Intelligent diagnosis system and method for mechanical fault of power transformer

Abstract

The invention relates to the field of intelligent fault diagnosis, and particularly discloses an intelligent diagnosis system and method for mechanical faults of a power transformer. An artificial intelligence technology in the field of deep learning is used to carry out feature extraction and coding on a plurality of vibration signals collected by a plurality of vibration sensors of a transformer and a topological matrix of the plurality of vibration sensors so as to obtain a result whether a fault exists in the transformer. Thus, the internal fault condition of the transformer is intelligently judged, and an online monitoring method is adopted, so that the condition of power failure during maintenance is avoided, and the maintenance cost and time are reduced.

Description

Technical Field

[0001] This application relates to the field of intelligent fault diagnosis, and particularly to an intelligent diagnosis system and method for mechanical faults of power transformers. Background Art

[0002] A power transformer is an electromagnetic device used to change the AC voltage or current level in a circuit. It operates based on the principle of electromagnetic induction, converting the input electrical energy into output electrical energy at different voltage or current levels. Power transformers play a crucial role in the power system. By changing voltage, current, and power levels, they ensure safe, efficient, and reliable power transmission and distribution. Once a fault occurs during the operation of a transformer, it will not only damage the transformer itself but also extremely likely cause large-scale power outages, resulting in huge economic losses and affecting people's normal lives. Transformer faults can be divided into internal tank faults and external tank faults according to their occurrence locations. Internal tank faults are mainly caused by factors such as oil aging, insulation damage, and winding loosening and deformation, leading to problems such as partial discharge and inter-turn short circuit, while external tank faults involve faults of components such as the tank surface and bushings. Among them, winding faults are one of the most common faults in transformers. Traditional methods for detecting mechanical faults in transformer windings include short-circuit impedance method, low-voltage pulse method, frequency response analysis method, etc. However, these methods are all off-line detection methods, which require power outage during maintenance, and have the disadvantages of long time and high cost.

[0003] Therefore, an optimized intelligent diagnosis system for mechanical faults of power transformers is expected. Summary of the Invention

[0004] To solve the above technical problems, this application is proposed. Embodiments of this application provide an intelligent diagnosis system and method for mechanical faults of power transformers, which use artificial intelligence technology in the field of deep learning to extract features and encode multiple vibration signals collected by multiple vibration sensors of the transformer and the topological matrix of the multiple vibration sensors to obtain the result of whether there is a fault inside the transformer. In this way, by intelligently judging the internal fault situation of the transformer, an on-line monitoring method is adopted, avoiding the situation of power outage during maintenance, and reducing the maintenance cost and time.

[0005] According to one aspect of this application, an intelligent diagnosis system for mechanical faults of power transformers is provided, which includes:

[0006] A transformer vibration signal acquisition module for acquiring multiple vibration signals collected by multiple vibration sensors deployed on the transformer;

[0007] A vibration signal analysis module for analyzing the multiple vibration signals to obtain a global multi-scale vibration signal time-frequency feature matrix;

[0008] A sensor topology matrix analysis module, configured to construct a topology matrix of the multiple vibration sensors and perform analysis to obtain a sensor topology feature matrix;

[0009] A transformer internal fault analysis module, configured to comprehensively analyze the global multi-scale vibration signal time-frequency feature matrix and the sensor topology feature matrix to obtain a result indicating whether there is a fault inside the transformer.

[0010] According to another aspect of the present application, there is also provided an intelligent diagnosis method for mechanical faults of a power transformer, which includes:

[0011] Obtain multiple vibration signals collected by multiple vibration sensors deployed on the transformer;

[0012] Analyze the multiple vibration signals to obtain a global multi-scale vibration signal time-frequency feature matrix;

[0013] Construct a topology matrix of the multiple vibration sensors and perform analysis to obtain a sensor topology feature matrix;

[0014] Comprehensively analyze the global multi-scale vibration signal time-frequency feature matrix and the sensor topology feature matrix to obtain a result indicating whether there is a fault inside the transformer.

[0015] In summary, the intelligent diagnosis system and method for mechanical faults of a power transformer provided by the present application use artificial intelligence technology in the field of deep learning to perform feature extraction and encoding on multiple vibration signals collected by multiple vibration sensors of the transformer and the topology matrix of the multiple vibration sensors, so as to obtain a result indicating whether there is a fault inside the transformer. In this way, by intelligently judging the internal fault condition of the transformer, an online monitoring method is adopted, avoiding the situation of power outage during maintenance, and reducing the maintenance cost and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application, and do not constitute a limitation to the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 It is a block diagram of an intelligent diagnosis system for mechanical faults of a power transformer according to an embodiment of the present application.

[0018] Figure 2It is a block diagram of a vibration signal analysis module in an intelligent diagnosis system for mechanical faults of a power transformer according to an embodiment of the present application.

[0019] Figure 3 It is a block diagram of a sensor topology matrix analysis module in an intelligent diagnosis system for mechanical faults of a power transformer according to an embodiment of the present application.

