A method and system for distributed collection and intelligent diagnosis of ultra-high frequency partial discharge signals
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI SHENGYUAN ELECTRICAL TECHNOLOGY CO LTD
- Filing Date
- 2026-04-16
- Publication Date
- 2026-08-07
AI Technical Summary
针对现有技术的不足，本发明提供了一种特高频局放信号分布式采集与智能诊断方法及系统，具备信号传输损耗低、诊断准确率高、定位精度高、运维便捷性强等优点，解决了现有特高频局放检测技术中信号传输损耗大、诊断准确率低、定位精度差、就地处理能力不足的问题
1、该特高频局放信号分布式采集与智能诊断方法及系统，通过分布式采集与就地预处理设计，减少原始信号的长距离传输，降低特高频信号的损耗与干扰，从而保障信号保真度，实现信号传输损耗低的作用。
Smart Images
Figure CN122525301A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power equipment condition monitoring technology, specifically to a method and system for distributed acquisition and intelligent diagnosis of ultra-high frequency partial discharge signals. Background Technology
[0002] Ultra-high frequency (UHF) partial discharge detection is a core technology for early warning of insulation degradation in power equipment. It captures UHF electromagnetic wave signals generated by partial discharge, enabling early detection of equipment faults. However, existing technologies suffer from three major shortcomings: First, in centralized acquisition modes, UHF signals experience significant transmission losses over long distances and are susceptible to electromagnetic interference, leading to signal distortion. Second, fault diagnosis often relies on single machine learning models, which lack the ability to extract features from complex discharge signals, resulting in low accuracy. Third, discharge point location depends solely on two-dimensional planar algorithms, neglecting the influence of the equipment's three-dimensional structure and dielectric losses, resulting in significant positioning errors. Furthermore, traditional systems have limited functionality in their local acquisition units, only capable of signal sampling and lacking real-time anti-interference and feature extraction capabilities, leading to high computational load on the server and poor diagnostic timeliness. Therefore, there is an urgent need to develop a distributed acquisition, intelligent anti-interference, and high-precision positioning online monitoring method and system for UHF partial discharge. Summary of the Invention
[0003] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a distributed acquisition and intelligent diagnosis method and system for UHF partial discharge signals. It has advantages such as low signal transmission loss, high diagnostic accuracy, high positioning accuracy, and convenient operation and maintenance, and solves the problems of high signal transmission loss, low diagnostic accuracy, poor positioning accuracy, and insufficient local processing capability in existing UHF partial discharge detection technologies.
[0004] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a distributed acquisition and intelligent diagnosis method for ultra-high frequency partial discharge signals, comprising the following steps: S1: Simultaneously capture 300MHz to 1.5GHz ultra-high frequency electromagnetic wave signals generated by partial discharge inside power equipment through at least 3 distributed ultra-high frequency sensors, with the sensor output impedance matched to a 50Ω radio frequency transmission link; S2: The local acquisition unit performs high-speed sampling of the analog signal transmitted by the sensor at a rate of more than 2GSps. The signal is then processed by FPGA to achieve dual-level anti-interference processing of LCL high-pass filtering and wavelet threshold denoising, and extracts 12-dimensional feature parameters such as peak value, pulse width, and spectral entropy of the signal. S3: The preprocessed feature parameters are transmitted to the monitoring server via industrial Ethernet. The server builds an improved CNN-LSTM hybrid model and inputs the feature parameters into the model to complete the identification of typical fault types such as corona discharge, surface discharge, and floating potential discharge. S4: Based on the Time Difference of Arrival (TDOA) of the sensor array and combined with the three-dimensional structural parameters of the power equipment, the spatial coordinates of the discharge point are calculated through a triangulation algorithm to achieve precise location of the discharge point.
[0005] Preferably, the dual-stage anti-interference processing in step S2 specifically involves: firstly filtering out low-frequency interference signals below 280MHz using an LCL high-pass filter, and then using an improved adaptive wavelet threshold denoising algorithm to suppress noise in the filtered signal. The threshold function is an asymmetric threshold function, and the threshold coefficient is dynamically adjusted according to the signal-to-noise ratio.
