PRPD image generation circuit and method and partial discharge type identification method

Abstract

The invention discloses a PRPD image generation circuit and method and a partial discharge type identification method, and belongs to the technical field of electrical equipment discharge detection. The generating circuit of the PRPD image comprises a power frequency generating unit used for generating a power frequency synchronizing signal, a counter used for starting to update a phase number at a power frequency falling edge of the power frequency synchronizing signal, a peak holding unit used for collecting a partial discharge signal, a first digital-to-analog converter used for updating a reference voltage, and a second digital-to-analog converter used for updating a reference voltage. The comparator is used for comparing partial discharge voltage of the partial discharge signal with reference voltage; and the data generation unit is used for generating and outputting an image matrix. The PRPD image generation circuit does not need to store, process and transmit a large batch of partial discharge data, and generates a PRPD image with amplitude and phase information by using a low-complexity and low-data-volume method. According to the partial discharge type identification method, the partial discharge amplitude and phase are analyzed based on the PRPD image, so that the partial discharge type is identified.

Description

Technical Field

[0001] The present invention relates to the technical field of discharge detection of electric power equipment, and in particular to a circuit and method for generating a PRPD image and a method for identifying a partial discharge type. Background Art

[0002] Partial discharge (PD) can be classified into different types, including tip discharge, particle discharge, air gap discharge, and suspended discharge. PD classification is a key component of electrical insulation fault diagnosis in high-voltage equipment. With the development of artificial intelligence (AI), deep learning has become a key method for PD type identification. Compared with traditional methods such as waveform recognition, PD pattern recognition methods based on deep learning offer advantages such as low complexity, high recognition accuracy, and a high degree of intelligence, as described in Chinese Patent Publication No. CN111814834B. In actual field PD classification and identification using deep learning models such as convolutional neural networks, PRPD (Phase Resolved Partial Discharge) images are first generated from collected PD data. Deep learning models are then used for on-site online pattern recognition. PRPD images provide key information for pattern recognition, including the phase and amplitude characteristics of the PD pulses. By describing the relationship between the amplitude, phase, and number of discharges in the discharge signal, PRPD images not only reveal how the discharge signal amplitude varies with phase, but also demonstrate the distribution of the number of discharges within different phase intervals.

[0003] Traditional ultra-high frequency (UHF) PD receivers digitize the signal using an analog-to-digital converter (ADC) after filtering, amplification, and detection. Correlation signal processing is then performed to extract amplitude information and simultaneously record the phase information of the PD pulse. Unlike conventional carrier-modulated wireless communication signals, PD signals are short, intermittent, pulsed, broadband baseband signals, with no useful information for the majority of the remaining time. With the development of artificial intelligence-based Internet of Things (AIOT) technology, implementing edge computing in sensor terminals is becoming a technological trend. Sensor terminals are data processing platforms with limited hardware resources, typically based on single-chip microcomputers. Traditional methods using ADCs to sample partial discharge signals in real time generate large amounts of data, much of which lacks useful information. Furthermore, high-speed ADCs consume high power, posing significant challenges for IoT sensor terminals with limited hardware resources and typically battery power. Therefore, a low-data-volume, low-complexity method that does not require an ADC is needed to generate PRPD images and extract PD pulse amplitude and phase characteristics, enabling cost-effective PD type identification on IoT sensor terminals. Summary of the Invention

[0004] To address the difficulties in identifying partial discharge (PD) types at the sensor end of the Internet of Things (IoT), the present invention provides a circuit and method for generating PRPD images, as well as a method for identifying PD types. The PRPD image generation circuit of the present invention eliminates the need to store, process, and transmit large amounts of PD data, instead using a low-complexity, low-data-volume method to generate PRPD images containing amplitude and phase information. Furthermore, the PD type identification method of the present invention analyzes the PD amplitude and phase based on the PRPD image to identify the PD type.

[0005] The invention objectives of this application can be achieved through the following technical means: A circuit for generating a PRPD image, comprising: A power frequency generating unit, used for generating a power frequency synchronization signal; A counter, used to update the phase number starting at the falling edge of the power frequency synchronization signal; A peak holding unit is used to collect partial discharge signals, and an input end of the peak holding unit is connected to the main circuit; a first digital-to-analog converter, configured to update a reference voltage; a comparator, for comparing a partial discharge voltage U1 of a partial discharge signal with a reference voltage U2, wherein a first input of the comparator is connected to the first digital-to-analog converter, a second input of the comparator is connected to the peak holding unit, and an output of the comparator is connected to the first digital-to-analog converter; A data generation unit is used to generate and output an image matrix, and the data generation unit is connected to the first digital-to-analog converter and the counter, wherein: If U1 is greater than λU2, the comparator outputs a rising edge of partial discharge to the first digital-to-analog converter, otherwise the comparator outputs a falling edge of partial discharge to the first digital-to-analog converter and the counter. λ is the accuracy parameter. When the first DAC receives the rising edge of the partial discharge, the reference voltage U2 increases. When the first DAC and the counter receive the falling edge of the partial discharge, the first DAC outputs the current reference voltage U2 and the counter outputs the current phase number. The data generation unit generates multiple arrays containing reference voltages and phase numbers, normalizes the reference voltages and phase numbers of the arrays according to the number of rows and columns of the image matrix, and then generates elements of the image matrix according to the number of rows and columns corresponding to each array.

