Power equipment partial discharge identification method and device, electronic equipment and storage medium

Abstract

The application provides a partial discharge identification method of a power equipment, comprising: performing feature fusion on a plurality of pre-acquired multispectral images to obtain a spectrum fusion image. The multispectral images comprise ultraviolet images, visible light images and near-infrared images. A discharge identification model is obtained by performing transfer learning on a preset general image dataset and the spectrum fusion image. The partial discharge image of the power equipment to be identified is identified according to the discharge identification model to obtain a discharge identification result. Thus, the accurate identification of the partial discharge of the power equipment is realized.

Description

Technical Field

[0001] This application relates to the field of power equipment fault identification technology, and in particular to a method, device, electronic device and storage medium for identifying partial discharge in power equipment. Background Technology

[0002] Early and accurate identification and warning of partial discharge (PD) in power equipment is crucial for preventing insulation breakdown and avoiding large-scale power outages. As a core precursor signal of equipment insulation degradation, partial discharge provides multi-dimensional evidence for detection through its physical effects (such as electrical, optical, acoustic, and thermal effects). Among these, multispectral imaging technology, with its complementary multi-band information characteristics, has become a core technical approach for partial discharge monitoring in complex scenarios. Multispectral images cover three key bands: ultraviolet (UV), visible (VIS), and near-infrared (NIR). The imaging principles and application value of each band differ significantly: ultraviolet images, based on ultraviolet photon radiation imaging generated by partial discharge, can accurately capture the location of discharge points at all times, including day and night, with a photon detection sensitivity of up to 10. 4 Ultraviolet (UV) images, with a photon / second resolution, can identify weak discharge signals ≤1pC; however, their low spatial resolution (only 320×240) and lack of equipment structural texture information make it difficult to distinguish between surface discharges and discharges caused by airborne dust interference. Visible light images, with their high spatial resolution of 1920×1080, can clearly present structural details such as insulator strings, bushings, and terminals, providing a basis for determining the correlation between discharge locations and equipment components; however, their imaging depends on ambient light, and in severe weather conditions such as heavy rain and dense fog, the signal-to-noise ratio (SNR) drops below 15dB, causing the discharge spot to be confused with background noise, resulting in a false negative rate exceeding 70%. Near-infrared (NIIR) images are based on the weak temperature rise effect of the discharge area (typically ≤5℃), possessing some ability to penetrate fog and haze, but their temperature resolution is only 0.5℃, making them susceptible to thermal interference from normal equipment heat dissipation and sunlight, resulting in a false positive rate as high as 60%.

[0003] Traditional multi-band stitching methods (such as pixel overlay and simple threshold segmentation) do not consider the different correlations between each band and partial discharge. For example, directly stitching ultraviolet and visible light images at a 1:1 ratio results in the discharge signal being submerged by the background texture. This leads to a false negative rate of over 80% for small targets such as discharge at the edge of a φ5mm insulator skirt, and the processing delay for 1080p images exceeds 300ms, which cannot meet the requirements for real-time early warning.

[0004] Recognition methods based on traditional machine learning (such as support vector machine (SVM) and random forest (RF) rely on manual extraction of discharge features (such as spot area and gray mean), resulting in poor feature robustness and weak generalization ability across different equipment types (transformers, switchgear, transmission lines). Furthermore, they do not incorporate insulation characteristic parameters of power equipment (such as insulation dielectric breakdown field strength and partial discharge standard threshold), leading to an accuracy of less than 65% in identifying weak discharges caused by insulation aging. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method, device, electronic device and storage medium for identifying partial discharge in power equipment.

[0006] Based on the above objectives, this application provides a method for identifying partial discharge in power equipment, comprising: performing feature fusion on a pre-acquired multispectral image to obtain a spectral fusion image; wherein the multispectral image includes an ultraviolet image, a visible light image, and a near-infrared image; performing transfer learning on a preset general image dataset and the spectral fusion image to obtain a discharge identification model; and identifying the partial discharge image of the power equipment to be identified based on the discharge identification model to obtain a discharge identification result.

[0007] In some embodiments, the feature fusion of the pre-acquired multispectral images to obtain a spectral fusion image specifically includes: extracting features from the ultraviolet image, the visible light image, and the near-infrared image to obtain ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features; the ultraviolet discharge features are determined by the following formula: ;in, This represents the intensity probability diagram of the discharged photons. Represents pixels ( x,y The normalized value of the ultraviolet photon count. This represents the average photon count in the background region of an ultraviolet image. The normalized value of the minimum detectable photon count for partial discharge is represented; the visible light structural characteristics are determined by the following formula: ;in, This represents the equipment structure weight diagram. This represents the structure mask of the power equipment. This represents the Gaussian gradient of a visible light image. The normalization function maps the result to the [0,1] interval; the near-infrared temperature rise characteristic is determined by the following formula: ;in, This represents the relative temperature rise coefficient. This represents the normalized near-infrared temperature value of pixel (x,y). This indicates the average temperature during normal operation of the equipment. The maximum allowable operating temperature of the device is indicated; the ultraviolet discharge characteristics, visible light structural characteristics, and near-infrared temperature rise characteristics are weighted and fused according to preset image allocation weights to obtain the spectral fusion image.

[0008] In some embodiments, the image allocation weights include ultraviolet weights, visible light weights, and near-infrared weights; before weighting and fusing the ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features according to preset image allocation weights to obtain the spectral fusion image, the method further includes: determining the ultraviolet weights based on the ultraviolet discharge features and visible light structural features; the ultraviolet weights are determined by the following formula: ;in, This represents the Sigmoid activation function. Represents the ultraviolet characteristic coefficient. The background suppression coefficient is represented; the near-infrared weight is determined based on the near-infrared temperature rise characteristics and the ultraviolet discharge characteristics; the near-infrared weight is determined by the following formula: ;in, Represents the characteristic coefficient of temperature rise. The discharge correlation coefficient is represented; the visible light weight is determined based on the ultraviolet weight and the near-infrared weight; the visible light weight is determined by the following formula: The ultraviolet weight, the visible light weight, and the near-infrared weight are solved according to the preset energy minimization constraint.