[0020] Figure 4 It is a flowchart of an intelligent diagnosis method for mechanical faults of a power transformer according to an embodiment of the present application. Detailed implementation manners

[0021] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. The technical solutions in the embodiments of the present application are clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Figure 1 It is a block diagram of an intelligent diagnosis system for mechanical faults of a power transformer according to an embodiment of the present application. As Figure 1 shown, an intelligent diagnosis system 100 for mechanical faults of a power transformer according to an embodiment of the present application includes: a transformer vibration signal acquisition module 110, configured to obtain a plurality of vibration signals collected by a plurality of vibration sensors deployed on the transformer; a vibration signal analysis module 120, configured to analyze the plurality of vibration signals to obtain a global multi-scale vibration signal time-frequency feature matrix; a sensor topology matrix analysis module 130, configured to construct and analyze a topology matrix of the plurality of vibration sensors to obtain a sensor topology feature matrix; and a transformer internal fault analysis module 140, configured to comprehensively analyze the global multi-scale vibration signal time-frequency feature matrix and the sensor topology feature matrix to obtain a result of whether there is a fault inside the transformer.

[0023] In the above intelligent diagnosis system 100 for mechanical faults of a power transformer, the transformer vibration signal acquisition module 110 is configured to obtain a plurality of vibration signals collected by a plurality of vibration sensors deployed on the transformer. As mentioned in the above background art, traditional methods for detecting mechanical faults in transformer windings include short-circuit impedance method, low-voltage pulse method, frequency response analysis method, etc. However, these methods are all offline detection methods, which require power outage during maintenance and have the disadvantages of long time and high cost. Therefore, an optimized intelligent diagnosis system for mechanical faults of a power transformer is expected.

[0024] To address the above technical problems, an intelligent diagnosis system for mechanical faults of power transformers is proposed. It uses artificial intelligence technology in the field of deep learning to extract features and encode multiple vibration signals collected by multiple vibration sensors of the transformer and the topological matrix of the multiple vibration sensors, so as to obtain the result of whether there are faults inside the transformer. In this way, by intelligently judging the internal fault situation of the transformer, an online monitoring method is adopted, avoiding the need for power outage during maintenance, and reducing the maintenance cost and time.

[0025] Specifically, the vibration analysis method is adopted in the technical solution of this application to detect the mechanical faults of power transformers. The vibration analysis method is a non-electric quantity diagnosis method for transformer mechanical faults. This method uses vibration sensors to measure the vibration signals on the surface of the transformer oil tank. By analyzing the vibration characteristics on the surface of the transformer oil tank, the winding deformation inside the transformer and the changes in the pressing force of the winding and the iron core can be diagnosed. It has the advantages of no direct electrical connection with the power system, safety and rapidity.

[0026] Furthermore, considering that during the operation of a power transformer, its vibration mode has different vibration characteristics at different positions. Therefore, multiple vibration sensors are further used to collect the vibration mode characteristics at multiple positions of the power transformer, and the topological feature information of the multiple vibration sensors is fused to detect abnormal vibration modes of the power transformer. In this way, the vibration mode of the power transformer can be monitored in real time, and when the monitored vibration mode does not conform to the normal mode, a prompt indicating whether there are faults inside the transformer is generated.

[0027] Currently, deep learning and neural networks have been widely applied in fields such as computer vision, natural language processing, and speech signal processing. In addition, deep learning and neural networks have also shown levels close to or even exceeding that of humans in fields such as image classification, object detection, semantic segmentation, and text translation.

[0028] In recent years, the development of deep learning and neural networks has provided new solutions and ideas for the intelligent diagnosis system of power transformer mechanical faults.

[0029] Specifically, first, multiple vibration signals collected by multiple vibration sensors deployed on the transformer are obtained. The vibration signals can reflect the operating state inside the transformer and possible fault situations. By analyzing these vibration signals, abnormal situations inside the transformer can be detected in a timely manner, potential faults can be prevented, and the reliability and safety of the power system can be improved. Multiple vibration sensors are deployed on key parts or key components of the transformer to collect the vibration signals inside the transformer in real time.

[0030] In the above-mentioned intelligent diagnosis system 100 for mechanical faults of power transformers, the vibration signal analysis module 120 is used to analyze the multiple vibration signals to obtain a global multi-scale vibration signal time-frequency feature matrix.

[0031] Figure 2 It is a block diagram of the vibration signal analysis module in the intelligent diagnosis system for mechanical faults of power transformers according to an embodiment of the present application. As Figure 2 shown, the vibration signal analysis module 120 includes: a time-domain transformation unit 121, configured to perform an S-transform on each of the multiple vibration signals respectively to obtain S-transform time-frequency diagrams of the multiple vibration signals; a vibration signal encoding unit 122, configured to respectively pass the S-transform time-frequency diagrams of the multiple vibration signals through a vibration signal feature extraction module including multiple hybrid convolutional layers to obtain multiple multi-scale vibration signal time-frequency feature vectors; and a matrixing unit 123, configured to perform two-dimensional matrix arrangement on the multiple multi-scale vibration signal time-frequency feature vectors to obtain a global multi-scale vibration signal time-frequency feature matrix.

[0032] More specifically, in order to transform the vibration signal from the time domain to the frequency domain, the multiple vibration signals are subjected to an S-transform to obtain a time-frequency diagram, so as to better understand the frequency components and time-domain characteristics of the signal. The S-transform can transform the signal from the time domain to the frequency domain and decompose the signal into components of different frequencies. By analyzing the time-frequency diagram, the energy distribution of the signal at different frequencies can be clearly seen, which helps to identify periodic or specific-frequency components in the signal. Signal processing and filtering in the frequency domain are usually more effective than in the time domain. The time-frequency diagram obtained by the S-transform can be subjected to frequency-domain filtering to remove noise or sudden interference, improving the signal quality and reliability.