[0006] Preferably, the dual-stage anti-interference processing in step S2 specifically involves: firstly filtering out low-frequency interference signals below 280MHz using an LCL high-pass filter, and then using an improved adaptive wavelet threshold denoising algorithm to suppress noise in the filtered signal. The threshold function is an asymmetric threshold function, and the threshold coefficient is dynamically adjusted according to the signal-to-noise ratio.
[0007] Preferably, the improved CNN-LSTM hybrid model described in step S3 introduces an attention mechanism in the CNN layer to enhance the extraction of key spectral features of the partial discharge signal, the LSTM layer adopts a bidirectional structure to capture the temporal correlation features of the signal, and the model training adopts a transfer learning method, which is pre-trained based on a public partial discharge dataset and then fine-tuned by combining field measurement data.
[0008] Preferably, the positioning error correction method of the triangulation algorithm in step S4 is as follows: Based on the dielectric loss parameters of the power equipment and the electromagnetic wave propagation speed, the positioning formula after compensating for the signal arrival time difference is:
[0009] Where (x, y, z) are the coordinates of the discharge point, (x_i, y_i, z_i) are the coordinates of the sensor, v is the propagation speed of the electromagnetic wave in the medium of the device, ti is the measured arrival time of the signal, and Δti is the dielectric loss compensation time.
[0010] Another technical problem to be solved by the present invention is to provide a distributed acquisition and intelligent diagnosis system for ultra-high frequency partial discharge signals, including a distributed ultra-high frequency sensor array, a local acquisition unit, an industrial Ethernet transmission module, and a monitoring server; The distributed ultra-high frequency sensor array includes three or more ultra-high frequency sensors, which are connected to the local acquisition unit via SYV-50-5 RF coaxial cables. The local acquisition unit includes a high-speed ADC module, an FPGA signal processing module, and an STM32 industrial control core board, which are electrically connected in sequence. The monitoring server has a built-in improved CNN-LSTM hybrid model and triangulation algorithm module to realize fault identification and discharge point location.
[0011] Preferably, the ultra-high frequency sensor is a circular plate-type capacitive coupling sensor with an effective sensing area ≥ 5 cm². 2 Sensitivity ≤0.01V / m, protection rating IP67, suitable for working environments from -40℃ to 55℃.
[0012] Preferably, the local acquisition unit adopts a rail-mounted structure and has a built-in overvoltage protection circuit, which can be installed and maintained while the power equipment is energized, and the transmission delay of the acquired data is ≤10ms.
[0013] Preferably, the monitoring server also includes a visualization module and a fault early warning module, which can display the partial discharge signal waveform, fault type, and discharge point location in real time, and issue three-level early warning signals based on the discharge amount and development trend. The early warning information can be pushed to the power grid dispatching system.
[0014] (III) Beneficial Effects Compared with existing technologies, this invention provides a method and system for distributed acquisition and intelligent diagnosis of ultra-high frequency partial discharge signals, which has the following beneficial effects: 1. The distributed acquisition and intelligent diagnosis method and system for UHF partial discharge signals reduces the long-distance transmission of the original signal and reduces the loss and interference of UHF signals through distributed acquisition and local preprocessing design, thereby ensuring signal fidelity and achieving low signal transmission loss.
[0015] 2. This method and system for distributed acquisition and intelligent diagnosis of UHF partial discharge signals introduces attention mechanism and transfer learning through an improved CNN-LSTM hybrid model, achieving an accuracy of ≥99.9% in identifying complex partial discharge signals, which is superior to the traditional single model, thus achieving high diagnostic accuracy.