[0009] In some embodiments, solving for the ultraviolet weights, visible light weights, and near-infrared weights according to a preset energy minimization constraint further includes: determining an ultraviolet feature fidelity term based on the ultraviolet weights and the ultraviolet discharge characteristics; the ultraviolet feature fidelity term is determined by the following formula: The near-infrared feature fidelity term is determined based on the near-infrared weights and the near-infrared discharge characteristics; the near-infrared feature fidelity term is determined by the following formula: The spatial smoothing term is determined based on the ultraviolet weight and the infrared weight; the spatial smoothing term is determined by the following formula: The total energy function is determined based on the ultraviolet feature fidelity, the near-infrared feature fidelity term, the spatial smoothing term, and a preset spatial smoothing coefficient. ;in, The spatial smoothing coefficient is represented by the equation; the ultraviolet weight, the visible light weight, and the near-infrared weight are solved based on the total energy function.

[0010] In some embodiments, the step of identifying the partial discharge image of the power equipment to be identified according to the discharge identification model to obtain the discharge identification result specifically includes: determining the spectral fusion image, the ultraviolet weight, and the near-infrared weight corresponding to the partial discharge image of the power equipment to be identified, to obtain the spectral fusion identification image, the ultraviolet identification weight, and the near-infrared identification weight; mapping the ultraviolet identification weight and the near-infrared identification weight into a pseudo-color heatmap, and superimposing the pseudo-color heatmap with the spectral fusion identification image to obtain a weight contribution heatmap; determining the discharge region in the spectral fusion identification image according to the discharge identification result, and annotating the discharge region with parameters to obtain a discharge location parameter map; and visually outputting the discharge identification result according to the weight contribution heatmap and the discharge location parameter map.

[0011] In some embodiments, the step of performing transfer learning based on a preset general image dataset and the spectral fusion image to obtain a discharge recognition model specifically includes: calculating the difference loss between the general image dataset and the spectral fusion image using a domain adversarial training method; the difference loss is determined by the following formula: ;in, This indicates the number of samples in the general image dataset. This indicates the number of samples in the spectral fusion image. This represents the characteristics of the general image dataset. The features representing the spectral fusion image, The probability that a feature belongs to the current domain is represented; a total loss function is determined based on a preset power prior loss and the difference loss; the discharge recognition model is obtained by performing transfer learning on the spectral fusion image based on the total loss function and the general image dataset.

[0012] A partial discharge identification device for power equipment includes: an image fusion module for performing feature fusion on pre-acquired multispectral images to obtain a spectral fusion image; wherein the multispectral images include ultraviolet images, visible light images, and near-infrared images; a model training module for performing transfer learning based on a preset general image dataset and the spectral fusion image to obtain a discharge identification model; and a discharge identification module for identifying the partial discharge image of the power equipment to be identified based on the discharge identification model to obtain a discharge identification result.

[0013] In some embodiments, the image fusion module specifically includes: a feature extraction module, used to extract features from the ultraviolet image, the visible light image, and the near-infrared image to obtain ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features; the ultraviolet discharge features are determined by the following formula: ;in, This represents the intensity probability diagram of the discharged photons. Represents pixels ( x,y The normalized value of the ultraviolet photon count. This represents the average photon count in the background region of an ultraviolet image. The normalized value of the minimum detectable photon count for partial discharge is represented; the visible light structural characteristics are determined by the following formula: ;in, This represents the equipment structure weight diagram. This represents the structure mask of the power equipment. This represents the Gaussian gradient of a visible light image. The normalization function maps the result to the [0,1] interval; the near-infrared temperature rise characteristic is determined by the following formula: ;in, This represents the relative temperature rise coefficient. This represents the normalized near-infrared temperature value of pixel (x,y). This indicates the average temperature during normal operation of the equipment. The maximum allowable operating temperature of the device is indicated; the feature fusion module is used to perform weighted fusion of the ultraviolet discharge feature, the visible light structural feature, and the near-infrared temperature rise feature according to the preset image allocation weights to obtain the spectral fusion image.

[0014] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any of the preceding descriptions.

[0015] A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform any of the methods described above.

[0016] As can be seen from the above, the partial discharge identification method for power equipment provided in this application fuses ultraviolet, visible, and near-infrared images, thereby accurately integrating images of different resolutions. This avoids information in lower-resolution images being covered by the textures of other images, leading to information loss. Furthermore, since partial discharge in power equipment is relatively rare, the sample size is also scarce. Transfer learning can reduce the required number of samples, preventing overfitting due to insufficient sample size. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of the partial discharge identification method for power equipment provided in the embodiments of this application;

[0019] Figure 2 This is a schematic diagram of the structure of the partial discharge identification device for power equipment provided in the embodiments of this application; Figure 3 This is a schematic diagram of a more specific electronic device hardware structure provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0021] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0022] Early and accurate identification and warning of partial discharge (PD) in power equipment is crucial for preventing insulation breakdown and avoiding large-scale power outages. As a core precursor signal of equipment insulation degradation, partial discharge provides multi-dimensional evidence for detection through its physical effects (such as electrical, optical, acoustic, and thermal effects). Among these, multispectral imaging technology, with its complementary multi-band information characteristics, has become a core technical approach for partial discharge monitoring in complex scenarios. Multispectral images cover three key bands: ultraviolet (UV), visible (VIS), and near-infrared (NIR). The imaging principles and application value of each band differ significantly: ultraviolet images, based on ultraviolet photon radiation imaging generated by partial discharge, can accurately capture the location of discharge points at all times, including day and night, with a photon detection sensitivity of up to 10. 4 Ultraviolet (UV) images, with a photon / second resolution, can identify weak discharge signals ≤1pC; however, their low spatial resolution (only 320×240) and lack of equipment structural texture information make it difficult to distinguish between surface discharges and discharges caused by airborne dust interference. Visible light images, with their high spatial resolution of 1920×1080, can clearly present structural details such as insulator strings, bushings, and terminals, providing a basis for determining the correlation between discharge locations and equipment components; however, their imaging depends on ambient light, and in severe weather conditions such as heavy rain and dense fog, the signal-to-noise ratio (SNR) drops below 15dB, causing the discharge spot to be confused with background noise, resulting in a false negative rate exceeding 70%. Near-infrared (NIIR) images are based on the weak temperature rise effect of the discharge area (typically ≤5℃), possessing some ability to penetrate fog and haze, but their temperature resolution is only 0.5℃, making them susceptible to thermal interference from normal equipment heat dissipation and sunlight, resulting in a false positive rate as high as 60%.