[0033] Specifically, in the embodiment of the present application, the time-domain transformation unit 121 is configured to: perform an S-transform on each of the multiple vibration signals respectively according to the following S-transform formula to obtain the S-transform time-frequency diagrams of the multiple vibration signals; where the S-transform formula is:

[0034]

[0035] where s(f,τ) represents the S-transform time-frequency diagram of each vibration signal in the S-transform time-frequency diagrams of the multiple vibration signals, τ is the time shift factor, x(t) represents each vibration signal in the multiple vibration signals, f represents the frequency, and t represents the time.

[0036] Specifically, in the embodiment of the present application, the vibration signal feature extraction module is a convolutional neural network model including multiple hybrid convolutional layers.

[0037] More specifically, the S-transform time-frequency diagrams of multiple vibration signals are passed through a vibration signal feature extraction module that includes multiple hybrid convolutional layers to obtain multiple multi-scale vibration signal time-frequency feature vectors. The vibration signal feature extraction module is a convolutional neural network model that includes multiple hybrid convolutional layers. Vibration signals usually contain features at multiple scales, and vibration components at different frequencies may exhibit different features at different scales. By using multiple hybrid convolutional layers, a convolutional neural network (CNN) model can extract the time-frequency features of vibration signals at different scales, thereby more comprehensively capturing the feature information of the signals. The hybrid convolutional layers can simultaneously consider the feature information at different scales and effectively fuse this information. This helps to comprehensively utilize the features extracted at different scales and improve the richness and accuracy of feature representation. The CNN model has strong learning ability when processing time-frequency diagram data and can learn the complex time-frequency feature representation of vibration signals through training. Multiple hybrid convolutional layers can increase the depth and complexity of the model and improve the model's ability to model time-frequency diagram data. At the same time, by inputting the time-frequency diagram into the CNN model and processing it through multiple hybrid convolutional layers, a higher-dimensional and richer feature vector representation can be obtained. Such feature vectors are more conducive to subsequent feature extraction, classification, or fault diagnosis tasks.

[0038] Specifically, in the embodiment of the present application, the vibration signal encoding unit 122 includes: a first convolutional branch sub-unit for performing convolutional encoding on the S-transform time-frequency diagram of the vibration signal using a first convolutional kernel having a first size to obtain a first feature matrix; a second convolutional branch sub-unit for performing convolutional encoding on the S-transform time-frequency diagram of the vibration signal using a second convolutional kernel having a first dilation rate to obtain a second feature matrix; a third convolutional branch sub-unit for performing convolutional encoding on the S-transform time-frequency diagram of the vibration signal using a third convolutional kernel having a second dilation rate to obtain a third feature matrix; a fourth convolutional branch sub-unit for performing convolutional encoding on the S-transform time-frequency diagram of the vibration signal using a fourth convolutional kernel having a third dilation rate to obtain a fourth feature matrix; a multi-scale feature fusion sub-unit for aggregating the first feature matrix, the second feature matrix, the third feature matrix, and the fourth feature matrix along the channel dimension to obtain a feature map; and a dimension adjustment sub-unit for performing global average pooling processing on the feature map along the channel dimension to obtain the multi-scale vibration signal time-frequency feature vector.

[0039] Specifically, in the embodiment of the present application, the first convolutional kernel, the second convolutional kernel, the third convolutional kernel, and the fourth convolutional kernel have the same size, and the second convolutional kernel, the third convolutional kernel, and the fourth convolutional kernel have different dilation rates.

[0040] More specifically, multiple multi-scale vibration signal time-frequency feature vectors are arranged in a two-dimensional matrix to obtain a global multi-scale vibration signal time-frequency feature matrix. By arranging multiple multi-scale vibration signal time-frequency feature vectors in matrix form, global information can be better retained. This helps capture the feature distribution of the signal in the entire time-frequency space, rather than being limited to the local information of a single feature vector. At the same time, the matrix arrangement can help the model better model the spatial relationship between time-frequency features. In the matrix, elements at adjacent positions usually represent the features of adjacent time-frequency points, which helps the model learn the local structure and correlation of the signal in the time-frequency space. By arranging multiple feature vectors into a matrix, the high-dimensional feature representation can be converted into the form of a two-dimensional matrix, thereby reducing the dimension of the data. This helps reduce the complexity and computational cost of the model while retaining important time-frequency feature information.

[0041] In the above-mentioned intelligent diagnosis system 100 for mechanical faults of power transformers, the sensor topology matrix analysis module 130 is configured to construct a topology matrix of the multiple vibration sensors and perform analysis to obtain a sensor topology feature matrix.

[0042] Figure 3 It is a block diagram of the sensor topology matrix analysis module in the intelligent diagnosis system for mechanical faults of power transformers according to an embodiment of the present application. As Figure 3 shown, the sensor topology matrix analysis module 130 includes: a sensor topology matrix construction unit 131 configured to construct a topology matrix of the multiple vibration sensors, where the value at each non-diagonal position in the topology matrix is the distance between the corresponding two sensors, and the value at each diagonal position in the topology matrix is zero; and a sensor topology matrix feature extraction unit 132 configured to obtain a sensor topology feature matrix by passing the topology matrix of the multiple vibration sensors through a vibration sensor topology feature extraction module.

