Partial discharge characteristic diagnosis positioning method and system for power transmission equipment

Abstract

The application provides a partial discharge feature diagnosis positioning method and system of a power transmission device, relates to the field of power device fault detection, and comprises the following steps: collecting a plurality of physical field signals through a distributed sensing array, performing cross-domain cross-correlation calculation to extract a coupling feature vector, mapping to a feature space to determine a discharge type and a severity level; positioning a discharge source three-dimensional coordinate according to a signal energy space attenuation gradient and an arrival time; establishing a discharge source file to record features and coordinates; and calculating a space migration rate to generate a warning signal. The application realizes accurate diagnosis and precise positioning of partial discharge, and improves the safe operation level of the power transmission device.

Description

Technical Field

[0001] This invention relates to power equipment fault detection technology, and more particularly to a method and system for diagnosing and locating partial discharge characteristics in power transmission equipment. Background Technology

[0002] As a core component of the power system, the operating status of power transmission equipment directly affects the reliability and security of power supply. Therefore, accurate diagnosis and location of partial discharge in power transmission equipment is of great significance for preventing major power accidents and improving the reliability of power grid operation.

[0003] Currently, partial discharge detection technologies for power transmission equipment mainly include ultrasonic testing, ultra-high frequency testing, extra-high frequency testing, and transient ground voltage detection. These methods collect signals such as sound waves and electromagnetic waves generated by partial discharges and analyze their characteristic parameters to determine the type and location of the discharge.

[0004] Existing technologies typically rely on single physical field signals for partial discharge characteristic analysis, failing to fully utilize the correlation between multiple physical field signals. This results in limited diagnostic accuracy, particularly in complex electromagnetic environments where they are susceptible to interference, making it difficult to accurately distinguish different types of discharge sources and their severity. Traditional location methods mostly employ time-difference positioning principles, but they do not consider the directional characteristics of signals during propagation or effectively compensate for the signal propagation path. This leads to low location accuracy in complex power transmission equipment structures, making precise three-dimensional spatial coordinate positioning difficult. Furthermore, the lack of effective management and utilization mechanisms for historical discharge source data makes it impossible to track the dynamic characteristics of the same discharge source over time or predict the development trend of potential faults, hindering early warning and risk assessment of insulation defects in power transmission equipment. Summary of the Invention

[0005] This invention provides a method and system for diagnosing and locating partial discharge characteristics in power transmission equipment, which can solve the problems in the prior art.

[0006] A first aspect of the present invention provides a method for diagnosing and locating partial discharge characteristics in power transmission equipment, comprising:

[0007] Multi-physical field signals of power transmission equipment are collected by a distributed sensor array. Cross-domain cross-correlation is performed on the multi-physical field signals to extract phase difference and amplitude ratio features, which are then combined to form a coupling feature vector.

[0008] The coupling feature vector is mapped to the feature space, and the discharge type identifier and severity level are determined based on the distribution position of the coupling feature vector in the feature space.

[0009] Extract the signal energy intensity of each node in the distributed sensor array, calculate the spatial attenuation gradient of the energy intensity of each node, determine the main radiation direction of the discharge source based on the spatial attenuation gradient, use the main radiation direction to perform directional compensation on the signal arrival time of each node, and determine the three-dimensional spatial coordinates of the local discharge source based on the compensated signal arrival time.

[0010] Establish a discharge source archive to record the coupled feature vector and three-dimensional spatial coordinates, calculate the feature distance and spatial distance between the newly detected discharge event and the data stored in the archive, and perform weighted fusion. Based on the fusion result, assign the discharge source and update the archive.

[0011] Extract the severity level and three-dimensional spatial coordinates of the same discharge source at different times from the updated discharge source file, calculate the spatial migration rate, and generate graded early warning signals.

[0012] Multi-physics field signals from power transmission equipment are acquired using a distributed sensor array. Cross-domain cross-correlation is performed on these signals to extract phase difference and amplitude ratio features, which are then combined to form a coupling feature vector, including:

[0013] The geometric delay difference of the signal propagation path is calculated based on the spatial position coordinates of each sensor node in the distributed sensor array. The geometric delay difference is then used to synchronously correct the time reference of the multi-physics field signals collected by each sensor node, thereby obtaining the time-domain aligned multi-physics field signals.

[0014] The time-domain aligned multiphysics signals are grouped according to the physical field type. Cross-domain cross-correlation is performed between signals of different physical field types. The peak position of the cross-correlation function between signals of different physical field types is calculated by using a sliding time window.

[0015] The time offset corresponding to the peak position of the cross-correlation function is used as the phase difference feature, and the amplitude ratio corresponding to the peak of the cross-correlation function is extracted as the amplitude ratio feature.

[0016] The extracted phase difference features and amplitude ratio features are encoded using feature vectorization. The phase difference features and amplitude ratio features are then arranged in series according to a preset physical field combination order to form a coupled feature vector with a multi-dimensional structure.

[0017] The coupling feature vectors are mapped to a feature space, and the discharge type and severity level are determined based on the distribution of the coupling feature vectors in the feature space, including:

[0018] The cross-coupling term between the phase difference feature and the amplitude ratio feature in the coupling feature vector is extracted. The cross-coupling feature matrix is ​​constructed by calculating the product of the phase difference feature and the amplitude ratio feature corresponding to different combinations of physical field types. The cross-coupling feature matrix is ​​decomposed by tensor to extract multi-order coupling modes. The coupling feature vector and the multi-order coupling modes are combined and projected into the feature space.

[0019] The nonlinear distribution surface formed by historical discharge events in the feature space is fitted to generate a discharge type discrimination manifold. At the same time, the temporal trajectory of the same discharge source in the feature space is tracked and the evolution direction vector field is extracted to construct the severity level evolution trajectory.

[0020] Calculate the geodesic distance from the projection point of the coupled feature vector in the feature space to each discharge type discrimination manifold, extract the discharge type identifier with the smallest geodesic distance, calculate the tangential component of the severity level evolution trajectory of the projection point along the corresponding discharge type discrimination manifold, and determine the severity level based on the normalized position of the tangential component on the evolution trajectory.

[0021] Extracting the signal energy intensity of each node in the distributed sensor array, calculating the spatial attenuation gradient of the energy intensity of each node, and determining the main radiation direction of the power source based on the spatial attenuation gradient include:

[0022] Spectral decomposition is performed on the multi-physics field signals collected by each node of the distributed sensor array, and the cumulative value of spectral energy in the discharge characteristic frequency band is extracted as the signal energy intensity of each node.

[0023] Calculate the ratio of the difference in signal energy intensity between adjacent nodes to the distance between nodes, obtain the energy attenuation rate of each node in three orthogonal directions in three-dimensional space, and combine the energy attenuation rates in the three orthogonal directions to form the spatial attenuation gradient of the energy intensity of each node.