[0023] Table 1 Comparison of core features of multispectral images

[0024] Multispectral fusion and intelligent recognition technology is a core direction for overcoming the limitations of single-band technologies. However, current mainstream solutions in the power industry have significant bottlenecks in terms of adaptability to complex scenarios, small-sample learning, and recognition reliability, making it difficult to meet actual operation and maintenance needs. Traditional multi-band stitching methods (such as pixel overlay and simple threshold segmentation) do not consider the different correlations between each band and partial discharge. For example, directly stitching ultraviolet and visible light images at a 1:1 ratio results in the discharge signal being submerged by background texture, leading to a false negative rate of over 80% for small targets such as discharges at the edge of φ5mm insulator skirts. Furthermore, the processing latency for 1080p images exceeds 300ms, failing to meet real-time early warning requirements. Recognition methods based on traditional machine learning (such as Support Vector Machine (SVM) and Random Forest (RF)) rely on manual extraction of discharge features (such as spot area and grayscale mean), resulting in poor feature robustness and weak generalization ability across different equipment types (transformers, switchgear, transmission lines). Moreover, they do not incorporate insulation characteristic parameters of power equipment (such as insulation dielectric breakdown field strength and partial discharge standard threshold), resulting in an accuracy of less than 65% for identifying weak discharges caused by insulation aging. Single deep learning models, such as convolutional neural networks (CNNs), require a large number of labeled discharge samples (usually ≥100,000) for training. However, partial discharge samples are scarce in the power industry and the labeling cost is high, leading to severe overfitting of the model. At the same time, the model is a "black box" structure and cannot explain the key basis for discharge identification (such as ultraviolet photon intensity and temperature rise), which contradicts the high reliability requirements of the power industry for "traceability and verifiability".

[0025] In summary, the core bottlenecks of existing technologies can be summarized into three points: First, poor scenario adaptability, as they do not integrate the insulation characteristics of power equipment with multi-band complementary information, making it difficult to cope with interference caused by severe weather and complex equipment structures; second, insufficient small-sample learning ability, as traditional deep learning models require a large number of labeled samples, which cannot adapt to the reality of scarce partial discharge samples in the power industry; and third, untraceable identification results, as the "black box" model cannot provide a basis for decision-making and does not meet the high reliability requirements of power operation and maintenance.

[0026] like Figure 1 As shown, this application provides a method for identifying partial discharge in power equipment, including: Step S1: Perform feature fusion on the pre-acquired multispectral images to obtain a spectral fusion image. The multispectral images include ultraviolet images, visible light images, and near-infrared images.

[0027] In this embodiment, ultraviolet images, visible light images, and near-infrared images are fused to ensure that images of different resolutions can be accurately fused, thus avoiding information loss caused by the texture of other images covering the information recorded in the lower resolution images.

[0028] As an optional implementation of this application, since the image clarity of ultraviolet and near-infrared bands acquired by existing devices is generally lower than that of images acquired by visible light bands, ultraviolet images, visible light images and near-infrared images can be standardized before image fusion to achieve registration of images of different bands.

[0029] Specifically, the input ultraviolet (320×240), visible light (1920×1080), and near-infrared (640×512) images are first standardized to ensure data consistency. Ultraviolet image: The photon counting signal was normalized to the [0,1] interval, and the resolution was upsampled to 640×512 (achieved through bilinear interpolation). The detection sensitivity threshold was set to 10. 4 Photons per second (corresponding to a normalized value of 0.3); Visible light images: resolution downsampled to 640×512, signal-to-noise ratio improved to ≥35dB (achieved through bilateral filtering noise reduction); Near-infrared images: temperature values ​​are normalized to the [0,1] range, and the resolution is uniformly 640×512.

[0030] The Scale-Invariant Feature Transform (SIFT) algorithm is used to achieve multispectral image registration: ≥800 key feature points (stable structures such as device edges and terminals) are extracted from each image, and the matching threshold is set to 0.65 to ensure that the registration error is ≤1 pixel, thus avoiding fusion distortion caused by image misalignment.

[0031] Step S2: Transfer learning is performed on the preset general image dataset and the spectral fusion image to obtain the discharge recognition model.

[0032] In this embodiment, since the phenomenon of partial discharge in power equipment is relatively rare, the sample size is also relatively scarce. By using transfer learning, the required number of samples can be reduced, thus avoiding model overfitting due to insufficient sample size.

[0033] Step S3: Based on the discharge recognition model, identify the partial discharge image of the power equipment to be identified, and obtain the discharge recognition result.

[0034] In some embodiments, step S1 specifically includes: Step S11: Extract features from the ultraviolet image, visible light image and near-infrared image to obtain ultraviolet discharge features, visible light structure features and near-infrared temperature rise features.

[0035] The characteristics of ultraviolet discharge are determined by the following formula: .

[0036] in, This represents the intensity probability map of the discharged photons, used for pixel quantization. x,y The probability of a true discharge signal is given at position ), with a value range of [0,1] (1 indicates a confirmed discharge, and 0 indicates no discharge). Represents pixels ( x,y The normalized value of the ultraviolet photon count (i.e., the relative number of ultraviolet photons detected by the pixel per unit time). This represents the average photon count in the background region of an ultraviolet image (calculated by extracting the background region through adaptive threshold segmentation). This represents the normalized value of the minimum detectable photon count for partial discharge.

[0037] In this embodiment, when hour, (No discharge); when hour, (Discharge confirmed).

[0038] The visible light structural characteristics are determined by the following formula: .

[0039] in, This represents the device structure weight map, used for pixel quantization. (x,y The probability that the device belongs to a critical area (insulator, bushing, terminal block) is within the range of [0,1]. This represents the structure mask of the power equipment (i.e., it indicates whether the identifier pixel belongs to the equipment area, 1 represents the equipment area, and 0 represents the background area, generated based on the power equipment model and image segmentation). The Gaussian gradient of a visible light image (characterizing the sharpness of structural edges). This represents the normalization function, which maps the result to the interval [0,1].