[0006] In the present invention, the output end of the comparator is also connected to the peak holding unit. When the comparator generates a rising edge of partial discharge, it outputs a high level to the peak holding unit. When the comparator generates a falling edge of partial discharge, it outputs a low level to the peak holding unit. When the comparator signal received by the peak hold unit remains at a low level, the peak hold unit enters the envelope tracking state. When the comparator signal received by the peak holding unit remains at a high level or jumps from a low level to a high level, the peak holding unit enters the peak detection state. When the comparator signal received by the peak holding unit jumps from a high level to a low level, the peak holding unit releases the current partial discharge signal.

[0007] A PRPD image generation method according to the PRPD image generation circuit comprises the following steps: Step 1: The data generation unit creates an image matrix, the power frequency generation unit generates a power frequency synchronization signal, and after the counter receives the power frequency falling edge of the power frequency synchronization signal, the phase number is reset and counting begins; Step 2: The peak hold unit enters the envelope tracking state, the first digital-to-analog converter initializes the reference voltage U2, and the peak hold unit collects the partial discharge signal; Step 3: The comparator compares the partial discharge voltage U1 of the partial discharge signal with the reference voltage U2. If U1 is greater than λU2, the process proceeds to step 4. Otherwise, the process returns to step 2. λ is the accuracy parameter. Step 4: The peak hold unit enters the peak detection state, the comparator outputs a partial discharge rising edge to the first digital-to-analog converter, and the reference voltage U2 increases; Step 5: The peak hold unit re-collects the partial discharge signal, and the comparator re-compares the partial discharge voltage U1 with the reference voltage U2. If U1 is greater than λU2, the process returns to step 4; otherwise, the process proceeds to step 6. Step 6: The comparator outputs a falling edge of the partial discharge to the first digital-to-analog converter and the counter. The first digital-to-analog converter outputs the current reference voltage U2, the counter outputs the current phase number, and the reference voltage U2 is reset. Step 7: Repeat steps 2 to 6 to generate multiple arrays containing reference voltages and phase numbers. Normalize the reference voltages and phase numbers of the arrays according to the number of rows and columns of the image matrix. Then, generate elements of the image matrix based on the normalized values ​​of the reference voltages and phase numbers, and output the image matrix.

[0008] In the present invention, in step 4, the increment step length U of the reference voltage U2 is step =0.001V to 0.005V, the image matrix is ​​a 32×32 two-dimensional matrix, the power frequency synchronization signal is a square wave signal with a period of 20ms, and the voltage of the power frequency synchronization signal is 3.3V.

[0009] In the present invention, in step 4, the peak holding unit receives the reference voltage U of the second digital-to-analog converter. ref , partial discharge voltage U1=U0-U ref , U0 is the current voltage of the partial discharge signal.

[0010] In the present invention, in step 7, the image matrix is ​​an N-row M-column matrix, and according to the maximum reference voltage U max Set the normalization ratio M / U max , M is the number of columns of the image matrix, and the normalized value of the reference voltage U2 is Floor(U2M / U max )+1, Floor() is the rounding down function.

[0011] In the present invention, the number of arrays K whose normalized value of the statistical phase number is n and the normalized value of the reference voltage is m is nm , element P in the image matrix nm =(1-0.02K nm / T)×255, T is the detection time, n≤N, m≤M.

[0012] A method for identifying a partial discharge type according to the PRPD image generation method comprises the following steps: Step 100: Generate at least two groups of partial discharge signals of partial discharge types, implement the PRPD image generation method, and generate multiple groups of image matrices; Step 200: training a partial discharge classification model based on multiple sets of image matrices and corresponding partial discharge types; Step 300: Generate an image matrix according to the PRPD image generation method, and identify the partial discharge type of the image matrix according to the partial discharge classification model.

[0013] In the present invention, in step 100, a partial discharge signal is generated based on the gas-insulated switchgear, and the types of partial discharge include tip discharge, particle discharge, air gap discharge and suspended discharge.

[0014] In the present invention, the partial discharge classification model includes multiple groups of alternately connected hourglass-shaped attention layers and transition layers. The hourglass-shaped attention layer is composed of a plurality of hourglass-shaped attention units stacked together. The hourglass-shaped attention unit has a lightweight convolution block and an attention block for adjusting the feature weights of the image matrix.

