Electrical equipment thermal defect assessment method

Abstract

The invention discloses a power equipment thermal defect assessment method, which belongs to the technical field of power equipment thermal defect assessment, and comprises the following steps: obtaining a surface temperature transient sequence of power equipment and corresponding microenvironment disturbance data; performing thermal response and environmental stress dynamic decoupling on the surface temperature transient sequence and the microenvironment disturbance data to obtain a thermal inertia characteristic spectrum; carrying out abnormal region clustering on the surface of the power equipment by adopting a graph neural network based on the thermal inertia characteristic spectrum, and identifying a potential defect region and a topological association relationship thereof; multi-modal signals of the potential defect area are collected in a targeted mode, and a defect evaluation result subjected to cross-modal verification is generated through fusion analysis. According to the method, the problem of evaluation errors caused by coupling interference of thermal response and environmental stress in thermal defect evaluation of the power equipment is solved.

Description

Technical Field

[0001] This invention relates to the field of thermal defect assessment technology for power equipment, and more particularly to a method for assessing thermal defects in power equipment. Background Technology

[0002] In recent years, the assessment of thermal defects in power equipment has mainly relied on single-mode temperature acquisition technologies such as infrared thermography and fiber optic sensing, supplemented by static threshold judgment or simple regression models based on historical data. For example, traditional methods use infrared thermal imagers to acquire the surface temperature distribution of equipment, and then manually correct it by combining it with micro-environmental parameters such as ambient temperature, humidity, and wind speed, or use a fixed thermal inertia coefficient model to fit the thermal response curve. Such methods can identify basic defects under steady-state environmental conditions, but they are limited by the strong coupling characteristics between dynamic environmental disturbances and thermal response. Micro-environmental disturbances such as gusts, sudden changes in sunlight, and equipment load fluctuations can directly change the heat exchange conditions on the equipment surface, resulting in a nonlinear deviation between the temperature measurement value and the actual thermal state, making it difficult to accurately reflect thermal anomalies caused by internal defects in the equipment.

[0003] The core flaw in existing technologies lies in the ineffective decoupling of thermal response from environmental stress. Specifically, this manifests as: insufficient adaptability to dynamic environments; micro-environmental disturbances directly alter the surface convection / radiation coefficients of equipment, leading to systematic evaluation errors in traditional static models or single-point correction methods. For example, a gust of wind might temporarily lower the surface temperature by 10-20°C, masking actual defect hotspots. Traditional methods, lacking dynamic decoupling capabilities, often misjudge such environmental disturbances as normal conditions. Furthermore, the extraction of thermal inertia features is lacking. Power equipment exhibits significant thermal inertia, with surface temperature changes lagging behind heat source changes caused by internal defects. Existing technologies often employ… Static thresholding or simple time-series averaging, without establishing a thermal inertia feature map, cannot distinguish between temperature fluctuations caused by environmental disturbances and continuous thermal anomalies caused by defects. This leads to early defects being missed due to thermal inertia smoothing or environmental interference being mistakenly identified as defects. Furthermore, the lack of multimodal correlation verification means that single temperature modal data is insufficient to cope with complex environmental coupling interference. Existing technologies lack targeted acquisition and cross-modal fusion verification mechanisms for multimodal signals. For example, relying solely on infrared thermometry is susceptible to environmental reflection interference, while combining acoustic signals can help distinguish between environmental noise and abnormal acoustic-thermal correlations caused by defects. However, the lack of multimodal collaborative analysis results in insufficient reliability of the evaluation results.

[0004] In summary, the evaluation error caused by the coupling interference between thermal response and environmental stress in the thermal defect assessment of power equipment is difficult to meet the engineering requirements of high accuracy and strong robustness. Summary of the Invention

[0005] The technical problem solved by this invention is the evaluation error caused by the coupling interference between thermal response and environmental stress in the thermal defect assessment of power equipment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] As a preferred embodiment of the thermal defect assessment method for power equipment described in this invention, wherein:

[0008] Acquire the transient sequence of surface temperature of power equipment and the corresponding micro-environmental disturbance data;

[0009] Thermal response and environmental stress dynamic decoupling were performed on the transient sequence of surface temperature and microenvironmental disturbance data to obtain thermal inertia characteristic maps;

[0010] Based on thermal inertial feature maps, graph neural networks are used to cluster abnormal regions on the surface of power equipment to identify potential defect regions and their topological relationships.

[0011] Multimodal signals from potential defect areas are collected in a targeted manner and generated through fusion analysis to produce defect assessment results that have been verified across modalities.

[0012] Furthermore, the transient sequence of surface temperature of the power equipment and the corresponding micro-environmental disturbance data are obtained, including:

[0013] The temperature distribution sequence on the surface of the power equipment was acquired using an infrared thermal imager;

[0014] The wind speed, light intensity, and ambient temperature data at the location of the power equipment are collected using environmental sensors.

[0015] The temperature distribution sequence is spatiotemporally aligned with the wind speed, light intensity, and ambient temperature data to generate a transient surface temperature sequence of the power equipment and corresponding micro-environmental disturbance data.

[0016] Furthermore, the step of dynamically decoupling the thermal response and environmental stress of the transient surface temperature sequence and microenvironmental disturbance data to obtain a thermal inertia characteristic map includes:

[0017] The transient sequence of surface temperature is decomposed into temperature change components caused by microenvironmental disturbances and temperature change components unrelated to microenvironmental disturbances;

[0018] Based on the temperature change components caused by microenvironmental disturbances and the temperature change components unrelated to microenvironmental disturbances, the thermal time constant and thermal sensitivity parameters of each part of the power equipment are calculated.