[0040] In this embodiment, The higher the value, the more critical the structure of the equipment is, and the greater the potential for discharge needs to be monitored.

[0041] The near-infrared temperature rise characteristics are determined by the following formula: .

[0042] in, Represents the relative temperature rise coefficient, used for quantizing pixels. x,y The degree of temperature deviation from the normal temperature of the equipment, with a value range of [0,1]. Represents the normalized near-infrared temperature value of pixel (x,y) (physical meaning: the relative value of the actual device temperature corresponding to this pixel). This indicates the average temperature during normal operation of the equipment. This indicates the maximum allowable operating temperature of the equipment (based on the DL / T 722-2019 standard, such as a maximum transformer winding temperature of 95℃).

[0043] In this embodiment, when hour, (No abnormal temperature rise); when hour, (Severe temperature rise).

[0044] Step S12: The ultraviolet discharge features, visible light structure features, and near-infrared temperature rise features are weighted and fused according to the preset image allocation weights to obtain a spectral fusion image.

[0045] In some embodiments, the image assignment weights include ultraviolet weights, visible light weights, and near-infrared weights.

[0046] Before step S11, the method further includes: Step S101: Determine the ultraviolet weight based on the ultraviolet discharge characteristics and visible light structure characteristics.

[0047] The ultraviolet weight is determined by the following formula: .

[0048] in, This represents the Sigmoid activation function. Represents the ultraviolet characteristic coefficient. This represents the background suppression coefficient.

[0049] In this embodiment, the purpose of using the Sigmoid activation function is to constrain the weights within the [0,1] interval. The ultraviolet feature coefficient is used to adjust the contribution intensity of the discharge photon features, and the background suppression coefficient is used to suppress ultraviolet interference in non-device regions.

[0050] Step S102: Determine the near-infrared weights based on the near-infrared temperature rise characteristics and ultraviolet discharge characteristics.

[0051] The near-infrared weights are determined by the following formula: .

[0052] in, Represents the characteristic coefficient of temperature rise. This represents the discharge correlation coefficient.

[0053] In this embodiment, the temperature rise characteristic coefficient is used to adjust the contribution intensity of the temperature rise characteristic, and the discharge correlation coefficient is used to ensure the correlation between the temperature rise characteristic and the discharge characteristic.

[0054] Step S103: Determine the visible light weight based on the ultraviolet weight and the near-infrared weight.

[0055] The visible light weight is determined by the following formula: .

[0056] Step S104: Solve for the ultraviolet weight, visible light weight, and near-infrared weight according to the preset energy minimization constraint.

[0057] In this embodiment, the ultraviolet weight, visible light weight and near-infrared weight are solved by introducing an energy minimization constraint, thereby avoiding block effects in the fused image caused by abrupt weight changes.

[0058] In some embodiments, step S104 further includes: Step S1041: Determine the ultraviolet feature fidelity term based on the ultraviolet weight and ultraviolet discharge characteristics.

[0059] The ultraviolet feature fidelity term is determined by the following formula: .

[0060] In this embodiment, the ultraviolet feature fidelity term is used to minimize the difference between the ultraviolet weight and the discharge probability, ensuring that the discharge region is dominated by ultraviolet radiation.

[0061] Step S1042: Determine the near-infrared feature fidelity term based on the near-infrared weight and near-infrared discharge characteristics.

[0062] The near-infrared feature fidelity term is determined by the following formula: .

[0063] In this embodiment, the near-infrared feature fidelity term is used to minimize the difference between the near-infrared weight and the temperature rise coefficient, ensuring that the near-infrared dominates in the temperature rise region.

[0064] Step S1043: Determine the spatial smoothing term based on the ultraviolet weight and the infrared weight.

[0065] The spatial smoothing term is determined by the following formula: .

[0066] In this embodiment, the spatial smoothing term avoids sudden changes in the weights of adjacent pixels by constraining the rate of change of the weight space.

[0067] Step S1044: Determine the total energy function based on the ultraviolet feature fidelity, near-infrared feature fidelity, spatial smoothing term and preset spatial smoothing coefficient.

[0068] .

[0069] in, This represents the spatial smoothing coefficient.

[0070] Step S1045: Solve for the ultraviolet weight, visible light weight, and near-infrared weight based on the total energy function.

[0071] In this embodiment, the ultraviolet feature fidelity term is used to ensure the accuracy of the ultraviolet weight in the discharge region, the near-infrared feature fidelity term is used to ensure the accuracy of the near-infrared weight in the temperature rise region, and the spatial smoothing term is used to suppress weight abrupt changes. The smoothing coefficient ranges from [0.2, 0.4] and is used to balance feature fidelity and spatial smoothness.

[0072] As an alternative implementation method, when solving for the ultraviolet weight, visible light weight and near-infrared weight, the Fast Fourier Transform (FFT) can be used to solve the energy minimization problem, which transforms the continuous domain energy function into a frequency domain algebraic equation, improving the solution efficiency by more than 50% compared with the traditional iterative method.

[0073] As an optional implementation, the steps for solving the ultraviolet weight, visible light weight, and near-infrared weight can be as follows: Step S10: Take the variation of the energy function and set it equal to 0 to derive the system of linear partial differential equations; Step S20, apply Neumann boundary conditions ( , n (as the boundary normal vector), to avoid boundary effects; Step S30: Perform a two-dimensional FFT transform on the partial differential equation to convert it into a frequency domain algebraic equation for solution. Step S40: Perform inverse FFT transformation and [0,1] interval truncation correction on the solution result to obtain the optimal weights. , , .

[0074] As an optional implementation, after obtaining the optimal weights, the registered multispectral images are then subjected to pixel-by-pixel weighted fusion based on these optimal weights, using the following formula:

[0075] In the formula, The image pixel values ​​are fused (physical meaning: the combined grayscale value of the pixel after fusion, which combines discharge signal and device structure information). The fused image resolution is 640×512, which highlights the ultraviolet and temperature rise characteristics of partial discharge while preserving the texture of the device structure, providing high-quality input for subsequent recognition.

[0076] In some embodiments, step S3 specifically includes: Step S31: Determine the spectral fusion image, ultraviolet weight, and near-infrared weight corresponding to the partial discharge image of the power equipment to be identified, and obtain the spectral fusion recognition image, ultraviolet recognition weight, and near-infrared recognition weight.