[0024] Extend the reverse vector of the spatial attenuation gradient of each node in three-dimensional space, calculate the spatial intersection density distribution of all reverse extension vectors, extract the peak position of the intersection density as the estimated discharge source position, calculate the line vector connecting the estimated discharge source position to each node, and sum the line vectors according to the signal energy intensity of each node to obtain the main radiation direction of the discharge source.

[0025] Directional compensation is performed on the signal arrival time of each node using the main radiation direction. The three-dimensional spatial coordinates of the partial discharge source are determined based on the compensated signal arrival time, including:

[0026] The main radiation direction is projected onto the line connecting each node and the estimated discharge source location, and the projection component of the main radiation direction in each line direction is calculated as the direction consistency coefficient of each node.

[0027] Based on the directional consistency coefficient, a time correction amount is constructed for the signal arrival time of each node, and the time correction amount is superimposed on the signal arrival time of each node to obtain the compensated signal arrival time.

[0028] Using the compensated signal arrival time as the time reference, establish isochronous surfaces for each node, extract the set of intersection lines of the isochronous surfaces corresponding to different nodes in three-dimensional space, and perform spatial curvature analysis on the set of intersection lines to identify the curvature abrupt change locations.

[0029] The spatial coordinates corresponding to the curvature change location are used as candidate points. The direction of the candidate line connecting each candidate point and each node is calculated. The main radiation direction is projected onto the direction of each candidate line to obtain the candidate projection component.

[0030] Calculate the deviation between the candidate projection component and the direction consistency coefficient corresponding to each candidate point in three-dimensional spatial coordinates, and select the candidate point with the smallest deviation value as the three-dimensional spatial coordinates of the local discharge source.

[0031] Establishing a discharge source archive records the coupling feature vector and three-dimensional spatial coordinates. Calculating and weighting the feature distance and spatial distance between newly detected discharge events and the data stored in the archive, and then assigning the discharge source based on the fusion result and updating the archive, includes:

[0032] The coupling feature vector is bound to the three-dimensional spatial coordinates of the local discharge source as a discharge source file entry and stored in the discharge source file formation file for data storage.

[0033] Extract the coupling feature vectors of the data stored in the archive at different detection times, fit them in the feature space to generate feature evolution curves, extract the tangent vectors of the feature evolution curves to construct feature evolution trend vectors, and calculate the deviation distance between the coupling feature vectors of the newly detected discharge event and the feature evolution trend vectors at the extended prediction points in the feature space as the feature distance.

[0034] Extract the three-dimensional spatial coordinates of the partial discharge source at different detection times from the data stored in the archive, fit the three-dimensional space to generate a spatial migration curve, extract the tangent vector of the spatial migration curve to construct a spatial migration trend vector, and calculate the deviation distance between the three-dimensional spatial coordinates of the partial discharge source of the newly detected discharge event and the extended prediction point of the spatial migration trend vector in the three-dimensional space as the spatial distance.

[0035] The fusion distance is obtained by weighted summation of the feature distance and the spatial distance. The assigned discharge source is determined based on the fusion distance. The discharge source file entries of newly detected discharge events are appended to the file corresponding to the assigned discharge source to store the data and form an updated discharge source file.

[0036] The severity level and three-dimensional spatial coordinates of the same discharge source at different times are extracted from the updated discharge source archive. The spatial migration rate is calculated and graded early warning signals are generated, including:

[0037] The severity levels of the same discharge source at different times are extracted from the updated discharge source archive, and the changes in severity levels between adjacent times are calculated to construct a severity level evolution sequence.

[0038] The three-dimensional spatial coordinates of the same discharge source at different times are extracted from the updated discharge source file. The instantaneous migration rate is obtained by calculating the ratio of the displacement of the three-dimensional spatial coordinates at adjacent times to the time interval. The instantaneous migration rate is then filtered in the time dimension to obtain the spatial migration rate.

[0039] A severity level evolution sequence is time-series fitted to generate a severity level evolution curve. The curvature change features of the severity level evolution curve are extracted to construct a severity aggravation factor. A spatial migration rate evolution curve is time-series fitted to generate a migration rate evolution curve. The curvature change features of the migration rate evolution curve are extracted to construct a spatial diffusion acceleration factor.

[0040] The cross-correlation coefficient between the severity level evolution sequence and the spatial migration rate in the time dimension is calculated. Based on the cross-correlation coefficient, a two-domain coupling strength index is constructed. The severity aggravation factor, spatial diffusion acceleration factor and two-domain coupling strength index are nonlinearly combined to obtain the early warning hazard measurement value.

[0041] Warning levels are classified according to the numerical range of the warning hazard measurement value, and graded warning signals are generated based on the warning level and the discharge type identifier of the same discharge source.

[0042] A second aspect of the present invention provides a partial discharge characteristic diagnosis and location system for power transmission equipment, comprising:

[0043] The first unit is used to collect multi-physical field signals of power transmission equipment through a distributed sensor array, perform cross-domain cross-correlation calculations on the multi-physical field signals to extract phase difference features and amplitude ratio features, and combine them to form a coupling feature vector.

[0044] The second unit is used to map the coupling feature vector to the feature space, and determine the discharge type identifier and severity level based on the distribution position of the coupling feature vector in the feature space;

[0045] The third unit is used to extract the signal energy intensity of each node in the distributed sensor array, calculate the spatial attenuation gradient of the energy intensity of each node, determine the main radiation direction of the discharge source based on the spatial attenuation gradient, use the main radiation direction to perform directional compensation on the signal arrival time of each node, and determine the three-dimensional spatial coordinates of the local discharge source based on the compensated signal arrival time.

[0046] The fourth unit is used to establish a discharge source archive, record the coupling feature vector and three-dimensional spatial coordinates, calculate the feature distance and spatial distance between the newly detected discharge event and the data stored in the archive, and perform weighted fusion. Based on the fusion result, the discharge source is assigned and the archive is updated.

[0047] The fifth unit is used to extract the severity level and three-dimensional spatial coordinates of the same discharge source at different times from the updated discharge source file, calculate the spatial migration rate, and generate graded early warning signals.

[0048] A third aspect of the present invention provides an electronic device, comprising:

[0049] processor;

[0050] Memory used to store processor-executable instructions;

[0051] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0052] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0053] In this embodiment, multi-physics field signals are acquired through a distributed sensor array, and coupled feature vectors are extracted, enabling accurate identification of discharge type and severity level, thus improving the comprehensiveness and reliability of diagnosis. The main radiation direction is determined by calculating the spatial attenuation gradient and directional compensation is performed. Combined with the time difference of arrival, precise three-dimensional positioning of the discharge source is achieved, solving the problem of insufficient positioning accuracy in complex electromagnetic environments using traditional methods. Establishing a discharge source profile and calculating a weighted fusion of characteristic distance and spatial distance enables continuous tracking and evaluation of the same discharge source. By calculating the spatial migration rate, a graded early warning signal is generated, providing a scientific basis for equipment maintenance decisions and effectively reducing power grid safety risks. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the method for diagnosing and locating partial discharge characteristics of power transmission equipment according to an embodiment of the present invention.

