GIS disconnector state intelligent monitoring method and device based on multi-dimensional sensing

Abstract

This disclosure presents an intelligent monitoring method and apparatus for the status of GIS disconnectors based on multi-dimensional sensing. One specific implementation of the method includes: performing electromagnetic partial discharge (EMI) detection on the GIS gas-insulated switchgear to be monitored, obtaining EMI signals; acquiring optical partial discharge (OPD) signals from the GIS gas-insulated switchgear based on the EMI signals, obtaining OPD signals; extracting deep features from the EMI signals and OPD signals respectively based on a disconnector status monitoring model, obtaining electromagnetic features and optical features; adaptively weighting and fusing the electromagnetic features and optical features to obtain fused features; enhancing the details of the fused features to obtain refined fused features; and determining the contact state of the disconnector contacts corresponding to the refined fused features. This implementation can ensure the accuracy of disconnector status monitoring even in scenarios with limited samples.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of partial discharge monitoring technology for GIS gas-insulated switchgear, specifically to a method and apparatus for intelligent monitoring of the status of GIS disconnect switches based on multi-dimensional sensing. Background Technology

[0002] Gas-insulated switchgear (GIS) is a core piece of equipment in power systems. Its disconnecting switch contacts are susceptible to mechanical damage from mechanical operation, electrodynamic shocks, and environmental stresses, leading to mechanical faults such as poor contact and contact loosening, often accompanied by partial discharge. Failure to diagnose and address these faults promptly can result in serious consequences, including insulation degradation, discharge breakdown, and even equipment shutdown. Therefore, research on the contact status of GIS disconnecting switch contacts is of significant engineering value and practical importance for the construction of a condition-based maintenance system for power systems and for grid risk prevention and control. Currently, the common method for condition monitoring of disconnecting switches within GIS equipment is a fault diagnosis approach that combines multiple sensors with artificial intelligence algorithms.

[0003] However, when using the above method to monitor the status of GIS disconnect switches, a common technical problem is that simple feature splicing lacks deep interaction with multimodal features, making it difficult to eliminate semantic differences between modes. Moreover, the number of field fault samples is small, making it difficult for artificial intelligence models to learn robust joint feature representations, resulting in insufficient generalization ability of the models and making it difficult to ensure the accuracy of GIS disconnect switch status monitoring in limited sample scenarios. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure propose a method and device for intelligent monitoring of the status of GIS disconnect switches based on multi-dimensional sensing, in order to solve the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide a method for intelligent monitoring of the status of a GIS disconnector based on multi-dimensional sensing. The method includes: performing electromagnetic partial discharge (EMD) detection on the GIS gas-insulated switchgear to be monitored, obtaining an EMD signal; acquiring an optical partial discharge (OPD) signal from the GIS gas-insulated switchgear based on the EMD signal, obtaining an OPD signal; performing deep feature extraction on the EMD signal and the OPD signal respectively based on a feature extraction layer included in a preset disconnector status monitoring model, obtaining electromagnetic features and optical features, wherein the disconnector status monitoring model includes: a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer; performing adaptive weighted fusion on the electromagnetic features and the optical features based on the adaptive weighted fusion layer, obtaining a fused feature; enhancing the details of the fused feature based on the detail enhancement layer, obtaining a refined fused feature; and determining the contact state of the disconnector contacts corresponding to the refined fused feature based on the classification layer.

[0007] Secondly, some embodiments of this disclosure provide a GIS disconnector switch status intelligent monitoring device based on multi-dimensional sensing. The device includes: an electromagnetic partial discharge detection unit configured to perform electromagnetic partial discharge detection on the GIS gas-insulated switchgear to be monitored, and obtain an electromagnetic partial discharge signal; an optical partial discharge signal acquisition unit configured to acquire an optical partial discharge signal from the GIS gas-insulated switchgear based on the electromagnetic partial discharge signal, and obtain an optical partial discharge signal; and a depth feature extraction unit configured to extract a feature from the electromagnetic partial discharge signal and the above-mentioned electromagnetic partial discharge signal based on a preset disconnector switch status monitoring model including a feature extraction layer. The optical partial discharge signals are subjected to depth feature extraction to obtain electromagnetic features and optical features. The disconnector switch state monitoring model includes: a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer. The adaptive weighted fusion unit is configured to perform adaptive weighted fusion of the electromagnetic features and the optical features based on the adaptive weighted fusion layer to obtain fused features. The detail enhancement unit is configured to perform detail enhancement on the fused features based on the detail enhancement layer to obtain refined fused features. The determination unit is configured to determine the contact state of the disconnector switch contacts corresponding to the refined fused features based on the classification layer.

[0008] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0009] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0010] The above-described embodiments of this disclosure have the following beneficial effects: The intelligent monitoring method for the status of GIS disconnect switches based on multi-dimensional sensing, as described in some embodiments of this disclosure, can ensure the accuracy of GIS disconnect switch status monitoring even in scenarios with limited samples. Specifically, the reason for the insufficient accuracy of GIS disconnect switch status monitoring is that simple feature stitching lacks deep interaction with multi-modal features, making it difficult to eliminate semantic differences between modalities. Furthermore, the limited number of on-site fault samples makes it difficult for artificial intelligence models to learn robust joint feature representations, resulting in insufficient generalization ability of the models. Based on this, the intelligent monitoring method for the status of GIS disconnect switches based on multi-dimensional sensing, as described in some embodiments of this disclosure, firstly performs electromagnetic partial discharge detection on the GIS gas-insulated switchgear to be monitored, obtaining electromagnetic partial discharge signals. Thus, high-frequency electromagnetic transient features in the GIS equipment can be captured through electromagnetic partial discharge detection. Secondly, based on the electromagnetic partial discharge signals, optical partial discharge signals are acquired from the GIS gas-insulated switchgear, obtaining optical partial discharge signals. Since optical partial discharge signals have strong anti-interference capabilities and complement electromagnetic signals, they can provide high-quality multi-modal basic information for subsequent feature learning. Subsequently, based on the feature extraction layer of the pre-defined disconnector switch status monitoring model, deep feature extraction is performed on the aforementioned electromagnetic partial discharge (EMD) and optical partial discharge (OPD) signals to obtain electromagnetic and optical features. The disconnector switch status monitoring model includes a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer. The deep network automatically learns the high-dimensional features of the EMD and OPD signals, avoiding the subjectivity and limitations of manually designed features and fully exploring the latent features in the electromagnetic and optical signals that are strongly correlated with the contact state of the contacts. Next, based on the adaptive weighted fusion layer, the electromagnetic and optical features are adaptively weighted and fused to obtain fused features. Through data-driven adaptive weight allocation, deep interaction and semantic alignment of multimodal features are achieved, avoiding the insufficient representational power caused by simple feature splicing. Then, based on the detail enhancement layer, the fused features are enhanced to obtain refined fused features. By strengthening important fine-grained details and suppressing background noise and redundant information, the model can still learn robust joint feature representations even in small sample scenarios. Finally, based on the above classification layer, the contact state of the disconnector switch corresponding to the refined fusion features is determined. This enables accurate mapping from multimodal raw signals to device states, effectively solving the problems of poor classification performance and weak adaptability caused by insufficient joint feature learning in existing technologies, thereby improving the model's recognition accuracy and generalization ability. Attached Figure Description

[0011] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0012] Figure 1 This is a schematic diagram of an application scenario of the intelligent monitoring method for the status of GIS disconnect switches based on multi-dimensional sensing, which is one of the embodiments of this disclosure.