[0015] The beneficial effects of implementing the PRPD image generation circuit, method and partial discharge type identification method of the present invention are: 1. The present invention greatly simplifies the overall circuit structure, can make full use of the built-in hardware circuit resources of the single-chip microcomputer, and adopts a single-chip partial discharge peak holding circuit solution. The system is also more stable and reliable and easy to implement.

[0016] 2. The present invention realizes a low-data-volume, low-complexity PRPD image generation method, which can realize the peak and phase measurement of the partial discharge pulse signal without the need for an ADC converter, thereby reducing system requirements and costs.

[0017] 3. The present invention combines the PRPD image generation circuit and the deep learning model to realize partial discharge pattern recognition on a single-chip microcomputer, which is suitable for application in resource-constrained power Internet of Things sensor terminals. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic diagram of an ultra-high frequency partial discharge signal; Figure 2 A block diagram of a circuit for generating a PRPD image according to the present invention; Figure 3 Schematic diagram of a circuit for generating a PRPD image according to the present invention; Figure 4 A circuit diagram of a peak holding unit of the present invention; Figure 5 Schematic diagram of the voltage of the subtraction amplifier circuit of the present invention; Figure 6 This is a communication principle diagram of the power frequency generation unit of the present invention; Figure 7 It is the working flow chart of the counter of the present invention; Figure 8 Schematic diagram of the PRPD image of the present invention; Figure 9 Schematic diagram of the PRPD image generated on the PC side by the tip discharge signal collected by the existing technology; Figure 10 Schematic diagram of the PRPD image generated on the PC side using the particle discharge signal collected by the existing technology; Figure 11 Schematic diagram of the PRPD image generated on the PC side by the air gap discharge signal collected by the existing technology; Figure 12 Schematic diagram of the PRPD image generated on the PC side by the suspended discharge signal collected by the existing technology; Figure 13 This is a schematic diagram of a PRPD image generated at the sensing end by the tip discharge signal collected by the present invention; Figure 14 This is a schematic diagram of a PRPD image generated at the sensing end by the particle discharge signal collected by the present invention; Figure 15 This is a schematic diagram of a PRPD image generated at the sensing end by the air gap discharge signal collected by the present invention; Figure 16 This is a schematic diagram of a PRPD image generated at the sensing end by the suspended discharge signal collected by the present invention; Figure 17 is a flow chart of a PRPD image generating method according to the PRPD image generating circuit of the present invention; Figure 18is a flow chart of a method for identifying a partial discharge type according to the present invention; Figure 19 Schematic diagram of the partial discharge classification model of the present invention. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] In the prior art, the identification of partial discharge models is usually completed on the PC side, and the data processing capabilities of the PC side are used to analyze the original partial discharge data. In the process of identifying ultra-high frequency partial discharges, the original partial discharge data contains a large amount of invalid information. With the development of power Internet of Things technology, it is necessary to realize the identification of ultra-high frequency partial discharge patterns through a single-chip microcomputer at the sensor side. In order to embed the recognition algorithm into the single-chip microcomputer, it is necessary to realize the lightweight of data acquisition and model. The PRPD image generation circuit, method and partial discharge type identification method of the present invention can make full use of the built-in hardware circuit resources of the single-chip microcomputer to realize the generation of PRPD images with low data volume and low complexity, and are suitable for application in resource-constrained power Internet of Things sensor terminals. Example 1

[0021] like Figures 1 to 16 The circuit for generating a PRPD image according to the present invention comprises a power frequency generation unit, a counter, a peak hold unit, a comparator, a first digital-to-analog converter (DAC), a data generation unit, and a second DAC. The present invention utilizes the ultra-low-power STM32L476RG microcontroller. This MCU operates at 80 MHz, and its built-in comparator COMP, DAC1, DAC2, and timer serve as the comparator, first DAC, second DAC, and counter, respectively. The STM32L476RG's built-in COMP can respond within a 55-80 ns timeframe, and the DAC1 and DAC2 DACs have a settling time of 1.4 μs, enabling rapid detection and tracking of changes in partial discharge signals. DAC1 and DAC2 can be configured with different bit counts; the present invention uses 8 bits.

[0022] The power frequency generation unit is used to generate a power frequency synchronization signal. Specifically, the power frequency generation unit converts 220V, 50Hz power frequency alternating current into a 3.3V, 50Hz power frequency synchronization signal (square wave signal) to achieve precise phase synchronization. The power frequency generation unit uses zero-crossing detection technology to extract the phase information of the power frequency voltage, and realizes the conversion of 220V power frequency alternating current to 3.3V square wave signal through an optocoupler isolation circuit. The power frequency generation unit can make the local discharge signal collected by the present invention consistent with the power frequency voltage phase of the power equipment, and provide a zero-starting reference point. In actual substations and other places, the 220V power frequency power socket is usually some distance away from the local discharge detection point. Therefore, a wireless transceiver circuit (wireless transmitting circuit, wireless receiving circuit) is required to transmit the power frequency synchronization signal generated by the power frequency generation unit to the detection point, such as Figure 6 The wireless transmitting circuit and the wireless receiving circuit use MAX1472 and MAX1470, which have low power consumption characteristics and simple and easy-to-implement modulation and demodulation methods, and are suitable for use in the embedded device of the present invention.