[0043] More specifically, a topology matrix of multiple vibration sensors is constructed, where the value at each non-diagonal position is the distance between the corresponding two sensors, and the value at each diagonal position is zero. The values at the non-diagonal positions in the topology matrix represent the distances between different sensors. This representation helps consider the spatial relationship between sensors in subsequent data processing and analysis, such as positioning. Through the distance values in the topology matrix, the spatial relationship between sensors in the sensor network can be better modeled. This is crucial for understanding the physical structure of the sensor layout, the signal propagation path, and the correlation between signal sensors. The value at the diagonal position being zero indicates that the distance of each sensor from itself is zero, because the distance of a sensor from itself does not exist. In this way, the construction of the topology matrix can provide important input for subsequent data processing and algorithms.

[0044] Furthermore, the topological matrices of multiple vibration sensors are processed by a vibration sensor topological feature extraction module to obtain a sensor topological feature matrix. The vibration sensor topological feature extraction module is a convolutional neural network model (CNN). Convolutional neural networks are good at extracting features from structured data. By inputting the topological matrix into the CNN model, the convolutional layer and pooling layer of the CNN can be used to automatically learn and extract the complex topological relationship features between sensors. The CNN can effectively capture spatial relationships when processing two-dimensional data. The sensor topological matrix can be regarded as two-dimensional data with a spatial structure, and the spatial relationship between sensors can be better modeled through the CNN model, thereby extracting richer topological features. At the same time, the CNN has the characteristics of parameter sharing and local connection, which makes the CNN effective in processing data with local correlation. The relationship between adjacent positions in the sensor topological matrix can be effectively utilized, and this local correlation can be better used to extract features through the CNN model. Processing the sensor topological matrix through the CNN model can transform high-dimensional topological information into a lower-dimensional feature representation, thereby reducing the complexity and redundancy of the data while retaining the key topological feature information.

[0045] Specifically, in the embodiment of the present application, the sensor topological matrix feature extraction unit 132 is configured to: perform convolutional processing on the input data using the convolutional units of each layer of the vibration sensor topological feature extraction module to obtain a convolutional feature map; perform pooling processing along the channel dimension on the convolutional feature map using the pooling units of each layer of the vibration sensor topological feature extraction module to obtain a pooled feature map; perform non-linear activation on the feature values at each position in the pooled feature map using the activation units of each layer of the vibration sensor topological feature extraction module to obtain an activation feature map; wherein, the output of the last layer of the vibration sensor topological feature extraction module is the sensor topological feature matrix.

[0046] In the above-mentioned intelligent diagnosis system 100 for mechanical faults of power transformers, the transformer internal fault analysis module 140 is configured to comprehensively analyze the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix to obtain a result indicating whether there is a fault inside the transformer.

[0047] Specifically, in the embodiment of the present application, the transformer internal fault analysis module 140 includes: a graph neural encoding unit, configured to obtain a topological global multi-scale transformer vibration time-frequency feature matrix by using a graph neural network for the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix; a feature matrix expansion unit, configured to expand the topological global multi-scale transformer vibration time-frequency feature matrix into a topological global multi-scale transformer vibration time-frequency feature vector; a feature correction unit, configured to perform fine internal feature correction based on matrix decomposition on the topological global multi-scale transformer vibration time-frequency feature vector to obtain a corrected topological global multi-scale transformer vibration time-frequency feature vector; and a transformer internal fault analysis unit, configured to obtain a classification result by passing the corrected topological global multi-scale transformer vibration time-frequency feature vector through a classifier, where the classification result is used to indicate whether there is a fault inside the transformer.

[0048] More specifically, the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix are passed through a graph neural network (GNN) to obtain a topological global multi-scale transformer vibration time-frequency feature matrix. Combining the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix can comprehensively consider the time-frequency features of the vibration signal and the topological relationship between sensors, thereby more comprehensively describing the vibration characteristics of the system. The sensor topological feature matrix can be regarded as a graph structure, where the relationship between sensors is represented by the distance information in the topological matrix. Through the GNN, graph data can be effectively processed, and the topological relationship between sensors can be used to learn the features of the sensor network. At the same time, the GNN can perform cross-layer information transmission in the graph structure, making full use of the topological relationship between sensors, so that information at different levels can interact and transmit with each other, thereby better capturing the complex relationships in the system. Through the GNN, the global multi-scale vibration signal time-frequency features and the sensor topological features can be fused to obtain a richer and more comprehensive feature representation. This helps to improve the feature representation ability and the performance of the system.

[0049] Specifically, in the embodiment of the present application, the graph neural network performs graph structure data encoding on the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix through learnable weight parameters to obtain the topological global multi-scale transformer vibration time-frequency feature matrix including sensor topological features and global multi-scale vibration signal time-frequency features.

[0050] More specifically, expand the topological global multi-scale transformer vibration time-frequency feature matrix into a topological global multi-scale transformer vibration time-frequency feature vector. The topological global multi-scale transformer vibration time-frequency feature matrix may contain a large amount of feature information, and sometimes high-dimensional data will increase the complexity of calculation. By expanding the topological global multi-scale transformer vibration time-frequency feature matrix into a topological global multi-scale transformer vibration time-frequency feature vector, dimensionality reduction processing of the data can be achieved, reducing the computational burden.

