mutual information entropy in each pre-divided working condition as the initial cluster center for secondary clustering; The module for determining the number of optimal divided working conditions is used to determine the number of working conditions by using clustering evaluation indicators to obtain the optimal number of divided working conditions; The detection module is used to establish a PCA detection model for blade attachments in each determined optimal division working condition and output the detection results.

7. The device for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 6, characterized in that: The preliminary clustering module performs the following steps: using the characteristic component containing the leaf attachment characteristics in the instantaneous frequency signal to randomly initialize the cluster number b and cluster center a b , perform a preliminary division of the current working condition and divide each sample into the corresponding initial working condition: Among them, γ (b) is the sample matrix divided into the bth working condition, u 1i is the i-th independent sample in the first signal component.

8. The device for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 6, characterized in that: The secondary clustering module performs the following steps: Step 3-1) Determine the pre-divided working conditions γ obtained by the preliminary clustering module (b) , where γ (b) Represents the sample matrix of the bth working condition; Step 3-2) Determine the mutual information entropy between samples in each pre-divided working condition: in, and are independent samples in the bth pre-divided working condition, K b is the number of samples in the bth operating condition, and Represents samples and The probability of Representation sample and The joint probability distribution of Step 3-3) Use the strategy of minimizing entropy to select the central sample, and obtain the sample with the minimum mutual information entropy in each pre-divided working condition as follows: Among them, v b is the central sample of the bth working condition; Step 3-4) The sample with the smallest mutual information entropy in each working condition is used as the initial working condition center, and secondary clustering is performed to obtain clustering results under different initialization cluster numbers.

9. The device for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 6, characterized in that: The clustering evaluation index is: S_Dbw=Scat+Dens_bw Among them, Scat and Dens_bw are the intra-class density and inter-class density of each cluster sample respectively.

10. The device for detecting attachments on ocean current generator blades under variable operating conditions based on mutual information clustering according to claim 6, characterized in that: The optimal number of divided working conditions is the number of divided working conditions corresponding to the minimum clustering evaluation index.

Images