[0023] The counter is used to update the phase count starting at the falling edge of the power frequency synchronization signal. The power frequency synchronization signal provides a reference point for the counter, and the counter begins counting at the falling edge of the power frequency synchronization signal. The counting period of the counter in the present invention is 20 ms, which is the range of the horizontal coordinate (phase direction) of the PRPD image. Each falling edge of the power frequency synchronization signal indicates the phase position of 0°, and the count value (CNT) corresponds to the specific horizontal coordinate value of the partial discharge signal in the image. The present invention can set the prescaler and counting period of the timer (counter) according to the main frequency of the single-chip microcomputer, so that the counter counts within a range of 1 to N. For example, for a counting range of 20, the counting interval is 1 ms. The prescaler divides the clock frequency into a lower counting frequency to meet the power frequency period requirement. By properly setting the prescaler and period, the counter count frequency can be an integer multiple of the power frequency, thereby achieving accurate phase acquisition.

[0024] The peak hold unit is used to collect partial discharge signals, and the input end of the peak hold unit is connected to the main circuit. The main circuit refers to the current circuit of the power equipment that needs to measure the type of partial discharge, such as a high-voltage power system, a mains power supply system, etc. Specifically, the peak hold unit is implemented by the HMC1120 chip, where pin 8 Mode_Sel is used to select the operating mode. When Mode_Sel is connected to a high level, it enters the peak detection state, which can capture and hold the peak value of the signal; when it is connected to a low level, the saved peak value is released and the envelope tracking state is entered. When pin 6 PH_Cap is directly grounded without an external capacitor, the peak value can be maintained for about 150us, which gives the microcontroller enough time to complete the partial discharge signal peak measurement and acquisition. When an external capacitor is connected, the peak hold time can be longer. The output range of pin 7 ETOUT is 0.95V-1.65V, 0.95V is a fixed DC component, and the dynamic range is 0-0.7V.

[0025] To ensure the accuracy of subsequent signal processing, the peak hold unit has a subtraction amplifier circuit to remove the fixed DC component and amplify the output peak signal. The subtraction amplifier circuit is implemented by the COS8092 chip, which has two built-in operational amplifiers. One operational amplifier is designed as a subtraction amplifier circuit, and the other operational amplifier is used as an emitter follower to isolate the peak hold circuit from the microcontroller. Figure 5 The input voltage of the subtraction amplifier circuit serves as the reference voltage of the second digital-to-analog converter (DAC2). The operational amplifier's gain can be adjusted based on actual needs. In this embodiment, the amplified output voltage is approximately 1.5V. The component parameters of the peak hold circuit are as follows: C1 = C2 = C4 = 1nF, C3 = 10nF, C5 = 0.1nF, C6 = C7 = 100nF, R1 = 50Ω, R2 = 604Ω, R3 = R6 = 5kΩ, and R4 = R5 = 15kΩ.

[0026] The first DAC is used to update the reference voltage. When the first DAC receives a rising edge of a partial discharge, the reference voltage U2 increases. When the first DAC and the counter receive a falling edge of a partial discharge, the first DAC outputs the current reference voltage U2, and the counter outputs the current phase number.

[0027] The comparator is used to compare the partial discharge voltage U1 of the partial discharge signal with a reference voltage U2. The first input of the comparator is connected to the first digital-to-analog converter, the second input is connected to the peak hold unit, and the output is connected to the first digital-to-analog converter. If U1 is greater than λU2, the comparator outputs a partial discharge rising edge to the first digital-to-analog converter. Otherwise, the comparator outputs a partial discharge falling edge to the first digital-to-analog converter and the counter.

[0028] The comparator's output is also connected to a peak hold unit. The comparator outputs a high level to the peak hold unit when a rising edge of a partial discharge is generated, and a low level to the peak hold unit when a falling edge of a partial discharge is generated. When the comparator signal received by the peak hold unit remains low, the peak hold unit enters an envelope tracking state. When the comparator signal received by the peak hold unit remains high or transitions from a low level to a high level, the peak hold unit enters a peak detection state. When the comparator signal received by the peak hold unit transitions from a high level to a low level, the peak hold unit releases the current partial discharge signal.