[0055] Figure 2 This is the three-dimensional positioning process for the partial discharge source in an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0058] Figure 1 This is a flowchart illustrating the partial discharge characteristic diagnosis and localization method for power transmission equipment according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] Multi-physical field signals of power transmission equipment are collected by a distributed sensor array. Cross-domain cross-correlation is performed on the multi-physical field signals to extract phase difference and amplitude ratio features, which are then combined to form a coupling feature vector.

[0060] The coupling feature vector is mapped to the feature space, and the discharge type identifier and severity level are determined based on the distribution position of the coupling feature vector in the feature space.

[0061] Extract the signal energy intensity of each node in the distributed sensor array, calculate the spatial attenuation gradient of the energy intensity of each node, determine the main radiation direction of the discharge source based on the spatial attenuation gradient, use the main radiation direction to perform directional compensation on the signal arrival time of each node, and determine the three-dimensional spatial coordinates of the local discharge source based on the compensated signal arrival time.

[0062] Establish a discharge source archive to record the coupled feature vector and three-dimensional spatial coordinates, calculate the feature distance and spatial distance between the newly detected discharge event and the data stored in the archive, and perform weighted fusion. Based on the fusion result, assign the discharge source and update the archive.

[0063] Extract the severity level and three-dimensional spatial coordinates of the same discharge source at different times from the updated discharge source file, calculate the spatial migration rate, and generate graded early warning signals.

[0064] In one optional implementation, multi-physics field signals from power transmission equipment are acquired using a distributed sensor array. Cross-domain cross-correlation operations are performed on the multi-physics field signals to extract phase difference and amplitude ratio features, which are then combined to form a coupling feature vector, including:

[0065] The geometric delay difference of the signal propagation path is calculated based on the spatial position coordinates of each sensor node in the distributed sensor array. The geometric delay difference is then used to synchronously correct the time reference of the multi-physics field signals collected by each sensor node, thereby obtaining the time-domain aligned multi-physics field signals.

[0066] The time-domain aligned multiphysics signals are grouped according to the physical field type. Cross-domain cross-correlation is performed between signals of different physical field types. The peak position of the cross-correlation function between signals of different physical field types is calculated by using a sliding time window.

[0067] The time offset corresponding to the peak position of the cross-correlation function is used as the phase difference feature, and the amplitude ratio corresponding to the peak of the cross-correlation function is extracted as the amplitude ratio feature.

[0068] The extracted phase difference features and amplitude ratio features are encoded using feature vectorization. The phase difference features and amplitude ratio features are then arranged in series according to a preset physical field combination order to form a coupled feature vector with a multi-dimensional structure.

[0069] In this embodiment, the distributed sensor array includes multiple sensor nodes, each equipped with various physical field sensing devices such as temperature sensors, acoustic sensors, electromagnetic wave sensors, and optical sensors. These sensor nodes are arranged at different spatial locations around the power transmission equipment, forming a spatially covering sensor network. The spatial location of the sensor nodes is represented by a three-dimensional coordinate system, with each node having definite coordinate values, facilitating subsequent calculation of the geometric time delay difference of the signal propagation path.

[0070] During the signal acquisition phase, each sensing node simultaneously acquires multi-physics field signals generated during the operation of the power transmission equipment. Because partial discharge phenomena simultaneously induce multiple physical effects such as temperature changes, sound wave propagation, electromagnetic radiation, and optical characteristics, different physical field signals carry different characteristic information about the discharge source. Signal acquisition employs a high-precision sampling mode, with sampling frequencies set according to the characteristics of different physical fields: 100Hz for temperature signals, 50kHz for sound signals, 10MHz for electromagnetic signals, and 1MHz for optical signals.

[0071] The geometric time delay difference of the signal propagation path is calculated for the acquired raw multiphysics signals. Using the suspected discharge source location as a reference point, the propagation time of each physical field signal from the source point to each sensing node is calculated. For electromagnetic wave signals, the propagation speed is close to the speed of light; for sound wave signals, the propagation speed is approximately 340 m / s; for temperature signals, the propagation speed depends on the medium material; for optical signals, the propagation speed is approximately equal to the speed of light in a vacuum. Based on the spatial coordinates of each sensing node and the distance to the suspected source point, combined with the propagation speed of each physical field signal, the geometric time delay difference matrix is ​​calculated.

[0072] The calculated geometric time delay difference is used to perform time reference synchronization correction on the multiphysics signals collected by each sensing node. The correction method is to shift the time axis of each signal so that the arrival time of the signal minus the time delay caused by the propagation path is aligned to the same moment. The corrected signals are aligned in the time domain, representing the behavior of the same event in different physical domains, which facilitates subsequent cross-domain correlation analysis.

[0073] The time-domain aligned multiphysics signals are grouped according to their physical field type, forming temperature signal group, sound wave signal group, electromagnetic wave signal group, and optical signal group. Cross-domain cross-correlation operations are performed between signal groups of different physical field types to discover implicit correlations between the signals. Specifically, a sliding time window with a width of 20 ms and a sliding step size of 5 ms is set, and the cross-correlation function between two different physical field signals is calculated within each window. The cross-correlation operation is implemented through time-domain convolution of the two signal sequences, calculating the similarity between the two signals at different time offsets.

[0074] For each pair of physical field signal combinations, the peak position and magnitude of the cross-correlation function are extracted. The time offset corresponding to the peak position of the cross-correlation function serves as the phase difference feature, representing the temporal relationship between the two physical field phenomena; the amplitude ratio corresponding to the peak of the cross-correlation function serves as the amplitude ratio feature, representing the relative relationship between the intensities of the two physical field phenomena. For example, cross-correlation analysis of sound wave and electromagnetic wave signals can yield their phase difference and amplitude ratio, reflecting the specific type characteristics of the discharge source.

[0075] The extracted phase difference and amplitude ratio features are encoded using feature vectorization, and the features of different physical field combinations are concatenated in a preset order. The preset order of physical field combinations is temperature-sound wave, temperature-electromagnetic wave, temperature-optics, sound wave-electromagnetic wave, sound wave-optics, and electromagnetic wave-optics, forming a total of six feature pairs. Each feature pair contains a phase difference value and an amplitude ratio value, and the combination of these two values ​​constitutes a two-dimensional feature sub-vector. The two-dimensional sub-vectors of the six feature pairs are concatenated in sequence to form a twelve-dimensional coupled feature vector.

[0076] The coupling feature vector undergoes data normalization to unify the feature values ​​of each dimension to the range of 0 to 1, eliminating the influence of numerical differences caused by different physical dimensions. The normalized coupling feature vector serves as a comprehensive representation of the partial discharge characteristics of power transmission equipment and is input into the subsequent diagnostic and localization model.

