Gallium nitride ultraviolet partial discharge monitoring device

Abstract

The invention discloses a gallium nitride ultraviolet partial discharge monitoring device, and relates to the technical field of online monitoring and fault diagnosis of power equipment. The device comprises a magnetic type monitoring probe, the magnetic type monitoring probe comprises a probe shell, a magnetic device is arranged on one side of the probe shell, and an optical sensing module, a signal processing module, an intelligent control and diagnosis module, a storage and communication module and a power supply module are integrated in an inner cavity of the probe shell; and the optical sensing module comprises a filtering lens and an AlGaN / GaN heterostructure ultraviolet detector array. The optical sensing module accurately senses ultraviolet radiation and shields background interference, the intelligent control and diagnosis module carries out rapid and accurate diagnosis on the edge side, original data does not need to be uploaded to the background, the response efficiency is greatly improved, the problems that in the prior art, monitoring is not accurate, and diagnosis delay is high are solved, and the monitoring accuracy is improved. And the partial discharge hidden danger of the high-voltage switch cabinet can be found in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of online monitoring and fault diagnosis of power equipment, and particularly relates to a gallium nitride ultraviolet partial discharge monitoring device. BACKGROUND

[0002] As the core electrical equipment for the safe and stable operation of the power system, the monitoring of the partial discharge of the high-voltage switch cabinet is crucial. Hidden dangers such as contamination of insulating parts, dampness, conductor connection defects, and metal burrs can easily cause partial discharge. If not monitored and handled in a timely manner, it can lead to insulation breakdown and further induce large-scale power outage accidents. Therefore, accurate and effective partial discharge monitoring technology is of great significance to ensure the stable operation of the power system. At present, various partial discharge monitoring technologies are applied in the power system. However, with the development of the power system and the increasing demand for monitoring, these technologies have gradually exposed some problems.

[0003] From the performance and application requirements of the monitoring technology, the existing mainstream partial discharge monitoring technology has significant shortcomings. The pulse current method has a low detection frequency and is extremely susceptible to electromagnetic interference in complex field environments, which makes it unable to accurately capture weak discharge signals, resulting in inaccurate monitoring results and the possible omission of important partial discharge information. Although the ultra-high frequency electromagnetic wave detection method has certain advantages in theory, it requires complex antenna design and signal demodulation circuit in practical application, which not only increases the difficulty of system integration but also significantly increases the cost, limiting its widespread application. At the same time, most existing technologies upload raw monitoring data to the background for analysis. This approach requires a very high communication bandwidth and has a high diagnostic delay, making it difficult to reflect the partial discharge situation in real time. In addition, existing technologies have difficulty in accurately locating the discharge point, which cannot meet the current distributed, small-sized, and low-cost field monitoring requirements and cannot well adapt to the increasingly complex monitoring environment of the power system.

[0004] Therefore, we provide a gallium nitride ultraviolet partial discharge monitoring device to solve the above problems. SUMMARY

[0005] The purpose of the present application is to provide a gallium nitride ultraviolet partial discharge monitoring device that solves the problems of monitoring signals being susceptible to interference, high diagnostic delay, and difficulty in meeting distributed low-cost installation requirements through the cooperation of the optical perception module, the intelligent control and diagnosis module, and the power supply module.

