A method, apparatus, device, and storage medium for detecting the status of an on-load tap changer.

Abstract

This application discloses a method, apparatus, device, and storage medium for detecting the status of an on-load tap changer. It acquires triaxial vibration signals from the on-load tap changer, obtaining vibration signals from multiple axes, thus increasing the amount of effective information acquired and improving the accuracy of status detection. The triaxial vibration signals are then denoised to ensure the accuracy of the vibration signals used in subsequent processing. Next, the denoised triaxial vibration signals are input into a multi-vibration sensor information fusion network for training, resulting in a multi-vibration sensor information network model and a feature-fused triaxial vibration signal. The multi-vibration sensor information fusion network can fuse vibration signal feature values ​​with features from other axes, providing vibration signals with both depth and breadth. Finally, the feature-fused triaxial vibration signal is input into the multi-vibration sensor information fusion network model for label processing to determine whether the on-load tap changer is in a normal or faulty state.

Description

Technical Field

[0001] This application belongs to the field of on-load tap changer technology, and particularly relates to an on-load tap changer status detection method, apparatus, equipment and storage medium. Background Technology

[0002] On-load tap changers are key components inside transformers, capable of altering the effective turns ratio of the taps in the transformer windings under load or excitation conditions, thereby regulating the output voltage without interruption of operation. On-load tap changers are widely used, especially in converter transformers of ultra-high voltage direct current (UHVDC) transmission projects, where they compensate for system voltage variations and optimize inverter control angles.

[0003] On-load tap changers require frequent switching of tap positions to achieve voltage regulation. During this switching process, internal components experience wear. Currently, transformer failures caused by on-load tap changers account for over 20% of all transformer failures at transformer stations, and this failure rate is on the rise. Mechanical failures are the primary type of on-load tap changer failure, accounting for over 60% of all such failures.

[0004] Currently, on-load tap changers mainly utilize a single sensor for feature extraction, followed by machine learning algorithms for condition detection and fault classification. However, the accuracy of known condition detection methods needs improvement. Summary of the Invention

[0005] In view of this, this application provides a method, apparatus, device, and storage medium for detecting the status of an on-load tap changer, in order to solve the problem of low accuracy in known on-load tap changer status detection methods.

[0006] To address the above problems, this application provides the following solution:

[0007] A method for detecting the status of an on-load tap changer includes:

[0008] Acquire the triaxial vibration signal of the on-load tap changer;

[0009] The triaxial vibration signal is denoised to obtain a denoised triaxial vibration signal.

[0010] The denoised triaxial vibration signal is input into a multi-vibration sensor information fusion network for training to obtain a multi-vibration sensor information fusion network model and a feature-fused triaxial vibration signal; the multi-vibration sensor information fusion network model is a first-level model.

[0011] The triaxial vibration signal fused by the aforementioned features is input into the multi-vibration sensor information fusion network model for label processing to determine whether the on-load tap changer is in a normal or faulty state.

[0012] Optional, also includes:

[0013] When the multi-vibration sensor information fusion network model determines that the on-load tap changer is in a fault state, the data corresponding to the fault state is input into the secondary model to determine the fault type of the on-load tap changer; the secondary model includes multiple primary models.

[0014] Optionally, the step of inputting the data corresponding to the fault state into the secondary model to determine the fault type of the on-load tap changer includes:

[0015] The secondary model uses one type of fault sample as a positive sample and other fault types and normal state data as negative samples for training, so that when one of the primary models in the secondary model outputs a positive sample, the secondary model outputs the fault type corresponding to the primary model.

[0016] Optionally, acquiring the triaxial vibration signal of the on-load tap changer includes:

[0017] The triaxial vibration signal at the top of the on-load tap changer is obtained by a triaxial sensor.

[0018] Optionally, the denoising process for the triaxial vibration signal includes:

[0019] The triaxial vibration signal is subjected to empirical mode decomposition to obtain the intrinsic mode function set;

[0020] The intrinsic mode function set is subjected to threshold filtering to obtain the denoised triaxial vibration signal.