[0019] Based on the thermal time constant and thermal sensitivity parameters, a thermal inertial feature map characterizing the thermal dynamic characteristics of various parts of the power equipment is generated.

[0020] Furthermore, based on the temperature change components caused by microenvironmental disturbances and the temperature change components unrelated to microenvironmental disturbances, the thermal time constants and thermal sensitivity parameters of various parts of the power equipment are calculated, including:

[0021] Based on the temperature change component that is independent of microenvironmental disturbance, an exponential decay model is used for curve fitting to extract the thermal time constant of each part of the power equipment.

[0022] Calculate the partial derivatives of the temperature change component caused by microenvironmental disturbance with respect to wind speed and light intensity in the microenvironmental disturbance data, and use the statistical characteristics of the partial derivatives as the thermal sensitivity parameter.

[0023] Furthermore, based on the thermal time constant and thermal sensitivity parameters, a thermal inertial characteristic map characterizing the thermal dynamics of various parts of the power equipment is generated, including:

[0024] The thermal time constant and thermal sensitivity parameters are mapped according to spatial location to construct a two-dimensional thermal inertial feature matrix of the surface of the power equipment;

[0025] The thermal inertia feature matrix is ​​normalized to eliminate the dimensional differences between different parameters, resulting in a normalized feature matrix.

[0026] The normalized feature matrix is ​​converted into a pseudo-color image to generate a visualized thermal inertia feature map.

[0027] In this thermal inertial feature map, different colors represent different combinations of thermal dynamic characteristics of the power equipment.

[0028] Furthermore, based on thermal inertial feature maps, graph neural networks are used to cluster abnormal regions on the surface of power equipment to identify potential defect regions and their topological relationships, including:

[0029] The thermal inertial feature map is discretized into multiple nodes, each node corresponding to a detection area on the surface of the power equipment;

[0030] Using the thermal time constant and thermal sensitivity parameters as node features, a graph structure is constructed on the surface of the power equipment.

[0031] Based on spatial proximity and feature similarity, the edge connection relationships between nodes in the graph structure are established;

[0032] A graph attention network is used to cluster abnormal regions in the graph structure using deep learning to identify node clusters with similar abnormal features;

[0033] Based on the distribution characteristics of the node community, potential defect areas and their topological relationships are determined.

[0034] Furthermore, a graph attention network is used to cluster abnormal regions in the graph structure using deep learning, identifying node clusters with similar abnormal features, including:

[0035] Calculate the attention coefficient between each node in the graph structure and its neighboring nodes, the attention coefficient being determined based on the similarity between node features;

[0036] The features of neighboring nodes are weighted and aggregated based on the attention coefficients, and the feature representation of each node is updated.

[0037] The updated node features are input into a multilayer perceptron to calculate the probability that each node belongs to an abnormal region.

[0038] Based on the probability that each node belongs to an anomalous region and the spatial connectivity between nodes, a community detection algorithm is used to identify node clusters with similar anomalous characteristics.

[0039] Furthermore, based on the probability that each node belongs to an anomalous region and the spatial connectivity between nodes, a community detection algorithm is used to identify node clusters with similar anomalous characteristics, including:

[0040] Treat each node as an independent initial community and calculate the modularity gain between initial communities;

[0041] Based on the modularity gain, initial communities that simultaneously meet preset conditions are preferentially merged.

[0042] The preset conditions are: the probability of a node belonging to an abnormal region exceeds a preset probability threshold, and the weight of the edge between the corresponding nodes is greater than or equal to a preset weight threshold.

[0043] By iteratively optimizing the modularity function, initial communities that meet the preset conditions are continuously merged to obtain the optimal community partitioning structure;

[0044] Based on the optimal community partitioning structure, each initially partitioned community is identified as a cluster of nodes with similar anomalous characteristics.

[0045] Furthermore, multimodal signals from potential defect areas are targeted and collected, and fusion analysis is used to generate cross-modal validated defect assessment results, including:

[0046] In the potential defect area corresponding to the identified node cluster, the partial discharge ultrasonic signal, mechanical vibration spectrum signal and leakage current high frequency signal of the potential defect area are collected.

[0047] The PRPD spectrum features of the partial discharge ultrasonic signal, the resonant frequency shift features of the mechanical vibration spectrum signal, and the harmonic distortion rate features of the high-frequency leakage current signal are extracted respectively.

[0048] Based on the DS evidence theory, information fusion is performed on the PRPD spectrum features, the resonance frequency shift features, and the harmonic distortion rate features to calculate the fusion confidence of various defects;

[0049] Based on the fusion confidence level, a defect assessment result containing defect type and defect severity classification is generated.

[0050] Furthermore, based on DS evidence theory, information fusion is performed on the PRPD spectrum features, the resonance frequency shift features, and the harmonic distortion rate features to calculate the fusion confidence of various defects, including:

[0051] Construct a diagnostic proposition set that includes poor contact, insulation degradation, mechanical loosening, and arc discharge;

[0052] Based on the PRPD spectral features, a first basic probability allocation function is determined. The first basic probability allocation function is used to characterize the degree of support of ultrasound evidence for each proposition in the diagnostic proposition set.