[0051] In particular, in the technical solution of this application, considering that the topological global multi-scale transformer vibration time-frequency feature vector, as a high-dimensional data representation of multi-sensor vibration signals after multi-scale time-frequency analysis, the high-order dependence relationship and non-linear coupling structure contained therein essentially originate from the complex physical mechanism of transformer mechanical vibration and the spatial interaction characteristics of the sensor network. Specifically, vibration signals caused by mechanical faults such as transformer winding looseness and inter-turn short circuits are non-linearly superimposed and modulated by multiple frequency components in the time-frequency domain, and there are non-independent interaction effects between the signal characteristics of different frequency components, time points, and sensor spatial positions, forming high-order statistical dependencies that go beyond simple linear correlations. This high-order dependence relationship makes the correlation between feature dimensions present a complex network structure, rather than a simple pattern of pairwise linear correlations, resulting in the difficulty of traditional second-order statistics-based feature analysis methods to effectively analyze its internal structure. Based on this, to solve this technical problem, in the technical solution of this application, perform fine internal feature correction based on matrix decomposition on the topological global multi-scale transformer vibration time-frequency feature vector to obtain a corrected topological global multi-scale transformer vibration time-frequency feature vector.

[0052] Specifically, in the embodiment of this application, the feature correction unit is used to: calculate the full-range fine self-correlation graph matrix of the topological global multi-scale transformer vibration time-frequency feature vector, which is expressed by the formula:

[0053]

[0054] M = D1 ⊙ D2

[0055] v i , v j ∈ V

[0056] where V represents the topological global multi-scale transformer vibration time-frequency feature vector, v i and v j respectively represent the i-th and j-th eigenvalues of the topological global multi-scale transformer vibration time-frequency feature vector, w1, w2, w3, and w4 respectively represent different weight hyperparameters, ⊙ represents matrix dot product, D1 represents the topological global multi-scale transformer vibration time-frequency feature forward weighting matrix, and D2 represents the topological global multi-scale transformer vibration time-frequency feature reverse weighting matrix. Denotes the value at the (i, j) position of the forward weighting matrix of the topological global multi-scale transformer vibration time-frequency characteristics, Denotes the value at the (i, j) position of the reverse weighting matrix of the topological global multi-scale transformer vibration time-frequency characteristics, where M represents the full-range fine self-correlation graph matrix.

[0057] That is, by calculating the full-range fine self-correlation graph matrix of the topological global multi-scale transformer vibration time-frequency feature vector, the complex correlation relationship between multi-dimensional vibration time-frequency characteristics and the sensor spatial layout can be systematically captured, and the structural information hidden in the topological global multi-scale transformer vibration time-frequency feature vector and the spatial dependence relationship of the sensor topology are transformed into an explicit mathematical representation that can be quantitatively analyzed. The full-range fine self-correlation graph matrix generated in this way provides a structured data basis with both global vision and detail accuracy for subsequent in-depth feature analysis based on matrix decomposition.

[0058] Specifically, the feature correction unit is further configured to: perform matrix decomposition on the full-range fine self-correlation graph matrix to obtain a set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component analysis vectors, which is expressed by the formula:

[0059]

[0060] where A represents a diagonal matrix, λ1 and λ m respectively represent the first and the m-th eigenvalues on the diagonal of the diagonal matrix, (·) T represents the transpose of the vector, U represents a set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component analysis vectors, and x1, x2 and x m respectively represent the first, the second and the m-th topological global multi-scale transformer vibration time-frequency feature fine eigen-component analysis vectors in the set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component analysis vectors.

[0061] That is, by performing matrix decomposition on the full-range fine autocorrelation map matrix, the complex correlation formed by the coupling of multi-scale vibration time-frequency features and sensor topological structures can be effectively deconstructed, and a set of topological global multi-scale transformer vibration time-frequency feature fine intrinsic component analytical vectors guided by the intrinsic structure of the data can be generated. Specifically, the feature correlation structure of high-dimensional coupling is converted into a set of fine intrinsic component analytical vectors of topological global multi-scale transformer vibration time-frequency features sorted by importance level, so as to realize the decoupling of complex mutual interference patterns in the topological global multi-scale transformer vibration time-frequency feature space. The set of fine intrinsic component analytical vectors of topological global multi-scale transformer vibration time-frequency features obtained by decomposition not only retains the core structural information of the topological global multi-scale transformer vibration time-frequency features, but also eliminates the redundant correlation between feature dimensions through orthogonalization processing, providing a low-coupling and high-interpretation feature expression form for the subsequent sparse processing of feature information and the directional extraction of key fault features, thereby supporting the targeted adjustment and optimization based on the feature independent mode, and effectively improving the fusion efficiency of multi-source heterogeneous features and the analytical accuracy of fault features in transformer mechanical fault diagnosis.

[0062] Specifically, the feature correction unit is further used to: perform information sparsification on each topological global multi-scale transformer vibration time-frequency feature fine intrinsic component analytical vector in the set of topological global multi-scale transformer vibration time-frequency feature fine intrinsic component analytical vectors to obtain a set of topological global multi-scale transformer vibration time-frequency feature fine intrinsic component optimization vectors, which is expressed by the formula:

[0063]

[0064] Among them, x i represents the analytic vector of the fine eigencomponent of the vibration time-frequency characteristics of the i-th topological global multi-scale transformer, ||·|| represents the Euclidean norm, and y i represents the optimized vector of refined intrinsic components of the i-th topological global multi-scale transformer vibration time-frequency characteristics.