[0006] To solve the above technical problems, the present application is realized by the following technical solutions: The application is a kind of gallium nitride ultraviolet partial discharge monitoring device, including a magnetic monitoring probe, the magnetic monitoring probe includes a probe shell, one side of the probe shell is provided with a magnetic attraction device, the inner cavity of the probe shell is integrated with an optical sensing module, a signal processing module, an intelligent control and diagnosis module, a storage and communication module and a power supply module; the optical sensing module includes a filter lens and an AlGaN / GaN heterostructure ultraviolet detector array, the filter lens only transmits 240-280nm solar blind ultraviolet band light; the control unit of the intelligent control and diagnosis module is a high-performance single-chip microcomputer with a built-in floating-point operation unit, and the high-performance single-chip microcomputer is embedded with a lightweight multi-layer perception artificial intelligence model, which can complete the recognition, intensity determination and direction estimation of partial discharge on the edge side, through the integration of the optical sensing module, the signal processing module, the intelligent control and diagnosis module, the storage and communication module and the power supply module, a complete and powerful monitoring system is formed, the optical sensing module can effectively shield background interference and accurately perceive ultraviolet radiation; the intelligent control and diagnosis module completes the recognition, intensity determination and direction estimation of partial discharge on the edge side, improves the response efficiency and reduces the bandwidth dependence, compared with the prior art, the problems of inaccurate monitoring, delayed response and high bandwidth requirement are solved.

[0007] The application is further provided that the AlGaN / GaN heterostructure ultraviolet detector array and the row and column gating readout circuit are integrated on the same substrate through a microprocessing technology, the AlGaN / GaN heterostructure ultraviolet detector array adopts a matrix gating readout architecture, the circuit board integrates corresponding row gating and column readout paths, and the control unit can activate the specified detector region in turn and read the photoelectric response signal in time through the row and column addressing mechanism, the AlGaN / GaN heterostructure ultraviolet detector array and the row and column gating readout circuit are integrated on the same substrate, and the matrix gating readout architecture enables the control unit to obtain the spatial distribution information of the ultraviolet spot through the row and column addressing mechanism, improving the accuracy and comprehensiveness of information acquisition, which helps to more accurately analyze the partial discharge condition.

[0008] The application is further provided that each detector in the AlGaN / GaN heterostructure ultraviolet detector array has independent addressing capability, and the output signal can be read individually or in combination, and the circuit components integrate a bias and decoupling circuit composed of resistors and capacitors, each detector in the AlGaN / GaN heterostructure ultraviolet detector array has independent addressing capability, and the output signal can be read individually or in combination, and the bias and decoupling circuit is integrated. This makes the detector more flexible in signal reading and processing, better adapts to different monitoring scenes, and improves the reliability and adaptability of monitoring.

[0009] The present invention is further configured such that a polarization electric field pointing towards the GaN layer exists between the AlGaN layer and the GaN layer of the AlGaN / GaN heterostructure. The GaN layer has a quantum well near the AlGaN interface, which has a two-dimensional electron gas. The AlGaN layer has negatively charged DX centers. When ultraviolet radiation is incident, electrons in the DX centers absorb photon energy and enter the conduction band. Under the action of the polarization electric field, they enter the quantum well, thereby modulating the density of the high-mobility two-dimensional electron gas at the heterostructure interface and realizing ultraviolet radiation detection. The AlGaN / GaN heterostructure realizes ultraviolet radiation detection by modulating the density of the two-dimensional electron gas with photogenerated electrons, which has higher sensitivity and can more effectively sense weak ultraviolet radiation, thereby improving the accuracy of partial discharge monitoring.

[0010] The present invention is further configured such that the signal processing module is composed of a micro-amplifier, an integrated filter bank, and a multi-channel analog-to-digital converter cascaded in sequence, which can perform time-division conditioning and digital conversion on the multi-channel signals output by the AlGaN / GaN heterostructure ultraviolet detector array. The signal processing module, composed of a micro-amplifier, an integrated filter bank, and a multi-channel analog-to-digital converter cascaded in sequence, can perform time-division conditioning and digital conversion on the multi-channel signals, ensuring the accuracy and stability of the signals and improving the quality of signal processing.