[0013] Figure 2 This is a flowchart of some embodiments of the intelligent monitoring method for the status of GIS disconnect switches based on multi-dimensional sensing according to the present disclosure;

[0014] Figure 3 This is a schematic diagram of the disconnector switch status prediction model according to the GIS disconnector switch status intelligent monitoring method based on multi-dimensional sensing disclosed herein;

[0015] Figure 4 This is a structural schematic diagram of some embodiments of the GIS disconnector status intelligent monitoring device based on multi-dimensional sensing according to the present disclosure;

[0016] Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0017] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0018] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0019] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0020] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0021] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0022] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] Figure 1 This is a schematic diagram of an application scenario of the intelligent monitoring method for the status of GIS disconnect switches based on multi-dimensional sensing, which is one of the embodiments of this disclosure.

[0024] exist Figure 1 In the application scenario, firstly, the computing device 101 can perform electromagnetic partial discharge detection on the GIS gas-insulated switchgear 102 to be monitored, obtaining an electromagnetic partial discharge signal 103. Next, the computing device 101 can acquire optical partial discharge signals from the GIS gas-insulated switchgear 102 based on the electromagnetic partial discharge signal 103, obtaining an optical partial discharge signal 104. Then, the electromagnetic partial discharge signal 103 and the optical partial discharge signal 104 are respectively input into the feature extraction layer 1051 of the disconnector switch status monitoring model 105 to obtain electromagnetic features 106 and optical features 107. The disconnector switch status monitoring model 105 includes: a feature extraction layer 1051, an adaptive weighted fusion layer 1052, a detail enhancement layer 1053, and a classification layer 1054. Afterwards, the computing device 101 can input the electromagnetic features 106 and the optical features 107 into the adaptive weighted fusion layer 1052 to obtain fused features 108. Then, the computing device 101 can input the above-mentioned fusion feature 108 into the detail enhancement layer 1053 to obtain the refined fusion feature 109. Finally, the computing device 101 can input the above-mentioned refined fusion feature 109 into the classification layer 1054 to obtain the contact state 110 of the disconnecting switch contact.

[0025] It should be noted that the aforementioned computing device 101 can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed in the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here. It should be understood that... Figure 1 The number of computing devices in the system can be arbitrary, depending on the implementation requirements.

[0026] Continue to refer to Figure 2The diagram illustrates a flow 200 of some embodiments of the intelligent monitoring method for the status of GIS disconnect switches based on multi-dimensional sensing according to this disclosure. This intelligent monitoring method for the status of GIS disconnect switches based on multi-dimensional sensing includes the following steps:

[0027] Step 201: Perform electromagnetic partial discharge detection on the GIS gas-insulated switchgear to be monitored to obtain electromagnetic partial discharge signals.

[0028] In some embodiments, the executing entity of the GIS disconnector status intelligent monitoring method based on multidimensional sensing (e.g., Figure 1 The computing device 101 shown can perform electromagnetic partial discharge detection on the GIS gas-insulated switchgear to be monitored, and obtain electromagnetic partial discharge signals. The GIS gas-insulated switchgear can be a complete set of power distribution equipment that encloses high-voltage electrical components such as circuit breakers, disconnectors, grounding switches, and busbars in a grounded metal casing, using sulfur hexafluoride gas as the insulating medium. The electromagnetic partial discharge signal can be a digital signal characterizing partial discharge information within the GIS equipment over a period of time. The duration of the electromagnetic partial discharge signal can be 1 second. In practice, firstly, an ultra-high frequency (UHF) sensor placed outside or inside the GIS gas-insulated switchgear can be used to capture the electromagnetic wave signal in the GIS gas-insulated switchgear and convert it into an analog electrical signal. Secondly, an analog-to-digital converter can be used to convert the analog electrical signal into a digital signal as the electromagnetic partial discharge signal.

[0029] In practice, a common technical challenge in detecting electromagnetic partial discharge (EMD) in GIS gas-insulated switchgear is that ultra-high frequency (UHF) sensors capture signals with a wide bandwidth that also includes background noise and partial discharge characteristics. This can easily lead to information redundancy, resulting in significant consumption of data storage, transmission, and computing resources, and reducing the real-time performance of signal processing. Therefore, the following solution is proposed.

[0030] Optionally, the aforementioned implementing entity performs electromagnetic partial discharge detection on the GIS gas-insulated switchgear to be monitored to obtain electromagnetic partial discharge signals, which may include the following steps:

[0031] The first step involves using an ultra-high frequency (UHF) sensor to detect electromagnetic partial discharge (EMD) in the aforementioned GIS gas-insulated switchgear, thereby obtaining the raw UHF signal. The UHF sensor can be a microstrip antenna sensor. In practice, the UHF sensor, positioned externally or internally within the GIS gas-insulated switchgear, can be used to detect EMD and capture the UHF electromagnetic wave signal. This UHF electromagnetic wave signal is then converted into an analog electrical signal, which serves as the raw UHF signal.

[0032] The second step involves bandpass filtering the original UHF signal to obtain a filtered UHF signal. This bandpass filter can be a bandpass filter operating in the 0.5 GHz to 3 GHz frequency range. The input of the bandpass filter can be electrically connected to the output of the UHF sensor. In practice, the original UHF signal can be input into the bandpass filter for filtering to obtain the filtered UHF signal.

[0033] In practice, the aforementioned 0.5GHz to 3GHz frequency band can effectively cover the typical UHF radiation frequency band generated by partial discharge in GIS equipment. Therefore, the aforementioned bandpass filter can filter out electromagnetic interference and background noise outside the frequency band of the original UHF signal, thereby retaining the effective partial discharge signal.

[0034] The third step involves amplifying the filtered UHF signal using a low-noise amplifier (LNA) to obtain an enhanced UHF signal. The input of the LNA can be electrically connected to the output of the bandpass filter. In practice, the filtered UHF signal can be input into the LNA to obtain the enhanced UHF signal. This increases the amplitude of weak partial discharge signals in the filtered UHF signal, ensuring the signal-to-noise ratio and detection accuracy in subsequent signal processing and preventing weak signals from being overwhelmed by noise.

