Multi-source partial discharge diagnosis system and method for power equipment

Abstract

The invention discloses a multi-source partial discharge diagnosis system and method for power equipment. The system comprises a signal acquisition unit and a signal processing unit, and the signal processing unit comprises a signal preprocessing module, an atlas generation module and an image generation module; in the method, a signal acquisition unit acquires an optical signal generated by partial discharge of multiple discharge sources of power equipment, converts the optical signal into a current signal and sends the current signal to a signal processing unit, and a signal preprocessing module in the signal processing unit obtains a time domain waveform signal and a PRPD statistical sequence based on the optical signal; the map generation module obtains a two-dimensional time-frequency map based on the time-frequency waveform signals and the PRPD statistical sequence, the image generation module generates an IFCNN network, and fusion feature maps of the optical signals generated by the discharge sources are output in a classified mode based on the time-frequency waveform signals and the two-dimensional time-frequency map through the IFCNN network. The partial discharge of different discharge sources of the power equipment can be effectively distinguished.

Description

Technical Field

[0001] The present invention relates to the field of partial discharge diagnosis of electric power equipment, and in particular to a system and method for diagnosing multi-source partial discharge of electric power equipment. Background Art

[0002] Epoxy fiberglass PCBs are widely used in communications equipment, power modules, and new energy power electronics due to their excellent dielectric properties and high-frequency stability. However, under high-voltage and high-frequency operating conditions, partial discharge (PD) is prone to occur within the PCB or in the interlayer insulation area due to concentrated electric fields at the conductor edges, microcracks in the insulation layer, or manufacturing defects. PD can carbonize the epoxy resin substrate and increase dielectric loss, further causing short circuits, signal distortion, and even equipment failure.

[0003] There are existing technologies for diagnosing PD using optical signals, such as the Chinese patent application "A Multi-spectral Weak-Light Detection Device and Method for Partial Discharge" with publication number CN111308289A. The device uses a photoelectric detection array to synchronously detect the multi-spectral signals generated by partial discharge in power equipment and convert them into multi-channel current signals. The multi-channel current signals are then synchronously processed. This device has the advantages of high sensitivity and strong anti-interference ability. However, it has the disadvantage of being unable to distinguish between multiple PD sources, making it difficult to apply to partial discharge diagnosis in power equipment. Summary of the Invention

[0004] The present invention provides a system and method for diagnosing multi-source partial discharge of electric equipment, so as to solve the problem that the existing method for detecting PD of electric equipment based on optical signals cannot distinguish between multiple PD sources.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A multi-source partial discharge diagnostic system for power equipment, comprising a signal acquisition unit and a signal processing unit;

[0007] The signal acquisition unit collects the optical signal generated by the partial discharge of the multi-discharge source of the power equipment, converts the optical signal into a current signal and then sends it to the signal processing unit;

[0008] The signal processing unit includes a signal preprocessing module, a spectrum generation module, and an image generation module; wherein:

[0009] The signal preprocessing module preprocesses the current signal corresponding to the optical signal to obtain a time domain waveform signal corresponding to the optical signal and a PRPD statistical sequence;

[0010] The spectrum generation module generates a two-dimensional time-frequency diagram based on the time domain waveform of the optical signal through wavelet transform, and constructs a three-dimensional feature matrix in combination with the PRPD statistical sequence to achieve comprehensive characterization of multi-source discharge;

[0011] The image generation module generates an IFCNN network, which includes a one-dimensional CNN, a two-dimensional CNN, a feature fusion layer, a fully connected layer, and an output layer. The one-dimensional CNN inputs the time domain waveform signal obtained by the signal preprocessing module and extracts the time domain local features of the optical signals generated by different discharge sources. The two-dimensional CNN inputs the two-dimensional time-frequency map obtained by the spectrum generation module and extracts the frequency domain features of the optical signals generated by different discharge sources. The feature fusion layer then fuses the time domain local features and the frequency domain features to obtain a fused feature map of the optical signals generated by different discharge sources. The fused feature map is then reduced in dimension by the fully connected layer. Finally, the output layer classifies and outputs the category label probabilities of different discharge sources, thereby distinguishing partial discharges from different discharge sources.