[0016] 3. The distributed acquisition and intelligent diagnosis method and system for ultra-high frequency partial discharge signals, by combining the triangular positioning algorithm of the three-dimensional structure of the equipment and dielectric loss compensation, achieves a positioning error of ≤3cm, which solves the shortcoming of the accuracy of traditional two-dimensional positioning and thus achieves high positioning accuracy.
[0017] 4. The distributed acquisition and intelligent diagnosis method and system for ultra-high frequency partial discharge signals supports live installation and maintenance through local acquisition units, eliminating the need for power outages, reducing operation and maintenance costs, improving power supply reliability, and thus achieving a high degree of convenience in operation and maintenance.
[0018] 5. The distributed acquisition and intelligent diagnosis method and system for ultra-high frequency partial discharge signals can push early warning information to the power grid dispatching system through real-time data processing and fault diagnosis, providing accurate basis for equipment maintenance, avoiding the expansion of insulation faults, and thus achieving the effect of good early warning timeliness. Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the detection circuit of a distributed acquisition and intelligent diagnosis method and system for ultra-high frequency partial discharge signals proposed in this invention. Figure 2 This is a structural diagram of the data acquisition unit of the distributed acquisition and intelligent diagnosis method and system for ultra-high frequency partial discharge signals proposed in this invention; Figure 3 This is a schematic diagram of the system architecture of a distributed acquisition and intelligent diagnosis method and system for ultra-high frequency partial discharge signals proposed in this invention. Detailed Implementation
[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] Please see Figure 1-3 This invention provides a method for distributed acquisition and intelligent diagnosis of ultra-high frequency partial discharge signals, comprising the following steps: S1: Simultaneously capture 300MHz to 1.5GHz ultra-high frequency electromagnetic wave signals generated by partial discharge inside power equipment through at least 3 distributed ultra-high frequency sensors, with the sensor output impedance matched to a 50Ω radio frequency transmission link; S2: The local acquisition unit performs high-speed sampling of the analog signal transmitted by the sensor at a rate of more than 2GSps. The signal is then processed by FPGA to achieve dual-level anti-interference processing of LCL high-pass filtering and wavelet threshold denoising, and extracts 12-dimensional feature parameters such as peak value, pulse width, and spectral entropy of the signal. S3: The preprocessed feature parameters are transmitted to the monitoring server via industrial Ethernet. The server builds an improved CNN-LSTM hybrid model and inputs the feature parameters into the model to complete the identification of typical fault types such as corona discharge, surface discharge, and floating potential discharge. S4: Based on the Time Difference of Arrival (TDOA) of the sensor array and combined with the three-dimensional structural parameters of the power equipment, the spatial coordinates of the discharge point are calculated through a triangulation algorithm to achieve precise location of the discharge point.
[0022] The two-stage anti-interference processing in step S2 is as follows: First, low-frequency interference signals below 280MHz are filtered out using an LCL high-pass filter. Then, an improved adaptive wavelet threshold denoising algorithm is used to suppress noise in the filtered signal. The threshold function is an asymmetric threshold function, and the threshold coefficient is dynamically adjusted according to the signal-to-noise ratio. In step S3, the improved CNN-LSTM hybrid model introduces an attention mechanism in the CNN layer to enhance the extraction of key spectral features of the partial discharge signal. The LSTM layer adopts a bidirectional structure to capture the temporal correlation features of the signal. The model training adopts a transfer learning method, pre-trained on a public partial discharge dataset, and then fine-tuned using field measurement data. In step S4, the positioning error correction method of the triangulation algorithm is as follows: Based on the dielectric loss parameters of the power equipment and the electromagnetic wave propagation speed, the positioning formula after compensating for the signal arrival time difference is:
[0023] Where (x, y, z) are the coordinates of the discharge point, (x_i, y_i, z_i) are the coordinates of the sensor, v is the propagation speed of the electromagnetic wave in the medium of the device, ti is the measured arrival time of the signal, and Δti is the dielectric loss compensation time.