[0065] That is, by performing information sparsification on the set of analytical vectors of the fine intrinsic components of the topological global multi-scale transformer vibration time-frequency characteristics, we can further explore higher-order dependencies beyond second-order statistics on the basis of the decoupled intrinsic components and enhance their discrimination ability. The set of optimized vectors of the fine intrinsic components of the topological global multi-scale transformer vibration time-frequency characteristics obtained by information sparsification focuses more on key features, reduces redundant information, and provides a more concise and effective feature expression for subsequent analysis and fault diagnosis, which helps to improve the accuracy and efficiency of online intelligent diagnosis of transformer mechanical faults.

[0066] Specifically, the feature correction unit is further configured to: calculate the internal feature dominant adjustment parameters of each topological global multi-scale transformer vibration time-frequency feature fine intrinsic component optimization vector in the set of topological global multi-scale transformer vibration time-frequency feature fine intrinsic component optimization vectors to obtain a set of internal feature dominant adjustment parameters, which is expressed by the formula:

[0067]

[0068] where α and β respectively represent different weight hyperparameters, ||·||1 represents the first norm, ||·||2 represents the second norm, L represents the length of the topological global multi-scale transformer vibration time-frequency feature fine intrinsic component optimization vector, and a i represents the internal feature dominant adjustment parameter corresponding to y i .

[0069] That is, calculating the internal feature dominant adjustment parameters of the topological global multi-scale transformer vibration time-frequency feature fine intrinsic component optimization vector can quantify a scalar value for each optimization vector to measure the value of its internal structure information after information sparsification processing. The obtained set of internal feature dominant adjustment parameters provides a dynamic and current-state-based importance basis for subsequent feature fusion.

[0070] Specifically, the feature correction unit is further configured to: perform a normalization process on the set of internal feature dominant adjustment parameters to obtain a set of internal feature dynamic balance coefficients, which is expressed by the formula:

[0071] w i = Softmax(a i )

[0072] where Softmax(·) represents the normalization function, and w i represents the internal feature dynamic balance coefficient corresponding to a i .

[0073] That is, the internal feature dominant adjustment parameters are transformed into internal feature dynamic balance coefficients that can be used for fusion. Specifically, by imposing a normalization constraint, it avoids misjudgment caused by numerical instability in subsequent comprehensive analysis and also prevents over-reliance on a few feature components. This normalizes the scale of the internal feature dominant adjustment parameters, enabling subsequent fusion to reasonably allocate weights according to the actual importance, ensuring that different vibration time-frequency features and sensor topological features can be fused in a balanced and controllable manner when analyzing internal faults of the transformer, thereby improving the reliability and stability of the fault diagnosis results.

[0074] Specifically, the feature correction unit is further configured to: based on the set of internal feature dynamic balance coefficients, perform fine integration on the set of optimized vectors of the fine eigen-components of the topological global multi-scale transformer vibration time-frequency features to obtain a corrected topological global multi-scale transformer vibration time-frequency feature vector, which is expressed by the formula:

[0075]

[0076] where V' represents the corrected topological global multi-scale transformer vibration time-frequency feature vector.

[0077] That is, using the internal feature dynamic balance coefficients obtained by normalization processing, perform fine integration on the set of optimized vectors of the fine eigen-components of the topological global multi-scale transformer vibration time-frequency features, so that the finally obtained corrected topological global multi-scale transformer vibration time-frequency feature vector condenses the key structural information in the original vibration and topological features, improves the representation ability through non-linear processing, is more robust and discriminative than the topological global multi-scale transformer vibration time-frequency feature vector, and can more effectively serve the subsequent task of intelligent diagnosis of transformer internal faults.

[0078] Furthermore, pass the corrected topological global multi-scale transformer vibration time-frequency feature vector through a classifier to obtain a classification result, which is used to indicate whether there is a fault inside the transformer. By extracting features and classifying the vibration signals of the transformer, it can help diagnose whether there are faults inside the transformer. Different types of faults may cause changes in the characteristics of the vibration signals. Through the classifier, these characteristics can be analyzed to judge the working state of the transformer. By automatically analyzing the vibration signals through the classifier, the burden of manual analysis can be reduced, and the efficiency and accuracy of fault diagnosis can be improved. At the same time, the situation of power outage during maintenance is avoided, and the maintenance cost and time are reduced.

[0079] Specifically, in the embodiment of the present application, the transformer internal fault analysis unit is configured to: use the classifier to process the corrected topological global multi-scale transformer vibration time-frequency feature vector with the following classification formula to obtain the classification result; where the classification formula is:

[0080] O = softmax{(W c , B c )|Project(F)}

[0081] where Project(F) represents projecting the corrected topological global multi-scale transformer vibration time-frequency feature vector into a vector, W c is the weight matrix, B c represents the bias vector, softmax represents the normalized exponential function, and O represents the classification result.

[0082] In summary, the intelligent diagnosis system for mechanical faults of power transformers according to the embodiments of the present application has been elucidated. It uses artificial intelligence technology in the field of deep learning to extract features and encode multiple vibration signals collected by multiple vibration sensors of the transformer and the topological matrix of the multiple vibration sensors, so as to obtain the result of whether there is a fault inside the transformer. In this way, by intelligently judging the internal fault situation of the transformer, an online monitoring method is adopted, avoiding the situation of power outage during maintenance, and reducing the maintenance cost and time.