[0053] Based on the resonant frequency shift characteristics, a second basic probability assignment function is determined. The second basic probability assignment function is used to characterize the degree of support of vibration evidence for each proposition in the diagnostic proposition set.

[0054] Based on the harmonic distortion rate characteristics, a third basic probability allocation function is determined, which is used to characterize the degree of support of electrical evidence for each proposition in the diagnostic proposition set.

[0055] The Dempster combination rule is used to combine the evidence of the first basic probability assignment function, the second basic probability assignment function and the third basic probability assignment function to obtain the fused diagnostic proposition set.

[0056] Calculate the confidence function value of each proposition in the fused diagnostic proposition set, and use the confidence function value as the fusion confidence of the corresponding defect.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] 1. This invention generates a thermal inertia feature map by dynamically decoupling thermal response and environmental stress, combines it with graph neural network to perform topological correlation clustering of abnormal areas on the surface of power equipment, and verifies it through multimodal signal targeted acquisition and cross-modal fusion. This effectively solves the evaluation error problem caused by the coupling interference between thermal response and environmental stress, and realizes high-precision and robust evaluation of thermal defects in power equipment.

[0059] 2. This invention decomposes the transient surface temperature sequence into micro-environmental disturbance-related and unrelated components, and calculates the thermal time constant and thermal sensitivity parameters to generate a thermal inertia feature spectrum. This achieves precise decoupling of thermal response and environmental stress, and can quantitatively distinguish between transient temperature fluctuations caused by environmental disturbances and continuous thermal anomalies caused by defects. This avoids the problems of early defect omission and environmental interference misjudgment caused by thermal inertia in traditional static threshold methods.

[0060] 3. This invention constructs a surface graph structure of equipment based on thermal inertial feature maps, and uses graph attention network combined with community detection algorithm to cluster abnormal regions. It can accurately identify node communities with similar abnormal features and their spatial topological relationships, breaking through the limitations of traditional single-point analysis and realizing global and accurate positioning of defect region distribution characteristics and relationships.

[0061] 4. This invention targets and collects partial discharge ultrasonic signals, mechanical vibration spectrum signals, and high-frequency leakage current signals in potential defect areas. By fusing features such as PRPD spectrum, resonant frequency shift, and harmonic distortion rate through DS evidence theory and calculating the fusion confidence level, it achieves cross-modal collaborative verification of defect type and severity, significantly improving the credibility and engineering applicability of the assessment results. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the basic process of a method for assessing thermal defects in power equipment, provided as an embodiment of the present invention. Detailed Implementation

[0063] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0064] Example 1

[0065] like Figure 1 As shown in the figure, this embodiment introduces a method for assessing thermal defects in power equipment, including:

[0066] Step 1: Obtain the transient sequence of surface temperature of the power equipment and the corresponding micro-environmental disturbance data.

[0067] This invention uses an infrared thermal imager and environmental sensors to simultaneously collect surface temperature distribution sequences of power equipment and micro-environmental disturbance data such as wind speed, light intensity, and ambient temperature. After spatiotemporal alignment processing, a transient surface temperature sequence and corresponding disturbance data are generated, ensuring the spatiotemporal consistency and multi-parameter correlation of the basic data. This provides high-precision and traceable raw data support for subsequent dynamic decoupling and avoids analytical bias caused by data mismatch.

[0068] Step 2: Dynamically decouple the thermal response and environmental stress of the transient surface temperature sequence and microenvironmental disturbance data to obtain the thermal inertia characteristic spectrum.

[0069] This invention decouples the transient surface temperature sequence into micro-environmental disturbance-related and unrelated components. By quantifying the thermal time constant, which reflects thermal inertia characteristics, and the thermal sensitivity parameter, which reflects the intensity of environmental stress response, a visualized thermal inertia feature map is generated. This achieves precise dynamic decoupling of thermal response and environmental stress, effectively distinguishing between transient environmental disturbances and continuous thermal anomalies caused by defects, and overcoming the problem of missed or false judgments caused by thermal inertia in traditional static threshold methods.

[0070] Step 3: Based on the thermal inertial feature map, a graph neural network is used to cluster abnormal areas on the surface of the power equipment to identify potential defect areas and their topological relationships.

[0071] This invention constructs a surface graph structure of a device based on thermal inertial feature maps, uses graph attention networks combined with community detection algorithms to cluster abnormal regions, identifies node clusters with similar abnormal features through node feature similarity and spatial proximity, and determines the topological association of potential defect regions. This breaks through the limitations of traditional single-point analysis and achieves global and accurate localization and association analysis of defect region distribution characteristics and associations.

[0072] Step 4: Targeted acquisition of multimodal signals from potential defect areas and generation of cross-modal validated defect assessment results through fusion analysis.

[0073] This invention targets and collects multimodal signals such as partial discharge ultrasound, mechanical vibration spectrum, and leakage current in potential defect areas, extracts features such as PRPD spectrum, resonant frequency shift, and harmonic distortion rate, and performs cross-modal information fusion and calculates fusion confidence through DS evidence theory. This enables cross-modal collaborative verification of defect type and severity, significantly improving the credibility and engineering applicability of defect assessment results, and meeting the requirements for high-precision and robust assessment.