[0024] A distributed acquisition and intelligent diagnostic system for ultra-high frequency (UHF) partial discharge signals includes a distributed UHF sensor array, a local acquisition unit, an industrial Ethernet transmission module, and a monitoring server. The distributed UHF sensor array comprises three or more UHF sensors, connected to the local acquisition unit via SYV-50-5 RF coaxial cables. The local acquisition unit includes a high-speed ADC module, an FPGA signal processing module, and an STM32 industrial control core board, electrically connected in sequence. The monitoring server incorporates an improved CNN-LSTM hybrid model and triangulation algorithm module to achieve fault identification and discharge point location. The UHF sensors are circular plate-type capacitively coupled sensors with an effective sensing area ≥5 cm². 2 With a sensitivity of ≤0.01V / m and an IP67 protection rating, it is suitable for working environments ranging from -40℃ to 55℃. The local acquisition unit adopts a rail-mounted structure and has a built-in overvoltage protection circuit, allowing for installation and maintenance while the power equipment is energized. The data transmission delay is ≤10ms. The monitoring server also includes a visualization module and a fault early warning module, which can display the partial discharge signal waveform, fault type, and discharge point location in real time. It can also issue three-level early warning signals based on the discharge quantity and development trend, and the early warning information can be pushed to the power grid dispatching system.
[0025] Example 1: Implementation of a low-cost acquisition system based on detector frequency reduction This embodiment addresses the issues of high frequency and high sampling cost associated with ultra-high frequency signals by employing a detector-frequency reduction circuit to process the signal.
[0026] The frequency of the electromagnetic wave signal released by partial discharge detected by the ultra-high frequency sensor is in the range of 300 MHz to 3 GHz. In this embodiment, a detection circuit is used to down-convert the ultra-high frequency signal. The principle of the detection circuit is as Figure 1 shown. The diode is selected as the detection diode 1N60, and its internal resistance R. < R. In this embodiment, the input discharge pulse width is 0.35 ns to 3 ns. After passing through the detection circuit, the narrow pulse is broadened to more than several hundred ns. A high-speed A / D with a sampling rate of 40 MHz (sampling period of 25 ns) can be used for signal acquisition.
[0027] When a partial discharge pulse arrives, the detection circuit charges, and the charging time constant τ charge = R.C; when the partial discharge pulse ends, the detection diode is reversely cut off, and the capacitor of the detection circuit discharges, and the discharge time constant τ discharge = R1C. Design the capacitance value of the capacitor C and the resistance value of the load R in the detection circuit so that τ discharge is much larger than the sampling period of the high-speed A / D, which can ensure the effective acquisition of the signal.
[0028] In this embodiment, the system adopts a typical data acquisition structure. The partial discharge signal collected by the sensor is completed by the A / D chip and the local cache. The system is independently completed by the FPGA, the LCP2292 single-chip microcomputer and the peripherals for data processing. The structure is as Figure 2 shown.
[0029] Embodiment 2: Implementation of a distributed acquisition and intelligent diagnosis system based on GIS equipment In this embodiment, the system of the present invention is applied to the on-line monitoring of partial discharge of GIS equipment with high voltage levels of ≥110 kV.
[0030] Distributed ultra-high frequency sensing array: Circular plate-shaped capacitive coupling ultra-high frequency sensors are respectively installed at three adjacent pot-type insulators of the GIS equipment, and the sensor spacing is about 2 to 3 meters. The effective sensing area of the sensor is 5 cm 2 , the sensitivity is 0.01 V / m, the protection level is IP67, and it is connected to the nearby installed on-site acquisition unit through a SYV-50-5 radio frequency coaxial cable.
[0031] On-site acquisition unit: Adopting a rail-mounted installation structure, it incorporates an AD9680 high-speed ADC module (sampling rate 2 GSps), a Xilinx Zynq-7020 FPGA signal processing module and an STM32 industrial control core board. The FPGA module first performs LCL high-pass filtering on the acquired signal to filter out low-frequency interference below 280 MHz; then adopts an improved adaptive wavelet threshold denoising algorithm, and the threshold coefficient is dynamically adjusted according to the signal-to-noise ratio to filter out random noise; finally, 12-dimensional characteristic parameters such as the peak value, pulse width, and spectral entropy of the signal are extracted. The STM32 core board packages the characteristic parameters and sends them through the industrial Ethernet module.
