A Multi-Sensor Collaborative Partial Discharge Waveform Recognition Method and System

Abstract

This invention discloses a multi-sensor collaborative method and system for partial discharge waveform recognition. The method includes extracting the envelope of the oscillating waveform from the acquired partial discharge signal using Hilbert transform; fitting each sensor signal with a Logistic function and an exponential decay function; using the feature parameters extracted from each sensor as node features in a graph structure; establishing edge connections in the graph based on the time difference and similarity between the maximum values ​​of the signals from different sensors; constructing a graph neural network model; using the graph structure for training and inference; and finally outputting a result indicating whether it is a partial discharge event. The system includes a signal conversion module, a feature extraction module, a data fusion module, and a partial discharge diagnosis module. This invention improves the accuracy and adaptability of partial discharge signal recognition, meeting the online monitoring needs of transformers in complex electromagnetic environments.

Description

Technical Field

[0001] This invention relates to the field of online monitoring of primary equipment in power systems, and in particular to a method and system for identifying partial discharge waveforms using multi-sensor collaboration. Background Technology

[0002] Partial discharge (PD) refers to a non-penetrating discharge phenomenon limited to a localized area in the insulation system of electrical equipment. It is one of the important precursors to insulation degradation in high-voltage electrical equipment. In high-voltage DC transmission and transformation equipment such as converter transformers, partial discharge is often caused by manufacturing defects, operational aging, or environmental stress, and manifests as weak but recurring electromagnetic waves, acoustic emission signals, and high-frequency current disturbances.

[0003] As a core component of high-voltage direct current (HVDC) transmission systems, the reliability of the converter transformer's insulation system directly affects the safe and stable operation of the entire HVDC transmission project. Due to its high operating voltage level, complex operating conditions, and long-term exposure to a combined AC / DC electric field, higher requirements are placed on the identification of partial discharge signals: not only is high-sensitivity acquisition capability required, but also strong anti-interference capability and multi-source information fusion mechanism.

[0004] Common methods in existing partial discharge detection technologies include: threshold-triggered pulse peak detection, which determines whether a partial discharge event has occurred by setting an amplitude or energy threshold; frequency domain analysis, which uses Fast Fourier Transform (FFT) or Wavelet Transform (WT) to extract the spectral features of the partial discharge signal; pattern recognition, which combines traditional machine learning algorithms (such as Support Vector Machine (SVM) and Random Forest (RF)) for classification; and multi-sensor collaborative detection, which uses UHF electromagnetic sensors, HFCT current sensors, and AE acoustic sensors to jointly determine partial discharge events.

[0005] However, the above methods have several problems in practical applications: low edge feature extraction accuracy: traditional methods usually rely on fixed windows to extract rising and falling edges, making it difficult to accurately capture key features such as steep rises and slow falls in partial discharge signals, resulting in a high false detection rate; imperfect multi-sensor fusion mechanism: existing studies mostly use result-level voting or feature splicing to process multi-source signals, lacking modeling of the physical consistency and time synchronization between sensors, which cannot effectively improve the robustness of recognition; poor anti-interference capability: in actual operating environments, converter transformer partial discharge signals are often accompanied by complex electromagnetic interference and environmental noise. Due to the weak and transient nature of partial discharge signals, traditional partial discharge identification methods are prone to false alarms or missed alarms when facing interference such as high-frequency noise, transient processes of equipment operation, and external electromagnetic coupling.

[0006] Therefore, there is an urgent need to propose a new method that can effectively extract key edge features from partial discharge waveforms and intelligently fuse them with data from multiple sensors. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to provide a multi-sensor collaborative partial discharge waveform recognition method and system, which aims to solve the problems of low waveform feature extraction accuracy, imperfect multi-sensor fusion mechanism and poor system robustness in the prior art.

[0008] Technical solution: The multi-sensor collaborative partial discharge waveform recognition method of the present invention includes the following steps:

[0009] Step 1: Install multiple sensors on the transformer tank wall to collect partial discharge signals inside the transformer during operation. The sensors include ultra-high frequency sensors, high frequency sensors, and acoustic sensors.