[0035] The fourth step involves using an envelope detection circuit to perform envelope detection on the enhanced UHF signal to obtain a low-frequency envelope signal. This envelope detection circuit can be a diode envelope detector. The input terminal of the envelope detection circuit can be electrically connected to the output terminal of the low-noise amplifier. In practice, the enhanced UHF signal can be input into the envelope detection circuit to obtain the low-frequency envelope signal.

[0036] In practice, because the bandwidth of the original electromagnetic partial discharge signal is too wide, envelope detection can convert the original signal into a low-frequency envelope signal that characterizes the change in discharge intensity, thereby reducing the computational complexity of signal processing and improving the real-time performance of GIS disconnector status monitoring.

[0037] The fifth step involves applying a second-order resistor-capacitor low-pass filter to the aforementioned low-frequency envelope signal to obtain a smoothed envelope signal. The input of this second-order resistor-capacitor low-pass filter can be electrically connected to the output of the envelope detection circuit. In practice, the low-frequency envelope signal can be input into the second-order resistor-capacitor low-pass filter to obtain the smoothed envelope signal. Here, the second-order resistor-capacitor low-pass filter removes residual high-frequency glitches and detection ripples from the envelope signal, further smoothing the envelope waveform and improving the accuracy of subsequent pulse discrimination.

[0038] Step 6: Synchronously input the smoothed envelope signal and the preset threshold voltage into a high-speed voltage comparator to obtain a partial discharge pulse sequence. The threshold voltage can be 10mV, and is not specifically limited here. The high-speed voltage comparator can be a high-speed CMOS voltage comparator or a high-speed comparator chip. The high-speed voltage comparator can include a signal input terminal and a reference input terminal. The signal input terminal can be electrically connected to the second-order resistor-capacitor low-pass filter. The reference input terminal can be electrically connected to the power supply that generates the threshold voltage. The partial discharge pulse sequence can be a pulse sequence consisting of alternating low and high levels. In practice, the smoothed envelope signal can be input into the signal input terminal of the high-speed voltage comparator, and the threshold voltage can be input into the reference input terminal of the high-speed voltage comparator to obtain the partial discharge pulse sequence.

[0039] Here, when the amplitude of the smooth envelope signal exceeds the threshold voltage, the high-speed voltage comparator can output an effective pulse signal, thereby separating the pulse sequence characterizing the partial discharge event from the continuous smooth envelope signal.

[0040] Step 7: Using an analog-to-digital converter (ADC), the smooth envelope signal is sampled in a triggered manner according to the partial discharge pulse sequence to obtain the electromagnetic partial discharge signal. The ADC may include a trigger control terminal and an analog signal input terminal. The trigger control terminal can be a trigger pin on the ADC. It can be electrically connected to the output of the high-speed voltage comparator. The analog signal input terminal can be electrically connected to the output of the second-order resistor-capacitor low-pass filter. In practice, the partial discharge pulse sequence can be used as a trigger signal to control the ADC to sample the smooth envelope signal only when a partial discharge pulse occurs, thus obtaining the electromagnetic partial discharge signal. Specifically, when the partial discharge pulse sequence is high, the ADC is triggered to sample the smooth envelope signal; when the partial discharge pulse sequence is low, the ADC remains in a waiting state and does not perform any sampling operation.

[0041] Here, hardware-based triggered sampling of the smooth envelope signal not only reduces the data storage volume, transmission pressure, and software computation load of the electromagnetic partial discharge signal, but also ensures that the acquired signal retains the key characteristic information of the partial discharge.

[0042] To avoid excessive consumption of data storage, transmission, and computing resources, and to improve the real-time performance of signal processing, the following steps are taken: First, bandpass filtering effectively removes out-of-band noise interference from the original electromagnetic signal. Next, a low-noise amplifier amplifies the filtered signal, increasing the amplitude of weak partial discharge signals without significantly introducing additional noise, preventing these signals from being overwhelmed by noise and improving the detection capability for minor partial discharges. Second, an envelope detection circuit converts the ultra-high frequency signal into a low-frequency envelope signal, reducing signal bandwidth and subsequent processing complexity, significantly reducing computing resource consumption, and improving the real-time performance and response speed of the GIS disconnector status online monitoring. Then, a second-order RC low-pass filter is used to perform secondary filtering on the envelope signal, effectively removing detection ripple and high-frequency glitches, resulting in a smoother envelope waveform. Finally, a high-speed voltage comparator generates a partial discharge pulse sequence corresponding to the smooth envelope waveform, which serves as a hardware trigger signal. This controls the ADC to sample only when a discharge occurs, and to enter a sleep state when there is no discharge, reducing invalid data acquisition at the source. Compared to software processing, this hardware sampling method can significantly reduce data storage, data transmission pressure, and computational load. While ensuring the complete preservation of key partial discharge characteristic information, it reduces sampling costs and power consumption, making it particularly suitable for long-term online status monitoring scenarios of GIS equipment.

[0043] Step 202: Based on the electromagnetic partial discharge signal, optical partial discharge signal acquisition is performed on the GIS gas-insulated switchgear to obtain the optical partial discharge signal.

[0044] In some embodiments, the aforementioned execution entity can acquire optical partial discharge signals from the aforementioned GIS gas-insulated switchgear based on the aforementioned electromagnetic partial discharge signals, thereby obtaining optical partial discharge signals. The aforementioned optical partial discharge signals can be digital signals representing partial discharge information within the aforementioned GIS gas-insulated switchgear over a period of time. The duration of the aforementioned optical partial discharge signals can be 1 second. In practice, while acquiring the aforementioned electromagnetic partial discharge signals, optical signals within the aforementioned GIS gas-insulated switchgear can be simultaneously acquired using optical sensors arranged on the observation window of the aforementioned GIS gas-insulated switchgear, and the optical signals can be converted into analog electrical signals. Then, an analog-to-digital converter can be used to convert the analog electrical signals into digital signals, which serve as the optical partial discharge signals. The aforementioned optical sensors may include, but are not limited to, photomultiplier tubes and silicon photodiodes.

[0045] Optionally, the aforementioned execution entity may acquire optical partial discharge signals from the aforementioned GIS gas-insulated switchgear based on the aforementioned electromagnetic partial discharge signals, thereby obtaining optical partial discharge signals, which may include the following steps:

[0046] The first step is to determine the estimated energy level of the partial discharge inside the aforementioned GIS gas-insulated switchgear based on the signal strength of the electromagnetic partial discharge signal. The signal strength of the electromagnetic partial discharge signal can be the peak voltage of the corresponding digital signal. In practice, the estimated energy level corresponding to the signal strength can be determined according to a preset signal strength-energy level mapping table. This mapping table characterizes the mapping relationship between signal strength and energy level. For example, a signal strength less than or equal to 0.5V indicates a low energy level; a signal strength greater than 0.5V and less than or equal to 1.5V indicates a medium energy level; and a signal strength greater than 1.5V indicates a high energy level.