[0021] Optionally, the step of inputting the denoised triaxial vibration signal into a multi-vibration sensor information fusion network for training to obtain a multi-vibration sensor information fusion network model and a feature-fused triaxial vibration signal includes:

[0022] The vibration signals of the three axial branches in the denoised triaxial vibration signal acquired by the same triaxial sensor are sequentially extracted in the convolution module and the Inception module;

[0023] The feature extraction results corresponding to the convolution module and the feature extraction results corresponding to the Inception module are normalized.

[0024] The normalized feature extraction results of the convolutional modules corresponding to vibration signals of different axes and the normalized feature extraction results of the Inception modules corresponding to different branches are fused together.

[0025] The feature fusion results corresponding to the three axis branches are input into the multi-vibration sensing information fusion network for training, thus obtaining the multi-vibration sensing information fusion network model.

[0026] Optionally, the step of inputting the triaxial vibration signal fused with the features into the multi-vibration sensor information fusion network model for label processing to determine whether the on-load tap changer is in a normal or faulty state includes:

[0027] The multi-vibration sensor information fusion network model processes the triaxial vibration signal labels fused with the features and outputs positive or negative samples.

[0028] When the output of the multi-vibration sensor information fusion network model is a positive sample, the on-load tap changer is determined to be in a normal state.

[0029] When the output of the multi-vibration sensor information fusion network model is a negative sample, the on-load tap changer is characterized as a fault state.

[0030] An on-load tap changer status detection device, comprising:

[0031] The unit consists of an acquisition unit, a denoising unit, a training fusion unit, and a determination unit, wherein:

[0032] The acquisition unit is used to acquire the triaxial vibration signal of the on-load tap changer;

[0033] The denoising unit is used to denoise the triaxial vibration signal to obtain a denoised triaxial vibration signal.

[0034] The training fusion unit is used to input the denoised triaxial vibration signal into the multi-vibration sensor information fusion network for training, so as to obtain the multi-vibration sensor information fusion network model and the feature-fused triaxial vibration signal; the multi-vibration sensor information fusion network model is a first-level model;

[0035] The determining unit is used to input the triaxial vibration signal fused by the features into the multi-vibration sensor information fusion network model, perform label processing, and determine whether the on-load tap changer is in a normal state or a fault state.

[0036] An on-load tap changer status detection device includes a memory and a processor;

[0037] The memory is used to store instructions;

[0038] The processor is used to execute instructions stored in the memory to implement the on-load tap changer status detection method described in any of the above descriptions.

[0039] A storage medium storing a computer program thereon, which, when executed by a processor, enables the on-load tap changer status detection method as described in any of the preceding claims.

[0040] As can be seen from the above scheme, the on-load tap changer status detection method, apparatus, device, and storage medium disclosed in this application acquire triaxial vibration signals of the on-load tap changer, acquiring vibration signals from multiple axes, thus increasing the amount of effective information obtained and improving the accuracy of status detection. The triaxial vibration signals are denoised to ensure the accuracy of the vibration signals participating in subsequent processing. Then, the denoised triaxial vibration signals are input into a multi-vibration sensor information fusion network for training, obtaining a multi-vibration sensor information network model and a feature-fused triaxial vibration signal. The multi-vibration sensor information fusion network can fuse vibration signal feature values ​​with features from other axes, providing vibration signals with both depth and breadth. Finally, the feature-fused triaxial vibration signal is input into the multi-vibration sensor information fusion network model for label processing to determine whether the on-load tap changer is in a normal or faulty state. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating one of the on-load tap changer status detection methods provided in this application;

[0043] Figure 2 This is a time-domain waveform diagram of the triaxial vibration signal of the on-load tap changer under normal conditions provided in this application;

[0044] Figure 3 This is a schematic diagram of the vibration signal fusion principle provided in this application;

[0045] Figure 4 This is a schematic diagram of the multi-vibration sensor information fusion network model provided in this application;

[0046] Figure 5 This is another flowchart illustrating the on-load tap changer status detection method provided in this application;

[0047] Figure 6 This is a detection example diagram of the on-load tap changer status detection method provided in this application;

[0048] Figure 7 This is a flowchart illustrating an on-load tap changer status detection method provided in this application.