[0083] As described above, the intelligent diagnosis system 100 for mechanical faults of power transformers according to the embodiments of the present application can be implemented in various terminal devices, such as an intelligent diagnosis server for mechanical faults of power transformers, etc. In one example, the intelligent diagnosis system 100 for mechanical faults of power transformers according to the embodiments of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the intelligent diagnosis system 100 for mechanical faults of power transformers can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the intelligent diagnosis system 100 for mechanical faults of power transformers can also be one of the many hardware modules of the terminal device.

[0084] Alternatively, in another example, the intelligent diagnosis system 100 for mechanical faults of power transformers and the terminal device can also be separate devices, and the intelligent diagnosis system 100 for mechanical faults of power transformers can be connected to the terminal device through a wired and / or wireless network, and transmit and interact information according to a predefined data format.

[0085] Based on the same inventive concept, the embodiments of the present application also provide an intelligent diagnosis method for mechanical faults of power transformers, which can be used to implement the system described in the above embodiments, as described in the following embodiments.

[0086] Figure 4 FIG. is a flowchart of the intelligent diagnosis method for mechanical faults of power transformers according to the embodiments of the present application. As Figure 4 shown, the intelligent diagnosis method for mechanical faults of power transformers according to the embodiments of the present application includes the steps of: S110, obtaining multiple vibration signals collected by multiple vibration sensors deployed on the transformer; S120, analyzing the multiple vibration signals to obtain a global multi-scale vibration signal time-frequency feature matrix; S130, constructing and analyzing the topological matrix of the multiple vibration sensors to obtain a sensor topological feature matrix; and S140, comprehensively analyzing the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix to obtain the result of whether there is a fault inside the transformer.

[0087] In the present specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. Here, for the intelligent diagnosis method of mechanical faults of power transformers disclosed in the embodiments, those skilled in the art can understand that the specific operations of each step in the above intelligent diagnosis method of mechanical faults of power transformers have been described in detail in the description of the Figures 1 to 3 intelligent diagnosis system of mechanical faults of power transformers above. Therefore, the description is relatively simple. For the relevant parts, reference can be made to the description of the intelligent diagnosis system of mechanical faults of power transformers, and thus, the repeated description thereof will be omitted.

[0088] In summary, the intelligent diagnosis method of mechanical faults of power transformers according to the embodiments of the present application has been clarified. It uses artificial intelligence technology in the field of deep learning to extract features and encode multiple vibration signals collected by multiple vibration sensors of the transformer and the topological matrix of the multiple vibration sensors to obtain the result of whether there are faults inside the transformer. In this way, by intelligently judging the internal fault situation of the transformer, an online monitoring method is adopted, avoiding the situation of power outage during maintenance, and reducing the maintenance cost and time.

[0089] In summary, after reading this detailed disclosure, those skilled in the art can understand that the foregoing detailed disclosure may be presented only by way of example and may not be restrictive. Although not explicitly stated here, those skilled in the art can understand that the present application is intended to encompass various reasonable changes, improvements, and modifications to the embodiments. These changes, improvements, and modifications are intended to be proposed by the present application and are within the spirit and scope of the exemplary embodiments of the present application.

[0090] In addition, certain terms in the present application have been used to describe the embodiments of the present application. For example, "one embodiment", "embodiment" and / or "some embodiments" mean that the specific features, structures, or characteristics described in connection with the embodiment may be included in at least one embodiment of the present application. Therefore, it should be emphasized and understood that two or more references to "embodiment" or "one embodiment" or "alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. In addition, the specific features, structures, or characteristics may be appropriately combined in one or more embodiments of the present application.

[0091] Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that an article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such article or device. In the absence of more restrictions, the elements defined by the phrase "comprising a ..." do not exclude the presence of additional identical elements in the article or device comprising the above elements.

[0092] It should be understood that in the foregoing description of the embodiments of the present application, in order to help understand a feature and for the purpose of simplifying the present application, the present application combines various features in a single embodiment, drawing or its description. However, this does not mean that the combination of these features is necessary. When reading the present application, it is entirely possible for a person skilled in the art to extract some of the features and understand them as separate embodiments. In other words, the embodiments in the present application can also be understood as the integration of multiple secondary embodiments. This is also true when the content of each secondary embodiment is less than all the features of a single aforementioned disclosed embodiment.

[0093] Finally, it should be understood that the embodiments of the application disclosed herein are explanations of the principles of the embodiments of the present application. Other modified embodiments are also within the scope of the present application. Therefore, the embodiments disclosed in the present application are only used as examples and are not limited. Those skilled in the art can adopt alternative configurations to realize the application in the present application according to the embodiments in the present application.

[0094] Therefore, the embodiments of the present application are not limited to the embodiments precisely described in the application.

Claims

1. An intelligent diagnosis system for mechanical faults of power transformers, characterized in that, Including: A transformer vibration signal acquisition module, configured to obtain a plurality of vibration signals collected by a plurality of vibration sensors deployed on the transformer; A vibration signal analysis module, configured to analyze the plurality of vibration signals to obtain a global multi-scale vibration signal time-frequency feature matrix; A sensor topology matrix analysis module, configured to construct and analyze a topology matrix of the plurality of vibration sensors to obtain a sensor topology feature matrix; A transformer internal fault analysis module, configured to comprehensively analyze the global multi-scale vibration signal time-frequency feature matrix and the sensor topology feature matrix to obtain a result of whether there is a fault inside the transformer.