[0077] The diagnostic and localization model employs a multilayer perceptron architecture. The input layer has 12 nodes, corresponding to the dimension of the coupled feature vectors. Two hidden layers are used, with 24 and 16 nodes respectively. The output layer has a combination of discharge type and possible location coordinates. The model training data comes from historical cases of simulated discharges in the laboratory and field data collection, containing feature vectors of various typical discharge types along with their corresponding discharge type and location labels. The model's parameters are optimized using backpropagation, and the loss function is cross-entropy, used to measure the difference between the predicted results and the actual labels.

[0078] This invention acquires multi-physics field signals through a distributed sensor array, achieving comprehensive perception of partial discharge phenomena. Through cross-domain cross-correlation operations, it extracts phase difference and amplitude ratio features between different physical field signals, constructing a coupled feature vector with rich diagnostic information. Based on the diagnostic and localization model of this coupled feature vector, it achieves accurate identification of discharge type and location. By fully utilizing the complementarity of multi-physics field signals, it overcomes the limitations of single-physics field signal diagnosis, improving the accuracy and reliability of diagnostic localization.

[0079] In one optional implementation, mapping the coupling feature vector to a feature space, and determining the discharge type identifier and severity level based on the distribution location of the coupling feature vector in the feature space, includes:

[0080] The cross-coupling term between the phase difference feature and the amplitude ratio feature in the coupling feature vector is extracted. The cross-coupling feature matrix is ​​constructed by calculating the product of the phase difference feature and the amplitude ratio feature corresponding to different combinations of physical field types. The cross-coupling feature matrix is ​​decomposed by tensor to extract multi-order coupling modes. The coupling feature vector and the multi-order coupling modes are combined and projected into the feature space.

[0081] The nonlinear distribution surface formed by historical discharge events in the feature space is fitted to generate a discharge type discrimination manifold. At the same time, the temporal trajectory of the same discharge source in the feature space is tracked and the evolution direction vector field is extracted to construct the severity level evolution trajectory.

[0082] Calculate the geodesic distance from the projection point of the coupled feature vector in the feature space to each discharge type discrimination manifold, extract the discharge type identifier with the smallest geodesic distance, calculate the tangential component of the severity level evolution trajectory of the projection point along the corresponding discharge type discrimination manifold, and determine the severity level based on the normalized position of the tangential component on the evolution trajectory.

[0083] In the process of coupled feature vector mapping, the cross-coupling term between the phase difference feature and the amplitude ratio feature is first extracted. The coupled feature vector contains the phase difference and amplitude ratio features of six combinations of physical fields, corresponding to combinations such as temperature-sound wave, temperature-electromagnetic wave, temperature-optics, sound wave-electromagnetic wave, sound wave-optics, and electromagnetic wave-optics. The cross-coupling feature is constructed by calculating the product of the phase difference feature and the amplitude ratio feature in each combination. Taking the temperature-sound wave combination as an example, its phase difference feature value is 0.25, and its amplitude ratio feature value is 1.75. The product of the two, 0.4375, is the cross-coupling term for this combination. The cross-coupling term is calculated for each of the six combinations, forming a 6×1 cross-coupling feature vector.

[0084] Furthermore, the cross-coupling eigenvectors are reconstructed into a cross-coupling eigenma matrix, which has a 4×4 structure and corresponds to the cross-combinations of four physical fields: temperature, sound waves, electromagnetic waves, and optics. The diagonal elements of the matrix are set to zero, and the off-diagonal elements are filled with the cross-coupling terms of the corresponding physical field combinations. Due to the matrix's symmetry, the upper and lower triangular elements have the same value. The cross-coupling eigenma matrix contains the multidimensional interaction relationships between different physical field signals, reflecting the physical coupling characteristics of the partial discharge source.

[0085] Tensor decomposition is performed on the cross-coupling feature matrix to extract multi-order coupling modes. The Tucker decomposition method is used to decompose the cross-coupling feature matrix into a combination of several low-rank components. The decomposition process retains 95% of the energy, typically yielding three main coupling modes. These coupling modes characterize the dominant modes of physical field-signal interaction, each represented by a 4×4 matrix, capturing the essential characteristics of the partial discharge physical process.

[0086] The original coupled feature vector is combined with the extracted multi-level coupling patterns to form an enhanced feature representation. The combination method involves flattening the matrix elements of the original 12-dimensional coupled feature vector and the three main coupling patterns, and then concatenating them to form an extended feature vector. The extended feature vector has a dimension of 12 + 3 × 16 = 60 dimensions, containing both the original features and multi-level coupling information.

[0087] A nonlinear dimensionality reduction method is employed to project the extended feature vectors into a three-dimensional feature space. The t-SNE algorithm is used for dimensionality reduction, with a learning rate of 100%, a perplexity parameter of 30, and 1000 iterations. The dimensionality reduction process preserves the local structural relationships between samples in the high-dimensional feature space, allowing samples with similar discharge types to remain clustered in the low-dimensional space. The resulting three-dimensional coordinates are used as projection points in the feature space, facilitating visualization analysis and subsequent processing.

[0088] In the feature space, a nonlinear distribution surface formed by historical discharge events is obtained and fitted to generate a discharge type discrimination manifold. The historical discharge event samples include typical discharge types such as internal air gap discharge, creeping discharge, levitation discharge, and corona discharge, with at least 50 sample points for each type. Manifold learning is performed on the distribution of each discharge sample point in the feature space using a locally linear embedding algorithm with a neighborhood of 15 points and a regularization parameter of 0.001. Through manifold learning, a two-dimensional surface is constructed for each discharge type; this surface is the discharge type discrimination manifold, which accurately represents the distribution region and boundaries of different discharge types in the feature space.

[0089] Simultaneously, the temporal trajectory of the same discharge source in the feature space is tracked, and the evolution direction vector field is extracted. For continuously monitored discharge data, the trajectory of the discharge source characteristics over time is recorded. Taking a certain air-gap discharge source as an example, its monitoring data over 30 days is continuously plotted in the feature space to form a temporal trajectory curve. The trajectory curve is differentially calculated to obtain the direction vector at each time point, constituting the evolution direction vector field. The vector field reflects the trend of discharge characteristics evolving over time, pointing in the direction of increasing discharge severity.

[0090] The geodesic distance from the projection point of the coupled feature vector in the feature space to the discriminant manifold for each discharge type is calculated. The geodesic distance calculation employs Dijkstra's algorithm, discretizing the feature space into a grid of points with a resolution of 0.01, constructing an adjacency graph, and connecting each grid point to its 26 neighbors. By searching for the shortest path, the geodesic distance from the projection point to the discriminant manifold for each discharge type is calculated. Geodesic distance measures the distance from the point to the surface along the curvature of the feature space, reflecting the similarity between the sample and each discharge type more accurately than Euclidean distance.