[0011] The present invention is further configured such that the lightweight multilayer perceptron artificial intelligence model includes an input layer, two hidden layers, and an output layer. The first hidden layer has 128 ReLU activated neurons, the second hidden layer has 64 ReLU activated neurons, and the output layer includes discharge confidence output neurons and partial discharge stage output neurons. The lightweight multilayer perceptron artificial intelligence model is trained using the Adam optimizer and a weighted combination loss function of mean square error and cross-entropy. The lightweight multilayer perceptron artificial intelligence model is trained using a specific network structure, activation function, optimizer, and loss function, which can more accurately identify and determine partial discharge, has higher diagnostic accuracy and efficiency, and improves the intelligent diagnostic capability of the monitoring device.

[0012] The present invention is further configured such that the feature dataset of the lightweight multilayer perceptron artificial intelligence model is constructed from laboratory simulated discharge experimental data and field measured data. The instantaneous response values ​​of the effective detection units of the AlGaN / GaN heterostructure ultraviolet detector array at a single sampling moment constitute the ultraviolet spot spatial intensity distribution matrix as the input feature of the lightweight multilayer perceptron artificial intelligence model. The feature dataset of the lightweight multilayer perceptron artificial intelligence model is constructed from laboratory simulated discharge experimental data and field measured data, and the ultraviolet spot spatial intensity distribution matrix is ​​used as the input feature, making the lightweight multilayer perceptron artificial intelligence model more closely resemble the actual monitoring situation, thereby improving the practicality and accuracy of the lightweight multilayer perceptron artificial intelligence model.

[0013] The present invention is further configured such that the storage and communication module includes a memory and a 4G wireless signal transmission module. The memory is used to temporarily store local monitoring data, and the 4G wireless signal transmission module can upload edge-side diagnostic results to the remote monitoring center in real time. The memory of the storage and communication module can temporarily store local monitoring data, and the 4G wireless signal transmission module can upload edge-side diagnostic results to the remote monitoring center in real time, thereby realizing effective data management and fast transmission, improving data security and transmission efficiency, and facilitating remote monitoring and management.

[0014] The present invention is further configured such that the power supply unit of the power supply module supports both built-in battery power supply and inductive power supply, and the magnetic attraction device for fixing the probe housing can be attached to the inner wall of the switch cabinet for quick deployment. The power supply module supports both built-in battery power supply and inductive power supply.

[0015] The present invention is further configured such that the edge-side diagnostic process is as follows: after power-on, a lightweight multilayer perceptron artificial intelligence model weight is loaded; a high-performance microcontroller cyclically collects signals from the AlGaN / GaN heterostructure ultraviolet detector array through a row and column addressing mechanism and constructs a feature vector; the feature vector is input into the lightweight multilayer perceptron artificial intelligence model to complete the forward propagation calculation; and the diagnostic result is output and uploaded to the remote monitoring center by a 4G wireless signal transmission module. The edge-side diagnostic process, by loading the lightweight multilayer perceptron artificial intelligence model weight, collecting signals from the AlGaN / GaN heterostructure ultraviolet detector array, constructing a feature vector, inputting the lightweight multilayer perceptron artificial intelligence model for calculation, and uploading the result, achieves fast and efficient real-time intelligent diagnosis, significantly improves response efficiency, reduces diagnostic latency, and better meets the real-time requirements of on-site monitoring.

[0016] The present invention has the following beneficial effects: This invention uses an optical sensing module to accurately sense ultraviolet radiation and shield background interference, and an intelligent control and diagnostic module to perform rapid and accurate diagnosis at the edge, eliminating the need to upload raw data to the backend. This greatly improves response efficiency and solves the problems of inaccurate monitoring and high diagnostic delays in existing technologies. It can promptly detect partial discharge hazards in high-voltage switchgear and provide strong protection for the stable operation of the power system.