[0029] Specifically, the comparator sets the output initial threshold according to actual needs to filter out some noise. In the absence of a partial discharge signal, the partial discharge voltage U1≤λU2, and the output of the comparator remains at a low level. At this time, the threshold setting serves to isolate the noise signal. λ is an accuracy parameter, usually ranging from 1 to 1.001. When a partial discharge signal is collected, the voltage of the partial discharge signal will exceed the threshold THRH (in this embodiment, THRH=0.1V). At this time, U1>λU2 (the partial discharge voltage is slightly greater than the reference voltage), and the output level of the first digital-to-analog converter jumps from a low level to a high level, generating a partial discharge rising edge. At this time, the first digital-to-analog converter updates the reference voltage U2, and U2 increases by a fixed step size U each time. step . During the output adjustment process of the first digital-to-analog converter, the comparator maintains a high-level output, and the high-level Mode_Sel puts the circuit in peak hold working mode. When the partial discharge voltage U1≤λU2 (the reference voltage U2 is slightly larger than the partial discharge voltage U1), the output of the comparator will switch from a high level to a low level, generating a partial discharge falling edge. The first digital-to-analog converter outputs a reference voltage U2, which is close to the partial discharge voltage U1, that is, the amplitude of the partial discharge signal. At the same time, the first digital-to-analog converter is reset to the initial threshold, and the low-level Mode_Sel causes the peak hold unit to switch to the envelope tracking state, ready to start collecting the next partial discharge signal.

[0030] The present invention monitors the signal changes in real time through a comparator, cooperates with the adjustment of the first digital-to-analog converter, ensures that the amplitude of the partial discharge signal can be captured, and gradually adjusts the reference voltage U2 to make it approach the amplitude, completing the collection and processing of the partial discharge amplitude. The accuracy of the measured amplitude depends on the set step size U step , U step The larger the step size, the worse the accuracy, and the shorter the time it takes for the reference voltage U2 to reach a value slightly greater than the partial discharge voltage U1, and vice versa. In addition, the smaller the step size, the smaller the accuracy parameter λ, thereby simultaneously improving the detection accuracy.

[0031] The data generation unit is configured to generate and output an image matrix. The data generation unit is connected to the first digital-to-analog converter and the counter. The data generation unit generates multiple arrays containing reference voltages and phase numbers, normalizes the reference voltages and phase numbers of the arrays according to the number of rows and columns of the image matrix, and then generates elements of the image matrix according to the number of rows and columns corresponding to each array.

[0032] Reference Figure 8 PRPD images are typically composed of arrays superimposed from multiple power frequency cycles. The pixel value of a PRPD image is determined by the number of PD signals occurring at the corresponding phase and amplitude. For example, the maximum number of occurrences within a single second is 50. In power equipment, different types of PD have distinct distribution characteristics depending on the power frequency phase. These distribution characteristics determine the different PRPD images produced by these different types of PD. Example 2

[0033] like Figures 8 to 17 As shown, a PRPD image generation method according to the PRPD image generation circuit of the present invention includes the following steps.

[0034] Step 1: The data generation unit creates an image matrix, and the power frequency generation unit generates a power frequency synchronization signal. Upon receiving the falling edge of the power frequency synchronization signal, the counter resets the phase count and begins counting. The number of rows and columns in the image matrix is ​​preset based on the microcontroller's capacity. Assuming an M×N PRPD image is required, the image matrix is ​​an M×N matrix. The phase corresponds to the horizontal axis of the PRPD image, so the counter (TIM) represents the different phase intervals within a power frequency cycle (20ms) by counting from 1 to N. Larger M and N values ​​increase system accuracy, while smaller M and N values ​​reduce system memory requirements.

[0035] Step 2: The peak hold unit enters the envelope tracking state, the first digital-to-analog converter initializes the reference voltage U2, and the peak hold unit collects the partial discharge signal. The initialized reference voltage U2 is the threshold THRH in the first embodiment.

[0036] Step 3: The comparator compares the partial discharge voltage U1 of the partial discharge signal with the reference voltage U2. If U1 is greater than λU2, the process proceeds to step 4; otherwise, the process returns to step 2, where λ is the accuracy parameter. This embodiment detects the comparator's Q level to determine whether to enter the peak detection state. The comparator's interrupt function is set to edge trigger mode. When the Q level transitions, the comparator's interrupt function is triggered and executed, and different processing flows are entered based on the transition type (partial discharge rising edge or partial discharge falling edge). If the Q level remains low, indicating that the partial discharge voltage has not exceeded the reference voltage, the process continues to monitor the Q level and returns to step 2 until a valid transition signal is detected.