[0091] Extract the discharge type identifier with the smallest geodesic distance as the result of the current discharge source type determination. If the geodesic distance from the projection point to the internal air gap discharge current shape is 0.15, the geodesic distance to the creepage discharge current shape is 0.42, and the geodesic distance to other discharge type current shapes is larger, then the current discharge source is determined to be an internal air gap discharge.

[0092] Calculate the tangential component of the severity level evolution trajectory of the projection point along the discrimination manifold corresponding to the discharge type. The severity level evolution trajectory is formed by sorting historical samples of this type of discharge according to severity. Project the position of the projection point on the discrimination manifold onto the evolution trajectory to obtain the tangential component. Determine the severity level based on the normalized position of the tangential component on the evolution trajectory. The starting point of the evolution trajectory corresponds to a minor discharge (level 1), the ending point corresponds to a severe discharge (level 10), and the intermediate points are divided proportionally. If the normalized position of the tangential component is 0.35, the severity level is determined to be level 4.

[0093] This invention enhances feature representation capabilities by extracting cross-coupling terms and multi-order coupling modes from coupled feature vectors; it maps high-dimensional features to an intuitive three-dimensional feature space using nonlinear dimensionality reduction techniques, facilitating analysis and judgment; and it achieves accurate assessment of discharge type and severity by constructing a discharge type discrimination manifold and severity level evolution trajectory based on historical discharge data. It can address changes in discharge characteristics under different operating environments and equipment conditions, providing a scientific basis for preventative maintenance and fault early warning of transmission equipment, and effectively improving the safe and reliable operation level of the power system.

[0094] In one optional implementation, the signal energy intensity of each node in the distributed sensor array is extracted, the spatial attenuation gradient of the energy intensity of each node is calculated, and the main radiation direction of the discharge source is determined based on the spatial attenuation gradient, including:

[0095] Spectral decomposition is performed on the multi-physics field signals collected by each node of the distributed sensor array, and the cumulative value of spectral energy in the discharge characteristic frequency band is extracted as the signal energy intensity of each node.

[0096] Calculate the ratio of the difference in signal energy intensity between adjacent nodes to the distance between nodes, obtain the energy attenuation rate of each node in three orthogonal directions in three-dimensional space, and combine the energy attenuation rates in the three orthogonal directions to form the spatial attenuation gradient of the energy intensity of each node.

[0097] Extend the reverse vector of the spatial attenuation gradient of each node in three-dimensional space, calculate the spatial intersection density distribution of all reverse extension vectors, extract the peak position of the intersection density as the estimated discharge source position, calculate the line vector connecting the estimated discharge source position to each node, and sum the line vectors according to the signal energy intensity of each node to obtain the main radiation direction of the discharge source.

[0098] When performing spectral decomposition on the multiphysics field signals collected by each node of the distributed sensor array, the time-domain signal is transformed to the frequency domain using a Fast Fourier Transform (FFT). This transformation process decomposes the original time-domain signal into a superposition of different frequency components, each with corresponding amplitude and phase information. Considering the typical characteristics of partial discharge signals from power transmission equipment, the 20kHz to 100kHz frequency band is selected as the discharge characteristic band. Within this band, the energy spectral density is obtained by squaring the spectral amplitude corresponding to each discrete frequency point. Subsequently, the energy spectral densities of all frequency points within the characteristic band are summed, and the accumulated value is the signal energy intensity of that node.

[0099] After obtaining the signal energy intensity of each node, it is necessary to calculate the energy attenuation rate between adjacent nodes in three-dimensional space. In a three-dimensional Cartesian coordinate system, the space is divided into three orthogonal directions: X-axis, Y-axis, and Z-axis. For any node, its nearest neighbor in each of the three orthogonal directions is found. The energy intensity difference between the current node and its nearest neighbor in the positive X-axis direction is calculated, and this difference is divided by the actual physical distance between the two nodes to obtain the energy attenuation rate of the node in the X-axis direction. The energy attenuation rates in the Y-axis and Z-axis directions are calculated using the same method. The energy attenuation rate values ​​in the three directions are combined to form a three-dimensional vector, which is the spatial attenuation gradient of the node's energy intensity. The direction of the attenuation gradient points to the direction of the fastest decrease in energy intensity, and its magnitude reflects the severity of energy attenuation. In a partial discharge location implementation of a switchgear, the energy attenuation rate of the sensing node located east of the discharge source is 0.4 joules per meter in the X-axis direction, 0.1 joules per meter in the Y-axis direction, and 0.05 joules per meter in the Z-axis direction.

[0100] The reverse vector of the spatial attenuation gradient at each node points in the direction of increasing energy intensity, theoretically pointing to the location of the discharge source. After reversing the attenuation gradient vector of each node, it is extended in three-dimensional space starting from that node. The extension process uses a ray tracing method, advancing forward with a fixed step size along the reverse gradient direction. The extension length is set to twice the maximum span of the sensing array to ensure coverage of possible discharge source areas. The three-dimensional space is divided into cubic grid cells with a side length of 0.1 meters. The count value of each grid cell traversed by the reverse extension vector is increased by 1. After traversing the reverse vectors of all nodes, the cumulative count value of each grid cell is calculated; this count value is the spatial convergence density at that location. Regions with high convergence density indicate that the reverse attenuation gradients of multiple nodes converge there, with a higher probability of the presence of a discharge source. The cell with the largest count value is found after traversing all grid cells; its geometric center coordinates are the estimated location of the discharge source.

[0101] After determining the estimated location of the discharge source, the spatial connection vector from this location to each sensing node is calculated. The starting point of the connection vector is the estimated location of the discharge source, and the ending point is the actual coordinate position of each node. Each connection vector contains three components, corresponding to the projected lengths along the X, Y, and Z axes, respectively. The connection vectors are weighted by the signal energy intensity of the corresponding nodes, with nodes having higher energy intensities receiving higher weighting coefficients. Specifically, the energy intensity of a node is divided by the sum of the energy intensities of all nodes to obtain the normalized weight of that node. This normalized weight is then multiplied by the three components of the corresponding connection vector to obtain the weighted vector components. The weighted vector components of all nodes are summed along the three coordinate axes to obtain three total component values. The vector formed by these three values ​​is the main radiation direction vector of the discharge source. This vector reflects the main propagation trend of the discharge energy in space, pointing towards the clustered area of ​​nodes receiving stronger signals. For example, the estimated location coordinates of the discharge source are X = 1.2 meters, Y = 0.8 meters, and Z = 2.5 meters. By weighted summation, the three components of the main radiation direction vector are obtained as X = 0.65, Y = 0.35, and Z = -0.20, indicating that the discharge energy mainly radiates along the east-northeast direction and slightly downwards.