[0017] The signal processing module of this invention ensures signal quality, the storage and communication module realizes effective data management and transmission, and the power supply module and magnetic suction device provide flexible and convenient power supply and installation methods. The modules work together to meet the needs of distributed, miniaturized, and low-cost on-site monitoring, and have higher market competitiveness and practical value compared with existing technologies. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0019] Figure 1 A front cross-sectional view of a gallium nitride ultraviolet partial discharge monitoring device; Figure 2 This is a side view of a gallium nitride ultraviolet partial discharge monitoring device. Figure 3 This is a schematic diagram of an AlGaN / GaN heterostructure ultraviolet detector array for a gallium nitride ultraviolet partial discharge monitoring device. Figure 4 A schematic diagram of a lightweight multilayer sensor artificial intelligence model structure for a gallium nitride ultraviolet partial discharge monitoring device; Figure 5 This is a data graph showing the pattern recognition results of a gallium nitride ultraviolet partial discharge monitoring device.

[0020] In the attached diagram: 1. Probe housing; 2. Magnetic suction device; 3. Filter lens; 4. AlGaN / GaN heterostructure ultraviolet detector array; 5. High-performance microcontroller; 6. Miniature amplifier; 7. Integrated filter bank; 8. Multi-channel analog-to-digital converter; 9. Memory; 10. 4G wireless signal transmission module; 11. Power supply module. Detailed Implementation

[0021] The technical solutions of the present invention will be described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments.

[0022] Please see Figures 1-5 This invention relates to a gallium nitride ultraviolet partial discharge monitoring device, comprising a magnetically attached monitoring probe. The magnetically attached monitoring probe includes a probe housing 1, with a magnetic attachment device 2 on one side of the probe housing 1. The inner cavity of the probe housing 1 integrates an optical sensing module, a signal processing module, an intelligent control and diagnostic module, a storage and communication module, and a power supply module 11. The optical sensing module includes a filter lens 3 and an AlGaN / GaN heterostructure ultraviolet detector array 4. The filter lens 3 transmits only 240-280nm solar-blind ultraviolet light. The control unit of the intelligent control and diagnostic module is a high-performance microcontroller 5 with a built-in floating-point arithmetic unit. The high-performance microcontroller 5 embeds a lightweight multilayer perceptron artificial intelligence model, which can complete the identification, intensity determination, and orientation estimation of partial discharge at the edge.

[0023] Specifically, by integrating an optical sensing module, a signal processing module, an intelligent control and diagnostic module, a storage and communication module, and a power supply module 11, a complete and powerful monitoring system is formed. The optical sensing module can effectively shield background interference and accurately sense ultraviolet radiation. The intelligent control and diagnostic module completes the identification, intensity determination, and orientation estimation of partial discharge at the edge side, improving response efficiency and reducing bandwidth dependence. Compared with existing technologies, it solves the problems of inaccurate monitoring, response delay, and high bandwidth requirements.

[0024] Please see Figures 1-5 Based on the first specific embodiment, the AlGaN / GaN heterostructure ultraviolet detector array 4 and the row and column gating readout circuit are integrated on the same substrate using microfabrication technology. The AlGaN / GaN heterostructure ultraviolet detector array 4 adopts a matrix gating readout architecture. The circuit board integrates the corresponding row gating and column readout paths. The control unit can sequentially activate the designated detector area and read the photoelectric response signal in a time-division manner through the row and column addressing mechanism. Each detector in the AlGaN / GaN heterostructure ultraviolet detector array 4 has independent addressing capability, and its output signal can be read individually or in combination. The circuit components integrate bias and decoupling circuits composed of resistors and capacitors. There is a polarization electric field pointing towards the GaN layer between the GaN layers. The GaN layer has a quantum well near the AlGaN interface, which has a two-dimensional electron gas. The AlGaN layer has negatively charged DX centers. When ultraviolet radiation is incident, the electrons in the DX centers absorb photon energy and enter the conduction band. Under the action of the polarization electric field, they enter the quantum well, thereby modulating the density of the high-mobility two-dimensional electron gas at the heterostructure interface, realizing ultraviolet radiation detection. The signal processing module is composed of a micro-amplifier 6, an integrated filter bank 7, and a multi-channel analog-to-digital converter 8 cascaded in sequence. It can perform time-division conditioning and digital conversion on the multi-channel signals output by the AlGaN / GaN heterostructure ultraviolet detector array 4.

