GIS disconnecting switch state monitoring method based on multi-modal feature fusion

Abstract

The invention provides a GIS isolation switch state monitoring method based on multi-modal feature fusion, and the method comprises the following steps: 1, collecting multi-modal data of a drive motor in each operation state, carrying out the preprocessing of the multi-modal data, extracting a feature matrix of each modal data, and obtaining a feature matrix of each modal data; the multi-modal data comprises power signal data and other modal data; 2, performing feature matrix fusion on the power signal data and other modal data one by one to generate a plurality of fusion feature matrixes; and step 3, carrying out fault classification on the fusion feature matrix by using a CNN. According to the method, feature fusion is carried out on modals, the correlation between features is dynamically calculated, a feature combination method is optimized, and the accuracy of fault diagnosis can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of GIS switch state detection, and particularly to a method for monitoring the state of a GIS disconnector based on multi-modal feature fusion. Background Art

[0002] Gas-insulated switchgear disconnectors (GIS) are widely used in modern power grids. Their core function is to isolate and switch on high-voltage power lines to ensure the safe and stable operation of the power system. However, due to the enclosed internal mechanism of GIS disconnectors, the diagnosis and monitoring of mechanical failures face many challenges. Typical mechanical failures include:

[0003] Jamming failure: Due to foreign object intrusion, poor lubrication, or aging of mechanical components, the disconnector may experience blocked or stuck movement during operation, resulting in failed switching or abnormally extended operation time.

[0004] Incomplete switching: In the case of abnormal motor drive systems or worn actuator mechanisms, the GIS disconnector may not reach the expected switching positions, affecting the electrical isolation effect and even leading to arc combustion and safety hazards.

[0005] Aging or loosening of mechanical components: Long-term operation may cause loosening and increased wear of moving parts, affecting the operation accuracy and reliability, and ultimately affecting the mechanical life and working stability of the switch.

[0006] Due to the complex mechanical structure and enclosed working environment of GIS disconnectors, traditional mechanical state detection methods (such as optical or mechanical sensors) are difficult to directly measure the operating state of internal components. The current mainstream fault diagnosis methods are mainly based on the analysis of motor power signals, and the health state of the switch mechanism is inferred by monitoring the power change characteristics of the driving motor.

[0007] The power signal reflects the energy input of the motor during the movement of the disconnector and contains information on the change of mechanism resistance during the switching process. Therefore, power signal analysis has become an important means for fault diagnosis of GIS disconnectors. The current diagnosis methods based on power signals mainly include:

[0008] Time-domain analysis method: By calculating time-domain statistical features such as the peak value, mean value, variance, and root mean square value of the power signal, it is judged whether there are abnormal resistances or power mutations during the switching process.

[0009] Problem: Time-domain features can only provide global trend information, cannot accurately capture short-time fault features, and are easily affected by noise interference.

[0010] Frequency-domain analysis method: Using Fourier transform (FFT) or wavelet transform to analyze the energy distribution of the power signal in different frequency ranges to identify abnormal frequency components.

[0011] Problem: Frequency domain analysis relies on fixed basis functions, making it difficult to adapt to non-stationary signals and having limited ability to identify complex mechanical fault patterns.

[0012] Time-frequency joint analysis method: Combine methods such as short-time Fourier transform (STFT), wavelet packet transform, or empirical mode decomposition (EMD) to analyze the time-varying frequency characteristics of power signals and extract fault characteristics at different time scales.

[0013] Problem: Some methods have the problem of mode aliasing, and the feature extraction results rely on manual experience and do not have the ability of adaptive optimization.

[0014] Although the above methods can identify the fault patterns of GIS disconnectors to a certain extent, there are still the following core limitations:

[0015] Feature extraction relies on empirical rules and has limited adaptability: Traditional feature extraction methods mainly rely on fixed statistical or transform features and cannot adaptively adjust feature selection according to different working conditions, resulting in insufficient adaptability to some complex fault patterns.

[0016] The correlation of time series features is not fully utilized: Most existing methods adopt an independent feature calculation method, ignoring the dynamic evolution law of power signals in the time series, making the diagnostic results vulnerable to signal fluctuations.

[0017] The diagnostic accuracy is greatly affected by signal noise: The power signals of GIS disconnectors may be affected by factors such as environmental noise and electromagnetic interference. Traditional signal processing methods have insufficient robustness in a noisy environment and are prone to misjudgment or missed judgment. Summary of the Invention

[0018] In view of this, the purpose of the present invention is to provide a method for monitoring the state of GIS disconnectors based on multi-modal feature fusion to solve at least the above problems.