[0102] This invention extracts effective feature information through frequency domain energy accumulation, avoiding the influence of noise and interference components in the time domain signal. Spatial attenuation gradient calculation fully utilizes the distributed characteristics of the sensor array, enabling the capture of the propagation pattern of the discharge signal in three-dimensional space. Reverse vector extension and intersection density analysis achieve effective inference from multi-node observation data to the location of a single discharge source, improving the robustness of the localization. The weighted vector summation method considers the differences in the received signal strength at each node, making the estimation of the main radiation direction more accurate and reliable.

[0103] like Figure 2 The diagram illustrates the three-dimensional positioning process of the partial discharge source in this embodiment.

[0104] In one optional implementation, the signal arrival time of each node is directionally compensated using the main radiation direction, and the three-dimensional spatial coordinates of the partial discharge source are determined based on the compensated signal arrival time, including:

[0105] The main radiation direction is projected onto the line connecting each node and the estimated discharge source location, and the projection component of the main radiation direction in each line direction is calculated as the direction consistency coefficient of each node.

[0106] Based on the directional consistency coefficient, a time correction amount is constructed for the signal arrival time of each node, and the time correction amount is superimposed on the signal arrival time of each node to obtain the compensated signal arrival time.

[0107] Using the compensated signal arrival time as the time reference, establish isochronous surfaces for each node, extract the set of intersection lines of the isochronous surfaces corresponding to different nodes in three-dimensional space, and perform spatial curvature analysis on the set of intersection lines to identify the curvature abrupt change locations.

[0108] The spatial coordinates corresponding to the curvature change location are used as candidate points. The direction of the candidate line connecting each candidate point and each node is calculated. The main radiation direction is projected onto the direction of each candidate line to obtain the candidate projection component.

[0109] Calculate the deviation between the candidate projection component and the direction consistency coefficient corresponding to each candidate point in three-dimensional spatial coordinates, and select the candidate point with the smallest deviation value as the three-dimensional spatial coordinates of the local discharge source.

[0110] The dot product of the three coordinate components of the main radiation direction vector and the corresponding components of the vector connecting to a node is obtained. Simultaneously, the magnitudes of the main radiation direction vector and the connecting vector are calculated. Dividing the dot product by the product of these two magnitudes yields the projection component of the main radiation direction onto the connecting vector, defined as the directional consistency coefficient of the node. The directional consistency coefficient ranges from -1 to +1. A coefficient close to +1 indicates the node is directly in front of the main radiation direction, close to -1 indicates it is in the opposite direction, and close to 0 indicates it is in the side direction. In a transformer bushing partial discharge monitoring scenario, the directional consistency coefficient of a sensing node directly in front of the main radiation direction is 0.92, while that of a node in the side direction is only 0.15.

[0111] The directional consistency coefficient reflects the degree of deviation between the discharge signal propagation path and the main radiation direction. This deviation leads to a systematic deviation in the signal arrival time. When constructing the time correction, the average directional consistency coefficient of all nodes is subtracted from the directional consistency coefficient of each node to obtain the relative deviation. This relative deviation is multiplied by a time correction factor, which is determined based on the propagation speed of the discharge signal in the medium. For electromagnetic wave signals propagating in air, the propagation speed is close to the speed of light, and the time correction factor can be set as the maximum span of the array divided by the speed of light. For acoustic wave signals propagating in solid insulating materials, the propagation speed is significantly reduced, and the time correction factor should be adjusted according to the acoustic speed characteristics of the specific material. The calculated time correction is superimposed on the original signal arrival time of each node. Nodes with positive deviations have earlier arrival times, and nodes with negative deviations have later arrival times, thus compensating for the directional propagation characteristics. In a cable joint discharge location, a node with a directional consistency coefficient of 0.75 has a time correction of positive 0.8 microseconds, an original arrival time of 15.3 microseconds, and a compensated arrival time of 14.5 microseconds.

[0112] After obtaining the compensated signal arrival time, an isochronous surface is established for each node using this time as a reference. A reference node is selected, and the arrival time difference between the reference node and all other nodes is calculated. For any point in space, assuming it is the location of the power source, the theoretical time difference can be obtained by dividing the distance difference from this point to the reference node and another node by the signal propagation speed. When the theoretical time difference equals the measured time difference, the spatial point lies on the corresponding isochronous surface. The isochronous surface exhibits a hyperboloid shape in three-dimensional space, its shape determined by the spatial position and time difference between the two nodes. Multiple isochronous surfaces are formed between different node pairs, and these surfaces intersect in three-dimensional space to form a complex set of intersection lines. When extracting the set of intersection lines, a three-dimensional spatial mesh scanning method is used to divide the space into cubic units with a side length of 0.05 meters. It is determined whether each unit simultaneously satisfies the constraints of multiple isochronous surfaces; the boundary of the unit that satisfies the conditions is the discrete point representation of the intersection line.

[0113] When performing spatial curvature analysis on the extracted set of intersecting lines, three adjacent discrete points are selected along the tangent direction of the intersecting line, and an arc is constructed passing through these three points. The reciprocal of the radius of this arc is the curvature value at that location; a larger curvature indicates a more severe bending of the intersecting line. The curvature value is calculated sequentially along the intersecting line. When the curvature difference between two adjacent points exceeds a set threshold, that location is determined to be a curvature abrupt change. Curvature abrupt changes reflect drastic changes in the geometric relationship of the intersection of isochronous surfaces, typically corresponding to the vicinity of the actual discharge source location. The curvature threshold is set based on the spatial resolution of the sensor array and the signal time measurement accuracy. Higher measurement accuracy allows for a smaller curvature threshold to improve positioning sensitivity.

[0114] The spatial coordinates corresponding to the identified curvature abrupt change locations are used as candidate discharge source locations. Each candidate point forms a new candidate connection vector with each sensing node. The three coordinate components of the candidate connection vector are calculated. The corresponding component of the main radiation direction vector is multiplied by each candidate connection vector component, and the sum is obtained. This sum is then divided by the product of the two vector magnitudes to obtain the projection component of the main radiation direction along the candidate connection direction, defined as the candidate projection component. For each candidate point, the deviation between the candidate projection components of all its corresponding nodes and the previously determined directional consistency coefficients of each node is calculated. The deviation is calculated using the sum of squared differences. The directional consistency coefficient of each node is subtracted from the candidate projection component, the difference is squared, and the squared values ​​of all nodes are summed. The square root of the sum is the overall deviation value of the candidate point. The smaller the deviation value, the higher the degree of matching between the candidate point location and the main radiation direction characteristics, and the more consistent it is with the spatial radiation law of the actual discharge source.

[0115] The deviation values ​​of all candidate points were calculated, and the candidate point with the smallest deviation value was selected as the final determined three-dimensional spatial coordinates of the local discharge source. In the discharge location implementation of a high-voltage cable terminal, five locations with abrupt curvature changes were identified as candidate points, with their coordinates distributed at different locations in the target area. By calculating the deviation values ​​of each candidate point, the deviation value of the third candidate point was 0.08, significantly lower than the 0.23 to 0.45 of the other candidate points. Therefore, the spatial coordinates of this candidate point, X = 2.35 meters, Y = 1.68 meters, and Z = 3.12 meters, were determined as the final location of the discharge source.