[0049] Figure 8 This is a structural diagram of the on-load tap changer status detection device provided in this application;

[0050] Figure 9 This is a hardware structure diagram of the on-load tap changer status detection device provided in this application. Detailed Implementation

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0053] It should be noted that the concepts of "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0054] It should be noted that the terms "a" and "a plurality of" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0055] The on-load tap changer is the only mechanically moving component in a transformer. The complex collisions and frictions during its operation cause its vibration signal to exhibit nonlinearity and non-stationarity. Therefore, feature extraction from the vibration signal of the on-load tap changer for state detection is a hot research topic in on-load tap changer research.

[0056] Currently, on-load tap changers primarily utilize a single sensor for feature extraction, followed by machine learning algorithms for status detection and fault classification. However, with the gradual evolution of on-load tap changer structures, more sensors are being installed to detect the switch's operating status. The inventors have discovered that relevant research indicates the accuracy of on-load tap changer status detection using a single sensor needs improvement.

[0057] To address the aforementioned problems, the inventors have discovered that multi-sensor signal data fusion is a feasible approach to improve the accuracy of on-load tap changer status detection. Therefore, this application provides an on-load tap changer status detection method, apparatus, device, and storage medium.

[0058] See Figure 1 The present application provides a schematic flowchart of an on-load tap changer status detection method, as shown in the figure. The on-load tap changer status detection method includes the following steps:

[0059] Step 101: Obtain the triaxial vibration signal of the on-load tap changer.

[0060] On-load tap changers have different states, mainly categorized into normal and fault states. Fault states primarily include loose contacts, spring fatigue, drive shaft jamming, and drive shaft loosening, among others. The triaxial vibration signal can be the triaxial vibration signal collected by a triaxial accelerometer corresponding to any of the above states.

[0061] It should be noted that the components of the on-load tap changer generate energy due to mutual friction. Considering that the effective information of the on-load tap changer may not be directly reflected in the vertical top cover component, this application collects vibration signals from different axes. This allows for the collection of more effective information. The coordinated identification of vibration signals from multiple axes can improve the accuracy of on-load tap changer status identification under changing operating conditions or equipment.

[0062] Optionally, a triaxial vibration signal from the top of the on-load tap changer can be acquired using a triaxial accelerometer.

[0063] Specifically, a triaxial accelerometer is vertically positioned on top of the on-load tap changer to collect one axis component perpendicular to the top cover and two mutually perpendicular axis components horizontal to the top cover. For example, these can be the x-axis, y-axis, and z-axis. The triaxial vibration signals generated during the switching process of the on-load tap changer in different states are collected.

[0064] It should be noted that these triaxial vibration signals under different mechanical conditions can be used to build a database. The triaxial vibration signals in this database can be used to train a multi-vibration sensor information fusion network model in the multi-vibration sensor information fusion network. This model can also be called a classifier. Its main function is to detect whether the on-load tap changer is in a normal state or a fault state based on the processed triaxial vibration signals.

[0065] For example, see Figure 2 The application provides a time-domain waveform diagram of the triaxial vibration signal of an on-load tap changer under normal conditions.

[0066] Step 102: Denoise the triaxial vibration signal to obtain the denoised triaxial vibration signal.

[0067] The process of acquiring triaxial vibration signals using a triaxial accelerometer is highly susceptible to noise interference. Therefore, it is necessary to denoise the acquired triaxial vibration signals. The denoising process mainly includes two steps: first, decomposition based on empirical modes, and then threshold filtering of the decomposed triaxial vibration signals.

[0068] Optionally, the triaxial vibration signal can be decomposed using the EMD (Empirical Mode Decomposition) algorithm to obtain multiple intrinsic mode functions. Then, the noise mixed in the intrinsic mode function components can be processed by threshold filtering.

[0069] For example, let the triaxial vibration signal be C. x (t), C y (t) and C z (t) The vibration signals of the three shaft branches are decomposed using the EMD algorithm to obtain K intrinsic mode functions (IMFs). Then, the noise mixed in the IMF components is processed by threshold filtering.