[0025] Specifically, the AlGaN / GaN heterostructure ultraviolet detector array 4 and the row and column gating readout circuit are integrated on the same substrate. The matrix gating readout architecture enables the control unit to obtain the spatial distribution information of the ultraviolet spot through the row and column addressing mechanism, which improves the accuracy and comprehensiveness of information acquisition and helps to analyze the partial discharge situation more accurately. Each detector in the AlGaN / GaN heterostructure ultraviolet detector array 4 has independent addressing capability, and the output signal can be read individually or in combination. It also integrates bias and decoupling circuits. This makes the detector more flexible in signal reading and processing, better adaptable to different monitoring scenarios, and improves the reliability and adaptability of monitoring. The AlGaN / GaN heterostructure in the AlGaN / GaN heterostructure ultraviolet detector array 4 achieves ultraviolet radiation detection by modulating the two-dimensional electron gas density with photogenerated electrons, which has higher sensitivity and can more effectively sense weak ultraviolet radiation, thereby improving the accuracy of partial discharge monitoring. The signal processing module is composed of a micro amplifier 6, an integrated filter bank 7, and a multi-channel analog-to-digital converter 8 cascaded in sequence. It can perform time-division conditioning and digital conversion of multi-channel signals, ensuring the accuracy and stability of the signals and improving the quality of signal processing.

[0026] Please see Figures 1-5 Based on specific embodiments one and two, the lightweight multilayer perceptron artificial intelligence model includes an input layer, two hidden layers, and an output layer. The first hidden layer has 128 ReLU activated neurons, and the second hidden layer has 64 ReLU activated neurons. The output layer includes discharge confidence output neurons and partial discharge stage output neurons. The lightweight multilayer perceptron artificial intelligence model is trained using the Adam optimizer and a weighted combination loss function of mean square error and cross-entropy. The feature dataset of the lightweight multilayer perceptron artificial intelligence model is constructed from laboratory simulated discharge experimental data and field measured data. The instantaneous response values ​​of the four effective detection units of the AlGaN / GaN heterostructure ultraviolet detector array at a single sampling moment constitute the ultraviolet spot spatial intensity distribution matrix as the lightweight... The multilayer perceptron artificial intelligence model input features, storage and communication module includes memory 9 and 4G wireless signal transmission module 10. Memory 9 is used to temporarily store local monitoring data, and 4G wireless signal transmission module 10 can upload edge-side diagnostic results to the remote monitoring center in real time. The power supply module 11 supports two power supply methods: built-in battery power supply or inductive power supply. The edge-side diagnostic process is as follows: after power-on, the lightweight multilayer perceptron artificial intelligence model weights are loaded. The high-performance microcontroller 5 collects the signals of AlGaN / GaN heterostructure ultraviolet detector array 4 in a loop through the row and column addressing mechanism and constructs feature vectors. The input is used to complete the forward propagation calculation of the lightweight multilayer perceptron artificial intelligence model. After outputting the diagnostic results, they are uploaded to the remote monitoring center by 4G wireless signal transmission module 10.