[0012] Furthermore, the signal acquisition unit is a SiPM sensor array.

[0013] Furthermore, the signal acquisition unit also includes an optical filter group and a focusing lens. The optical filter group includes multiple filters with different allowed wavelengths. The optical signal generated by the partial discharge of the multi-discharge source of the power equipment is filtered by each filter and then converged to the SiPM sensor array through the focusing lens.

[0014] Furthermore, in the signal processing unit, the signal preprocessing module includes a transimpedance amplifier, a noise suppression circuit, and an ADC. The transimpedance amplifier converts the current signal corresponding to the optical signal into a voltage signal and amplifies it, and the noise suppression circuit suppresses the noise in the voltage signal. Finally, the ADC collects the voltage signal to form the time domain waveform signal and obtains the PRPD statistical sequence.

[0015] Furthermore, in the signal processing unit, the spectrum generation module obtains a time-frequency energy distribution diagram based on the time-frequency waveform signal, and maps the PRPD statistical sequence into a pseudo-color grayscale image, and then maps the phase-amplitude-discharge number in the PRPD statistical sequence into a pseudo-color channel, and then superimposes it on the time-frequency energy distribution diagram to form a two-dimensional RGB image.

[0016] Furthermore, the two-dimensional CNN adopts a cross-layer feature fusion strategy to splice the shallow high-resolution features and the deep high-semantic features to obtain the time-frequency features.

[0017] A method for diagnosing multi-source partial discharge of power equipment based on the above-mentioned multi-source partial discharge diagnosis system for power equipment has the following process:

[0018] Collect the optical signal generated by partial discharge of multiple power sources in power equipment, convert the optical signal into a time domain waveform signal, and obtain the corresponding PRPD statistical sequence;

[0019] The spectrum generation module obtains the time-frequency energy distribution map based on the time-frequency waveform signal, maps the PRPD statistical sequence into a pseudo-color grayscale image, and then maps the phase-amplitude-discharge number in the PRPD statistical sequence into a pseudo-color channel, which is then superimposed on the time-frequency energy distribution map to form a two-dimensional RGB image;

[0020] An IFCNN network is generated, which includes a one-dimensional CNN, a two-dimensional CNN, a feature fusion layer, a fully connected layer, and an output layer; wherein the one-dimensional CNN inputs the time domain waveform signal obtained by the signal preprocessing module and extracts the time domain local features of the optical signals generated by different discharge sources, the two-dimensional CNN inputs the two-dimensional time-frequency map obtained by the spectrum generation module and extracts the frequency domain features of the optical signals generated by different discharge sources, then the feature fusion layer fuses the time domain local features and the frequency domain features to obtain a fused feature map of the optical signals generated by different discharge sources, then the fused feature map is reduced in dimension by the fully connected layer, and finally the output layer classifies and outputs the category label probabilities of different discharge sources, thereby distinguishing partial discharges from different discharge sources.

[0021] This invention uses a SiPM sensor array to capture multi-band optical signals generated by multiple power sources in power equipment. Combining time-frequency analysis with an improved convolutional neural network, it achieves high-precision, real-time detection and classification of multi-source PDs. This invention has broad application prospects in areas such as partial discharge monitoring, fault diagnosis, and safety monitoring of power equipment. Compared with existing technologies, this invention has the following advantages:

[0022] 1) The present invention uses a multi-source atlas fusion method to map the phase-amplitude discharge counts in the PRPD statistical sequence into pseudo-color RGB channels to generate a pseudo-color image with discharge statistical characteristics. At the same time, the time-frequency waveform is transformed into a time-frequency energy distribution map through wavelet transformation. The pseudo-color statistical information is superimposed on the time-frequency energy basis through pixel-level weighted fusion to form a fused image with both time-frequency characteristics and discharge statistical characteristics, significantly improving the representation and discrimination of multi-source discharge patterns.

[0023] 2) The present invention uses convolutional neural networks to establish a lightweight IFCNN network, designs a one-dimensional CNN to process time domain waveform signals, and a two-dimensional CNN to process PRPD. Through cross-layer feature fusion and channel compression, the number of parameters is reduced while improving recognition accuracy, and it can effectively identify the characteristics of multiple power sources.