[0074] Example 2

[0075] Similar to the inventive concept of Embodiment 1, this embodiment describes the implementation steps of a method for assessing thermal defects in power equipment, including:

[0076] Step 1: Obtain the transient sequence of surface temperature of the power equipment and the corresponding micro-environmental disturbance data.

[0077] Step 1.1: Collect the temperature distribution sequence on the surface of the power equipment using an infrared thermal imager.

[0078] This embodiment uses an infrared thermal imager to non-contactly acquire the surface temperature distribution sequence of power equipment. Utilizing its high spatial resolution and real-time imaging characteristics, it accurately captures subtle temperature differences and hot spot distribution on the equipment surface, providing high-precision, full-view basic temperature data for subsequent thermal response analysis, and avoiding equipment status interference or measurement blind spots that may be caused by contact temperature measurement.

[0079] Step 1.2: Collect wind speed, light intensity, and ambient temperature data at the location of the power equipment using environmental sensors.

[0080] This embodiment synchronously collects micro-environmental disturbance data such as wind speed, light intensity, and ambient temperature through environmental sensors, quantifies the changes in thermal convection / radiation conditions of the environment in which the equipment is located, provides key environmental stress parameters for the dynamic decoupling of thermal response and environmental stress, and ensures the traceability and controllability of environmental factors in the decoupling analysis.

[0081] Step 1.3: Perform spatiotemporal alignment processing on the temperature distribution sequence with the wind speed, light intensity and ambient temperature data to generate the transient surface temperature sequence of the power equipment and the corresponding micro-environmental disturbance data.

[0082] This embodiment uses spatiotemporal alignment processing to precisely match the temperature distribution sequence with the microenvironmental disturbance data in the time dimension, such as the synchronization of sampling timestamps, and in the spatial dimension, such as the mapping of coordinates on the surface of power equipment. This generates a surface temperature transient sequence with strong spatiotemporal consistency and corresponding microenvironmental disturbance data, eliminating spatiotemporal misalignment errors in the data acquisition process and laying a high-confidence data foundation for subsequent thermal inertia feature map generation and dynamic decoupling analysis.

[0083] Step 2: Dynamically decouple the thermal response and environmental stress of the transient surface temperature sequence and microenvironmental disturbance data to obtain the thermal inertia characteristic map.

[0084] Step 2.1: Decompose the transient surface temperature sequence into temperature change components caused by microenvironmental disturbances and temperature change components unrelated to microenvironmental disturbances.

[0085] This embodiment decomposes the transient surface temperature sequence into micro-environmental disturbance-related and unrelated components, achieving preliminary physical separation between thermal response and environmental stress. This eliminates the direct impact of transient environmental disturbances on the surface temperature of power equipment, providing interference-free benchmark data for the subsequent accurate calculation of thermal inertia parameters, and avoiding analytical biases caused by the coupling of environmental factors and equipment thermal response in traditional methods.

[0086] Step 2.2: Based on the temperature change components caused by microenvironmental disturbances and the temperature change components unrelated to microenvironmental disturbances, calculate the thermal time constant and thermal sensitivity parameters of each part of the power equipment.

[0087] This embodiment quantifies the thermal time constant and thermal sensitivity parameters based on the decomposed temperature components. By fitting an exponential decay model and extracting statistical features of partial derivatives, it achieves a numerical characterization of thermal dynamics, providing a quantifiable and comparable parameter basis for the generation of thermal inertia feature maps, and overcoming the problem of fuzzy thermal inertia parameters in traditional empirical models.

[0088] Step 2.2.1: Based on the temperature change component that is independent of microenvironmental disturbance, use an exponential decay model to perform curve fitting and extract the thermal time constant of each part of the power equipment.

[0089] This embodiment uses an exponential decay model to fit the temperature change components that are independent of microenvironmental disturbances, accurately extracts the thermal time constants of each part, and quantitatively characterizes the response delay characteristics of the equipment surface temperature change to the internal defect heat source, providing key time dimension parameter support for distinguishing between instantaneous environmental disturbances and continuous thermal anomalies of defects.

[0090] Step 2.2.2: Calculate the partial derivatives of the temperature change component caused by microenvironmental disturbance with respect to the wind speed and light intensity in the microenvironmental disturbance data, and use the statistical characteristics of the partial derivatives as the thermal sensitivity parameter.

[0091] This embodiment calculates the partial derivatives of the temperature change component caused by micro-environmental disturbance with respect to wind speed / light intensity, and extracts its statistical characteristics as a thermal sensitivity parameter. This quantitatively characterizes the sensitivity of the equipment surface heat exchange conditions to environmental disturbances, providing numerical basis for the visualization of environmental stress response characteristics in thermal inertia feature maps.

[0092] Step 2.3: Based on the thermal time constant and thermal sensitivity parameters, generate a thermal inertia characteristic map that characterizes the thermal dynamic properties of various parts of the power equipment.

[0093] This embodiment constructs a two-dimensional thermal inertia feature matrix based on the spatial mapping of thermal time constant and thermal sensitivity parameters. The difference in parameter dimensions is eliminated through normalization processing, and the matrix is ​​converted into a pseudo-color image to generate a visualized thermal inertia feature map. This enables an intuitive spatial distribution display of the thermal dynamic characteristics of various parts of the equipment surface, assists in the identification of thermal inertia anomaly areas and environmentally sensitive areas, and improves the intuitiveness and analysis efficiency of defect location.