[0027] Specifically, the lightweight multilayer perceptron AI model is trained using a specific network structure, activation function, optimizer, and loss function, enabling more accurate identification and judgment of partial discharges. This results in higher diagnostic accuracy and efficiency, enhancing the intelligent diagnostic capabilities of the monitoring device. The feature dataset of the lightweight multilayer perceptron AI model is constructed from laboratory simulated discharge experimental data and field measured data, using the ultraviolet spot spatial intensity distribution matrix as input features. This makes the lightweight multilayer perceptron AI model more closely reflect actual monitoring conditions, improving its practicality and accuracy. The storage and communication module's memory 9 can temporarily store local monitoring data, while the 4G wireless signal transmission module 10 can upload edge-side diagnostic results to the remote monitoring center in real time. This system enables effective data management and rapid transmission, improving data security and transmission efficiency, and facilitating remote monitoring and management. The power supply module 11 supports both built-in battery and inductive power supply. The probe housing 1 can be quickly deployed by attaching to the inner wall of the switch cabinet via the magnetic suction device 2, improving the flexibility and convenience of the device and facilitating the construction of a distributed monitoring network. The edge-side diagnostic process achieves rapid and efficient real-time intelligent diagnosis by loading the weights of the lightweight multilayer perceptron artificial intelligence model, acquiring signals from the AlGaN / GaN heterostructure ultraviolet detector array 4, constructing feature vectors, inputting the lightweight multilayer perceptron artificial intelligence model for calculation, and uploading the results. This significantly improves response efficiency, reduces diagnostic latency, and better meets the real-time requirements of on-site monitoring.

[0028] The filter lens 3 is made of quartz material, and after coating, it achieves high transmittance in the 240-280nm wavelength band and visible light cutoff rate ≥99%. The AlGaN / GaN heterostructure ultraviolet detector array 4 adopts a 4×4 matrix layout. The substrate is made of Si material, and the AlGaN / GaN heterostructure is grown using MOCVD technology. The two-dimensional electron gas mobility is [not specified]. ; The row and column gating readout circuits are integrated on the substrate using CMOS technology, with a row gating voltage of 5V and a column readout circuit bandwidth of 1MHz. The high-performance microcontroller 5 uses the STM32H7 series, with a built-in FPU unit and a main frequency of ≥400MHz, which can meet the real-time computing needs of lightweight multilayer perceptron artificial intelligence models. The signal processing module features a micro-amplifier 6 with adjustable gain (20-60dB), an integrated filter bank 7 containing 10kHz high-pass and 1MHz low-pass filter paths, and a multi-channel ADC with a sampling accuracy of 12bit and a sampling rate ≥1MSPS. The storage module is equipped with a 16GB Flash memory 9, and the 4G wireless signal transmission module 10 uses the Cat.1 standard and supports full network compatibility; The power supply module has a built-in 3.7V / 5000mAh lithium battery and a reserved inductive power input interface, which can draw power from the switch cabinet busbar to achieve continuous power supply. The magnetic housing is made of ferrite permanent magnet material with an attraction force of ≥10N and a protection level of IP65, making it suitable for complex environments inside switch cabinets.

[0029] The lightweight multilayer perceptron artificial intelligence model and its implementation method include the following key contents: Data from the AlGaN / GaN heterostructure ultraviolet detector array 4 was obtained through laboratory simulated discharge experiments and field measurements. At a single sampling moment, the instantaneous response values ​​of all effective detection units in the AlGaN / GaN heterostructure ultraviolet detector array 4 were read to form a matrix representing the spatial intensity distribution of the ultraviolet spot at that moment.

[0030] The lightweight multilayer perceptron AI model comprises one input layer, two hidden layers, and one output layer. The number of neurons in the input layer matches the dimension of the fused feature vector. The first hidden layer contains 128 neurons, and the second hidden layer contains 64 neurons, both using the ReLU activation function to introduce non-linearity and accelerate training convergence. The output layer contains two types of neurons: one type outputs the firing confidence level in the range of 0-1 with a step size of 0.1, and the other type outputs the local discharge levels of early, mid, and late partial discharges. This lightweight multilayer perceptron AI model uses backpropagation for gradient calculation and employs the Adam optimizer to adaptively adjust the network parameters. The loss function is designed as a weighted combination of mean squared error and cross-entropy to synergistically optimize performance for regression and classification tasks.