[0047] The second step involves determining the sensor's operating mode based on the assumption that the estimated energy level meets preset energy conditions. The energy conditions can be either a medium or high energy level. The optical sensor used for localized optical signal acquisition can be an optical sensor array. This array may include equally spaced optical sensors. The first operating mode allows each optical sensor in the array to operate with low sensitivity. This first operating mode can refer to adjusting the operating parameters of the optical sensor to reduce its response to incident photons, thus adapting to strong light signal scenarios. For example, when the optical sensor is a photomultiplier tube (PMT), its sensitivity is reduced by lowering its operating voltage. In practice, when the estimated energy level is medium or high, the first operating mode can be determined as the sensor's operating mode.

[0048] Thirdly, in response to the determination that the estimated energy level does not meet the energy conditions, a preset second operating mode is established as the sensor operating mode. This second operating mode allows each optical sensor in the optical sensor array to operate with high sensitivity. For example, when the optical sensor is a photomultiplier tube (PMT), its sensitivity is increased by raising its operating voltage. In practice, when the estimated energy level is low, the second operating mode can be established as the sensor operating mode.

[0049] The fourth step involves acquiring optical signals from the GIS gas-insulated switchgear using an optical sensor array deployed on the switchgear, based on the sensor's operating mode, to obtain optical partial discharge signals. The optical sensor array corresponds to a spatial location information set. This spatial location information can be coordinates representing the position of the optical sensors.

[0050] As an example, assume that the aforementioned optical sensors are arranged on the front window of the GIS test chamber, located in the four directions of up, down, left, and right. Each sensor has specific spatial coordinates, such as (0, 1), (0, -1), (-1, 0), and (1, 0). When the sensors are in the first operating mode, the operating voltage of each optical sensor can be reduced, or a neutral density filter can be added at the front end to attenuate the incident light intensity and prevent the sensor from entering the saturation region. When the sensors are in the second operating mode, the operating voltage of the sensors can be increased, or single-photon counting technology can be used, outputting a pulse for each detected photon to ensure that no weak signals are missed. Optical partial discharge (OPD) signals are collected by each optical sensor in the aforementioned optical sensor array to obtain an OPD signal set, which serves as the optical partial discharge signal. The OPD signals in the aforementioned OPD signal set can correspond one-to-one with the spatial position information in the aforementioned spatial position information set.

[0051] Step 203: Based on the feature extraction layer included in the preset disconnector switch status monitoring model, deep feature extraction is performed on the electromagnetic partial discharge signal and the optical partial discharge signal respectively to obtain electromagnetic features and optical features.

[0052] In some embodiments, the aforementioned execution entity can perform deep feature extraction on the electromagnetic partial discharge signal and the optical partial discharge signal respectively, based on the feature extraction layer included in the preset disconnector switch status monitoring model, to obtain electromagnetic features and optical features. The disconnector switch status monitoring model can be a deep learning network model for identifying the contact state of GIS disconnector switches. The disconnector switch status monitoring model can include: a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer. The feature extraction layer can be a network layer for extracting features from the original input signal. The feature extraction layer can include two parallel feature extraction networks. The two parallel feature extraction networks have the same structure but different parameters. Both feature extraction networks can be CSPNEXt networks. The adaptive weighted fusion layer can be a network layer for fusing the electromagnetic features and optical features output from the feature extraction layer. The adaptive weighted fusion layer can be a Feature Weighted Fusion Module (FWFM). The detail enhancement layer can be a network layer for enhancing the details of the fused features output from the adaptive weighted fusion layer. The aforementioned detail enhancement layer can be a Feature Refinement Module (FRM). The aforementioned classification layer can be used to map the features output by the detail enhancement layer to the contact categories of the disconnector switch contacts and output the corresponding probability distribution. These contact categories can include: closed state, open state, and abnormal contact state. The aforementioned classification layer can include, but is not limited to: at least one fully connected layer, an activation function (e.g., Softmax), and an output layer. The aforementioned output layer can be used to output the contact category of the disconnector switch contact with the highest probability.

[0053] In practice, the electromagnetic partial discharge signal and the optical partial discharge signal can be input into the two parallel feature extraction networks included in the feature extraction layer to obtain electromagnetic features and optical features.

[0054] Step 204: Based on the adaptive weighted fusion layer, the electromagnetic features and optical features are adaptively weighted and fused to obtain the fused features.

[0055] In some embodiments, the execution entity may perform adaptive weighted fusion of the electromagnetic features and the optical features based on the adaptive weighted fusion layer to obtain fused features. In practice, the electromagnetic features and the optical features can be input into the adaptive weighted fusion layer to obtain fused features.

[0056] Optionally, the execution entity performs adaptive weighted fusion of the electromagnetic features and the optical features based on the adaptive weighted fusion layer to obtain fused features, which may include the following steps:

[0057] The first step involves adaptive attention fusion of the aforementioned electromagnetic and optical features to obtain initial fused features. In practice, the initial fused features can be obtained by fusing the electromagnetic and optical features based on a preset adaptive attention mechanism. This adaptive attention mechanism can be a deformable attention module or a gated attention fusion mechanism.

[0058] The second step involves convolving the initial fused features to obtain two-channel convolutional features. In practice, two convolutional kernels can be used to convolve the initial fused features separately, resulting in a first convolutional feature and a second convolutional feature. The first convolutional feature is then concatenated with the second convolutional feature along the channel dimension to obtain the two-channel convolutional features.

[0059] As an example, assume the initial fused feature is a 512×512×1 dimensional feature. Here, 512 represents the width and height of the feature, and 1 represents the number of channels. Assume the two convolutional kernels include a first convolutional kernel and a second convolutional kernel. Both the first and second convolutional kernels are 3×3 dimensional, but with different parameters. The first convolutional kernel is used to convolve the initial fused feature with a stride of 2 and padding of 1, resulting in a 256×256×1 dimensional first convolutional feature. Here, 256 represents the width and height of the first convolutional feature, and 1 represents the number of channels. The second convolutional kernel is then used to convolve the initial fused feature with a stride of 2 and padding of 1, resulting in a 256×256×1 dimensional second convolutional feature. The concatenated two-channel convolutional feature can then have a dimension of 256×256×2.

[0060] The third step is to normalize the two-channel convolutional features to obtain normalized features. In practice, for the two-channel convolutional features, a preset activation function can be applied along the channel dimension to achieve channel-dimensional normalization and obtain normalized features. The preset activation function can be the Softmax function or the Sigmoid function.

[0061] The fourth step is to decompose the normalized features to obtain the first adaptive weight and the second adaptive weight. In practice, the normalized features can be decomposed along the channel dimension to obtain the first adaptive weight and the second adaptive weight.