2. The intelligent diagnosis system for mechanical faults of a power transformer according to claim 1, characterized in that The vibration signal analysis module includes: A time-domain transformation unit, configured to perform an S-transform on each of the plurality of vibration signals respectively to obtain a plurality of vibration signal S-transform time-frequency diagrams; A vibration signal encoding unit, configured to respectively pass the plurality of vibration signal S-transform time-frequency diagrams through a vibration signal feature extraction module including a plurality of hybrid convolutional layers to obtain a plurality of multi-scale vibration signal time-frequency feature vectors; A matrixization unit, configured to perform two-dimensional matrix arrangement on the plurality of multi-scale vibration signal time-frequency feature vectors to obtain a global multi-scale vibration signal time-frequency feature matrix.

3. The intelligent diagnosis system for mechanical faults of a power transformer according to claim 2, characterized in that, The time-domain transformation unit is configured to: Perform an S-transform on each of the plurality of vibration signals respectively according to the following S-transform formula to obtain the plurality of vibration signal S-transform time-frequency diagrams; Wherein, the S-transform formula is: Wherein, s(f,τ) represents each vibration signal S-transform time-frequency diagram in the plurality of vibration signal S-transform time-frequency diagrams, τ is a time shift factor, x(t) represents each vibration signal in the plurality of vibration signals, f represents frequency, and t represents time.

4. The intelligent diagnosis system for mechanical faults of a power transformer according to claim 3, wherein, The vibration signal feature extraction module is a convolutional neural network model including a plurality of hybrid convolutional layers.

5. The intelligent diagnosis system for mechanical faults of a power transformer according to claim 4, characterized in that, The sensor topology matrix analysis module includes: A sensor topology matrix construction unit, configured to construct a topology matrix of the plurality of vibration sensors, where the values at each non-diagonal position in the topology matrix are the distances between the corresponding two sensors, and the values at each diagonal position in the topology matrix are zero; A sensor topology matrix feature extraction unit, configured to pass the topology matrix of the plurality of vibration sensors through a vibration sensor topology feature extraction module to obtain a sensor topology feature matrix.

6. The intelligent diagnosis system for mechanical faults of a power transformer according to claim 5, characterized in that, The sensor topology matrix feature extraction unit is configured to: Perform convolution processing on the input data using the convolution units of each layer of the vibration sensor topology feature extraction module to obtain a convolution feature map; Perform pooling processing along the channel dimension on the convolution feature map using the pooling units of each layer of the vibration sensor topology feature extraction module to obtain a pooling feature map; Perform non-linear activation on the feature values at each position in the pooling feature map using the activation units of each layer of the vibration sensor topology feature extraction module to obtain an activation feature map; Wherein, the output of the last layer of the vibration sensor topology feature extraction module is the sensor topology feature matrix.

7. The intelligent diagnosis system for mechanical faults of a power transformer according to claim 6, characterized in that, The transformer internal fault analysis module includes: A graph neural coding unit, which is used to obtain a topological global multi-scale transformer vibration time-frequency feature matrix by means of a graph neural network for the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix; A feature matrix expansion unit, which is used to expand the topological global multi-scale transformer vibration time-frequency feature matrix into a topological global multi-scale transformer vibration time-frequency feature vector; A feature correction unit, which is used to perform fine internal feature correction based on matrix decomposition on the topological global multi-scale transformer vibration time-frequency feature vector to obtain a corrected topological global multi-scale transformer vibration time-frequency feature vector; A transformer internal fault analysis unit, which is used to pass the corrected topological global multi-scale transformer vibration time-frequency feature vector through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a fault inside the transformer.

8. The intelligent diagnosis system for mechanical faults of a power transformer according to claim 7, characterized in that The feature correction unit is used for: Calculating a full-range fine self-correlation graph matrix of the topological global multi-scale transformer vibration time-frequency feature vector; Performing matrix decomposition on the full-range fine self-correlation graph matrix to obtain a set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component analysis vectors; Performing information sparsification on each topological global multi-scale transformer vibration time-frequency feature fine eigen-component analysis vector in the set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component analysis vectors to obtain a set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component optimization vectors; Calculating internal feature dominant adjustment parameters for each topological global multi-scale transformer vibration time-frequency feature fine eigen-component optimization vector in the set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component optimization vectors to obtain a set of internal feature dominant adjustment parameters; Performing normalization processing on the set of internal feature dominant adjustment parameters to obtain a set of internal feature dynamic balance coefficients; Based on the set of internal feature dynamic balance coefficients, performing fine integration on the set of topological global multi-scale transformer vibration time-frequency feature fine eigen-component optimization vectors to obtain a corrected topological global multi-scale transformer vibration time-frequency feature vector.

9. An intelligent diagnosis method for mechanical faults of power transformers, characterized in that, It includes: Obtaining a plurality of vibration signals collected by a plurality of vibration sensors deployed on the transformer; Analyzing the plurality of vibration signals to obtain a global multi-scale vibration signal time-frequency feature matrix; Constructing a topological matrix of the plurality of vibration sensors and analyzing it to obtain a sensor topological feature matrix; Comprehensively analyzing the global multi-scale vibration signal time-frequency feature matrix and the sensor topological feature matrix to obtain a result of whether there is a fault inside the transformer.

Images