[0031] After a gallium nitride ultraviolet partial discharge monitoring device is powered on, the weight parameters of the trained lightweight multilayer perceptron artificial intelligence model are loaded into the memory of a high-performance microcontroller 5. The high-performance microcontroller 5 uses a row and column addressing mechanism to cyclically collect signals from each unit of the AlGaN / GaN heterostructure ultraviolet detector array 4 at a predetermined sampling rate, and constructs an input feature vector which is input into the loaded lightweight multilayer perceptron artificial intelligence model for forward propagation calculation. After the calculation is completed, the lightweight multilayer perceptron artificial intelligence model outputs the recognition result, and the control unit uploads the recognition result to the remote monitoring center through a 4G wireless module.

[0032] The working principle of this invention is as follows: There is a polarized electric field pointing towards the GaN layer between the AlGaN layer and the GaN layer. The GaN layer has a quantum well near the AlGaN interface, which has a two-dimensional electron gas. The AlGaN layer has negatively charged DX centers. The specially designed filter lens 3 at the front end of the probe only allows light in the 240-280nm solar-blind ultraviolet band to pass through, which can effectively shield background interference such as visible light and infrared light. When the switch cabinet experiences partial discharge, it will radiate ultraviolet light. After this light is incident on the AlGaN / GaN heterostructure ultraviolet detector array 4, the electrons in the DX center absorb photon energy and enter the conduction band. Under the action of the polarized electric field, they enter the quantum well, thereby modulating the density of the high-mobility two-dimensional electron gas at the heterostructure interface. Under the action of external bias, a detectable current change is formed, realizing high sensitivity to ultraviolet radiation. In addition, the AlGaN / GaN heterostructure ultraviolet detector array 4 adopts a matrix-gated readout architecture. The control unit can activate the detectors in the designated area in sequence through the row and column addressing mechanism, and read the photoelectric response signal of each unit in a time-division manner to obtain the spatial distribution information of the ultraviolet spot. The original electrical signal output by the detector will be amplified by the micro amplifier 6, filtered by the integrated filter bank 7, and digitized by the multi-channel analog-to-digital converter 8 in sequence. This completes the time-division conditioning of the multi-channel signal, converting the analog signal into a digital signal that can be processed by the high-performance microcontroller 5, ensuring the accuracy and stability of the signal. The high-performance microcontroller 5 with built-in floating-point unit is preloaded with a lightweight multilayer perceptron artificial intelligence model trained by laboratory simulation and field measurement data. The high-performance microcontroller 5 will construct the feature vector of the spatial intensity distribution matrix of the ultraviolet spot from the digitized AlGaN / GaN heterostructure ultraviolet detector array 4 signal. The lightweight multilayer perceptron artificial intelligence model will perform forward propagation calculation and simultaneously output the discharge confidence (0-1 interval) and partial discharge stage (early, middle and late stages). At the same time, it will estimate the discharge point orientation by combining the spatial distribution information of AlGaN / GaN heterostructure ultraviolet detector array 4, and complete real-time intelligent diagnosis at the edge side without uploading the original data to the backend, which greatly improves the response efficiency and reduces bandwidth dependence. The diagnostic results are uploaded to the remote monitoring center in real time via the 4G wireless transmission module, and the local storage 9 can temporarily store the monitoring data. The device is powered by either a built-in battery or inductive power. The magnetic shell allows the probe to be flexibly attached to the inner wall of the switch cabinet, which is convenient for rapid deployment and the establishment of a distributed monitoring network.

[0033] The preferred embodiments of the present invention disclosed above are only for the purpose of illustrating the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation described herein. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can better understand and utilize the present invention.