[0062] The fifth step is to determine the weighted electromagnetic feature by multiplying the first adaptive weight and the electromagnetic feature mentioned above.

[0063] The sixth step is to determine the weighted optical feature by multiplying the second adaptive weight and the optical feature mentioned above.

[0064] Step 7: Add the weighted electromagnetic features and weighted optical features element-wise to obtain the fused features. In practice, the weighted electromagnetic features and weighted optical features can be added element-wise to obtain the fused features. The adaptive weighted fusion process corresponding to steps 1 to 7 above can be described as follows: Figure 3 As shown.

[0065] Step 205: Based on the detail enhancement layer, perform detail enhancement on the fused features to obtain refined fused features.

[0066] In some embodiments, the execution entity can enhance the details of the fused features based on the detail enhancement layer to obtain refined fused features. In practice, the fused features can be input into the detail enhancement layer to obtain refined fused features.

[0067] Optionally, the execution entity may perform detail enhancement on the fused features based on the aforementioned detail enhancement layer to obtain refined fused features, which may include the following steps:

[0068] The first step is to extract channel attention features from the above fused features. In practice, channel attention features can be extracted from the above fused features using a channel attention mechanism (CAM).

[0069] Optionally, the execution entity may extract channel attention features from the fused features to obtain channel attention features, which may include the following steps:

[0070] The first sub-step involves max pooling the aforementioned fused features to obtain max pooled features. In practice, firstly, for each channel of the fused features, the maximum value of all elements at all spatial locations within that channel can be used as the max pooling value. Then, the determined max pooling values ​​can be combined in channel order to form the max pooled features.

[0071] The second sub-step involves performing average pooling on the aforementioned fused features to obtain average pooled features. In practice, firstly, for each channel included in the aforementioned fused features, the average value of the elements at all spatial locations of the channel can be determined as the average pooling value. Then, the determined average pooling values ​​are combined in channel order to form the average pooling feature. Both the aforementioned average pooling feature and the aforementioned max pooling feature can have a dimension of 1×1×C. Here, 1 represents the width and height of the feature, and C represents the number of channels in the feature.

[0072] The third sub-step involves performing convolution processing on the aforementioned max-pooling features and average-pooling features respectively to obtain convolutional max-pooling features and convolutional average-pooling features. In practice, firstly, two pre-defined convolutional layers can be used to perform convolution processing on the aforementioned max-pooling features and average-pooling features respectively to obtain initial max-pooling features and initial average-pooling features. The convolutional kernel of each of the two convolutional layers can be 1×1. Next, the aforementioned initial average-pooling features and initial max-pooling features can be input into the ReLU activation function respectively to obtain convolutional max-pooling features and convolutional average-pooling features.

[0073] The fourth sub-step involves adding the above convolutional max pooling features and the above convolutional average pooling features element-wise to obtain the channel attention features.

[0074] The second step is to extract spatial attention features from the fused features. In practice, the fused features can be input into the Softmax activation function to normalize the element values ​​of each channel at each spatial location in the fused features, thus obtaining the spatial attention features.

[0075] The third step is to determine the refined electromagnetic features by multiplying the aforementioned channel attention features with the aforementioned weighted electromagnetic features.

[0076] The fourth step is to determine the product between the aforementioned spatial attention features and the aforementioned weighted optical features as the refined optical features.

[0077] The fifth step is to add the above-mentioned refined electromagnetic features and refined optical features element by element to obtain the refined fused features.

[0078] In practice, deep correlations between features of different modalities can be learned through channel attention feature extraction and spatial attention feature extraction. Furthermore, through weighted fusion, the model can focus on core features and remove background noise. This enables the model to learn robust feature representations even in scenarios with limited sample sizes, addressing the pain point of deep learning models relying on large amounts of data. This allows it to work accurately even in industrial scenarios like power systems where obtaining a large number of fault samples is difficult.

[0079] Step 206: Based on the classification layer, determine the contact state of the disconnector contacts corresponding to the refined fusion features.

[0080] In some embodiments, the execution entity can determine the contact state of the disconnecting switch corresponding to the refined fusion features based on the classification layer. In practice, the refined fusion features can be input into the classification layer to obtain the contact state of the disconnecting switch.

[0081] Optionally, the execution entity, based on the above classification layer, determines the contact state of the disconnecting switch corresponding to the above refined fusion features, which may include the following steps:

[0082] The first step is to determine the contact state probability distribution corresponding to the refined fusion features mentioned above. This contact state probability distribution can include the contact categories of each disconnector contact and their corresponding probability values. For example, the contact state probability distribution could be: "Closed state: 0.62, Opened state: 0.07, Abnormal contact state: 0.31". In practice, firstly, the refined fusion features are input into the fully connected layer in the classification layer to map the high-dimensional refined fusion features into a low-dimensional vector through linear transformation. Finally, the low-dimensional vector can be input into the Softmax activation function for probability normalization to obtain the contact state probability distribution.

[0083] The second step is to generate the contact state of the disconnecting switch contacts based on the above contact state probability distribution. In practice, the output layer included in the above classification layer can determine the contact category of the disconnecting switch contact with the highest probability value in the above contact state probability distribution as the contact state of the disconnecting switch contact.

[0084] Optionally, before performing deep feature extraction on the electromagnetic partial discharge signal and the optical partial discharge signal respectively based on the feature extraction layer of the preset disconnector state monitoring model according to the electromagnetic partial discharge signal to obtain electromagnetic features and optical features, the execution entity may also perform the following steps:

[0085] The first step is to obtain the disconnector switch status dataset and the initial disconnector switch status monitoring model. The disconnector switch status data in the dataset includes: sample electromagnetic partial discharge signals, sample optical partial discharge signals, and disconnector switch status labels. These labels can include: closed state, open state, and abnormal contact state. The initial disconnector switch status monitoring model has the same structure as the previous disconnector switch status monitoring model, but different parameters. This initial model can include: an initial feature extraction layer, an initial adaptive weighted fusion layer, an initial detail enhancement layer, and an initial classification layer.

[0086] The second step involves training the initial disconnector status monitoring model based on the status data of each disconnector in the aforementioned disconnector status dataset, using the following steps:

[0087] The first sub-step involves sequentially inputting the sample electromagnetic partial discharge (EMI) signal and sample optical partial discharge (OPD) signal included in the aforementioned disconnector state data into the initial feature extraction layer, initial adaptive weighted fusion layer, and initial detail enhancement layer of the initial disconnector state monitoring model to generate refined sample fusion features. In practice, firstly, the sample EMI and OPD signals included in the aforementioned disconnector state data can be input into the initial feature extraction layer of the initial disconnector state monitoring model to obtain sample electromagnetic features and sample optical features. Next, the sample electromagnetic features and sample optical features can be input into the initial adaptive weighted fusion layer to obtain sample fusion features. Finally, the sample fusion features can be input into the initial detail enhancement layer to obtain refined sample fusion features.