[0094] Step 2.3.1: Map the thermal time constant and thermal sensitivity parameters according to spatial location to construct a two-dimensional thermal inertia feature matrix of the surface of the power equipment.

[0095] This embodiment constructs a two-dimensional matrix by mapping the thermal time constant and thermal sensitivity parameters according to the spatial position on the equipment surface, realizing the spatial structured integration of thermal dynamic characteristic parameters, providing feature inputs with spatial topological relationships for subsequent graph neural network analysis, and enhancing the accuracy of defect region clustering and association analysis.

[0096] Step 2.3.2: Normalize the thermal inertia feature matrix to eliminate the dimensional differences between different parameters and obtain the normalized feature matrix.

[0097] This embodiment normalizes the thermal inertia feature matrix to eliminate the dimensional differences between the thermal time constant and the thermal sensitivity parameter, ensuring that different parameters are comparable in the feature spectrum, avoiding analytical bias caused by dimensional inconsistencies, and improving the numerical stability and reliability of the feature spectrum.

[0098] Step 2.3.3: Convert the normalized feature matrix into a pseudo-color image to generate a visualized thermal inertia feature map.

[0099] This embodiment converts the normalized feature matrix into a pseudo-color image, and uses color differences to intuitively represent the combination of thermal dynamic characteristics of different parts of the power equipment, realizing the visualization of thermal inertia characteristics. This helps maintenance personnel to quickly locate potential defect areas and identify abnormal thermal characteristic patterns, improving the intuitiveness and interpretability of defect assessment.

[0100] In this thermal inertial feature map, different colors represent different combinations of thermal dynamic characteristics of the power equipment.

[0101] Step 3: Based on the thermal inertial feature map, a graph neural network is used to cluster abnormal areas on the surface of the power equipment to identify potential defect areas and their topological relationships.

[0102] Step 3.1: Discretize the thermal inertial feature map into multiple nodes, each node corresponding to a detection area on the surface of the power equipment.

[0103] This embodiment discretizes the thermal inertial feature map into nodes corresponding to the detection area on the device surface, realizing a structured transformation of thermal dynamic characteristics from continuous images to discrete nodes. This provides computable node units for graph neural networks, ensuring that each node can accurately associate with the thermal inertial features of a specific area on the device surface, such as thermal time constant / thermal sensitivity parameters, supporting the accuracy of subsequent graph structure construction and abnormal region clustering analysis.

[0104] Step 3.2: Using the thermal time constant and thermal sensitivity parameters as node features, construct a graph structure of the power equipment surface.

[0105] This embodiment uses thermal time constant and thermal sensitivity parameters as node features to construct a graph structure containing information on the thermal dynamic characteristics of the equipment surface. This allows the node features of the graph structure to directly reflect the essential thermal response characteristics of each part of the equipment, providing an interpretable and quantifiable feature input basis for deep learning clustering of graph neural networks and enhancing the physical meaning correlation of abnormal area identification.

[0106] Step 3.3: Based on spatial proximity and feature similarity, establish the edge connection relationships between nodes in the graph structure.

[0107] This embodiment establishes edge connections between nodes based on spatial proximity relationships that reflect physical location associations and feature similarity that reflects the consistency of thermal dynamic characteristics. This forms a graph structure that conforms to both the spatial topology of the equipment surface and reflects the similarity of thermal characteristics. The edge connections integrate information from both spatial location and thermal characteristics, improving the graph structure's ability to represent the distribution patterns of thermal defects on the equipment surface and providing more reasonable network topology support for the clustering analysis of graph neural networks.

[0108] Step 3.4: Use a graph attention network to cluster abnormal regions in the graph structure using deep learning to identify node clusters with similar abnormal features.

[0109] This embodiment employs a graph attention network to perform deep learning clustering on graph structures. By adaptively learning the attention weights between nodes and aggregating neighborhood features, it effectively identifies node clusters with similar anomalous characteristics, such as high thermal sensitivity accompanied by a short thermal time constant. This achieves accurate clustering from local node anomalies to global regional defects, breaking through the limitations of traditional single-point analysis, accurately locating potential defect regions and revealing their topological relationships, thereby improving the global and correlational analysis capabilities of thermal defect assessment.

[0110] Step 3.4.1: Calculate the attention coefficient between each node in the graph structure and its neighboring nodes. The attention coefficient is determined based on the similarity between node features.

[0111] This embodiment calculates the attention coefficient through node feature similarity, realizing adaptive quantization of the association strength between nodes in the graph structure. This allows nodes with similar thermal dynamic characteristics to receive higher weights, improving the graph neural network's ability to capture abnormal feature association patterns and avoiding the problem of insufficient sensitivity to feature similarity in traditional fixed-weight models.

[0112] Step 3.4.2: Perform weighted aggregation of the features of neighboring nodes based on the attention coefficients, and update the feature representation of each node.

[0113] This embodiment uses attention coefficients to weighted aggregate the features of neighboring nodes, thereby achieving local information integration and dynamic updating of node features. This enhances the ability of node features to represent the thermal dynamic characteristics of the surrounding area, provides richer feature inputs for the anomaly probability calculation of the multilayer perceptron, and improves the accuracy of anomaly region identification.

[0114] Step 3.4.3: Input the updated node features into the multilayer perceptron and calculate the probability that each node belongs to an abnormal region.