[0077] Step S32: Map the ultraviolet recognition weights and near-infrared recognition weights into a pseudo-color heatmap, and overlay the pseudo-color heatmap with the spectral fusion recognition image to obtain a weight contribution heatmap.

[0078] In this embodiment, the optimal weights are... , The mapping is a pseudo-color heatmap (blue→green→red, corresponding to weights 0→0.5→1), which is superimposed on the fused image to intuitively present the contribution areas of ultraviolet and near-infrared features to the recognition results (the red area is the core discharge feature area).

[0079] Step S33: Determine the discharge region in the spectral fusion recognition image based on the discharge recognition result, and annotate the parameters of the discharge region to obtain the discharge location parameter map.

[0080] In this embodiment, the attention map of the model classification layer is extracted using the Grad-CAM algorithm, highlighting the image region that the model focuses on when identifying discharge, and labeling the key feature parameters of the region (normalized value of ultraviolet photon intensity, temperature rise coefficient).

[0081] Step S34: Visualize the discharge identification results based on the weighted contribution heatmap and discharge location parameter map.

[0082] In this embodiment, a text report can also be automatically generated based on the weighted contribution heatmap and discharge location parameter map, including the recognition results, confidence level, and core feature parameters (such as...). , This ensures that decisions are traceable.

[0083] The above output methods enable the visualization and interpretation of results, providing a basis for identification for operations and maintenance personnel.

[0084] As an optional implementation, a three-level early warning system can also be established based on the content of the discharge identification results and the importance of the equipment. The early warning information includes the discharge level, the scope of impact, and handling suggestions, such as: Level 1 Warning (Severe Discharge): The identification result is severe discharge, with a confidence level ≥ 0.85; Warning method: local audible and visual alarm + remote platform push (SMS + APP); Handling suggestion: immediately shut down the machine for maintenance, focusing on checking the insulation status of the equipment; Level 2 Warning (Weak Discharge): The identification result is a weak discharge with a confidence level ≥ 0.8; Warning method: local indicator light alarm + remote platform recording; Handling suggestion: arrange inspection within 24 hours to track the discharge development trend; Level 3 warning (no discharge): The identification result is no discharge, with a confidence level ≥ 0.9; output normal operating status information, no additional processing required.

[0085] The early warning triggering conditions strictly follow the "Technical Guidelines for Partial Discharge Monitoring of Power Equipment" (DL / T 268-2019) to ensure the scientific nature and seriousness of the early warning.

[0086] For edge device adaptation and deployment, model optimization and deployment adaptation are performed based on the hardware characteristics of edge devices such as FPGAs and edge computing boxes: Model Quantization Optimization: Quantization-Aware Training (QAT) is used to compress the model size and improve inference speed; Computation graph optimization: Redundant computations are removed by fusing convolution and BN layers to reduce the number of device instruction executions; pipelined parallel scheduling is adopted to decompose the "fusion-identification-early warning" process into parallel tasks, thereby reducing the overall latency; Hardware interface adaptation: Optimize data storage format (row-first storage) and computing task allocation for the characteristics of FPGA on-chip memory (BRAM) and computing unit (DSP) to ensure stable operation of the model on the FPGA and meet the requirements of real-time monitoring and early warning.

[0087] In some embodiments, step S2 specifically includes: Step S21: Calculate the difference loss between the general image dataset and the spectral fusion image according to the domain adversarial training method.

[0088] The difference loss is determined by the following formula: .

[0089] in, This represents the number of samples in a general image dataset. This indicates the number of samples in the spectral fusion image. Features representing a general image dataset, Features representing spectral fusion images This represents the probability that a feature belongs to the current domain.

[0090] In this embodiment, to address the distribution differences between the source domain (the general image dataset ImageNet) and the target domain (the partial discharge dataset), a domain adversarial training mechanism is introduced. A domain classifier (a two-layer fully connected network) is constructed, and the loss is minimized by minimizing the distribution difference loss between the source and target domain features. This enables domain-adaptive transfer of features. The domain classifier output represents the probability that a feature belongs to the source domain. Through domain adversarial training, the features extracted by the model possess both generality and adaptability to power scenarios.

[0091] Step S22: Determine the total loss function based on the preset prior power loss and differential loss.

[0092] In this embodiment, to embed the insulation characteristics of power equipment and the standard parameters of partial discharge, the extracted features are constrained and optimized. The prior power loss is defined. The formula is: ; In the formula, Let k be the feature vector of the k-th sample; This is a feature mapping function (mapping high-dimensional features into discharge probability values); Let be the prior discharge probability of the k-th sample. Ensure that the model recognition results meet the professional standards of the power industry and improve the reliability of recognition.

[0093] Step S23: Perform transfer learning on the spectral fusion image based on the total loss function and the general image dataset to obtain the discharge recognition model.

[0094] In this embodiment, the total loss function of the model is: ; In the formula, Domain adaptation weights, These are the prior weights for electricity, determined through sample verification.

[0095] As an optional implementation, the model uses the lightweight convolutional neural network MobileNetV3 as its backbone (with only 4.2MB of parameters), and introduces a domain adaptation (DA) module and an electric power prior constraint layer. The overall architecture consists of three main modules: feature extraction, domain adaptation, and classification / recognition. Feature extraction module: Employs lightweight convolutional blocks from MobileNetV3 (depthiably separable convolution + SE attention mechanism) to process the fused image generated in stage one. Feature extraction is performed, and a high-dimensional feature map with dimensions of 64×64×128 is output. This module enhances the response of discharge-related features (such as light spots and temperature rise areas) through the SE attention mechanism, thereby suppressing background interference.

[0096] The classification and recognition module employs a 2-layer fully connected network and a Softmax activation function to map the optimized features into three output classes: no discharge (0), weak discharge (1), and severe discharge (2). The classification loss is... The cross-entropy loss function is used.

[0097] As an optional implementation, the model training process may include: Step S101, Pre-training: Train the backbone network on the source domain dataset, initialize the parameters of the feature extraction module, and obtain general image feature extraction capabilities; Step S102, Domain Adaptive Training: Input the target domain dataset, train the domain adaptive module and classifier, and minimize... and ; Step S103, Fine-tuning Training: Unfreeze all parameters of the backbone network and add electrical prior loss. Using a small learning rate (1e) -5 Fine-tune the model until the total loss is reached. convergence; Step S104, Model Compression: Quantization and channel pruning techniques are used to compress the number of model parameters to meet the deployment requirements of edge devices.