[0024] 3) The present invention utilizes a multi-spectral SiPM array design to collect the optical signal generated by PD in different bands through multiple filters. Combined with the high gain characteristics of SiPM, it can improve the detection sensitivity of weak optical signals, and the multi-spectral band design helps to suppress background noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1It is a schematic diagram of the system structure of an embodiment of the present invention.

[0026] Figure 2 This is a diagram of the IFCNN architecture of an embodiment of the present invention.

[0027] Figure 3 This is a diagram of the conventional CNN architecture on which the IFCNN in the embodiment of the present invention is based. DETAILED DESCRIPTION

[0028] The present invention will be further described below with reference to the accompanying drawings and examples.

[0029] like Figure 1 As shown, this embodiment discloses a multi-source partial discharge diagnosis system for power equipment, including a signal acquisition unit and a signal processing unit.

[0030] In this embodiment, the signal acquisition unit includes an optical filter set, a focusing lens, and a SiPM sensor array. The optical filter set includes filters that allow transmission of UV, VIS, and NIR wavelengths. The optical signals generated by partial discharge from multiple discharge sources in power equipment are filtered by each filter before being focused by the focusing lens onto the SiPM sensor array. This enables the acquisition of the optical signals generated by the partial discharge from each discharge source in separate wavelength bands. The SiPM sensor array then converts the optical signals into current signals, which are then fed into the signal processing unit.

[0031] The SiPM sensor array is composed of a plurality of SiPM sensors (silicon photomultiplier tubes) distributed in an array. As a new type of solid-state photon counting device, the working principle of the silicon photomultiplier tube is based on the avalanche effect of the avalanche photodiode. Each silicon photomultiplier tube is composed of multiple micro units, each of which contains an avalanche photodiode and a quenching resistor, working in parallel. Whenever a photon is captured, a weak electronic signal is generated. SiPM sensors have the characteristics of high gain, low driving voltage and fast response, and can effectively capture the weak light signals accompanying PD. By combining the multi-spectral band design of the optical filter group, the SiPM sensor array in this embodiment can further distinguish the spectral signal characteristics of different PD types, providing rich information for multi-source discharge diagnosis. Its advantages are as follows:

[0032] 1. High gain: The gain of SiPM is usually around 10 5 and 10 6 It can amplify weak light signals;

[0033] 2. Low drive voltage: Compared to the hundreds of volts of traditional photomultiplier tubes, SiPMs typically only have a drive voltage of around 30V.

[0034] 3. Response efficiency: With nanosecond response speed, it can be used to respond to PD signals in real time;

[0035] 4. High spatial resolution: Accurately locate the discharge signal position, suitable for high-precision spatial resolution detection.

[0036] In this embodiment, the SiPM sensor array is arranged in the upper part of a support frame, and the epoxy glass fiber PCB board to be detected is placed in the lower part of the support frame.

[0037] In this embodiment, the signal processing unit includes a signal preprocessing module, a spectrum generation module, and an image generation module; wherein:

[0038] The signal preprocessing module of this embodiment includes a transimpedance amplifier, a noise suppression circuit, and an ADC. The transimpedance amplifier converts the current signal corresponding to the optical signal output by the SiPM sensor array into a voltage signal and amplifies it. The noise suppression circuit suppresses the noise in the voltage signal. Finally, the ADC collects the voltage signal to form the time domain waveform signal and obtain the PRPD statistical sequence.

[0039] In the spectrum generation module of this embodiment, a complex Morlet wavelet transform is performed on the time-frequency waveform signal to obtain a time-frequency energy distribution diagram. The calculation formula of the complex Morlet wavelet transform is as follows:

[0040]

[0041] The basis function of complex Morlet wavelet transform is shown as follows:

[0042]

[0043] Among them: a is the scale factor; b is the translation factor; is the result of scaling and shifting the wavelet function; W(x, y) is the transformation result, f(t) is the input signal, For my mother Xiaobo, is the conjugate of the mother wavelet, and t is the time variable.