[0032] Industrial Ethernet transmission module: It adopts Huawei S5720 industrial switch to transmit data to the monitoring server through optical fiber, with a transmission delay of ≤8ms.
[0033] Monitoring Server: The server incorporates an improved CNN-LSTM hybrid model. The CNN layer introduces an attention mechanism to enhance spectral feature extraction, while the LSTM layer employs a bidirectional structure to capture temporal correlation features. After transfer learning training, the model achieves a 99.95% accuracy rate in identifying corona discharge, surface discharge, and floating potential discharge. The server also runs a triangulation algorithm module, which calculates the discharge point coordinates based on the arrival time difference of three sensor signals, combined with the 3D structural parameters of the GIS equipment and the dielectric loss compensation formula, achieving a positioning error ≤2.5cm. The visualization module displays the discharge waveform, fault type, and 3D positioning results in real time. When the discharge exceeds a threshold, a level-three warning is triggered, and the warning information is pushed to the power grid dispatch system.
[0034] Example 3: Implementation of a Transformer-Based UHF Partial Discharge Monitoring System This embodiment applies the system of the present invention to the online monitoring of partial discharge in oil-immersed transformers with high voltage levels of 110kV and above.
[0035] Distributed UHF sensor array: Taking a 220kV transformer as an example, UHF sensors are built-in at each of the four oil valves of the transformer. The sensors cover all four quadrants inside the transformer, eliminating monitoring blind spots. The sensor output impedance is matched to 50Ω and connected to a local acquisition unit outside the transformer body via a coaxial cable.
[0036] Local acquisition unit: The structure is the same as in Example 2, but the FPGA module adds feature parameter extraction dimensions to target the discharge signal characteristics in transformer oil. In addition to the 12 basic parameters, features such as the rising edge steepness and oscillation frequency of the signal are extracted.
[0037] Monitoring Server: The server employs the same CNN-LSTM hybrid model, but with secondary fine-tuning through transfer learning to address the characteristics of discharges in transformer oil. The training data includes typical transformer discharge samples. Real-world data shows that the model achieves a 99.8% accuracy rate in identifying internal transformer discharges. The localization algorithm, combined with the three-dimensional structure of the transformer tank, accurately calibrates the propagation speed of electromagnetic waves in the oil (approximately 1.5 × 10^8 m / s), with a localization error controlled within 3 cm.
[0038] In summary, this distributed acquisition and intelligent diagnosis method and system for UHF partial discharge signals, through the deep integration of a distributed UHF sensor array, a local acquisition unit, an industrial Ethernet transmission module, and a monitoring server, constructs a partial discharge online monitoring system covering the entire chain from "signal acquisition to local preprocessing to cloud-based intelligent diagnosis to three-dimensional precise positioning." This effectively solves the technical problems of high signal transmission loss, low diagnostic accuracy, and poor positioning accuracy in traditional UHF online monitoring technologies. It achieves excellent performance with an identification accuracy of ≥99.9% and a positioning error of ≤3cm, providing a reliable and efficient means of insulation status detection for high-voltage electrical equipment such as GIS, transformers, and switchgear.