[0098] As an optional implementation, after the model training is completed, the spectral fusion image obtained in step S1 can be re-input into the trained model to output the recognition result and recognition confidence.

[0099] As can be seen from the above embodiments of this application, the partial discharge identification method for power equipment provided by this application fuses ultraviolet images, visible light images, and near-infrared images, thereby enabling accurate fusion of images with different resolutions. This avoids information recorded in images with lower resolution being covered by the textures of other images, thus preventing information loss. Furthermore, since partial discharge phenomena in power equipment are relatively rare, the sample size is also scarce. Transfer learning can reduce the required number of samples, avoiding overfitting of the model due to insufficient sample size.

[0100] The partial discharge identification method for power equipment provided in this application has the following advantages: Strong scene adaptability and high recognition accuracy: Deeply embedding the insulation characteristic parameters of power equipment and industry standards, constructing a power-guided multispectral adaptive fusion model, and strengthening discharge characteristics and suppressing interference through pixel-level dynamic weight allocation, solving the problems of missed detection and misjudgment under severe weather and complex equipment structures.

[0101] It exhibits outstanding small-sample learning capabilities and good generalization: Employing a transfer learning framework, it transfers prior knowledge from general image recognition to the partial discharge scenario in the power industry, addressing the pain point of scarce discharge samples in the power sector. Through a domain-adaptive module and power prior constraints, it improves the model's generalization accuracy across equipment types, adapting to various power equipment such as transformers, switchgear, and transmission lines.

[0102] Traceable decision-making and reliable early warning: The innovative generation of weighted contribution heatmaps, model attention maps, and identification basis reports clearly presents the core criteria for discharge identification (such as ultraviolet photon intensity and temperature rise coefficient), solving the "black box" problem of deep learning and fully meeting the high reliability requirements of the power industry for "traceability and verifiability." The established three-level early warning system is scientific and standardized, providing clear guidance for operation and maintenance decisions and effectively reducing the risk of equipment failure.

[0103] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0104] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0105] Based on the same inventive concept, corresponding to any of the above embodiments, this application also provides a partial discharge identification device for power equipment.

[0106] refer to Figure 2 A partial discharge identification device for power equipment, comprising: The image fusion module 100 is used to perform feature fusion on pre-acquired multispectral images to obtain a spectral fusion image. The multispectral images include ultraviolet images, visible light images, and near-infrared images.

[0107] The model training module 200 is used to perform transfer learning based on a preset general image dataset and a spectral fusion image to obtain a discharge recognition model.

[0108] The discharge recognition module 300 is used to recognize the partial discharge image of the power equipment to be identified according to the discharge recognition model, and obtain the discharge recognition result.

[0109] In some embodiments, the image fusion module specifically includes: The feature extraction module is used to extract features from ultraviolet images, visible light images, and near-infrared images to obtain ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features.

[0110] The characteristics of ultraviolet discharge are determined by the following formula: .

[0111] in, This represents the intensity probability diagram of the discharged photons. Represents pixels ( x,y The normalized value of the ultraviolet photon count. This represents the average photon count in the background region of an ultraviolet image. This represents the normalized value of the minimum detectable photon count for partial discharge.

[0112] The visible light structural characteristics are determined by the following formula: .

[0113] in, This represents the equipment structure weight diagram. This represents the structure mask of the power equipment. This represents the Gaussian gradient of a visible light image. This represents the normalization function, which maps the result to the interval [0,1].

[0114] The near-infrared temperature rise characteristics are determined by the following formula: .

[0115] in, This represents the relative temperature rise coefficient. This represents the normalized near-infrared temperature value of pixel (x,y). This indicates the average temperature during normal operation of the equipment. This indicates the maximum allowable operating temperature of the device.

[0116] The feature fusion module is used to perform weighted fusion of ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features according to preset image weights to obtain a spectral fusion image.

[0117] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0118] The apparatus of the above embodiments is used to implement the corresponding power equipment partial discharge identification method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0119] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the power equipment partial discharge identification method described in any of the above embodiments.

[0120] Figure 3 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.

[0121] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0122] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0123] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0124] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0125] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.

[0126] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0127] The electronic devices described above are used to implement the corresponding partial discharge identification method for power equipment in any of the foregoing embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0128] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium that stores computer instructions for causing the computer to execute the power equipment partial discharge identification method as described in any of the above embodiments.

[0129] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0130] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the power equipment partial discharge identification method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0131] It should be noted that the embodiments of this application can also be further described in the following ways: A method for identifying partial discharge in power equipment, comprising: Feature fusion is performed on pre-acquired multispectral images to obtain a spectral fusion image; wherein, the multispectral images include ultraviolet images, visible light images, and near-infrared images; A discharge recognition model is obtained by performing transfer learning on the preset general image dataset and the spectral fusion image; The discharge recognition model is used to identify the partial discharge image of the power equipment to be identified, and the discharge recognition result is obtained.

[0132] Optionally, the step of performing feature fusion on the pre-acquired multispectral image to obtain a spectral fusion image specifically includes: Feature extraction is performed on the ultraviolet image, the visible light image, and the near-infrared image to obtain ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features; The ultraviolet discharge characteristics are determined by the following formula: ; in, This represents the intensity probability diagram of the discharged photons. Represents pixels ( x,y The normalized value of the ultraviolet photon count. This represents the average photon count in the background region of an ultraviolet image. This represents the normalized value of the minimum detectable photon count for partial discharge. The visible light structural features are determined by the following formula: ; in, This represents the equipment structure weight diagram. This represents the structure mask of the power equipment. This represents the Gaussian gradient of a visible light image. This represents a normalization function that maps the result to the interval [0,1]. The near-infrared temperature rise characteristic is determined by the following formula: ; in, This represents the relative temperature rise coefficient. This represents the normalized near-infrared temperature value of pixel (x,y). This indicates the average temperature during normal operation of the equipment. Indicates the maximum allowable operating temperature of the equipment; The ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features are weighted and fused according to preset image allocation weights to obtain the spectral fusion image.