[0019] The technical solution adopted by the present invention is as follows:

[0020] A method for monitoring the state of GIS disconnectors based on multi-modal feature fusion, the method includes the following steps:

[0021] Step 1: Collect multi-modal data of the driving motor in various operating states, preprocess the multi-modal data, and respectively extract the feature matrices of each modal data. The multi-modal data includes power signal data and other modal data;

[0022] Step 2: Perform feature matrix fusion on the power signal data and other modal data one by one to generate multiple fusion feature matrices;

[0023] Step 3: Use CNN to perform fault classification on the fused feature matrix.

[0024] Furthermore, the multimodal data collected in Step 1 all have time series characteristics.

[0025] Furthermore, the preprocessing of the multimodal data in Step 1 is specifically as follows: perform mean removal and normalization on the multimodal data, and the normalization process is: normalize the multimodal data to the range of [-1, 1].

[0026] Furthermore, the specific method for extracting the feature matrix of each modal data in Step 1 is as follows:

[0027] Segment the signal data of each modality in the multimodal data by time, extract 16 features for each segment, and form a 16*16 feature matrix.

[0028] Furthermore, the power calculation formula for the power signal data in Step 1 is:

[0029] P(t) = V(t)I(t)cosφ

[0030] where P(t) represents the instantaneous power of the motor, V(t) represents the instantaneous voltage at the input end of the motor, and I(t) represents the input current of the motor.

[0031] Furthermore, Step 2 includes the following steps:

[0032] Step S1: Select a modal data from the other modal data, perform intra-modal interaction between the power signal data and this modal data, so that the power signal data and this modal data generate corresponding self-interaction feature matrices;

[0033] Step S2: Calculate the contrastive attention between the power signal data and this modal data through the self-interaction feature matrix to perform inter-modal interaction, and generate a cross-modal interaction feature matrix;

[0034] Step S3: Perform multi-modal feature fusion on the self-interaction feature matrix and the cross-modal interaction feature matrix to generate fused features.

[0035] Furthermore, the formula for performing intra-modal interaction on the collected modal data in Step S1 is:

[0036]

[0037] where X' A and X' B respectively represent the self-interaction feature matrices of the power signal data and the selected modal data after intra-modal interaction; σ() is a non-linear activation function; W A and W ALearnable intra-modal interaction weights representing power signal data and selected modal data, respectively, and W A , W B ∈ i 16×16 ; X A and X B represent the 16*16 feature matrices of power signal data and selected modal data, respectively; b A and b B are bias terms, and b A , b B ∈ i 16×16 ;

[0038] Furthermore, the specific formula for step S2 is:

[0039]

[0040] where S AB and S BA represent the cross-modal contrast feature matrices of power signal data and selected modal data, respectively; W AB and W BA represent the modal conversion matrices of power signal data and selected modal data, respectively, and W AB , W BA ∈ i 16×16 ; b AB and b BA are cross-modal bias terms, and b AB , b BA ∈ i 16×16 ; A AB represents the inter-modal adaptive attention matrix.

[0041] Furthermore, step S3 is specifically:

[0042] Calculate the fusion weights of power signal data and selected modal data, and the formula is:

[0043]

[0044] where W A and W B represent the fusion weight matrices of power signal data and selected modal data, respectively; and represent the learnable high-dimensional projection matrices of power signal data and selected modal data, respectively.

[0045] Calculate the fusion feature matrix, and the formula is:

[0046]

[0047] where F represents the final fusion feature matrix; represents element-wise multiplication.

[0048] Further, step 3 is specifically as follows:

[0049] Input the fused feature matrix into the CNN. The CNN extracts local patterns through the convolutional layer, reduces the dimension and enhances key features through the pooling layer, and integrates information through the fully connected layer for fault classification.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] The present invention provides a GIS disconnector status monitoring method based on multi-modal feature fusion. By extracting 16 features from modal data including power signal data in each time period to form a 16*16 feature matrix, the ability to characterize temporal dynamic changes can be enhanced; by dynamically calculating the fusion weight matrix to replace the fixed fusion ratio, the contribution degrees of different modalities change with the fault situation; through the interaction within and between modalities and introducing the cross-modal attention mechanism, the fault features of the signal can be enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only the preferred embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0053] Figure 1 is the overall method flow chart provided by the present invention;

[0054] Figure 2 is the specific method flow chart of step 2 provided by the present invention;

[0055] Figure 3 is the schematic diagram of the 16*16 feature matrix constructed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] The following further elaborates on the technical solutions of the present invention in conjunction with the accompanying drawings of the specification and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In the following description, the expression "some embodiments" describes a subset of all possible embodiments. However, it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0057] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. However, it will be apparent to one skilled in the art that the present invention may be practiced without one or more of these specific details. In other instances, well-known features of the art are not described in order to avoid obscuring the present invention.