[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for distributed acquisition and intelligent diagnosis of ultra-high frequency partial discharge signals, characterized in that, Includes the following steps: S1: Simultaneously capture 300MHz to 1.5GHz ultra-high frequency electromagnetic wave signals generated by partial discharge inside power equipment through at least 3 distributed ultra-high frequency sensors, with the sensor output impedance matched to a 50Ω radio frequency transmission link; S2: The local acquisition unit performs high-speed sampling of the analog signal transmitted by the sensor at a rate of more than 2GSps. The signal is then processed by FPGA to achieve dual-level anti-interference processing of LCL high-pass filtering and wavelet threshold denoising, and extracts 12-dimensional feature parameters such as peak value, pulse width, and spectral entropy of the signal. S3: The preprocessed feature parameters are transmitted to the monitoring server via industrial Ethernet. The server builds an improved CNN-LSTM hybrid model and inputs the feature parameters into the model to complete the identification of typical fault types such as corona discharge, surface discharge, and floating potential discharge. S4: Based on the Time Difference of Arrival (TDOA) of the sensor array and combined with the three-dimensional structural parameters of the power equipment, the spatial coordinates of the discharge point are calculated through a triangulation algorithm to achieve precise location of the discharge point.
2. The method for distributed acquisition and intelligent diagnosis of UHF partial discharge signals according to claim 1, characterized in that, The two-stage anti-interference processing described in step S2 is as follows: first, low-frequency interference signals below 280MHz are filtered out by an LCL high-pass filter, and then an improved adaptive wavelet threshold denoising algorithm is used to suppress noise in the filtered signal. The threshold function adopts an asymmetric threshold function, and the threshold coefficient is dynamically adjusted according to the signal-to-noise ratio.
3. The method for distributed acquisition and intelligent diagnosis of UHF partial discharge signals according to claim 1, characterized in that, The improved CNN-LSTM hybrid model described in step S3 introduces an attention mechanism in the CNN layer to enhance the extraction of key spectral features of the partial discharge signal. The LSTM layer adopts a bidirectional structure to capture the temporal correlation features of the signal. The model training adopts a transfer learning method, which is pre-trained based on a public partial discharge dataset and then fine-tuned by combining field measurement data.
4. The method for distributed acquisition and intelligent diagnosis of UHF partial discharge signals according to claim 1, characterized in that, The positioning error correction method of the triangulation algorithm described in step S4 is as follows: Based on the dielectric loss parameters of the power equipment and the propagation speed of electromagnetic waves, the positioning formula after compensating for the signal arrival time difference is: Where (x, y, z) are the coordinates of the discharge point, (x_i, y_i, z_i) are the coordinates of the sensor, v is the propagation speed of the electromagnetic wave in the medium of the device, ti is the measured arrival time of the signal, and Δti is the dielectric loss compensation time.
5. A distributed acquisition and intelligent diagnostic system for ultra-high frequency partial discharge signals, characterized in that, It includes a distributed ultra-high frequency sensor array, a local acquisition unit, an industrial Ethernet transmission module, and a monitoring server; The distributed ultra-high frequency sensor array includes three or more ultra-high frequency sensors, which are connected to the local acquisition unit via SYV-50-5 RF coaxial cables. The local acquisition unit includes a high-speed ADC module, an FPGA signal processing module, and an STM32 industrial control core board, which are electrically connected in sequence. The monitoring server has a built-in improved CNN-LSTM hybrid model and triangulation algorithm module to realize fault identification and discharge point location.
6. The UHF partial discharge signal distributed acquisition and intelligent diagnostic system according to claim 5, characterized in that, The ultra-high frequency sensor is a circular plate capacitive coupling sensor with an effective sensing area ≥5cm², sensitivity ≤0.01V / m, protection level IP67, and is suitable for working environments from -40℃ to 55℃.
7. The UHF partial discharge signal distributed acquisition and intelligent diagnostic system according to claim 5, characterized in that, The local data acquisition unit adopts a rail-mounted structure and has a built-in overvoltage protection circuit. It can be installed and maintained while the power equipment is energized, and the data transmission delay is ≤10ms.
8. The UHF partial discharge signal distributed acquisition and intelligent diagnostic system according to claim 5, characterized in that, The monitoring server also includes a visualization module and a fault early warning module, which can display the partial discharge signal waveform, fault type, and discharge point location in real time, and issue three-level early warning signals based on the discharge amount and development trend. The early warning information can be pushed to the power grid dispatching system.