[0133] Optionally, the image assignment weights include ultraviolet weights, visible light weights, and near-infrared weights; Before performing weighted fusion of the ultraviolet discharge features, the visible light structural features, and the near-infrared temperature rise features according to preset image allocation weights to obtain the spectral fusion image, the method further includes: The ultraviolet weight is determined based on the ultraviolet discharge characteristics and the visible light structural characteristics; The ultraviolet weight is determined by the following formula: ; in, This represents the Sigmoid activation function. Represents the ultraviolet characteristic coefficient. Indicates the background suppression coefficient; The near-infrared weights are determined based on the near-infrared temperature rise characteristics and the ultraviolet discharge characteristics. The near-infrared weights are determined by the following formula: ; in, Represents the characteristic coefficient of temperature rise. Indicates the discharge correlation coefficient; The visible light weight is determined based on the ultraviolet weight and the near-infrared weight; The visible light weight is determined by the following formula: ; The ultraviolet weight, the visible light weight, and the near-infrared weight are solved according to the preset energy minimization constraint.

[0134] Optionally, the step of solving for the ultraviolet weight, the visible light weight, and the near-infrared weight according to the preset energy minimization constraint further includes: The ultraviolet feature fidelity term is determined based on the ultraviolet weight and the ultraviolet discharge characteristics; The ultraviolet feature fidelity term is determined by the following formula: ; The near-infrared feature fidelity term is determined based on the near-infrared weights and the near-infrared discharge characteristics. The near-infrared feature fidelity term is determined by the following formula: ; The spatial smoothing term is determined based on the ultraviolet weight and the infrared weight; The spatial smoothing term is determined by the following formula: ; The total energy function is determined based on the ultraviolet feature fidelity, the near-infrared feature fidelity term, the spatial smoothing term, and a preset spatial smoothing coefficient. ; in, This represents the spatial smoothing coefficient; The ultraviolet weight, the visible light weight, and the near-infrared weight are solved based on the total energy function.

[0135] Optionally, the step of identifying the partial discharge image of the power equipment to be identified based on the discharge identification model to obtain the discharge identification result specifically includes: The spectral fusion image, the ultraviolet weight, and the near-infrared weight corresponding to the partial discharge image of the power equipment to be identified are determined to obtain the spectral fusion identification image, the ultraviolet identification weight, and the near-infrared identification weight. The ultraviolet recognition weights and the near-infrared recognition weights are mapped to a pseudo-color heatmap, and the pseudo-color heatmap is superimposed on the spectral fusion recognition image to obtain a weight contribution heatmap. Based on the discharge identification results, the discharge region in the spectral fusion identification image is determined, and the discharge region is parameter-labeled to obtain a discharge location parameter map; The discharge identification results are visualized and output based on the weighted contribution heatmap and the discharge location parameter map.

[0136] Optionally, the step of performing transfer learning based on a preset general image dataset and the spectral fusion image to obtain a discharge recognition model specifically includes: The difference loss between the general image dataset and the spectral fusion image is calculated using a domain adversarial training method. The difference loss is determined by the following formula: ; in, This indicates the number of samples in the general image dataset. This indicates the number of samples in the spectral fusion image. This represents the characteristics of the general image dataset. The features representing the spectral fusion image, This indicates the probability that a feature belongs to the current domain. The total loss function is determined based on the preset prior power loss and the difference loss; The discharge recognition model is obtained by performing transfer learning on the spectral fusion image based on the total loss function and the general image dataset.

[0137] A partial discharge identification device for power equipment, comprising: An image fusion module is used to perform feature fusion on pre-acquired multispectral images to obtain a spectral fusion image; wherein, the multispectral images include ultraviolet images, visible light images, and near-infrared images; The model training module is used to perform transfer learning based on a preset general image dataset and the spectral fusion image to obtain a discharge recognition model; The discharge recognition module is used to recognize the partial discharge image of the power equipment to be recognized according to the discharge recognition model, and obtain the discharge recognition result.

[0138] Optionally, the image fusion module specifically includes: The feature extraction module is used to extract features from the ultraviolet image, the visible light image and the near-infrared image to obtain ultraviolet discharge features, visible light structural features and near-infrared temperature rise features; The ultraviolet discharge characteristics are determined by the following formula: ; in, This represents the intensity probability diagram of the discharged photons. Represents pixels ( x,y The normalized value of the ultraviolet photon count. This represents the average photon count in the background region of an ultraviolet image. This represents the normalized value of the minimum detectable photon count for partial discharge. The visible light structural features are determined by the following formula: ; in, This represents the equipment structure weight diagram. This represents the structure mask of the power equipment. This represents the Gaussian gradient of a visible light image. This represents a normalization function that maps the result to the interval [0,1]. The near-infrared temperature rise characteristic is determined by the following formula: ; in, This represents the relative temperature rise coefficient. This represents the normalized near-infrared temperature value of pixel (x,y). This indicates the average temperature during normal operation of the equipment. Indicates the maximum allowable operating temperature of the equipment; The feature fusion module is used to perform weighted fusion of the ultraviolet discharge features, the visible light structure features, and the near-infrared temperature rise features according to preset image weights to obtain the spectral fusion image.

[0139] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any of the preceding descriptions.

[0140] A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform any of the methods described above.

[0141] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0142] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0143] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0144] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A method for identifying partial discharge in power equipment, comprising: Feature fusion is performed on pre-acquired multispectral images to obtain a spectral fusion image; wherein, the multispectral images include ultraviolet images, visible light images, and near-infrared images; A discharge recognition model is obtained by performing transfer learning on the preset general image dataset and the spectral fusion image; The discharge recognition model is used to identify the partial discharge image of the power equipment to be identified, and the discharge recognition result is obtained.