[0010] Step 2: Perform Hilbert transform on the acquired signals to extract the envelope of the oscillating waveform;

[0011] Step 3: Apply the Logistic function to the rising edge region of the envelope of each sensor for nonlinear fitting, and extract the steepness and start time of the Logistic function as edge features;

[0012] Step 4: Apply an exponential decay function to the falling edge region of the envelope of each sensor for nonlinear fitting, and extract the time constant as the energy dissipation feature;

[0013] Step 5: Use the feature parameters extracted by each sensor as node features in the graph structure;

[0014] Step 6: Calculate the edge weights based on the start time difference and similarity of the signals between the sensors, obtain the edge attributes between nodes, and establish the edge connection relationships in the graph structure;

[0015] Step 7: Construct a graph neural network model and use the graph structure for training and inference to achieve collaborative fusion of information from multiple sensor sources;

[0016] Step 8: Output the identification result of whether it is a partial discharge event.

[0017] Furthermore, step 3 employs the Logistic function for nonlinear fitting, the Logistic function being expressed as follows:

[0018]

[0019] in, Indicates the platform height. Indicates the steepness of the edge. Indicates the start time.

[0020] Furthermore, step 4 employs an exponential decay function for nonlinear fitting, the exponential decay function being expressed as follows:

[0021]

[0022] in, Indicates the initial amplitude. This represents the time constant of the falling edge.

[0023] Furthermore, the node features in the graph structure in step 5 consist of parameters extracted by the Logistic function and the exponential decay function, including steepness, start time, and time constant.

[0024] Furthermore, the calculation of edge weights in step 6 includes:

[0025] Step 6.1: Calculate the start time difference between different sensor signals ;

[0026] Step 6.2: If the time difference is less than the threshold, then establish an edge between the two corresponding nodes;

[0027] Step 6.3: The edge weights are dynamically assigned based on the time difference. The smaller the time difference, the larger the edge weight, indicating a higher degree of consistency between the two.

[0028] Furthermore, the weights of the edges in step 6.2 not only consider the time difference but also incorporate the correlation score between sensors. This correlation score is derived from statistical analysis of previous historical data and is calculated as follows:

[0029]

[0030] in, , This represents the signal amplitude collected by sensors m and n at time t. , This represents the average signal values ​​of sensors m and n. This represents the signal correlation between sensors m and n, with a value range of [-1, 1].

[0031] Furthermore, step 6.2 introduces an edge attribute mechanism into the graph neural network, combining the time difference and similarity into an edge attribute vector:

[0032]

[0033] It is then fed into a graph neural network layer that supports edge attributes for training and inference.

[0034] Furthermore, in step 7, the graph neural network uses a graph isomorphic network framework to model the graph structure, and the graph isomorphic network framework uses GINEConv layers as graph convolutional layers.

[0035] The update rules for the GINEConv layer are as follows:

[0036]

[0037] in, It is the first The node at the th Layer embedding representation; It is a node The set of neighbors; It is an edge attribute vector; It is a learnable edge feature mapping function; It is a shared multilayer perceptron; It is a trainable scaling factor.

[0038] Furthermore, step 8 introduces a global average pooling operation into the output of the graph neural network to obtain a graph-level embedding vector, and inputs it into a classifier to determine whether a partial discharge event has occurred.

[0039] The multi-sensor collaborative partial discharge waveform recognition system of the present invention includes:

[0040] The signal conversion module is used to perform Hilbert transform on the acquired signals and extract the envelope of the oscillating waveform.

[0041] The feature extraction module is used to apply the Logistic function to the rising edge region of the envelope of each sensor for nonlinear fitting, and extract the Logistic function steepness and start time as edge features; and to apply the exponential decay function to the falling edge region of the envelope of each sensor for nonlinear fitting, and extract the time constant as energy dissipation feature.

[0042] The data fusion module is used to use the extracted edge features and energy dissipation features as node features in the graph structure to build a graph neural network model, and to use the graph structure for training and inference to achieve collaborative fusion of information from multiple sensor sources.

[0043] The partial discharge diagnosis module is used to introduce the output of the graph neural network model into a global average pooling operation to obtain a graph-level embedding vector, and then input it into a classifier to determine whether a partial discharge event has occurred.

[0044] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention uses function fitting to extract key parameters with physical meaning, avoiding the problem of information loss and improving the accuracy of edge recognition; the introduction of edge attribute mechanism into the graph neural network model enables the model to consider time synchronization and sensor correlation at the same time, realizing true feature-level fusion, breaking through the limitations of traditional "result voting" or "feature splicing", making the final recognition result more stable and consistent, and suitable for online monitoring tasks of transformers in complex electromagnetic environments on site. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the multi-sensor collaborative partial discharge waveform recognition method of the present invention.