[0088] The second sub-step involves inputting the refined and fused features of the aforementioned samples into the initial classification layer of the initial disconnector state monitoring model to generate a predicted disconnector state. This predicted disconnector state can be the contact category of the disconnector contact and its corresponding confidence level. For example, the predicted disconnector state could be: "Closed state, 0.78".

[0089] The third sub-step involves determining the cross-entropy loss value between the predicted disconnector state and the corresponding disconnector state label. In practice, this cross-entropy loss value can be determined based on a preset cross-entropy loss function.

[0090] The fourth sub-step involves determining the center loss value between the refined fusion features of the samples and the learnable class center features corresponding to the disconnector switch status labels. These learnable class center features can be trainable feature vectors assigned to each disconnector contact category during model training. The closing state, the opening state, and the contact anomaly state can each correspond to a learnable class center feature. The learnable class center feature represents the class center of its corresponding category in the feature space. In practice, the center loss value between the refined fusion features of the samples and the learnable class center features corresponding to the disconnector switch status labels can be determined based on a preset center loss function.

[0091] The fifth sub-step involves determining the total loss value by the weighted sum of the aforementioned center loss value and the aforementioned cross-entropy loss value. In practice, the total loss value can be obtained by adding the cross-entropy loss value to the product of the center loss value and the preset weights. The preset weights can be hyperparameters, for example, 0.1. No specific limitation is made here.

[0092] The sixth sub-step involves updating the initial disconnector state monitoring model based on the total loss value, resulting in an updated disconnector state monitoring model. This updated model is then used as the initial disconnector state monitoring model, and the training steps are executed again. In practice, the backpropagation algorithm can be used to update the initial disconnector state monitoring model based on the total loss value, and a moving average method can be used to update the center features of each learnable category. Afterward, the updated disconnector state monitoring model can be used as the initial disconnector state monitoring model, and the training steps are executed again until all disconnector state data in the disconnector state dataset have completed the training process.

[0093] The third step involves determining the updated disconnector state monitoring model corresponding to the total loss value as the disconnector state monitoring model if the final total loss value is less than or equal to a preset loss threshold. This loss threshold can be a numerical value, such as 0.2. No specific limitation is made here. In practice, when the final total loss value obtained in the above training steps is less than or equal to the loss threshold, the updated disconnector state monitoring model corresponding to the total loss value can be determined as the disconnector monitoring model. When the final total loss value is greater than the loss threshold, the finally obtained updated disconnector state monitoring model can be used as the initial disconnector model, and the above training steps can be executed again until the total loss value is less than or equal to the loss threshold.

[0094] In practice, a common technical challenge in monitoring the status of GIS disconnect switches is the difficulty in accurately predicting the location of defects in gas-insulated switchgear. Therefore, the following solution is proposed.

[0095] Optionally, the aforementioned implementing entity may also perform the following steps:

[0096] The first step is to perform a light intensity ratio analysis on the optical partial discharge signals based on the spatial location information set corresponding to the aforementioned optical sensor array, thereby obtaining the light intensity ratio coordinates. In practice, firstly, for each optical sensor in the aforementioned optical sensor array, the maximum amplitude of the pulse in the optical partial discharge signal acquired by the optical sensor can be determined as the target amplitude. The time interval between the rising and falling edges of the pulse can be determined as the pulse duration. Secondly, the ratio of the target amplitude to the pulse duration can be determined as the optical signal intensity. The optical signal intensity can be characterized by monitoring the slope of the signal pulse waveform. The optical signal intensity can correspond to a location label. For example, when the optical sensor is located above the array, the location label can be "above". Next, for each determined optical signal intensity, the sum of the optical signal intensity labeled "above" and the optical signal intensity labeled "below" can be determined as the vertical light intensity. Finally, the ratio of the "above" optical signal intensity to the vertical light intensity can be determined as the vertical light intensity ratio. For each determined optical signal intensity, the sum of the "right" optical signal intensity and the "left" optical signal intensity can be defined as the horizontal light intensity. Next, the ratio of the "right" optical signal intensity to the aforementioned horizontal light intensity can be defined as the horizontal light intensity percentage. Then, the horizontal light intensity percentage can be used as the abscissa, and the vertical light intensity percentage as the ordinate, to generate a light intensity percentage coordinate system.

[0097] In practice, a light intensity ratio coordinate system can be generated by using the horizontal light intensity ratio as the x-axis and the vertical light intensity ratio as the y-axis. For the same defect type and the same location, the coordinate points obtained from multiple measurements will cluster within a stable region of the light intensity ratio coordinate system; however, for defects at different locations, the coordinate points will be distributed in different regions.

[0098] The second step involves matching the aforementioned light intensity ratio coordinates with a pre-set defect location coordinate database to obtain the estimated spatial orientation of the defect. This defect location coordinate database can be a database representing the mapping relationship between defect locations and historical light intensity ratio coordinates. The aforementioned defect location can characterize the orientation of the defect within the aforementioned gas-insulated switchgear. For example, on the basin-type insulator at the bottom of the GIS test chamber, a location is set every 45° circumferentially. The aforementioned defect locations can include: 0° (facing the observation window), 45°, 90°, 135°, 180° (facing away from the observation window). The aforementioned defect location coordinate database can include: defect location: 0°, historical light intensity ratio coordinates: (0.58, 0.53); defect location: 45°, historical light intensity ratio coordinates: (0.59, 0.50).

[0099] In practice, for the aforementioned light intensity proportion coordinates, the Euclidean distance between these coordinates and each historical light intensity proportion coordinate in the defect location coordinate database is determined, generating a set of Euclidean distances. The historical light intensity proportion coordinate corresponding to the smallest Euclidean distance in the set is determined as the target historical light intensity proportion coordinate, and the defect location corresponding to the target historical light intensity proportion coordinate is determined as the preset defect spatial orientation. The preset defect orientation can be an angle value, for example, 45°.

[0100] The third step is to generate a first correction coefficient and a second correction coefficient based on the estimated spatial orientation of the defect. In practice, the first and second correction coefficients corresponding to the estimated spatial orientation of the defect can be determined according to a preset mapping table of spatial orientation and correction coefficients. The mapping table of spatial orientation and correction coefficients can be a one-to-one correspondence table between spatial orientation and the first and second correction coefficients. For example, the mapping table of spatial orientation and correction coefficients may include: "Spatial orientation: 20°, first correction coefficient: 0.15, second correction coefficient: 1.85; Spatial orientation: 45°, first correction coefficient: 0.23, second correction coefficient: 1.77; Spatial orientation: 160°, first correction coefficient: 1.94, second correction coefficient: 0.06".