[0044] Compared to the short-time Fourier transform (STFT), the wavelet transform overcomes the frequency-invariance issue. A suitable wavelet transform produces a well-focused time-frequency spectrum that is more realistic than other transforms, such as the Hilbert-Huang transform. The complex Morlet wavelet, due to its waveform's high similarity to the PD pulse, effectively captures the local time-frequency characteristics of the signal.

[0045] In this embodiment, the spectrum generation module generates a two-dimensional time-frequency diagram based on the optical signal's time-domain waveform through wavelet transform. A three-dimensional feature matrix is constructed in conjunction with the PRPD statistical sequence to comprehensively characterize multi-source discharges. The core of the two-dimensional time-frequency diagram is generated by performing time-frequency analysis (wavelet transform) on the time-domain signal, reflecting the time-frequency relationship. The PRPD statistical sequence records the phase, amplitude, and discharge of the PD pulse within the power frequency cycle.

[0046] Specifically, the spectrum generation module generates a time-frequency energy distribution map based on the time-frequency waveform signal, maps the PRPD statistical sequence into a pseudo-color grayscale image, and maps the phase-amplitude-discharge count in the PRPD statistical sequence into pseudo-color channels (RGB represents phase, amplitude, and discharge count, respectively). These are then superimposed on the time-frequency energy distribution map to form a two-dimensional RGB image. The PRPD phase is mapped to the red channel, the normalized amplitude is mapped to the green channel, and the discharge count is mapped to the blue channel. These are then weightedly fused with the time-frequency energy distribution map at the pixel level.

[0047] The spectrum generation module includes a time-frequency basis generation unit, a PRPD pseudo-color mapping unit, and a pixel-level fusion unit. The time-frequency basis generation unit performs wavelet transform on the time domain waveform to generate a time-frequency energy distribution diagram, including a PRPD pseudo-color mapping unit that encodes the phase / amplitude / discharge number into three RGB channels respectively; the pixel-level fusion unit performs matrix-weighted superposition of the pseudo-color channel and the time-frequency basis.

[0048] The image generation module of this embodiment generates an IFCNN network, such as Figure 2 As shown in the figure, the IFCNN network includes one-dimensional CNN, two-dimensional CNN, feature fusion layer, fully connected layer, and output layer.

[0049] One-dimensional CNN includes conventional CNN, such as Figure 3 As shown, the conventional CNN includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, and a fully connected layer. In order to improve the nonlinear expression capability, a 1×1 convolutional layer is added after the convolutional layer of the conventional CNN, thereby forming a one-dimensional CNN of this embodiment. Specifically, the one-dimensional CNN includes a first convolutional layer 1d-1 (conventional convolutional layer), a 1×1 second convolutional layer 1d-2, a first pooling layer 1-1, a third convolutional layer 1d-3 (conventional convolutional layer), a 1×1 fourth convolutional layer 1d-4, a second pooling layer 1-2, and a fully connected layer 1.

[0050] The two-dimensional CNN includes a conventional CNN and adopts a cross-layer feature fusion strategy to combine high-resolution features in shallow layers with high-semantic features in deep layers, thereby forming the two-dimensional CNN of this embodiment. Specifically, the two-dimensional CNN includes a first convolutional layer 2d-1, a first pooling layer 2-1, a second convolutional layer 2d-2, a second pooling layer 2-2, a third convolutional layer 2d-3, a third pooling layer 2-3, a fourth convolutional layer 2d-4, a fully connected layer 2, and a fully connected layer 3.

[0051] The one-dimensional convolutional branch (CNN-1D) inputs the wavelet-transformed PD waveform signal, which is the time-frequency waveform of the original optical pulse sequence. The processing flow involves extracting local time-frequency features through a convolutional layer (1d-1 / 2). The feature map is then downsampled through a pooling layer (1-1 / 2) to retain key waveform features. Subsequent convolutional layers (1d-3 / 4) perform cross-channel feature recombination to enhance the robustness of the time-frequency features. Finally, the fully connected layer 1 outputs a high-dimensional vector.