[0115] In this embodiment, the updated node features are input into a multilayer perceptron to calculate the anomaly probability. By leveraging the nonlinear mapping capability of the deep learning model, the probability of each node belonging to an abnormal region is accurately quantified, providing a reliable probability input for subsequent community detection algorithms and ensuring the accuracy and interpretability of node community identification.

[0116] Step 3.4.4: Based on the probability that each node belongs to an anomalous region and the spatial connectivity between nodes, a community detection algorithm is used to identify node clusters with similar anomalous characteristics, including:

[0117] Treat each node as an independent initial community and calculate the modularity gain between initial communities.

[0118] Based on the modularity gain, initial communities that simultaneously meet preset conditions are preferentially merged.

[0119] The preset conditions are: the probability that a node belongs to an abnormal region exceeds a preset probability threshold, and the weight of the edge between the corresponding nodes is greater than or equal to a preset weight threshold.

[0120] By iteratively optimizing the modularity function, initial communities that meet the preset conditions are continuously merged to obtain the optimal community partitioning structure.

[0121] Based on the optimal community partitioning structure, each initially partitioned community is identified as a cluster of nodes with similar anomalous characteristics.

[0122] This embodiment combines node anomaly probability with spatial connectivity, uses a community detection algorithm to identify node clusters with similar anomaly features, and achieves topological association clustering of anomaly regions through modularity optimization. This breaks through the limitations of traditional single-point analysis, accurately locates the distribution range and association relationships of potential defect regions, and improves the global and correlation analysis capabilities of thermal defect assessment.

[0123] Step 3.5: Based on the distribution characteristics of the node community, determine the potential defect areas and their topological relationships.

[0124] This embodiment, based on the distribution characteristics of node communities, such as spatial clustering and consistency of thermal dynamic characteristics, accurately determines the physical boundaries and topological relationships of potential defect areas, realizing the correlation mapping from local node anomalies to global area defects. Through spatial distribution pattern analysis of thermal time constants and thermal sensitivity parameters, the differences in thermal characteristics between defective and normal areas can be revealed, and the spatial adjacency or functional coupling relationships between defective areas can be identified. This provides high-precision spatial topological information support for the cause analysis, impact range assessment, and maintenance strategy formulation of equipment thermal defects, significantly improving the globality and interpretability of thermal defect assessment.

[0125] Step 4: Target and collect multimodal signals from potential defect areas and generate cross-modal validated defect assessment results through fusion analysis.

[0126] Step 4.1: In the potential defect area corresponding to the identified node cluster, collect the partial discharge ultrasonic signal, mechanical vibration spectrum signal and leakage current high frequency signal of the potential defect area.

[0127] This embodiment achieves precise allocation of detection resources by targeting and acquiring partial discharge ultrasonic signals, mechanical vibration spectrum signals, and high-frequency leakage current signals in potential defect areas. This avoids invalid data acquisition in non-defect areas, significantly improves detection efficiency and data quality, and provides high signal-to-noise ratio and multi-dimensional basic data support for subsequent cross-modal feature fusion, thereby enhancing the comprehensiveness of defect assessment.

[0128] Step 4.2: Extract the PRPD spectrum features of the partial discharge ultrasonic signal, the resonant frequency shift features of the mechanical vibration spectrum signal, and the harmonic distortion rate features of the high-frequency leakage current signal, respectively.

[0129] This embodiment extracts PRPD spectrum features reflecting partial discharge modes for identifying arc discharge / insulation degradation defects, resonant frequency shift features reflecting changes in the dynamic characteristics of mechanical structures for identifying mechanical loosening / poor contact defects, and harmonic distortion rate features reflecting electrical signal anomalies for identifying poor contact / insulation aging defects. This achieves feature decoupling of multi-mode signals, provides comparable and calculable feature inputs for DS evidence theory fusion, and improves the pertinence and accuracy of defect type identification.

[0130] Step 4.3: Based on the DS evidence theory, information fusion is performed on the PRPD spectrum features, the resonance frequency shift features, and the harmonic distortion rate features to calculate the fusion confidence of various defects.

[0131] This embodiment integrates information from features such as PRPD spectrum, resonant frequency shift, and harmonic distortion rate based on DS evidence theory. It quantifies the degree of support of each modal evidence for diagnostic propositions such as poor contact / insulation degradation through the basic probability allocation function, and uses Dempster combination rules to achieve multi-evidence collaborative reasoning. This effectively handles the uncertainty and conflict between modal information, generates defect diagnosis results with higher fusion confidence, and significantly improves the robustness and reliability of defect type judgment.

[0132] Step 4.3.1: Construct a set of diagnostic propositions that include poor contact, insulation degradation, mechanical loosening, and arc discharge.

[0133] This embodiment constructs a diagnostic proposition set including poor contact, insulation degradation, mechanical loosening, and arc discharge, clarifying the defect type classification framework. This provides standardized diagnostic targets for subsequent evidence allocation and fusion, avoiding diagnostic bias caused by ambiguous type definitions, and ensuring the comparability, interpretability, and accurate correspondence between the evaluation results and actual engineering defect types.

[0134] Step 4.3.2: Based on the PRPD spectral features, determine the first basic probability allocation function, which is used to characterize the degree of support of ultrasound evidence for each proposition in the diagnostic proposition set.