[0101] In practice, when the defect is located on the front of the sensor (azimuth angle of 0°), the optical signal reception intensity is high, and the weight of optical features can be increased. When the defect is located on the back of the sensor (azimuth angle of 180°), the optical signal is mainly reflected light, which is attenuated significantly. In this case, the weight of electromagnetic features should be increased.

[0102] The fourth step involves correcting the first adaptive weight and the second adaptive weight using the first correction coefficient and the second correction coefficient, resulting in a first corrected weight and a second corrected weight. In practice, the product of the first adaptive weight and the first correction coefficient can be used to determine the first corrected weight, and the product of the second adaptive weight and the second correction coefficient can be used to determine the second corrected weight.

[0103] Fifth, based on the first and second correction weights, the electromagnetic and optical features are fused to obtain the corrected fused feature. In practice, the product of the first correction weight and the electromagnetic feature can be used to determine the corrected electromagnetic feature. The product of the second correction weight and the optical feature can be used to determine the corrected optical feature. Finally, the corrected electromagnetic and optical features can be added element-wise to obtain the corrected fused feature.

[0104] Step 6: Based on the preset defect location prediction layer, determine the defect location corresponding to the modified fusion features. The defect location prediction layer can be a network layer used to predict the defect location. This layer can include, but is not limited to, fully connected layers and activation functions (e.g., softmax). The defect location can be the relative position between the predicted defect and the sensor. This location can include azimuth and distance. For example, the defect location could be: "Angle: 45°, Distance: 6cm". In practice, the modified fusion features can be input into the preset defect location prediction layer to obtain the defect location.

[0105] Specifically, each disconnector status data in the aforementioned disconnector status dataset can also correspond to a defect location label. The aforementioned disconnector status prediction model can also include an initial defect location prediction layer. A schematic diagram of the aforementioned disconnector status prediction model can be shown below. Figure 3 As shown. For each disconnector state data in the aforementioned disconnector state dataset, the sample electromagnetic partial discharge signal and sample optical partial discharge signal from the disconnector state data can be used as inputs to the disconnector state prediction model, and the defect location label corresponding to the disconnector state data can be used as the expected output of the disconnector state prediction model to train the initial defect location prediction layer in the disconnector state prediction model, thus obtaining the defect location prediction layer. During the training process, the parameters of the feature extraction layer and the adaptive weighted fusion layer in the disconnector state prediction model can be kept fixed, and only the parameters of the defect location prediction layer are updated.

[0106] To improve the accuracy of defect location prediction in GIS gas-insulated switchgear, firstly, an intensity ratio map is generated based on the position of each optical sensor in the optical sensor array and the intensity of the optical signals collected by each sensor. This provides an intuitive and structured data foundation for subsequent defect spatial analysis. Secondly, using an existing defect location coordinate trajectory database, the defect orientation corresponding to the coordinate points in the intensity ratio map is initially determined. Next, based on the defect orientation, the weights of the optical and electromagnetic signals are adjusted to generate a corrected fusion feature that includes physical location information. This allows for dynamic adjustment of the adaptive weighted fusion ratio based on the signal propagation path and attenuation, improving the accuracy of defect location prediction. Finally, a pre-trained defect location prediction layer is used to learn the complex mapping relationship between the corrected fusion feature and the defect location, thereby improving the accuracy of defect location prediction.

[0107] Further reference Figure 4As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a GIS disconnector status intelligent monitoring device based on multi-dimensional sensing. These device embodiments are similar to... Figure 2 Corresponding to the method embodiments shown, this GIS disconnector status intelligent monitoring device based on multidimensional sensing can be specifically applied to various electronic devices.

[0108] like Figure 4 As shown, some embodiments of the GIS disconnect switch status intelligent monitoring device 400 based on multi-dimensional sensing include: an electromagnetic partial discharge detection unit 401, an optical partial discharge signal acquisition unit 402, a depth feature extraction unit 403, an adaptive weighted fusion unit 404, a detail enhancement unit 405, and a determination unit 406. The system includes: an electromagnetic partial discharge detection unit 401, configured to detect electromagnetic partial discharge in the GIS gas-insulated switchgear to be monitored and obtain electromagnetic partial discharge signals; an optical partial discharge signal acquisition unit 402, configured to acquire optical partial discharge signals from the GIS gas-insulated switchgear based on the electromagnetic partial discharge signals and obtain optical partial discharge signals; a depth feature extraction unit 403, configured to extract depth features from the electromagnetic partial discharge signals and the optical partial discharge signals based on a preset feature extraction layer included in the disconnector switch status monitoring model, and obtain electromagnetic features and optical features, wherein the disconnector switch status monitoring model includes: a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer; an adaptive weighted fusion unit 404, configured to perform adaptive weighted fusion of the electromagnetic features and the optical features based on the adaptive weighted fusion layer to obtain fused features; a detail enhancement unit 405, configured to enhance the details of the fused features based on the detail enhancement layer to obtain refined fused features; and a determination unit 406, configured to determine the contact state of the disconnector switch contacts corresponding to the refined fused features based on the classification layer.

[0109] It is understandable that the various units recorded in the GIS disconnector status intelligent monitoring device 400 based on multi-dimensional sensing are related to the reference... Figure 2 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method also apply to the GIS disconnector status intelligent monitoring device 400 based on multi-dimensional sensing and the units contained therein, and will not be repeated here.

[0110] The following is for reference. Figure 5 It shows a schematic diagram of the structure of an electronic device (e.g., a computing device) 500 suitable for implementing some embodiments of the present disclosure. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0111] like Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory 502 or a program loaded from a storage device 508 into a random access memory 503. The random access memory 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, the read-only memory 502, and the random access memory 503 are interconnected via a bus 504. An input / output interface 505 is also connected to the bus 504.

[0112] Typically, the following devices can be connected to the input / output interface 505: input devices 506 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 508 including, for example, magnetic tape, hard disk, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0113] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a read-only memory 502. When the computer program is executed by the processing device 501, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0114] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0115] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0116] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: perform electromagnetic partial discharge detection on the GIS gas-insulated switchgear to be monitored, obtaining an electromagnetic partial discharge signal; based on the electromagnetic partial discharge signal, acquire optical partial discharge signals from the GIS gas-insulated switchgear, obtaining an optical partial discharge signal; based on a preset disconnector switch status monitoring model including a feature extraction layer, perform deep feature extraction on the electromagnetic partial discharge signal and the optical partial discharge signal respectively, obtaining electromagnetic features and optical features, wherein the disconnector switch status monitoring model includes: a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer; based on the adaptive weighted fusion layer, perform adaptive weighted fusion on the electromagnetic features and the optical features to obtain fused features; based on the detail enhancement layer, perform detail enhancement on the fused features to obtain refined fused features; based on the classification layer, determine the contact state of the disconnector switch contacts corresponding to the refined fused features.