[0070] Since the IMF component is a modulated sinusoidal signal with a mean of 0 and numerous zero-crossing points, an interval threshold is used, i.e.:

[0071]

[0072] Where, d j Let z be the discrete signal of the j-th IMF, and let ε be the interval width. i = (i, i+1, ..., i+ε-1), i = 1, 1+ε, ..., 1+N-ε, N, where N is the length of the signal x0(t), and ε can take any value; in this embodiment, it can be 100. When a certain interval z i All values ​​are less than or equal to the set threshold T j When the threshold is set to 0, all points within that region are set to 0. The threshold calculation formula is as follows:

[0073]

[0074] Where C is a constant, its value can be 0.6, and E j The value is determined by the following formula:

[0075]

[0076] Among them, E1, E j For IMF1, IMFj The energy; β and ρ are determined by the number of screening iterations. In this embodiment, β = 0.8 and ρ = 2. Each IMF is denoised and then summed to obtain the denoised signal.

[0077] Step 103: Input the denoised triaxial vibration signal into the multi-vibration sensor information fusion network for training to obtain the multi-vibration sensor information fusion network model and the feature-fused triaxial vibration signal; the multi-vibration sensor information fusion network model is a first-level model.

[0078] Optionally, features can first be extracted from the denoised triaxial vibration signal using the corresponding modules in a convolutional neural network. A convolutional neural network is a machine learning technique; its feature extraction method involves performing calculations using different convolutional kernels, which can reduce the dimensionality of the signal.

[0079] The convolutional neural network in this embodiment mainly includes a convolutional module, an Inception module, a global pooling layer, and a fully connected layer. The convolutional module contains three convolutional layers and one max pooling layer, and the Inception module contains two Inception modules and one max pooling layer.

[0080] The vibration signals of the three branches in the denoised triaxial vibration signal are extracted using a convolution module and an Inception module. Then, the algorithm proposed in this application can be used to directly fuse the vibration signals of different branches. It should be noted that vibration signals acquired by a single triaxial accelerometer can be directly fused because they have the same length and sampling rate.

[0081] Then, the feature extraction results corresponding to the convolution module and the Inception module are normalized. The normalized feature extraction results of the convolution module and the Inception module corresponding to the vibration signals of different axes are fused. Finally, the feature fusion results corresponding to the three axes are input into the multi-vibration sensing information fusion network for training to obtain the multi-vibration sensing information fusion network model.

[0082] Optional, see Figure 3 The vibration signal fusion principle diagram provided in this application is shown in the corresponding diagram. Figure 4 A schematic diagram of the multi-vibration sensor information fusion network model provided in this application.

[0083] The denoised vibration signals of the three axial branches are input into a convolutional neural network, and C is processed accordingly. x (t), C y (t) and C z(t) performs feature extraction. After feature extraction in different modules, the feature representation can be in matrix form. Using C... x (t), C y (t) is an example. Let the characteristic matrices of the two branches be X and Y, and their corresponding row vectors be [x1; x2; ...; x...]. n ]、[y1;y2;...;y n ].

[0084] Because the measurement values ​​of different axes differ significantly, a normalization operation is performed first during the fusion process. This reduces the computational cost of calculating the feature matrix. The normalization operation for matrices X and Y can be performed as follows:

[0085]

[0086]

[0087] Where, x i ' represents the i-th row vector x in matrix X. i y i ' represents the i-th row vector y in matrix Y. i x i,max x i,min y i,max y i,min They represent row vectors x and x respectively. i y i The maximum and minimum values ​​in the range.

[0088] Furthermore, it is necessary to obtain the weight matrix ω. x(y) , representing the weights obtained from the feature matrix x to y, and the weight matrix ω x(y) medium element ω i,j The calculation formula is as follows:

[0089]

[0090] To ensure the non-negativity of the weights, the SoftMax function is used to process the weight matrix, which can significantly highlight the weights of important components in the weight matrix. The specific processing formula is as follows:

[0091]

[0092] Where n is the number of rows in the matrix, W i,j and ω i,j The final weight matrix W is shown below. x(y) and weight matrix ω x(y) The elements in.