[0116] This invention eliminates the influence of non-uniform discharge signal radiation on time measurement through directional consistency compensation, improving the quality of basic data for time difference positioning. Isochronous surface intersection analysis fully utilizes the spatial geometric constraints of multi-node observation data, avoiding the uncertainty of solving a single time difference equation. The curvature abrupt change identification method can accurately extract key feature locations from complex spatial intersection lines, significantly narrowing the search range. The candidate point selection process combines time information and energy radiation direction information, achieving multi-physical quantity fusion discrimination and enhancing the reliability of the positioning results.

[0117] In one optional implementation, a discharge source profile is established, recording the coupled feature vector and three-dimensional spatial coordinates. The feature distance and spatial distance between the newly detected discharge event and the data stored in the profile are calculated and weighted, and the profile is then assigned to a discharge source based on the fusion result and updated. This includes:

[0118] The coupling feature vector is bound to the three-dimensional spatial coordinates of the local discharge source as a discharge source file entry and stored in the discharge source file formation file for data storage.

[0119] Extract the coupling feature vectors of the data stored in the archive at different detection times, fit them in the feature space to generate feature evolution curves, extract the tangent vectors of the feature evolution curves to construct feature evolution trend vectors, and calculate the deviation distance between the coupling feature vectors of the newly detected discharge event and the feature evolution trend vectors at the extended prediction points in the feature space as the feature distance.

[0120] Extract the three-dimensional spatial coordinates of the partial discharge source at different detection times from the data stored in the archive, fit the three-dimensional space to generate a spatial migration curve, extract the tangent vector of the spatial migration curve to construct a spatial migration trend vector, and calculate the deviation distance between the three-dimensional spatial coordinates of the partial discharge source of the newly detected discharge event and the extended prediction point of the spatial migration trend vector in the three-dimensional space as the spatial distance.

[0121] The fusion distance is obtained by weighted summation of the feature distance and the spatial distance. The assigned discharge source is determined based on the fusion distance. The discharge source file entries of newly detected discharge events are appended to the file corresponding to the assigned discharge source to store the data and form an updated discharge source file.

[0122] This implementation first establishes a discharge source archive recording system. This archive system stores data in the form of entries, each containing a coupling feature vector and its corresponding three-dimensional spatial coordinates. The coupling feature vector consists of measured parameters such as amplitude, phase, pulse width, rise time, and repetition rate, and can be represented as a set of multidimensional feature values. The three-dimensional spatial coordinates are obtained through a positioning algorithm, accurate to the millimeter level. For example, the archive entry for a discharge source at time t1 can be recorded as: feature vector [0.85mV, 75°, 25ns, 10ns, 120Hz], spatial coordinates [125.3mm, 87.6mm, 356.2mm]. The archive storage adopts a relational database structure, with each discharge source corresponding to a unique identifier, facilitating rapid retrieval and updates.

[0123] For newly detected discharge events, it is necessary to calculate their characteristic distance from existing records. First, the coupling characteristic vectors of the same discharge source at different times t1, t2, ..., tn are extracted from the records to construct a characteristic evolution sequence. These discrete points are then fitted in the feature space using cubic spline interpolation to generate a continuous characteristic evolution curve. Taking a certain discharge source as an example, the characteristic vectors at times t1, t2, and t3 are [0.85mV, 75°, 25ns, 10ns, 120Hz], [0.92mV, 73°, 27ns, 11ns, 125Hz], and [1.05mV, 70°, 30ns, 13ns, 135Hz], respectively. A smooth curve can be obtained through spline interpolation. Differentiating this curve at the most recent time tn yields the characteristic evolution trend vector, representing the direction and rate of characteristic change. Based on this trend vector, the characteristic position at the next time tn+1 can be predicted in the feature space, i.e., the extended prediction point. The Euclidean distance between the feature vector of the newly detected discharge event and the predicted point is calculated as the feature distance. For example, if the predicted point is [1.15mV, 68°, 32ns, 14ns, 142Hz], and the feature vector of the newly detected event is [1.18mV, 67°, 33ns, 14.5ns, 145Hz], the calculated feature distance is 5.1 units.

[0124] Calculating spatial distance follows a similar principle. The three-dimensional spatial coordinates of the same discharge source at different times are extracted from the archives to construct a spatial position sequence. The least squares method is used for three-dimensional spatial curve fitting to generate the spatial migration curve of the discharge source. Taking a certain discharge source as an example, the spatial coordinates at times t1, t2, and t3 are [125.3mm, 87.6mm, 356.2mm], [126.1mm, 88.2mm, 355.8mm], and [127.4mm, 89.5mm, 355.2mm], respectively, resulting in a three-dimensional curve. The derivative of this curve at the most recent time tn is taken to obtain the spatial migration trend vector, representing the direction and rate of the discharge source's movement. Based on this trend vector, the spatial position at the next time moment is predicted, i.e., the spatial extension prediction point. The Euclidean distance between the spatial coordinates of the newly detected discharge event and this prediction point is calculated as the spatial distance. For example, the spatial prediction point is [128.5mm, 90.7mm, 354.5mm], while the coordinates of the newly detected event are [128.7mm, 91.0mm, 354.3mm], and the calculated spatial distance is 0.5mm.

[0125] The fusion distance is calculated using a weighted summation method: Fusion Distance = α × Feature Distance + β × Spatial Distance, where α and β are weighting coefficients, satisfying α + β = 1. The weighting coefficients can be dynamically adjusted according to the actual application scenario; generally, α = 0.6 and β = 0.4 can be set. For example, if the calculated feature distance is 5.1 units and the spatial distance is 0.5 mm, then the fusion distance = 0.6 × 5.1 + 0.4 × 0.5 = 3.26.

[0126] The attribution determination uses a threshold method, setting a fusion distance threshold T. When the fusion distance is less than T, the newly detected event is considered to belong to the known discharge source; otherwise, it is considered a new discharge source. The threshold T can be determined through statistical learning methods, with a typical value of 10. For example, if the calculated fusion distance of 3.26 is less than the threshold of 10, the newly detected event is determined to belong to a known discharge source.

[0127] The process of updating the archive involves appending information about newly detected discharge events to the archive of the associated discharge source. If the discharge source is identified as a known source, a new entry (characteristic vector and spatial coordinates) is added to the archive sequence of that discharge source; if a new discharge source is identified, a new discharge source archive is created and the entry is added. Simultaneously, the status information of the discharge source is updated, including the most recent detection time, cumulative detection count, characteristic change rate, and spatial migration velocity. For example, new detection events [1.18mV, 67°, 33ns, 14.5ns, 145Hz] and [128.7mm, 91.0mm, 354.3mm] are appended to the archive of discharge source number PD035, updating its most recent detection time to 2023-06-10 14:30:25 and increasing the cumulative detection count to 47.