[0117] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0118] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0119] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0120] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for intelligent monitoring of the status of GIS disconnect switches based on multi-dimensional sensing, characterized in that, include: Electromagnetic partial discharge detection was performed on the GIS gas-insulated switchgear to be monitored to obtain electromagnetic partial discharge signals; Based on the electromagnetic partial discharge signal, optical partial discharge signal acquisition is performed on the GIS gas-insulated switchgear to obtain the optical partial discharge signal. Based on the feature extraction layer of the pre-set disconnector status monitoring model, deep feature extraction is performed on the electromagnetic partial discharge signal and the optical partial discharge signal respectively to obtain electromagnetic features and optical features. The disconnector status monitoring model includes: a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer. Based on the adaptive weighted fusion layer, the electromagnetic features and the optical features are adaptively weighted and fused to obtain the fused features; Based on the detail enhancement layer, the fused features are enhanced in detail to obtain refined fused features; Based on the classification layer, the contact state of the disconnecting switch corresponding to the refined fusion features is determined.

2. The method according to claim 1, characterized in that, The step of acquiring optical partial discharge signals from the GIS gas-insulated switchgear based on the electromagnetic partial discharge signals to obtain optical partial discharge signals includes: Based on the signal strength of the electromagnetic partial discharge signal, the estimated energy level of the partial discharge inside the GIS gas-insulated switchgear is determined. In response to determining that the estimated energy level meets the preset energy conditions, the preset first working mode is determined as the sensor working mode; In response to determining that the estimated energy level does not meet the energy condition, the preset second working mode is determined as the sensor working mode; By using an optical sensor array arranged on the GIS gas-insulated switchgear, optical signals are acquired from the GIS gas-insulated switchgear based on the sensor's operating mode to obtain optical partial discharge signals. The optical sensor array corresponds to a spatial location information set.

3. The method according to claim 1, characterized in that, The adaptive weighted fusion layer performs adaptive weighted fusion of the electromagnetic features and the optical features to obtain fused features, including: Adaptive attention fusion is performed on the electromagnetic features and the optical features to obtain initial fused features; The initial fused features are convolved to obtain two-channel convolutional features; The two-channel convolutional features are normalized to obtain normalized features; The normalized features are split to obtain the first adaptive weight and the second adaptive weight; The product between the first adaptive weight and the electromagnetic feature is determined as the weighted electromagnetic feature; The product between the second adaptive weight and the optical feature is determined as the weighted optical feature; The weighted electromagnetic features and the weighted optical features are added element by element to obtain the fused features.

4. The method according to claim 3, characterized in that, The step of enhancing the details of the fused features based on the detail enhancement layer to obtain refined fused features includes: Channel attention features are extracted from the fused features to obtain channel attention features; Spatial attention features are extracted from the fused features to obtain spatial attention features; The product between the channel attention feature and the weighted electromagnetic feature is determined as the refined electromagnetic feature; The product between the spatial attention feature and the weighted optical feature is determined as the refined optical feature; The refined electromagnetic features and the refined optical features are added element by element to obtain the refined fused features.

5. The method according to claim 4, characterized in that, The process of extracting channel attention features from the fused features to obtain channel attention features includes: Max pooling is performed on the fused features to obtain max pooled features; The fused features are then subjected to average pooling to obtain average pooled features; The max pooling feature and the average pooling feature are respectively convolved to obtain convolutional max pooling feature and convolutional average pooling feature; The channel attention features are obtained by adding the convolutional max pooling features and the convolutional average pooling features element by element.

6. The method according to claim 1, characterized in that, The step of determining the contact state of the disconnector switch corresponding to the refined fusion features based on the classification layer includes: Determine the contact state probability distribution corresponding to the refined fusion feature; Based on the probability distribution of the contact states, the contact states of the disconnecting switch contacts are generated.

7. The method according to claim 1, characterized in that, Before performing deep feature extraction on the electromagnetic partial discharge signal and the optical partial discharge signal respectively to obtain electromagnetic features and optical features in the feature extraction layer of the pre-set disconnector switch status monitoring model, the method further includes: A disconnector switch status dataset and an initial disconnector switch status monitoring model are obtained. The disconnector switch status data in the disconnector switch status dataset includes: sample electromagnetic partial discharge signals, sample optical partial discharge signals, and disconnector switch status labels. The initial disconnector switch status monitoring model includes: an initial feature extraction layer, an initial adaptive weighted fusion layer, an initial detail enhancement layer, and an initial classification layer. Based on each disconnector status data in the disconnector status dataset, the following training steps are performed on the initial disconnector status monitoring model: The sample electromagnetic partial discharge signal and sample optical partial discharge signal included in the disconnector status data are sequentially input into the initial feature extraction layer, initial adaptive weighted fusion layer and initial detail enhancement layer of the initial disconnector status monitoring model to generate sample refined fusion features. The refined and fused features of the samples are input into the initial classification layer of the initial disconnect switch state monitoring model to generate a predicted disconnect switch state. Determine the cross-entropy loss value between the predicted disconnector state and the corresponding disconnector state label; Determine the central loss value between the refined fusion features of the samples and the learnable category center features corresponding to the disconnector status labels; The total loss value is determined by the weighted sum of the center loss value and the cross-entropy loss value. Based on the total loss value, the initial disconnector state monitoring model is updated to obtain an updated disconnector state monitoring model, and the updated disconnector state monitoring model is used as the initial disconnector state monitoring model to perform the training step again. In response to the final total loss value being less than or equal to a preset loss threshold, the updated disconnector state monitoring model corresponding to the total loss value is determined as the disconnector state monitoring model.

8. A GIS disconnector switch status intelligent monitoring device based on multi-dimensional sensing, characterized in that, include: The electromagnetic partial discharge detection unit is configured to perform electromagnetic partial discharge detection on the GIS gas-insulated switchgear to be monitored and obtain electromagnetic partial discharge signals. An optical partial discharge signal acquisition unit is configured to acquire optical partial discharge signals from the GIS gas-insulated switchgear based on the electromagnetic partial discharge signal, thereby obtaining an optical partial discharge signal. The deep feature extraction unit is configured to perform deep feature extraction on the electromagnetic partial discharge signal and the optical partial discharge signal respectively based on the feature extraction layer of the preset disconnector switch status monitoring model to obtain electromagnetic features and optical features. The disconnector switch status monitoring model includes: a feature extraction layer, an adaptive weighted fusion layer, a detail enhancement layer, and a classification layer. An adaptive weighted fusion unit is configured to perform adaptive weighted fusion of the electromagnetic features and the optical features based on the adaptive weighted fusion layer to obtain fused features; The detail enhancement unit is configured to enhance the details of the fused features based on the detail enhancement layer to obtain refined fused features; The determining unit is configured to determine the contact state of the disconnector switch corresponding to the refined fusion features based on the classification layer.

9. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.

10. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images