[0093] Multiplying the weight matrix by the feature matrix X yields the reconstructed feature matrix of the heat map matrix X:

[0094] X Y =W x(y) ·X

[0095] After the convolutional modules are fused, their features become X′=X Y +X, such as Figure 4 As shown, the Inception layer can be continuously fed in for training.

[0096] Similarly, the feature fusion result of the three vibration signals can be obtained, and the input of the next-scale feature extraction module becomes X+X. Y +X Z The other two branches can be merged to obtain Y+Y X +Y Z Z+Z X +Z Y .

[0097] It should be noted that the information feature coupling method used in the embodiments of this application can adaptively extract the feature coupling relationship between signals of different sizes and types, realize deep signal feature fusion, and improve the robustness of on-load tap changer status identification in this application.

[0098] Step 104: Input the triaxial vibration signal fused by the feature into the multi-vibration sensor information fusion network model, perform label processing, and determine whether the on-load tap changer is in a normal state or a fault state.

[0099] The multi-vibration information coupling network (MICN) model in this application can use the triaxial vibration signal of the on-load tap changer under normal conditions as positive samples and the triaxial vibration signal of the on-load tap changer under fault conditions as negative samples.

[0100] In this embodiment, the MICN model is a first-level model, which can be regarded as a binary classification model: after the model is trained by the MICN of the first-level model, it only outputs the normal state and the fault state: when the MICN of the first-level model outputs a positive sample, it is determined that the on-load tap changer is in the normal state; if the MICN of the first-level model outputs a negative sample, it is determined that the on-load tap changer is in the fault state.

[0101] Optionally, the feature-fused triaxial vibration signal is input into a multi-vibration sensor information fusion network model for labeling. The model labels the feature-fused triaxial vibration signal as a positive or negative sample and outputs it. When the multi-vibration sensor information fusion network model outputs a positive sample, it indicates that the on-load tap changer is in a normal state; when the multi-vibration sensor information fusion network model outputs a negative sample, it indicates that the on-load tap changer is in a fault state.

[0102] It should be noted that the first-level model used in this implementation can only determine whether the on-load tap changer is in a normal state or a fault state, but cannot determine the fault type of the on-load tap changer in a fault state.

[0103] In summary, the on-load tap changer status detection method disclosed in this application acquires triaxial vibration signals from the on-load tap changer, obtaining vibration signals from multiple axes, thus increasing the amount of effective information acquired and improving the accuracy of status detection. Subsequently, the triaxial vibration signals are denoised to ensure the accuracy of the vibration signals used in subsequent processing. Then, the denoised triaxial vibration signals are input into a multi-vibration sensor information fusion network for training, resulting in a multi-vibration sensor information network model and a feature-fused triaxial vibration signal. The multi-vibration sensor information fusion network can fuse vibration signal feature values ​​with features from other axes, providing vibration signals with both depth and breadth. Finally, the feature-fused triaxial vibration signal is input into the multi-vibration sensor information fusion network model for label processing to determine whether the on-load tap changer is in a normal or faulty state.

[0104] Optional, see Figure 5 Another flowchart of the on-load tap changer status detection method provided in this application is shown. The on-load tap changer status detection method further includes the following steps:

[0105] Step 201: When the multi-vibration sensor information fusion network model determines that the on-load tap changer is in a fault state, the data corresponding to the fault state is input into the secondary model to determine the fault type of the on-load tap changer; the secondary model includes multiple primary models.

[0106] The Multi-Vibration Sensor Information Network (MICN) model provided in the previous embodiment can only determine whether the on-load tap changer is in a fault state, but it cannot determine the specific fault type. On-load tap changers exhibit numerous states during actual operation, and the limited types of data acquired in the laboratory restrict the range of fault types that can be covered. Therefore, this application designs a secondary model to address the above problem. This secondary model can identify sample types of unknown faults outside the training set. When the triaxial vibration signal fused by feature grading is determined as a negative sample by the primary model, indicating that the on-load tap changer is in a fault state, the data corresponding to the fault state is input into the secondary model. The secondary model consists of multiple MICNs, each of which can act as a binary classifier to identify different fault types.