[0058] It should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present invention to those skilled in the art. The purpose of the terminology used herein is to describe particular embodiments only and is not intended to be limiting of the present invention. As used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the terms "comprising" and / or "including", when used in this specification, specify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. As used herein, the term "and / or" includes any and all combinations of the associated listed items.

[0059] It should be further noted that when an element is referred to as being "fixed to" another element, it can be directly on the other element or there can also be an intermediate element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "inner", "outer", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only embodiments.

[0060] To thoroughly understand the present invention, detailed structures will be presented in the following description to illustrate the technical solutions proposed by the present invention. The alternative embodiments of the present invention are described in detail as follows. However, in addition to these detailed descriptions, the present invention can also have other embodiments.

[0061] Refer to Figure 1 , the present invention provides a GIS disconnector status monitoring method based on multi-modal feature fusion. The method includes the following steps:

[0062] Step 1: Collect multi-modal data of the driving motor in various operating states, preprocess the multi-modal data, and extract the feature matrices of each modal data respectively. The multi-modal data includes power signal data and other modal data;

[0063] The multi-modal data collected in Step 1 all have time series characteristics.

[0064] Exemplarily, other modal data includes: current, power, torque, stroke signal, vibration signal, etc.

[0065] The preprocessing of the multi-modal data in step 1 is specifically as follows: the multi-modal data is de-meaned and normalized, and the normalization process is: normalizing the multi-modal data to the range of [-1, 1].

[0066] Exemplarily, de-meaning the multi-modal data can remove the DC component and ensure that the signal fluctuates around zero mean; the multi-modal data can also be passed through a band-pass filter to remove low-frequency drift and high-frequency noise.

[0067] The specific process of extracting the feature matrix of each modal data in step 1 is as follows:

[0068] The modal signal data in the multi-modal data is segmented by time, and 16 features are extracted from each segment to form a 16 * 16 feature matrix.

[0069] Exemplarily, by constructing a 16 * 16 feature matrix, the feature matrix can cover various aspects such as time domain, frequency domain, and non-linear dynamic features, and can comprehensively reflect the changes of data under different mechanical states. For modal data other than power signal data, the method of constructing a 16 * 16 feature matrix is the same as that of power signal data. Mark the power signal data as X, then the 16 features of the power signal data are:

[0070] Peak value:

[0071] Peak = max(X)

[0072] The peak value of the power signal reflects the change of the maximum power during the closing and opening process, and is usually related to the peak value of the mechanical resistance when the disconnector contact touches or separates.

[0073] Average value:

[0074]

[0075] The average value characterizes the overall level of the power signal and reflects the overall load state of the mechanical movement.

[0076] Variance:

[0077]

[0078] The variance describes the degree of fluctuation of the power signal and can reflect the stability or abnormal jitter in the mechanical action.

[0079] Root mean square:

[0080]

[0081] The root mean square is used to reflect the energy level of the signal and is related to the overall energy output of the mechanical load.

[0082] Kurtosis:

[0083]

[0084] Kurtosis is used to measure the sharpness of the signal and reflects the abnormally sharp power fluctuations during mechanical failures.

[0085] Skewness:

[0086]

[0087] Skewness describes the symmetry of the signal and helps to determine whether there is abnormal behavior deviating from the center in the power signal.

[0088] Form factor:

[0089]

[0090] The form factor reflects the change in the signal shape and is used to identify periodic fluctuations.

[0091] Pulse factor:

[0092]

[0093] The pulse factor is used to measure the impact of transient peaks on the overall level in the power signal and is commonly used to detect sudden mechanical shocks.

[0094] Margin factor:

[0095]

[0096] The margin factor characterizes the ratio of the signal peak value to the root mean square value and is sensitive to abnormal pulses.