[0046] Figure 2 This is a diagram illustrating the effect of curve fitting and waveform feature extraction in this invention.

[0047] Figure 3 A graph structure diagram illustrating the feature node modeling of this invention;

[0048] Figure 4 This is a technical roadmap for implementing partial discharge waveform recognition in this invention. Detailed Implementation

[0049] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0050] like Figure 1 As shown, the multi-sensor collaborative partial discharge waveform recognition method of the present invention includes the following steps:

[0051] Step 1: Install multiple sensors on the transformer tank wall to collect partial discharge signals inside the transformer during operation. The sensors include ultra-high frequency sensors, high frequency sensors, and acoustic sensors.

[0052] Ultra-high frequency (UHF) sensors: used to detect electromagnetic wave signals;

[0053] High-frequency current sensor (HFCT): used to detect partial discharge induced current in windings;

[0054] Acoustic sensor (AE): Used to detect ultrasonic signals caused by partial discharge.

[0055] The three types of sensors mentioned above are deployed in a distributed manner according to the actual equipment structure and monitoring requirements, forming a multi-channel acquisition system. For example, in one specific embodiment, a total of 14 sensor channels are deployed, including 4 UHF channels, 4 HFCT channels, and 6 AE channels, to simultaneously acquire multi-source information of partial discharge signals.

[0056] The raw signals collected by each sensor are time-domain oscillating waveforms, exhibiting non-stationary and non-linear characteristics, and containing varying degrees of noise interference and electromagnetic coupling effects.

[0057] Step 2: Perform Hilbert transform on the acquired signals to extract the envelope of the oscillating waveform;

[0058] The raw oscillation signals acquired by each sensor are processed by Hilbert transform to construct an analytical signal, and its magnitude is extracted as the envelope of the signal.

[0059] Specifically, for the signal acquired by any sensor Its analytic signal is defined as:

[0060]

[0061] in, It is the result of the Hilbert transform of the signal. It is an analytic signal in complex form.

[0062] Then the magnitude of the analytic signal is calculated:

[0063]

[0064] This envelope reflects the overall energy change trend of the signal, eliminates the influence of high-frequency oscillations, and facilitates subsequent edge identification and function fitting.

[0065] Step 3: Apply the Logistic function to the rising edge region of the envelope of each sensor for nonlinear fitting, and extract the steepness and start time of the Logistic function as edge features;

[0066] After the envelope extraction is completed, the main rising edge region of each sensor signal is identified.

[0067] Specifically, gradient analysis or thresholding methods are used to identify the interval where the signal amplitude rapidly rises from the noise floor to the peak value. Then, a nonlinear least-squares fit is performed on this region using the Logistic function.

[0068]

[0069] in, Indicates the platform height. Indicates the steepness of the edge. Indicates the start time of the Logistic fit, such as Figure 2 The blue fitted line is shown.

[0070] These parameters have clear physical meanings and can be used to characterize the rising phase of partial discharge signals. Furthermore, the initial parameters are automatically estimated based on the maximum signal value and time axis information, improving the algorithm's stability and adaptability.

[0071] Step 4: Apply an exponential decay function to the falling edge region of the envelope of each sensor for nonlinear fitting, and extract the time constant as the energy dissipation feature;

[0072] The falling edge region of each sensor signal is identified, representing the process by which the signal slowly declines from its peak to low noise. An exponential decay function is then applied to this region for nonlinear fitting.

[0073]

[0074] in, Indicates the initial amplitude. The time constant representing the falling edge, such as Figure 2 The green fitted line is shown.

[0075] This time constant can be used as an important input feature for subsequent fusion models.

[0076] Step 5: Use the feature parameters extracted by each sensor as node features in the graph structure;

[0077] Treating all sensor acquisition channels as nodes in a graph structure, with each node corresponding to one sensor channel, specifically, the feature of each node is the steepness fitted by a Logistic function. Start time and the time constant of the exponential decay function fitting constitute.