[0107] Specifically, each MICN uses one type of fault sample as a positive sample and samples of other fault types and normal data as negative samples for network training. If a MICN outputs a positive sample, the secondary model outputs the fault type corresponding to that model. It should be noted that in actual operation, the vibration signal of the on-load tap changer may be affected by the vibration interference of the tap selector, which may result in multiple secondary models outputting positive samples. In this case, it can be identified as an unknown sample.

[0108] For example, see Figure 6 The detection example diagram of the on-load tap changer status detection method provided in this application is shown in the figure:

[0109] The primary diagnostic model marks the triaxial vibration signal fused by feature fusion as a positive or negative sample. When the primary diagnostic model outputs a negative sample, that is, when the on-load tap changer is diagnosed as being in a fault state, the negative sample is output to the secondary diagnostic model to identify the specific fault type.

[0110] In summary, the on-load tap changer status detection method provided in this application uses the triaxial vibration signal after EMD denoising as the input vector of a convolutional neural network. A trained deep coupled network is used to obtain the feature values ​​of the vibration signal, which are then fused with features from other axes to obtain a fusion result of multiple vibration information. This fusion result is used to assess the status of the Zhuzhou tap changer. This application implements joint training of the neural network with coupled features and original signal features, ensuring that the feature weights from different sensors are comprehensively considered during the assessment process. This application also proposes a two-level model, which can perform binary classification using all neural networks. This reduces computational load, improves recognition accuracy, and can also identify fault types that can be simulated outside the laboratory, marking them as unknown samples.

[0111] For example, see Figure 7 The following is a flowchart illustrating an example of on-load tap changer status detection provided in this application:

[0112] First, triaxial vibration signals from the on-load tap changer are collected. Then, the vibration signals are denoised using EMD (Electronic Dynamics Method). A multi-vibration sensor information fusion network is used to extract and fuse the deep features of the signals. A binary classification model is then trained as the primary model to determine whether the data is normal or abnormal. If the binary classification model determines the data is normal, it indicates that the mechanical state of the on-load tap changer is normal. If the binary classification model determines the data is abnormal, the abnormal samples output by this model are further analyzed using a multi-vibration sensor information fusion binary network to determine the mechanical state of the on-load tap changer malfunction.

[0113] Regarding the above-mentioned on-load tap changer status detection method, this application also provides an on-load tap changer status detection device, the composition of which is as follows: Figure 8 As shown.

[0114] The system comprises an acquisition unit 10, a denoising unit 20, a training fusion unit 30, and a determination unit 40, wherein:

[0115] The acquisition unit 10 is used to acquire the triaxial vibration signal of the on-load tap changer;

[0116] The denoising unit 20 is used to denoise the triaxial vibration signal to obtain a denoised triaxial vibration signal.

[0117] The training fusion unit 30 is used to input the denoised triaxial vibration signal into the multi-vibration sensor information fusion network for training, so as to obtain the multi-vibration sensor information fusion network model and the feature-fused triaxial vibration signal; the multi-vibration sensor information fusion network model is a first-level model;

[0118] The determining unit 40 is used to input the triaxial vibration signal fused by the feature into the multi-vibration sensing information fusion network model, perform label processing, and determine whether the on-load tap changer is in a normal state or a fault state.

[0119] In one embodiment, the above-described apparatus further includes a secondary model processing unit, used for:

[0120] When the multi-vibration sensor information fusion network model determines that the on-load tap changer is in a fault state, the data corresponding to the fault state is input into the secondary model to determine the fault type of the on-load tap changer; the secondary model includes multiple primary models.

[0121] In one embodiment, the secondary model processing unit is specifically used for:

[0122] The secondary model uses one type of fault sample as a positive sample and other fault types and normal state data as negative samples for training, so that when one of the primary models in the secondary model outputs a positive sample, the secondary model outputs the fault type corresponding to the primary model.

[0123] In one embodiment, the acquisition unit 10 is specifically used for:

[0124] The triaxial vibration signal at the top of the on-load tap changer is obtained by a triaxial sensor.

[0125] In one embodiment, the noise reduction unit 20 is specifically used for:

[0126] The triaxial vibration signal is subjected to empirical mode decomposition to obtain the intrinsic mode function set;

[0127] The intrinsic mode function set is subjected to threshold filtering to obtain the denoised triaxial vibration signal.