[0037] Step 4: The peak hold unit enters the peak detection state. The comparator outputs a rising edge of partial discharge to the first digital-to-analog converter, and the reference voltage U2 increases. If the Q level jumps from a low level to a high level (partial discharge rising edge), it means that the amplitude of the partial discharge signal exceeds the analog value output by the first digital-to-analog converter, and the program will enter the peak search process (peak detection state). In this stage, the present invention calculates the peak value according to the preset step size U. step The reference voltage is gradually increased and the updated reference voltage is output via the first digital-to-analog converter. By gradually increasing the analog value output by the first digital-to-analog converter, the present invention can approximate the peak value of the partial discharge signal. This process is iterative. After each increase in the reference voltage, the Q level is re-checked to determine whether the peak value of the partial discharge signal has been reached. This is the partial discharge signal peak search and approximation stage.

[0038] The increment step size U of the reference voltage U2 step =0.001V to 0.005V, the image matrix can be a 32×32 two-dimensional matrix. In a preferred embodiment, the peak holding unit receives the reference voltage U of the second digital-to-analog converter. ref , partial discharge voltage U1=U0-U ref , U0 is the current voltage of the partial discharge signal. Reference voltage U ref =0.95V. The reference voltage acts as a noise reducer and is used in conjunction with the subtraction amplifier circuit. To ensure data comparability, the reference voltage can also undergo the same subtraction process after the partial discharge voltage has been subtracted.

[0039] Step 5: The peak hold unit re-collects the PD signal, and the comparator re-compares the PD voltage U1 with the reference voltage U2. If U1 is greater than λU2, the process returns to step 4; otherwise, it proceeds to step 6. Specifically, if the Q level changes from high to low (partial discharge falling edge), it indicates that the analog value output by the first DAC has gradually increased to a value higher than the amplitude of the PD signal, and the peak value is saved, entering step 6.

[0040] Step 6: The comparator outputs a falling edge of the partial discharge to the first DAC and counter. The first DAC outputs the current reference voltage U2, the counter outputs the current phase number, and the reference voltage U2 is reset. The present invention saves the current reference voltage in an array to record the peak value of the partial discharge signal. After saving the peak value data, the present invention resets the reference voltage to the threshold THRH and releases the partial discharge peak value through the Mode_Sel terminal to enter the envelope tracking state and resume detection of the next partial discharge signal.

[0041] Step 7: Repeat steps 2 to 6 to generate multiple arrays containing reference voltages and phase numbers, normalize the reference voltage (voltage amplitude of partial discharge) and phase number of the array according to the number of rows and columns of the image matrix, and then generate the elements of the image matrix according to the normalized values ​​of the reference voltage and phase number, and output the image matrix. The number of repetitions depends on the number of arrays required. For example, if the number of power frequency cycles within 1s is 50, 50 arrays can be collected accordingly. The arrays collected by the present invention are subjected to amplitude normalization and pixel normalization. Amplitude normalization is to unify partial discharge signals of different amplitudes into the same range to avoid affecting the image quality due to excessive amplitude differences. Pixel normalization is to convert the processed data into an image format suitable for pattern recognition.

[0042] Reference Figure 8 , the image matrix is ​​an N-row M-column matrix, according to the maximum phase number W max Set the normalization ratio N / W max , N is the number of rows in the image matrix, and the normalized value of the phase number W is n=Floor(WN / W max )+1, according to the maximum reference voltage U max Set the normalization ratio M / U max , M is the number of columns of the image matrix, the normalized value of the reference voltage U2 is m=Floor(U2M / U max )+1, Floor() is a floor rounding function. The number of arrays K whose normalized value of the statistical phase number is n and the normalized value of the reference voltage is m nm , element P in the image matrix nm =(1-0.02K nm / T)×255, where T is the detection time. n≤N, m≤M.

[0043] The image matrix corresponds to the PRPD image, and the element values ​​of the image matrix are the pixel values ​​of the corresponding pixels. The smaller the pixel value, the more frequent partial discharges occurred. The PRPD image is a superposition of the amplitude and phase of an array acquired over multiple power frequency cycles. If the PRPD image size is M×N=10×10, the amplitude and phase are divided into 10 equal parts, with each phase interval being 36°. The minimum pixel value of the PRPD image is 0, indicating that all partial discharge signals are present at the corresponding phase and amplitude. The maximum value is 255, indicating that no partial discharge signals are present at the phase and amplitude. Example 3

[0044] like Figures 8 to 19 The present invention provides a method for identifying a local discharge type according to the PRPD image generation method, comprising the following steps.

[0045] Step 100: Generate at least two sets of partial discharge signals of different partial discharge types, and implement the PRPD image generation method to generate multiple sets of image matrices. Typically, the laboratory's GIS (Gas Insulated Switchgear) partial discharge test platform has four built-in partial discharge models, which generate corresponding partial discharge signals based on gas-insulated switchgear. Partial discharge types include tip discharge, particle discharge, air gap discharge, and suspension discharge.