[0135] This embodiment determines the first basic probability allocation function based on PRPD spectral features, quantifies the degree of support of ultrasound evidence for each diagnostic proposition, and utilizes the advantages of PRPD spectral features in partial discharge pattern recognition to accurately characterize the ultrasound features of arc discharge / insulation degradation defects, thereby improving the pertinence and specificity of ultrasound evidence in defect type judgment.

[0136] Step 4.3.3: Based on the resonant frequency shift characteristics, determine the second basic probability allocation function, which is used to characterize the degree of support of vibration evidence for each proposition in the diagnostic proposition set.

[0137] This embodiment determines the second basic probability allocation function based on the resonant frequency shift characteristics, quantifies the degree of support of vibration evidence for each diagnostic proposition, and utilizes the sensitivity of resonant frequency shift to changes in the dynamic characteristics of mechanical structures to accurately identify the vibration characteristics of mechanical loosening / poor contact defects, thereby enhancing the effectiveness of vibration evidence in the judgment of mechanical defects.

[0138] Step 4.3.4: Based on the harmonic distortion rate characteristics, determine the third basic probability allocation function, which is used to characterize the degree of support of electrical evidence for each proposition in the diagnostic proposition set.

[0139] This embodiment determines the third basic probability allocation function based on harmonic distortion rate characteristics, quantifies the degree of support of electrical evidence for each diagnostic proposition, utilizes the characterization ability of harmonic distortion rate for electrical signal anomalies, accurately identifies electrical characteristics of poor contact / insulation aging defects, and improves the reliability of electrical evidence in the judgment of electrical defects.

[0140] Step 4.3.5: Use Dempster's combination rule to combine the evidence of the first basic probability allocation function, the second basic probability allocation function and the third basic probability allocation function to obtain the fused diagnostic proposition set.

[0141] This embodiment uses the Dempster combination rule to combine evidence from three basic probability assignment functions. This rule handles the uncertainty and conflict between multiple pieces of evidence, realizes collaborative reasoning of ultrasonic, vibration and electrical evidence, effectively integrates multimodal information, and improves the robustness, anti-interference ability and adaptability to complex defect scenarios of the fusion results.

[0142] Step 4.3.6: Calculate the confidence function value of each proposition in the fused diagnostic proposition set, and use the confidence function value as the fusion confidence of the corresponding defect.

[0143] This embodiment calculates the trust function value of each proposition in the fused diagnostic proposition set and uses it as the fusion confidence level. The trust function quantifies the comprehensive credibility of each defect type, providing a quantitative basis for defect type judgment, ensuring the credibility, interpretability and direct correlation of the assessment results with maintenance decisions, and supporting defect severity classification and maintenance strategy formulation.

[0144] Step 4.4: Generate defect assessment results that include defect type and defect severity classification based on the fusion confidence level.

[0145] This embodiment generates assessment results based on fusion confidence, including defect types such as poor contact / insulation degradation and severity classification, to achieve quantitative characterization and visual output of defect status. This provides maintenance personnel with clear defect location, type diagnosis and maintenance priority information, directly guiding the formulation and implementation of equipment maintenance strategies. It meets the high-precision and highly interpretable engineering requirements of thermal defect assessment of power equipment and enhances the practical application value of defect assessment results.

[0146] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for assessing thermal defects in power equipment, characterized in that, include: Acquire the transient sequence of surface temperature of power equipment and the corresponding micro-environmental disturbance data; Thermal response and environmental stress dynamic decoupling were performed on the transient sequence of surface temperature and microenvironmental disturbance data to obtain thermal inertia characteristic maps; Based on thermal inertial feature maps, graph neural networks are used to cluster abnormal regions on the surface of power equipment to identify potential defect regions and their topological relationships. Multimodal signals from potential defect areas are collected in a targeted manner and generated through fusion analysis to produce defect assessment results that have been verified across modalities.

2. The method for assessing thermal defects in power equipment as described in claim 1, characterized in that, Acquire the transient sequence of surface temperature of power equipment and the corresponding micro-environmental disturbance data, including: The temperature distribution sequence on the surface of the power equipment was acquired using an infrared thermal imager; The wind speed, light intensity, and ambient temperature data at the location of the power equipment are collected using environmental sensors. The temperature distribution sequence is spatiotemporally aligned with the wind speed, light intensity, and ambient temperature data to generate a transient surface temperature sequence of the power equipment and corresponding micro-environmental disturbance data.

3. The method for assessing thermal defects in power equipment as described in claim 2, characterized in that, The process of dynamically decoupling the thermal response and environmental stress of the transient surface temperature sequence and microenvironmental disturbance data to obtain a thermal inertia characteristic map includes: The transient sequence of surface temperature is decomposed into temperature change components caused by microenvironmental disturbances and temperature change components unrelated to microenvironmental disturbances; Based on the temperature change components caused by microenvironmental disturbances and the temperature change components unrelated to microenvironmental disturbances, the thermal time constant and thermal sensitivity parameters of each part of the power equipment are calculated. Based on the thermal time constant and thermal sensitivity parameters, a thermal inertial feature map characterizing the thermal dynamic characteristics of various parts of the power equipment is generated.

4. The method for assessing thermal defects in power equipment as described in claim 3, characterized in that, Based on the temperature change components caused by microenvironmental disturbances and the temperature change components unrelated to microenvironmental disturbances, the thermal time constant and thermal sensitivity parameters of various parts of the power equipment are calculated, including: Based on the temperature change component that is independent of microenvironmental disturbance, an exponential decay model is used for curve fitting to extract the thermal time constant of each part of the power equipment. Calculate the partial derivatives of the temperature change component caused by microenvironmental disturbance with respect to wind speed and light intensity in the microenvironmental disturbance data, and use the statistical characteristics of the partial derivatives as the thermal sensitivity parameter.