[0128] In one embodiment, the training fusion unit 30 is specifically used for:

[0129] The vibration signals of the three axial branches in the denoised triaxial vibration signal acquired by the same triaxial sensor are sequentially extracted in the convolution module and the Inception module;

[0130] The feature extraction results corresponding to the convolution module and the feature extraction results corresponding to the Inception module are normalized.

[0131] The normalized feature extraction results of the convolutional modules corresponding to vibration signals of different axes and the normalized feature extraction results of the Inception modules corresponding to different branches are fused together.

[0132] The feature fusion results corresponding to the three axis branches are input into the multi-vibration sensing information fusion network for training, thus obtaining the multi-vibration sensing information fusion network model.

[0133] In one embodiment, the determining unit 40 is specifically used for:

[0134] The multi-vibration sensor information fusion network model processes the triaxial vibration signal labels fused with the features and outputs positive or negative samples.

[0135] When the output of the multi-vibration sensor information fusion network model is a positive sample, the on-load tap changer is determined to be in a normal state.

[0136] When the output of the multi-vibration sensor information fusion network model is a negative sample, the on-load tap changer is characterized as a fault state.

[0137] Optionally, this application also provides an on-load tap changer status detection device, such as... Figure 9 The hardware structure block diagram of the on-load tap changer status detection device provided in this application shown mainly includes a processor 1 and a memory 3, and at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4.

[0138] The memory 3 is used to store instructions;

[0139] The processor 1 is used to execute the instructions stored in the memory to implement the various processing steps of the on-load tap changer status detection method.

[0140] In addition, this application embodiment also provides a storage medium storing a computer program thereon, which, when executed by a processor, can implement various processing flows of the on-load tap changer status detection method.

[0141] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0142] For ease of description, the above systems or devices are described separately as various modules or units based on their functions. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware components.

[0143] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0144] Finally, it should be noted that in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0145] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method of detecting a state of a load tap changer, characterized by, The method comprises the following steps: obtaining three-axis vibration signals of an on-load tap changer; performing empirical mode decomposition on the three-axis vibration signals to obtain an intrinsic mode function set; performing threshold filtering on the intrinsic mode function set to obtain denoised three-axis vibration signals; inputting the denoised three-axis vibration signals into a multi-vibration sensing information fusion network for training to obtain a multi-vibration sensing information fusion network model and a feature-fused three-axis vibration signal, which comprises the following steps: extracting features from the denoised three-axis vibration signals through a convolutional neural network; the convolutional neural network comprises a convolution module, an Inception module for extracting scale features, a global pooling layer, and a fully connected layer; wherein the three-axis branch vibration signals in the denoised three-axis vibration signals collected by the same three-axis sensor are sequentially subjected to feature extraction in the convolution module and the Inception module; the feature extraction results corresponding to the convolution module and the feature extraction results corresponding to the Inception module are subjected to normalization processing; the normalized feature extraction results of the convolution module corresponding to different-axis branch vibration signals are subjected to fusion processing, and the normalized feature extraction results of the Inception module corresponding to different branches are subjected to fusion processing; Specifically, for the X-axis branch, after the convolution module fusion, the input of the Inception module is X+X Y +X Z ; similarly, the other two-axis branches can be obtained after the convolution module fusion: Y+Y X +Y Z , Z+Z X +Z Y ; wherein, X, Y, Z respectively represent the feature matrix of the three-axis branch after the convolution module feature extraction, and the corresponding row vectors are respectively: [x1; x2; …; x n ], [y1; y2; …; y n ], [z1; z2; …; z n ]; X Y =W X(Y) ·X, wherein, W X(Y) represents the maximum weight matrix; inputting the feature fusion results corresponding to the three-axis branches into the multi-vibration sensing information fusion network for training to obtain a multi-vibration sensing information fusion network model; inputting the feature-fused three-axis vibration signal into the multi-vibration sensing information fusion network model for label processing to determine whether the on-load tap changer is in a normal state or a fault state; the multi-vibration sensing information fusion network model is a primary model.