[0097] Mean absolute deviation:

[0098]

[0099] Measures the degree of data dispersion, is more robust to outliers than variance, and is commonly used for feature scaling or outlier detection.

[0100] Number of zero crossings:

[0101]

[0102] Energy:

[0103]

[0104] The signal energy directly reflects the intensity change of the power signal.

[0105] Main frequency:

[0106]

[0107] The main frequency is used to identify the main frequency components of mechanical actions.

[0108] Spectrum centroid:

[0109]

[0110] The spectrum centroid reflects the concentration degree of the spectrum and is used to diagnose the dynamic characteristics of the mechanical state.

[0111] Frequency standard deviation:

[0112]

[0113] The frequency standard deviation characterizes the dispersion degree of the signal frequency distribution.

[0114] Sample entropy:

[0115]

[0116] The sample entropy reflects the complexity and uncertainty of the signal and is used to identify the chaotic characteristics of the mechanical state. The power calculation formula for the power signal data in step 1 is:

[0117] P(t) = V(t)I(t)cosφ

[0118] Where, P(t) represents the instantaneous power of the motor, V(t) represents the instantaneous voltage at the input end of the motor, and I(t) represents the input current of the motor.

[0119] Step 2: Perform feature matrix fusion on the power signal data and other modal data one by one to generate multiple fusion feature matrices.

[0120] Exemplarily, fusing other modal data in the multi-modal data with the power signal data can enhance the features of the power signal data.

[0121] Step 2 includes the following steps:

[0122] Step S1: Select a modal data from other modal data, perform intra-modal interaction on the power signal data and this modal data, so that the power signal data and this modal data generate corresponding self-interaction feature matrices;

[0123] Exemplarily, intra-modal interaction can make the features within the same modality better influence each other and can perform non-linear enhancement on the original features.

[0124] The formula for performing intra-modal interaction on each collected modal data in step S1 is:

[0125]

[0126] Among them, X' A and X' B respectively represent the self-interaction feature matrices of the power signal data and the selected modal data after intra-modal interaction; σ() is a non-linear activation function; W A and W A respectively represent the learnable intra-modal interaction weights of the power signal data and the selected modal data, and W A , W B ∈i 16×16 ; X A and X B respectively represent the 16*16 feature matrices of the power signal data and the selected modal data; b A and b B are bias terms, and b A , b B ∈i 16×16 ;

[0127] Step S2: Calculate the contrastive attention between the power signal data and the modal data through the self-interaction feature matrix to perform inter-modal interaction, and generate a cross-modal interaction feature matrix;

[0128] Exemplarily, step S2 can adjust the feature contribution degrees between different modalities through an adaptive contrast vector by means of a high-dimensional feature interaction mechanism.

[0129] The specific formula for step S2 is:

[0130]

[0131] Among them, S AB and S BA respectively represent the cross-modal contrast feature matrices of the power signal data and the selected modal data; W AB and W BA respectively represent the modal transformation matrices of the power signal data and the selected modal data, and W AB , W BA ∈i 16×16 ; b AB and b BA are cross-modal bias terms, and b AB , b BA ∈i 16×16 ; A AB represents the inter-modal adaptive attention matrix.

[0132] Step S3: Perform multi-modal feature fusion on the self-interaction feature matrix and the cross-modal interaction feature matrix to generate a fused feature.

[0133] Step S3 specifically is as follows:

[0134] Calculate the fusion weights of the power signal data and the selected modal data. The formula is:

[0135]

[0136] where, W A and W B respectively represent the fusion weight matrices of the power signal data and the selected modal data; and respectively represent the learnable high-dimensional projection matrices of the power signal data and the selected modal data.

[0137] Calculate the fused feature matrix. The formula is:

[0138]

[0139] where, F represents the final fused feature matrix; represents element-wise multiplication.

[0140] Exemplarily, through W A and W B , different weights can be dynamically assigned to each feature channel; in different fault modes, this method can automatically amplify or reduce the influence of a certain mode, rather than simply fixing the fusion ratio, so that the fusion process not only depends on the features of a single mode, but also takes into account the complementary information between modes.

[0141] Step 3: Use CNN to perform fault classification on the fused feature matrix.

[0142] Step 3 specifically is as follows:

[0143] Input the fused feature matrix into CNN. CNN extracts local patterns through the convolutional layer, reduces the dimension and enhances key features through the pooling layer, and integrates information through the fully connected layer to perform fault classification.