[0078] Step 6: Calculate edge weights based on the start time difference and similarity of signals between sensors to obtain edge attributes between nodes, and establish edge connection relationships in the graph structure, such as... Figure 3 As shown, the edge connections in the graph are established based on the start time difference of the signals between the sensors. Specifically, the start time of the Logistic function is extracted. Calculate the initial time difference between any two sensors m and n. When the time difference is less than a set threshold (1ms), the two signals are considered to represent the same partial discharge. An edge is established between the corresponding two nodes. The weight of this edge considers not only the time difference but also the correlation score between the sensors. The correlation score is derived from statistical analysis of previous historical data, and the specific calculation method is as follows:

[0079]

[0080] in, , This represents the signal amplitude collected by sensors m and n at time t. , This represents the average signal values ​​of sensors m and n. This represents the signal correlation between sensors m and n, with a value range of [-1, 1]. An edge attribute mechanism is introduced into the graph neural network to combine the time difference and similarity into an edge attribute vector:

[0081]

[0082] The data is then fed into a graph neural network layer that supports edge attributes for training and inference. Edge weights are dynamically assigned based on the time difference; the smaller the time difference, the larger the edge weight, indicating higher consistency between the two.

[0083] Step 7: Construct a graph neural network model and use the graph structure for training and inference to achieve collaborative fusion of information from multiple sensor sources;

[0084] like Figure 4 As shown, after completing the graph structure modeling, a graph neural network model is constructed to achieve the collaborative fusion of information from multiple sensor sources and output the identification results of partial discharge events. In the embodiments of this invention, the GINEConv (GraphIsomorphism Network with Edge features) layer is used as the core component of the graph neural network. This layer is an extended version of the GraphIsomorphism Network (GIN), capable of accepting and processing edge attribute information, and is very suitable for the scenario of fusing time difference and similarity scoring in this invention.

[0085] The update rules for the GINEConv layer are as follows:

[0086]

[0087] in, It is the first The node at the th Layer embedding representation; It is a node The set of neighbors; It is an edge attribute vector that includes time difference and relevance score; It is a learnable edge feature mapping function; It is a shared multilayer perceptron used to aggregate neighbor information; It is a trainable scaling factor.

[0088] Using the above formula, GINEConv can simultaneously consider the node's own state and the information propagated along the edges, thereby achieving more refined graph-level modeling.

[0089] The propagation process of a multi-layer GNN is as follows:

[0090] 1. First layer GINEConv:

[0091] Input: Original node features (such as Logistic and exponential fitting parameters);

[0092] Output: Node embeddings after initial aggregation of neighbor information;

[0093] 2. Second layer GINEConv:

[0094] Input: Node embedding of the previous layer + edge attributes;

[0095] Output: A high-dimensional representation after further information propagation;

[0096] 3. Subsequent layers:

[0097] Depending on the actual scenario, GINEConv layers can be stacked to improve the model's ability to understand graph structures;

[0098] Each layer of GINEConv updates the node representation, enabling each sensor channel to progressively perceive the state of its neighboring sensors, and combines the time difference and correlation score in the edge attributes to form a unified graph structure representation.

[0099] In the last layer of the graph neural network, global mean pooling is used to obtain a unified embedding vector at the graph level:

[0100]

[0101] Where N is the total number of sensor nodes; It is the node embedding output of the last layer of the graph neural network; It is a graph-level embedding vector that represents the overall information of the entire graph.

[0102] Step 8: Output the identification result of whether it is a partial discharge event.

[0103] Subsequently, the embedded vector obtained in step 7 is input into the classifier, which outputs the probability value of whether the current event is a partial discharge event:

[0104]

[0105] in, , It is the weight matrix of the classifier; , It is a bias term; It is a sigmoid function that outputs a probability value (0~1). When the output probability is greater than a set threshold (e.g., 0.5), it is determined to be a partial discharge event; otherwise, it is considered a false alarm or background noise.