[0052] The two-dimensional convolutional branch (CNN-2D) inputs the two-dimensional time-frequency map of the fused PRPD statistical sequence (generating an RGB time-frequency map). The processing flow includes 2d-1 / 2 / 3 / 4 to form a multi-scale feature fusion, taking into account both high-frequency details and low-frequency global patterns. The pooling layer (2-1 / 2 / 3) uses a dynamic step size to preserve the distribution characteristics of the time-frequency map. Finally, the fully connected layer 2 outputs the feature vector. In the fusion layer, the outputs of the fully connected layers 1 and 2 are aligned across modal features using convolution to achieve channel dimension splicing and form a multimodal joint feature matrix. Then, the fully connected layer 3 performs nonlinear mapping to learn the feature combination patterns of different discharge sources. Finally, the softmax output outputs the discharge category probability distribution.

[0053] The time-domain local features obtained by the one-dimensional CNN and the frequency-domain features obtained by the two-dimensional CNN are fused by the feature fusion layer to obtain a fused feature map of the optical signals generated by different discharge sources. This fused feature map is then reduced in dimension by a fully connected layer and fed into the output layer. The output layer uses a Softmax classifier to output the class label probabilities of different discharge sources, thereby distinguishing partial discharges from different discharge sources.

[0054] The IFCNN network generated by the image generation module of this embodiment needs to be trained. The training dataset construction includes collecting time domain waveforms and PRPD data of different discharge sources (such as corona, surface, and internal discharges), generating time-frequency graphs, and annotating category labels. The cross entropy loss function is as follows:

[0055]

[0056] Where C is the number of discharge types, y is the true label, and p is the predicted probability.

[0057] The Adam optimizer is used for training, with an initial learning rate of 0.001 and a decay of 50% every 10 epochs.

[0058] Based on the above-mentioned multi-source partial discharge diagnosis system for power equipment, the method process for diagnosing multi-source partial discharge of power equipment in this embodiment is as follows:

[0059] Collect the optical signal generated by partial discharge of multiple power sources in power equipment, convert the optical signal into a time domain waveform signal, and obtain the corresponding PRPD statistical sequence;

[0060] A two-dimensional time-frequency diagram is generated using wavelet transforms based on the optical signal's time-domain waveform. Combined with the PRPD statistical sequence, a three-dimensional feature matrix is constructed to comprehensively characterize multi-source discharges. The core of the two-dimensional time-frequency diagram is generated through time-frequency analysis (wavelet transform) of the time-domain signal, reflecting the time-frequency relationship. The PRPD statistical sequence records the phase, amplitude, and discharge of the PD pulse within the power frequency cycle.

[0061] An IFCNN network is generated, which includes a one-dimensional CNN, a two-dimensional CNN, a feature fusion layer, a fully connected layer, and an output layer; wherein the one-dimensional CNN inputs the time domain waveform signal obtained by the signal preprocessing module and extracts the time domain local features of the optical signals generated by different discharge sources, the two-dimensional CNN inputs the two-dimensional time-frequency map obtained by the spectrum generation module and extracts the frequency domain features of the optical signals generated by different discharge sources, then the feature fusion layer fuses the time domain local features and the frequency domain features to obtain a fused feature map of the optical signals generated by different discharge sources, then the fused feature map is reduced in dimension by the fully connected layer, and finally the output layer classifies and outputs the category label probabilities of different discharge sources, thereby distinguishing partial discharges from different discharge sources.

[0062] The preferred embodiments of the present invention are described in detail above with reference to the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner unless there is any contradiction. Such combinations should also be regarded as the contents disclosed in this disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0063] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the scope of the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by those skilled in the art should fall within the scope of protection of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.