[0144] Exemplarily, use the constructed fused feature matrix as the input of the convolutional neural network CNN. CNN can automatically learn the feature patterns in the matrix and obtain efficient and accurate classification results.

[0145] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A GIS disconnector status monitoring method based on multi-modal feature fusion, characterized in that The method includes the following steps: Step 1: Collect multi-modal data of the drive motor in various operating states, preprocess the multi-modal data, and extract the feature matrices of each modal data respectively. The multi-modal data includes power signal data and other modal data; Step 2: Feature matrix fusion is performed on the power signal data and other modal data one by one to generate multiple fused feature matrices; Step 3: Use CNN to classify faults for the fused feature matrices.

2. The method for monitoring the state of a GIS disconnector based on multi-modal feature fusion according to claim 1, wherein The multi-modal data collected in Step 1 all have time series characteristics.

3. A GIS disconnector status monitoring method based on multi-modal feature fusion according to claim 2, characterized in that The preprocessing of the multi-modal data in Step 1 is specifically: de-mean and standardize the multi-modal data. The standardization processing is: normalize the multi-modal data to between [-1, 1].

4. A GIS disconnector status monitoring method based on multi-modal feature fusion according to claim 3, characterized in that, The specific method of extracting the feature matrices of each modal data in Step 1 is: Segment the signal data of each modality in the multi-modal data by time, extract 16 features for each segment, and form a 16*16 feature matrix.

5. A method for monitoring the state of a GIS disconnector based on multi-modal feature fusion according to claim 4, characterized in that The power calculation formula of the power signal data in Step 1 is: P(t) = V(t)I(t)cosφ where P(t) represents the instantaneous power of the motor, V(t) represents the instantaneous voltage at the input end of the motor, and I(t) represents the input current of the motor.

6. The method for monitoring the state of a GIS disconnector based on multimodal feature fusion according to claim 5, characterized in that, Step 2 includes the following steps: Step S1: Select a modal data from the other modal data, perform intra-modal interaction on the power signal data and this modal data, so that the power signal data and this modal data generate corresponding self-interaction feature matrices; Step S2: Calculate the contrastive attention between the power signal data and this modal data through the self-interaction feature matrix to perform inter-modal interaction and generate cross-modal interaction feature matrices; Step S3: Perform multi-modal feature fusion on the self-interaction feature matrix and the cross-modal interaction feature matrix to generate fused features.

7. A GIS disconnector status monitoring method based on multi-modal feature fusion according to claim 6, characterized in that, The formula for performing intra-modal interaction on the collected modal data in Step S1 is: Among them, X' A and X' B respectively represent the self-interaction feature matrices of the power signal data and the selected modal data after intra-modal interaction; σ() is a non-linear activation function; W A and W A respectively represent the learnable intra-modal interaction weights of the power signal data and the selected modal data, and W A , W B ∈i 16×16 ; X A and X B respectively represent the 16*16 feature matrices of the power signal data and the selected modal data; b A and b B are bias terms, and b A , b B ∈i 16×16 .

8. A method for monitoring the state of a GIS disconnector based on multi-modal feature fusion according to claim 7, characterized in that The specific formula for Step S2 is: Among them, S AB and S BA respectively represent the cross-modal comparison feature matrices of power signal data and selected modal data; W AB and W BA respectively represent the modal conversion matrices of power signal data and selected modal data, and W AB , W BA ∈i 16×16 ; b AB and b BA are cross-modal bias terms, and b AB , b BA ∈i 16×16 ; A AB represents the inter-modal adaptive attention matrix.

9. A GIS disconnector status monitoring method based on multi-modal feature fusion according to claim 8, characterized in that Step S3 is specifically: Calculate the fusion weights of the power signal data and the selected modal data, and the formula is: Among them, W A and W B represent the fusion weight matrices of the power signal data and the selected modal data, respectively; and represent the learnable high-dimensional projection matrices of the power signal data and the selected modal data, respectively. Calculate the fused feature matrix, and the formula is: Among them, F represents the final fused feature matrix; represents element-wise multiplication.

10. A GIS disconnector status monitoring method based on multi-modal feature fusion according to claim 9, characterized in that, Step 3 is specifically: Input the fused feature matrix into the CNN. The CNN extracts local patterns through the convolutional layer, reduces the dimension and enhances key features through the pooling layer, and integrates information through the fully connected layer to perform fault classification.

Images