Claims

1. A partial discharge waveform recognition method of multi-sensor cooperation, characterized in that, Includes the following steps: Step 1: Install multiple sensors on the transformer tank wall to collect partial discharge signals inside the transformer during operation. The sensors include ultra-high frequency sensors, high frequency sensors, and acoustic sensors. Step 2: Perform Hilbert transform on the acquired signals to extract the envelope of the oscillating waveform; Step 3: Apply the Logistic function to the rising edge region of the envelope of each sensor for nonlinear fitting, and extract the steepness and start time of the Logistic function as edge features; Step 4: Apply an exponential decay function to the falling edge region of the envelope of each sensor for nonlinear fitting, and extract the time constant as the energy dissipation feature; Step 5: Use the feature parameters extracted by each sensor as node features in the graph structure; Step 6: Calculate the edge weights based on the start time difference and similarity of the signals between the sensors, obtain the edge attributes between nodes, and establish the edge connection relationships in the graph structure; Step 7: Construct a graph neural network model and use the graph structure for training and inference to achieve collaborative fusion of information from multiple sensor sources; Step 8: Output the identification result of whether it is a partial discharge event.

2. The multi-sensor collaborative partial discharge waveform recognition method according to claim 1, characterized in that, Step 3 uses the Logistic function for nonlinear fitting. The Logistic function is expressed as follows: ， in, Indicates the platform height. Indicates the steepness of the edge. Indicates the start time.

3. The multi-sensor collaborative partial discharge waveform recognition method according to claim 1, characterized in that, Step 4 employs an exponential decay function for nonlinear fitting. The exponential decay function is expressed as follows: ， in, Indicates the initial amplitude. This represents the time constant of the falling edge.

4. The multi-sensor collaborative partial discharge waveform recognition method according to claim 1, characterized in that, The node features in the graph structure in step 5 consist of parameters extracted by the Logistic function and the exponential decay function, including steepness, start time, and time constant.

5. The multi-sensor collaborative partial discharge waveform recognition method according to claim 1, characterized in that, The construction of edge connection relationships in step 6 includes: Step 6.1: Calculate the start time difference between different sensor signals ; Step 6.2: If the time difference is less than the threshold, then establish an edge between the two corresponding nodes; Step 6.3: The edge weights are dynamically assigned based on the time difference. The smaller the time difference, the larger the edge weight, indicating a higher degree of consistency between the two.

6. The multi-sensor collaborative partial discharge waveform recognition method according to claim 5, characterized in that, The weights of the edges in step 6.2 not only consider the time difference, but also the correlation score between sensors. The correlation score is derived from the statistical analysis of previous historical data, and the calculation method is as follows: ， in, , This represents the signal amplitude collected by sensors m and n at time t. , This represents the average signal values ​​of sensors m and n. This represents the signal correlation between sensors m and n, with a value range of [-1, 1].

7. The multi-sensor collaborative partial discharge waveform recognition method according to claim 5, characterized in that, Step 6.2 introduces an edge attribute mechanism into the graph neural network, combining time difference and similarity into an edge attribute vector: ， It is then fed into a graph neural network layer that supports edge attributes for training and inference.

8. The multi-sensor collaborative partial discharge waveform recognition method according to claim 1, characterized in that, Step 7, the graph neural network, uses a graph isomorphic network framework to model the graph structure. The graph isomorphic network framework uses GINEConv layers as graph convolutional layers. The update rules for the GINEConv layer are as follows: ， in, It is the first The node at the th Layer embedding representation; It is a node The set of neighbors; It is an edge attribute vector; It is a learnable edge feature mapping function; It is a shared multilayer perceptron; It is a trainable scaling factor.

9. The multi-sensor collaborative partial discharge waveform recognition method according to claim 1, characterized in that, Step 8 introduces a global average pooling operation into the output of the graph neural network to obtain a graph-level embedding vector, and inputs it into a classifier to determine whether a partial discharge event has occurred.

10. A multi-sensor collaborative partial discharge waveform recognition system, characterized in that, include: The signal conversion module is used to perform Hilbert transform on the acquired signals and extract the envelope of the oscillating waveform. The feature extraction module is used to apply the Logistic function to the rising edge region of the envelope of each sensor for nonlinear fitting, and extract the Logistic function steepness and start time as edge features. For the falling edge region of the envelope of each sensor, an exponential decay function is applied for nonlinear fitting, and the time constant is extracted as an energy dissipation feature. The data fusion module is used to use the extracted edge features and energy dissipation features as node features in the graph structure to build a graph neural network model, and to use the graph structure for training and inference to achieve collaborative fusion of information from multiple sensor sources. The partial discharge diagnosis module is used to introduce the output of the graph neural network model into a global average pooling operation to obtain a graph-level embedding vector, and then input it into a classifier to determine whether a partial discharge event has occurred.

Images