2. The on-load tap changer status detection method of claim 1, wherein, Further comprising: when the multi-vibration sensing information fusion network model determines that the on-load tap changer is in a fault state, inputting data corresponding to the fault state into a secondary model to determine the fault type of the on-load tap changer; the secondary model comprises a plurality of primary models.

3. The on-load tap changer status detection method of claim 2, wherein, The step of inputting the data corresponding to the fault state into the secondary model to determine the fault type of the on-load tap changer comprises: the secondary model trains a type of fault sample as a positive sample and the remaining fault type samples and normal state data as negative samples, so that when a primary model in the secondary model outputs a positive sample, the secondary model outputs the fault type corresponding to the primary model.

4. The on-load tap changer status detection method of claim 1, wherein, The step of obtaining three-axis vibration signals of an on-load tap changer comprises: obtaining three-axis vibration signals at the top of the on-load tap changer through a three-axis sensor.

5. The on-load tap changer status detection method of claim 1, wherein, The step of inputting the feature-fused three-axis vibration signal into the multi-vibration sensing information fusion network model for label processing to determine whether the on-load tap changer is in a normal state or a fault state comprises: the multi-vibration sensing information fusion network model outputs a positive sample or a negative sample after label processing of the feature-fused three-axis vibration signal; when the multi-vibration sensing information fusion network model outputs a positive sample, it indicates that the on-load tap changer is in a normal state. When the multi-vibration sensing information fusion network model outputs a negative sample, it is determined that the on-load tap changer is in a fault state.

6. A status detection device for a load tap changer, characterized in that The method comprises the following steps: An acquisition unit, a denoising unit, a training fusion unit, and a determination unit are provided, wherein: The acquisition unit is configured to acquire three-axis vibration signals of an on-load tap changer. The denoising unit is configured to perform empirical mode decomposition processing on the three-axis vibration signals to obtain an intrinsic mode function set, and perform threshold filtering processing on the intrinsic mode function set to obtain denoised three-axis vibration signals. The training fusion unit is configured to input the denoised three-axis vibration signals into a multi-vibration sensing information fusion network for training to obtain a multi-vibration sensing information fusion network model and feature-fused three-axis vibration signals, comprising: The denoised three-axis vibration signals are subjected to feature extraction through a convolutional neural network, wherein the convolutional neural network comprises a convolution module, an Inception module for extracting scale features, a global pooling layer, and a full connection layer. The three-axis branch vibration signals in the denoised three-axis vibration signals collected by the same three-axis sensor are subjected to feature extraction in the convolution module and the Inception module in sequence, the feature extraction results corresponding to the convolution module and the Inception module are subjected to normalization processing, and the normalized feature extraction results of the convolution module corresponding to different-axis branch vibration signals are subjected to fusion processing, and the normalized feature extraction results of the Inception module corresponding to different branches are subjected to fusion processing. Specifically, for the X-axis branch, after the convolution module fusion, the input of the Inception module is X+X Y +X Z ; similarly, the other two-axis branches can be obtained after the convolution module fusion: Y+Y X +Y Z , Z+Z X +Z Y ; wherein, X, Y, Z respectively represent the feature matrix of the three-axis branch after the convolution module feature extraction, and the corresponding row vectors are respectively: [x1; x2; …; x n ], [y1; y2; …; y n ], [z1; z2; …; z n ]; X Y =W X(Y) ·X, wherein, W X(Y) represents the maximum weight matrix; The feature fusion results corresponding to the three-axis branches are input into the multi-vibration sensing information fusion network for training to obtain a multi-vibration sensing information fusion network model. The determination unit is configured to input the feature-fused three-axis vibration signals into the multi-vibration sensing information fusion network model for label processing to determine whether the on-load tap changer is in a normal state or a fault state, and the multi-vibration sensing information fusion network model is a primary model.

7. A load tap changer status detection apparatus, characterized by, The method comprises the following steps: The memory is configured to store instructions. The processor is configured to execute the instructions stored in the memory to implement the on-load tap changer state detection method of any one of claims 1-5.

8. A storage medium having stored thereon a computer program, characterized in that The computer program, when executed by the processor, can implement the on-load tap changer state detection method of any one of claims 1-5.

Images