5. The method for assessing thermal defects in power equipment as described in claim 4, characterized in that, Based on the thermal time constant and thermal sensitivity parameters, a thermal inertial characteristic map characterizing the thermal dynamics of various parts of the power equipment is generated, including: The thermal time constant and thermal sensitivity parameters are mapped according to spatial location to construct a two-dimensional thermal inertial feature matrix of the surface of the power equipment; The thermal inertia feature matrix is ​​normalized to eliminate the dimensional differences between different parameters, resulting in a normalized feature matrix. The normalized feature matrix is ​​converted into a pseudo-color image to generate a visualized thermal inertia feature map. In this thermal inertial feature map, different colors represent different combinations of thermal dynamic characteristics of the power equipment.

6. The method for assessing thermal defects in power equipment as described in claim 5, characterized in that, Based on thermal inertial feature maps, a graph neural network is used to cluster abnormal regions on the surface of power equipment to identify potential defect regions and their topological relationships, including: The thermal inertial feature map is discretized into multiple nodes, each node corresponding to a detection area on the surface of the power equipment; Using the thermal time constant and thermal sensitivity parameters as node features, a graph structure is constructed on the surface of the power equipment. Based on spatial proximity and feature similarity, the edge connection relationships between nodes in the graph structure are established; A graph attention network is used to cluster abnormal regions in the graph structure using deep learning to identify node clusters with similar abnormal features; Based on the distribution characteristics of the node community, potential defect areas and their topological relationships are determined.

7. The method for assessing thermal defects in power equipment as described in claim 6, characterized in that, A graph attention network is used to cluster abnormal regions in the graph structure using deep learning, identifying node clusters with similar abnormal features, including: Calculate the attention coefficient between each node in the graph structure and its neighboring nodes, the attention coefficient being determined based on the similarity between node features; The features of neighboring nodes are weighted and aggregated based on the attention coefficients, and the feature representation of each node is updated. The updated node features are input into a multilayer perceptron to calculate the probability that each node belongs to an abnormal region. Based on the probability that each node belongs to an anomalous region and the spatial connectivity between nodes, a community detection algorithm is used to identify node clusters with similar anomalous characteristics.

8. The method for assessing thermal defects in power equipment as described in claim 7, characterized in that, Based on the probability that each node belongs to an anomalous region and the spatial connectivity between nodes, a community detection algorithm is used to identify node clusters with similar anomalous characteristics, including: Treat each node as an independent initial community and calculate the modularity gain between initial communities; Based on the modularity gain, initial communities that simultaneously meet preset conditions are preferentially merged. The preset conditions are: the probability of a node belonging to an abnormal region exceeds a preset probability threshold, and the weight of the edge between the corresponding nodes is greater than or equal to a preset weight threshold. By iteratively optimizing the modularity function, initial communities that meet the preset conditions are continuously merged to obtain the optimal community partitioning structure; Based on the optimal community partitioning structure, each initially partitioned community is identified as a cluster of nodes with similar anomalous characteristics.

9. The method for assessing thermal defects in power equipment as described in claim 8, characterized in that, Multimodal signals from potential defect regions are targeted and collected, and fusion analysis is used to generate cross-modal validated defect assessment results, including: In the potential defect area corresponding to the identified node cluster, the partial discharge ultrasonic signal, mechanical vibration spectrum signal and leakage current high frequency signal of the potential defect area are collected. The PRPD spectrum features of the partial discharge ultrasonic signal, the resonant frequency shift features of the mechanical vibration spectrum signal, and the harmonic distortion rate features of the high-frequency leakage current signal are extracted respectively. Based on the DS evidence theory, information fusion is performed on the PRPD spectrum features, the resonance frequency shift features, and the harmonic distortion rate features to calculate the fusion confidence of various defects; Based on the fusion confidence level, a defect assessment result containing defect type and defect severity classification is generated.

10. The method for assessing thermal defects in power equipment as described in claim 9, characterized in that, Based on the DS evidence theory, information fusion is performed on the PRPD spectrum features, the resonance frequency shift features, and the harmonic distortion rate features to calculate the fusion confidence of various defects, including: Construct a diagnostic proposition set that includes poor contact, insulation degradation, mechanical loosening, and arc discharge; Based on the PRPD spectral features, a first basic probability allocation function is determined. The first basic probability allocation function is used to characterize the degree of support of ultrasound evidence for each proposition in the diagnostic proposition set. Based on the resonant frequency shift characteristics, a second basic probability assignment function is determined. The second basic probability assignment function is used to characterize the degree of support of vibration evidence for each proposition in the diagnostic proposition set. Based on the harmonic distortion rate characteristics, a third basic probability allocation function is determined, which is used to characterize the degree of support of electrical evidence for each proposition in the diagnostic proposition set. The Dempster combination rule is used to combine the evidence of the first basic probability assignment function, the second basic probability assignment function and the third basic probability assignment function to obtain the fused diagnostic proposition set. Calculate the confidence function value of each proposition in the fused diagnostic proposition set, and use the confidence function value as the fusion confidence of the corresponding defect.

Images