[0046] Step 200: Train a partial discharge classification model based on multiple sets of image matrices and corresponding partial discharge types, and deploy the model to a single-chip microcontroller. The partial discharge classification model of the present invention is a lightweight deep learning model suitable for classifying PRPD images. The model comprises multiple sets of alternating hourglass attention layers (Sandglass-SE layers) and transition layers. The hourglass attention layers are composed of a stack of multiple hourglass attention units (Sandglass-SE blocks), each of which has a lightweight convolution block and an attention block for adjusting the feature weights of the image matrix. In this patent application, the attention block can be composed of a convolution operation.

[0047] Reference Figure 19 , the main part of the local discharge classification model of this embodiment is composed of Sandglass-SE layers and transition layers alternately connected, and convolution layers and maximum pooling layers are added between the input and the main body, and global pooling layers and fully connected layers are connected after the main body. The Sandglass-SE layer is composed of multiple Sandglass-SE blocks stacked together, and lightweight convolution blocks and optimized depth-separable convolutions are used to reduce the number of model parameters. The SE attention mechanism can enhance the focus on important features and improve the ability of feature expression by adaptively adjusting the weights of each feature channel during the feature map recovery process, so as to more accurately capture important feature information. The lightweight model is deployed on the STM32L476RG microcontroller of Example 1, occupying only 73.36KB of Flash and 41.09KB of RAM.

[0048] Step 300: Generate an image matrix using the PRPD image generation method, and identify the PD type of the image matrix using the PD classification model. The PD type identification result can be sent to a host computer or to other sensor terminals via a wireless communication module.

[0049] Figures 9 to 12 PRPD images of different partial discharge types are generated on the PC side using traditional high-speed ADC and digital processor technology, corresponding to a 32×32 image matrix. Figures 13 to 16The PRPD images of different partial discharge types are generated by the single chip microcomputer partial discharge peak value and phase acquisition method of the present invention, corresponding to a 32×32 image matrix. Figures 9 to 12 and Figures 13 to 16 The PRPD image distribution characteristics of the two images are similar. Among them, the tip discharge is mainly concentrated in the first quadrant of the image, showing a clear phase distribution feature; the particle discharge shows a more dispersed feature, its discharge phenomenon has no obvious phase distribution feature, and the discharge frequency is relatively large; the air gap discharge mainly appears in the first and third quadrants of the image, and its discharge characteristics overlap with the tip discharge; the suspension discharge is also distributed in the first and third quadrants, but it is mainly concentrated in the upper middle area of ​​the image.

[0050] By deploying the trained lightweight partial discharge classification model on a single-chip microcomputer, pattern recognition can be performed on the generated PRPD images, determining the type of partial discharge, and enabling on-site monitoring of insulation faults in power equipment. The following table compares the results of partial discharge type identification on a PC (existing technology) and on a single-chip microcomputer (the present invention). Classification and identification on the PC uses traditional high-speed ADC and digital processor technology to generate PRPD images and train classification on the PC. Classification and identification on a single-chip microcomputer uses the PRPD image generation method of the present invention to identify partial discharge types. The table below shows that the classification accuracy of the present invention deployed on a single-chip microcomputer is close to that of a PC, validating the effectiveness of the present invention's low-complexity method for acquiring partial discharge feature data and generating PRPD images. It also demonstrates the portability of the lightweight partial discharge classification model on single-chip embedded platforms.

[0051] PRPD image generation method Accuracy Recall F1 score Accuracy Existing technology / PC side 96.97% 96.88% 96.90% 96.88% The present invention / single chip microcomputer 93.22% 93.00% 93.05% 93.00% The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A circuit for generating a PRPD image, characterized in that: include: A power frequency generating unit, used for generating a power frequency synchronization signal; A counter, used to update the phase number starting at the falling edge of the power frequency synchronization signal; A peak holding unit is used to collect partial discharge signals, and an input end of the peak holding unit is connected to the main circuit; a first digital-to-analog converter, configured to update a reference voltage; a comparator, for comparing a partial discharge voltage U1 of a partial discharge signal with a reference voltage U2, wherein a first input of the comparator is connected to the first digital-to-analog converter, a second input of the comparator is connected to the peak holding unit, and an output of the comparator is connected to the first digital-to-analog converter; A data generation unit is used to generate and output an image matrix, and the data generation unit is connected to the first digital-to-analog converter and the counter, wherein: If U1 is greater than λU2, the comparator outputs a rising edge of partial discharge to the first digital-to-analog converter, otherwise the comparator outputs a falling edge of partial discharge to the first digital-to-analog converter and the counter. λ is the accuracy parameter. When the first DAC receives the rising edge of the partial discharge, the reference voltage U2 increases. When the first DAC and the counter receive the falling edge of the partial discharge, the first DAC outputs the current reference voltage U2 and the counter outputs the current phase number. The data generation unit generates multiple arrays containing reference voltages and phase numbers, normalizes the reference voltages and phase numbers of the arrays according to the number of rows and columns of the image matrix, and then generates elements of the image matrix according to the number of rows and columns corresponding to each array.