Claims

1. A multi-source partial discharge diagnostic system for power equipment, characterized in that: It includes a signal acquisition unit and a signal processing unit; The signal acquisition unit collects the optical signal generated by the partial discharge of the multi-discharge source of the power equipment, converts the optical signal into a current signal and then sends it to the signal processing unit; The signal processing unit includes a signal preprocessing module, a spectrum generation module, and an image generation module; wherein: The signal preprocessing module preprocesses the current signal corresponding to the optical signal to obtain a time domain waveform signal corresponding to the optical signal and a PRPD statistical sequence; The spectrum generation module generates a two-dimensional time-frequency diagram based on the time domain waveform of the optical signal through wavelet transform, and constructs a three-dimensional feature matrix in combination with the PRPD statistical sequence to achieve comprehensive characterization of multi-source discharge; The image generation module generates an IFCNN network, which includes a one-dimensional CNN, a two-dimensional CNN, a feature fusion layer, a fully connected layer, and an output layer. The one-dimensional CNN inputs the time domain waveform signal obtained by the signal preprocessing module and extracts the time domain local features of the optical signals generated by different discharge sources. The two-dimensional CNN inputs the two-dimensional time-frequency map obtained by the spectrum generation module and extracts the frequency domain features of the optical signals generated by different discharge sources. The feature fusion layer then fuses the time domain local features and the frequency domain features to obtain a fused feature map of the optical signals generated by different discharge sources. The fused feature map is then reduced in dimension by the fully connected layer. Finally, the output layer classifies and outputs the category label probabilities of different discharge sources, thereby distinguishing partial discharges from different discharge sources.

2. The multi-source partial discharge diagnostic system for power equipment according to claim 1, characterized in that: The signal acquisition unit is a SiPM sensor array.

3. The multi-source partial discharge diagnostic system for power equipment according to claim 2, characterized in that: The signal acquisition unit also includes an optical filter group and a focusing lens. The optical filter group includes multiple filters with different allowed wavelengths. The optical signal generated by the partial discharge of the multi-discharge source of the power equipment is filtered by each filter and then converged to the SiPM sensor array through the focusing lens.

4. The multi-source partial discharge diagnostic system for power equipment according to claim 1, characterized in that: In the signal processing unit, the signal preprocessing module includes a transimpedance amplifier, a noise suppression circuit, and an ADC. The transimpedance amplifier converts the current signal corresponding to the optical signal into a voltage signal and amplifies it, and the noise suppression circuit suppresses the noise in the voltage signal. Finally, the ADC collects the voltage signal to form the time domain waveform signal and obtain the PRPD statistical sequence.

5. The multi-source partial discharge diagnostic system for power equipment according to claim 1, characterized in that: In the signal processing unit, the spectrum generation module obtains a time-frequency energy distribution diagram based on the time-frequency waveform signal, maps the PRPD statistical sequence into a pseudo-color grayscale image, and then maps the phase-amplitude-number of discharges in the PRPD statistical sequence into a pseudo-color channel, which is then superimposed on the time-frequency energy distribution diagram to form a two-dimensional RGB image.

6. The multi-source partial discharge diagnostic system for power equipment according to claim 1, characterized in that: The two-dimensional CNN adopts a cross-layer feature fusion strategy to splice the shallow high-resolution features and the deep high-semantic features to obtain the time-frequency features.

7. A method for diagnosing multi-source partial discharge of power equipment based on the multi-source partial discharge diagnostic system for power equipment according to any one of claims 1 to 6, characterized in that: The process is as follows: Collect the optical signal generated by partial discharge of multiple power sources in power equipment, convert the optical signal into a time domain waveform signal, and obtain the corresponding PRPD statistical sequence; The spectrum generation module obtains the time-frequency energy distribution map based on the time-frequency waveform signal, maps the PRPD statistical sequence into a pseudo-color grayscale image, and then maps the phase-amplitude-discharge number in the PRPD statistical sequence into a pseudo-color channel, which is then superimposed on the time-frequency energy distribution map to form a two-dimensional RGB image; Generate an IFCNN network, wherein the IFCNN network includes a one-dimensional CNN, a two-dimensional CNN, a feature fusion layer, a fully connected layer, and an output layer; The one-dimensional CNN inputs the time domain waveform signal obtained by the signal preprocessing module and extracts the time domain local features of the optical signals generated by different discharge sources. The two-dimensional CNN inputs the two-dimensional time-frequency map obtained by the spectrum generation module and extracts the frequency domain features of the optical signals generated by different discharge sources. The feature fusion layer then fuses the time domain local features and the frequency domain features to obtain a fused feature map of the optical signals generated by different discharge sources. The fused feature map is then reduced in dimension by a fully connected layer. Finally, the output layer classifies and outputs the category label probabilities of different discharge sources, thereby distinguishing partial discharges from different discharge sources.

Images