Claims

1. A gallium nitride ultraviolet partial discharge monitoring device, comprising a magnetically attached monitoring probe, characterized in that: The magnetic monitoring probe includes a probe housing (1), a magnetic suction device (2) is provided on one side of the probe housing (1), and an optical sensing module, a signal processing module, an intelligent control and diagnosis module, a storage and communication module and a power supply module (11) are integrated in the inner cavity of the probe housing (1). The optical sensing module includes a filter lens (3) and an AlGaN / GaN heterostructure ultraviolet detector array (4). The filter lens (3) transmits only 240-280nm solar-blind ultraviolet light. The control unit of the intelligent control and diagnostic module is a high-performance microcontroller (5) with a built-in floating-point arithmetic unit. The high-performance microcontroller (5) embeds a lightweight multilayer perceptron artificial intelligence model, which can complete the identification, intensity determination and orientation estimation of partial discharge at the edge.

2. The gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The AlGaN / GaN heterostructure ultraviolet detector array (4) and the row and column gating readout circuit are integrated on the same substrate through microfabrication technology. The AlGaN / GaN heterostructure ultraviolet detector array (4) adopts a matrix gating readout architecture. The circuit board integrates the corresponding row gating and column readout paths. The control unit can activate the designated detector area in sequence and read the photoelectric response signal in time-division by the row and column addressing mechanism.

3. The gallium nitride ultraviolet partial discharge monitoring device according to claim 2, characterized in that: Each detector in the AlGaN / GaN heterostructure ultraviolet detector array (4) has independent addressing capability, and its output signal can be read individually or in combination. The circuit components integrate bias and decoupling circuits composed of resistors and capacitors.

4. The gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The AlGaN / GaN heterostructure ultraviolet detector array (4) has a polarization electric field pointing towards the GaN layer between the AlGaN layer and the GaN layer. The GaN layer has a quantum well near the AlGaN interface, which has a two-dimensional electron gas. When ultraviolet radiation is incident, the electrons at the DX center absorb photon energy and enter the conduction band. Under the action of the polarization electric field, they enter the quantum well, thereby modulating the density of the high-mobility two-dimensional electron gas at the heterostructure interface and realizing ultraviolet radiation detection.

5. The gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The signal processing module consists of a micro amplifier (6), an integrated filter bank (7), and a multi-channel analog-to-digital converter (8) cascaded together, which can perform time-division conditioning and digital conversion on the multi-channel signals output by the AlGaN / GaN heterostructure ultraviolet detector array (4).

6. The gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The lightweight multilayer perceptron artificial intelligence model includes an input layer, two hidden layers, and an output layer. The first hidden layer has 128 ReLU activated neurons, the second hidden layer has 64 ReLU activated neurons, and the output layer includes discharge confidence output neurons and partial discharge stage output neurons. The lightweight multilayer perceptron artificial intelligence model is trained using the Adam optimizer and a weighted combination loss function of mean square error and cross-entropy.

7. The gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The feature dataset of the lightweight multilayer perceptron artificial intelligence model is constructed from laboratory simulated discharge experimental data and field measured data. The instantaneous response value of the effective detection unit of the AlGaN / GaN heterostructure ultraviolet detector array (4) at a single sampling moment constitutes the ultraviolet spot spatial intensity distribution matrix as the input feature of the lightweight multilayer perceptron artificial intelligence model.

8. The gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The storage and communication module includes a memory (9) and a 4G wireless signal transmission module (10). The memory (9) is used to temporarily store local monitoring data, and the 4G wireless signal transmission module (10) can upload edge-side diagnostic results to the remote monitoring center in real time.

9. A gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The power supply module (11) supports two power supply methods: built-in battery power supply or inductive power supply.

10. A gallium nitride ultraviolet partial discharge monitoring device according to claim 1, characterized in that: The edge-side diagnostic process is as follows: After power-on, the weights of the lightweight multilayer perceptron artificial intelligence model are loaded. The high-performance microcontroller (5) collects the signals of the AlGaN / GaN heterostructure ultraviolet detector array (4) in a loop through the row and column addressing mechanism and constructs the feature vector. The input is the lightweight multilayer perceptron artificial intelligence model to complete the forward propagation calculation. After outputting the diagnostic results, they are uploaded to the remote monitoring center by the 4G wireless signal transmission module (10).

Images