2. The PRPD image generation circuit according to claim 1, characterized in that: The output end of the comparator is also connected to the peak holding unit. When the comparator generates a rising edge of partial discharge, it outputs a high level to the peak holding unit. When the comparator generates a falling edge of partial discharge, it outputs a low level to the peak holding unit. When the comparator signal received by the peak hold unit remains at a low level, the peak hold unit enters the envelope tracking state. When the comparator signal received by the peak holding unit remains at a high level or jumps from a low level to a high level, the peak holding unit enters the peak detection state. When the comparator signal received by the peak holding unit jumps from a high level to a low level, the peak holding unit releases the current partial discharge signal.

3. A PRPD image generation method using the PRPD image generation circuit according to claim 1 or 2, characterized in that: The following steps are involved: Step 1: The data generation unit creates an image matrix, the power frequency generation unit generates a power frequency synchronization signal, and the counter resets the phase number and starts counting after receiving the power frequency falling edge of the power frequency synchronization signal; Step 2: The peak hold unit enters the envelope tracking state, the first digital-to-analog converter initializes the reference voltage U2, and the peak hold unit collects the partial discharge signal; Step 3: The comparator compares the partial discharge voltage U1 of the partial discharge signal with the reference voltage U2. If U1 is greater than λU2, the process proceeds to step 4. Otherwise, the process returns to step 2. λ is the accuracy parameter. Step 4: The peak hold unit enters the peak detection state, the comparator outputs a partial discharge rising edge to the first digital-to-analog converter, and the reference voltage U2 increases; Step 5: The peak hold unit re-collects the partial discharge signal, and the comparator re-compares the partial discharge voltage U1 with the reference voltage U2. If U1 is greater than U2, the process returns to step 4; otherwise, the process proceeds to step 6. Step 6: The comparator outputs a falling edge of the partial discharge to the first digital-to-analog converter and the counter. The first digital-to-analog converter outputs the current reference voltage U2, the counter outputs the current phase number, and the reference voltage U2 is reset. Step 7: Repeat steps 2 to 6 to generate multiple arrays containing reference voltages and phase numbers. Normalize the reference voltages and phase numbers of the arrays according to the number of rows and columns of the image matrix. Then, generate elements of the image matrix based on the normalized values ​​of the reference voltages and phase numbers, and output the image matrix.

4. The PRPD image generation method according to claim 3, characterized in that: In step 4, the reference voltage U2 is increased by a step size U step =0.001V to 0.005V, the image matrix is ​​a 32×32 two-dimensional matrix, the power frequency synchronization signal is a square wave signal with a period of 20ms, and the voltage of the power frequency synchronization signal is 3.3V.

5. The PRPD image generation method according to claim 3, characterized in that: In step 4, the peak holding unit receives the reference voltage U of the second DAC. ref , partial discharge voltage U1=U0-U ref , U0 is the current voltage of the partial discharge signal.

6. The PRPD image generation method according to claim 3, characterized in that: In step 7, the image matrix is ​​an N-row M-column matrix, and according to the maximum reference voltage U max Set the normalization ratio M / U max , M is the number of columns of the image matrix, the normalized value of the reference voltage U2 is m=Floor(U2M / U max )+1, Floor() is the rounding down function.

7. The PRPD image generation method according to claim 6, characterized in that: The number of arrays K whose normalized value of the statistical phase number is n and whose normalized value of the reference voltage is m nm , element P in the image matrix nm =(1-0.02K nm / T)×255, T is the detection time, n≤N, m≤M.

8. A method for identifying partial discharge types in the PRPD image generation method according to claim 3, characterized in that: The following steps are involved: Step 100: Generate at least two groups of partial discharge signals of partial discharge types, implement the PRPD image generation method, and generate multiple groups of image matrices; Step 200: training a partial discharge classification model based on multiple sets of image matrices and corresponding partial discharge types; Step 300: Generate an image matrix according to the PRPD image generation method, and identify the partial discharge type of the image matrix according to the partial discharge classification model.

9. The method for identifying partial discharge types according to claim 8, characterized in that: In step 100 , a partial discharge signal is generated based on a gas-insulated switchgear, where the types of partial discharge include tip discharge, particle discharge, air gap discharge, and suspended discharge.

10. The method for identifying partial discharge types according to claim 8, characterized in that: The partial discharge classification model includes multiple groups of alternately connected hourglass-shaped attention layers and transition layers. The hourglass-shaped attention layer is composed of a plurality of hourglass-shaped attention units stacked together. The hourglass-shaped attention unit has a lightweight convolution block and an attention block for adjusting the feature weights of the image matrix.

Images