2. The method for identifying partial discharge in power equipment according to claim 1, wherein, The step of fusing features from a pre-acquired multispectral image to obtain a spectral fusion image specifically includes: Feature extraction is performed on the ultraviolet image, the visible light image, and the near-infrared image to obtain ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features; The ultraviolet discharge characteristics are determined by the following formula: ； in, This represents the intensity probability diagram of the discharged photons. Represents pixels ( x,y The normalized value of the ultraviolet photon count. This represents the average photon count in the background region of an ultraviolet image. This represents the normalized value of the minimum detectable photon count for partial discharge. The visible light structural features are determined by the following formula: ； in, This represents the equipment structure weight diagram. This represents the structure mask of the power equipment. This represents the Gaussian gradient of a visible light image. This represents a normalization function that maps the result to the interval [0,1]. The near-infrared temperature rise characteristic is determined by the following formula: ； in, This represents the relative temperature rise coefficient. This represents the normalized near-infrared temperature value of pixel (x,y). This indicates the average temperature during normal operation of the equipment. Indicates the maximum allowable operating temperature of the equipment; The ultraviolet discharge features, visible light structural features, and near-infrared temperature rise features are weighted and fused according to preset image allocation weights to obtain the spectral fusion image.

3. The method for identifying partial discharge in power equipment according to claim 2, wherein, The image assignment weights include ultraviolet weights, visible light weights, and near-infrared weights; Before performing weighted fusion of the ultraviolet discharge features, the visible light structural features, and the near-infrared temperature rise features according to preset image allocation weights to obtain the spectral fusion image, the method further includes: The ultraviolet weight is determined based on the ultraviolet discharge characteristics and the visible light structural characteristics; The ultraviolet weight is determined by the following formula: ； in, This represents the Sigmoid activation function. Represents the ultraviolet characteristic coefficient. Indicates the background suppression coefficient; The near-infrared weights are determined based on the near-infrared temperature rise characteristics and the ultraviolet discharge characteristics. The near-infrared weights are determined by the following formula: ； in, Represents the characteristic coefficient of temperature rise. Indicates the discharge correlation coefficient; The visible light weight is determined based on the ultraviolet weight and the near-infrared weight; The visible light weight is determined by the following formula: ； The ultraviolet weight, the visible light weight, and the near-infrared weight are solved according to the preset energy minimization constraint.

4. The method for identifying partial discharge in power equipment according to claim 3, wherein, The step of solving for the ultraviolet weight, the visible light weight, and the near-infrared weight according to the preset energy minimization constraint also includes: The ultraviolet feature fidelity term is determined based on the ultraviolet weight and the ultraviolet discharge characteristics; The ultraviolet feature fidelity term is determined by the following formula: ； The near-infrared feature fidelity term is determined based on the near-infrared weights and the near-infrared discharge characteristics. The near-infrared feature fidelity term is determined by the following formula: ； The spatial smoothing term is determined based on the ultraviolet weight and the infrared weight; The spatial smoothing term is determined by the following formula: ； The total energy function is determined based on the ultraviolet feature fidelity, the near-infrared feature fidelity term, the spatial smoothing term, and a preset spatial smoothing coefficient. ； in, This represents the spatial smoothing coefficient; The ultraviolet weight, the visible light weight, and the near-infrared weight are solved based on the total energy function.

5. The method for identifying partial discharge in power equipment according to claim 3, wherein, The step of identifying the partial discharge image of the power equipment to be identified based on the discharge identification model to obtain the discharge identification result specifically includes: The spectral fusion image, the ultraviolet weight, and the near-infrared weight corresponding to the partial discharge image of the power equipment to be identified are determined to obtain the spectral fusion identification image, the ultraviolet identification weight, and the near-infrared identification weight. The ultraviolet recognition weights and the near-infrared recognition weights are mapped to a pseudo-color heatmap, and the pseudo-color heatmap is superimposed on the spectral fusion recognition image to obtain a weight contribution heatmap. Based on the discharge identification results, the discharge region in the spectral fusion identification image is determined, and the discharge region is parameter-labeled to obtain a discharge location parameter map; The discharge identification results are visualized and output based on the weighted contribution heatmap and the discharge location parameter map.

6. The method for identifying partial discharge in power equipment according to claim 1, wherein, The step of performing transfer learning based on a preset general image dataset and the spectral fusion image to obtain a discharge recognition model specifically includes: The difference loss between the general image dataset and the spectral fusion image is calculated using a domain adversarial training method. The difference loss is determined by the following formula: ； in, This indicates the number of samples in the general image dataset. This indicates the number of samples in the spectral fusion image. This represents the characteristics of the general image dataset. The features representing the spectral fusion image, This indicates the probability that a feature belongs to the current domain. The total loss function is determined based on the preset prior power loss and the difference loss; The discharge recognition model is obtained by performing transfer learning on the spectral fusion image based on the total loss function and the general image dataset.

7. A partial discharge identification device for power equipment, comprising: An image fusion module is used to perform feature fusion on pre-acquired multispectral images to obtain a spectral fusion image; wherein, the multispectral images include ultraviolet images, visible light images, and near-infrared images; The model training module is used to perform transfer learning based on a preset general image dataset and the spectral fusion image to obtain a discharge recognition model; The discharge recognition module is used to recognize the partial discharge image of the power equipment to be recognized according to the discharge recognition model, and obtain the discharge recognition result.

8. The power equipment partial discharge identification device according to claim 6, wherein, The image fusion module specifically includes: The feature extraction module is used to extract features from the ultraviolet image, the visible light image and the near-infrared image to obtain ultraviolet discharge features, visible light structural features and near-infrared temperature rise features; The ultraviolet discharge characteristics are determined by the following formula: ； in, This represents the intensity probability diagram of the discharged photons. Represents pixels ( x,y The normalized value of the ultraviolet photon count. This represents the average photon count in the background region of an ultraviolet image. This represents the normalized value of the minimum detectable photon count for partial discharge. The visible light structural features are determined by the following formula: ； in, This represents the equipment structure weight diagram. This represents the structure mask of the power equipment. This represents the Gaussian gradient of a visible light image. This represents a normalization function that maps the result to the interval [0,1]. The near-infrared temperature rise characteristic is determined by the following formula: ； in, This represents the relative temperature rise coefficient. This represents the normalized near-infrared temperature value of pixel (x,y). This indicates the average temperature during normal operation of the equipment. Indicates the maximum allowable operating temperature of the equipment; The feature fusion module is used to perform weighted fusion of the ultraviolet discharge features, the visible light structure features, and the near-infrared temperature rise features according to preset image weights to obtain the spectral fusion image.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as claimed in any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1 to 6.

Images