[0128] This invention establishes a complete lifecycle tracking mechanism for discharge sources, enabling continuous monitoring from initial detection to ongoing evolution. The fitting methods for feature evolution curves and spatial migration curves fully leverage the development patterns inherent in historical data, improving the accuracy of attribution determination for newly detected events. The introduction of extended prediction points makes attribution determination forward-looking, adapting to dynamic changes in discharge source characteristics and location. The weighted fusion of feature distance and spatial distance integrates multiple dimensions of similarity measurement, enhancing the robustness and reliability of attribution identification. The continuous updating mechanism of the archives ensures the cumulative effect of monitoring data, enabling long-term trend analysis and degradation early warning.

[0129] In one alternative implementation, extracting the severity level and three-dimensional spatial coordinates of the same discharge source at different times from the updated discharge source archive, calculating the spatial migration rate, and generating a graded early warning signal includes:

[0130] The severity levels of the same discharge source at different times are extracted from the updated discharge source archive, and the changes in severity levels between adjacent times are calculated to construct a severity level evolution sequence.

[0131] The three-dimensional spatial coordinates of the same discharge source at different times are extracted from the updated discharge source file. The instantaneous migration rate is obtained by calculating the ratio of the displacement of the three-dimensional spatial coordinates at adjacent times to the time interval. The instantaneous migration rate is then filtered in the time dimension to obtain the spatial migration rate.

[0132] A severity level evolution sequence is time-series fitted to generate a severity level evolution curve. The curvature change features of the severity level evolution curve are extracted to construct a severity aggravation factor. A spatial migration rate evolution curve is time-series fitted to generate a migration rate evolution curve. The curvature change features of the migration rate evolution curve are extracted to construct a spatial diffusion acceleration factor.

[0133] The cross-correlation coefficient between the severity level evolution sequence and the spatial migration rate in the time dimension is calculated. Based on the cross-correlation coefficient, a two-domain coupling strength index is constructed. The severity aggravation factor, spatial diffusion acceleration factor and two-domain coupling strength index are nonlinearly combined to obtain the early warning hazard measurement value.

[0134] Warning levels are classified according to the numerical range of the warning hazard measurement value, and graded warning signals are generated based on the warning level and the discharge type identifier of the same discharge source.

[0135] The severity levels of the same discharge source at different times are extracted from the updated discharge source archive and arranged chronologically to form a time series. The change in severity level between adjacent time points is calculated by subtracting the severity level of the previous time point from the severity level of the later time point, yielding the increment value of the severity level. These increment values ​​are then arranged chronologically to construct a severity level evolution sequence. This sequence reflects the changing trend of local discharge source severity over time, providing basic data for subsequent early warning analysis.

[0136] The updated discharge source archive extracts the three-dimensional spatial coordinates of the same discharge source at different times. These coordinates are typically represented in a Cartesian coordinate system with a reference point of the transmission equipment as the origin. The displacement of the three-dimensional spatial coordinates between adjacent times is calculated, i.e., the Euclidean distance between the two time points is calculated. The displacement is divided by the corresponding time interval to obtain the instantaneous migration rate. The instantaneous migration rate reflects the movement speed of a local discharge source in a short period of time, but may contain noise and fluctuations. The instantaneous migration rate is filtered in the time dimension using a moving average filtering method, selecting an appropriate time window length for averaging, to eliminate the influence of short-term fluctuations and obtain a smoother spatial migration rate.

[0137] A time-series fitting process was performed on the severity level evolution sequence to generate a severity level evolution curve. A polynomial fitting method was used, and an appropriate polynomial order was selected to ensure that the fitted curve reflects the overall trend of the data without overfitting. The resulting polynomial function is the severity level evolution curve. The curvature of this curve at each point was calculated. The curvature value represents the degree of curvature at that point, and the change in curvature reflects the acceleration or deceleration trend of the severity level change rate. Features of curvature change, such as the average value, maximum value, and rate of change of curvature, were extracted to construct a severity aggravation factor. This factor is a scalar value; the larger the value, the more significant the aggravation of the severity of partial discharge.

[0138] A time-series fitting method was used to generate a migration rate evolution curve for the spatial migration rate, employing a polynomial fitting approach. Curvature change features were extracted from the fitted curve to construct a spatial diffusion acceleration factor. This factor reflects the degree of acceleration of the local discharge source in space; a larger value indicates a more pronounced acceleration and a stronger diffusion trend.

[0139] The cross-correlation coefficient between the severity evolution sequence and the spatial migration rate over time is calculated. The two sequences are aligned to the same time point, and the Pearson correlation coefficient is calculated. The cross-correlation coefficient reflects the correlation between changes in discharge severity and spatial migration rate; a correlation coefficient close to 1 indicates a strong positive correlation, close to -1 indicates a strong negative correlation, and close to 0 indicates a weak correlation. A two-domain coupling strength index is constructed based on the cross-correlation coefficient. The absolute value of the cross-correlation coefficient can be used as the index, or a nonlinear transformation can be introduced to enhance the difference.

[0140] A warning hazard metric is obtained by nonlinearly combining the severity aggravation factor, the spatial diffusion acceleration factor, and the dual-domain coupling strength index. The nonlinear combination employs a weighted product form, assigning different weights to the three factors. These weights are determined statistically based on historical data from different types of discharge sources. The warning hazard metric is a comprehensive indicator reflecting the hazard level of a partial discharge source, taking into account changes in discharge severity, spatial mobility characteristics, and the coupling relationship between the two.

[0141] Warning levels are categorized into four levels—low risk, medium risk, high risk, and extremely high risk—based on the numerical range of the warning hazard measurement value, each corresponding to a different numerical interval. The warning level is combined with the discharge type identifier to generate a graded warning signal containing both warning level and discharge type information. This graded warning signal can guide subsequent maintenance decisions, with different warning levels corresponding to different maintenance priorities and handling strategies.

[0142] This implementation method achieves early warning of potential faults in power transmission equipment by analyzing the severity changes and spatial migration characteristics of partial discharge sources. It comprehensively considers the evolution trend of discharge severity and spatial migration characteristics, overcoming the limitations of traditional methods that rely on only a single feature. By introducing severity aggravation factors and spatial diffusion acceleration factors, the dynamic evolution process of discharge activity is accurately captured. The construction of a dual-domain coupling strength index reveals the intrinsic correlation between discharge severity and spatial migration, enhancing the accuracy of early warning. A nonlinear combination method is used to generate early warning hazard metrics, improving the characterization ability of complex discharge behaviors. The establishment of a graded early warning mechanism provides a scientific basis for power transmission equipment maintenance, realizing a shift from passive response to proactive prevention, effectively reducing equipment failure risks, extending the service life of power transmission equipment, and improving the safe and stable operation level of the power system.

[0143] A second aspect of the present invention provides a partial discharge characteristic diagnosis and localization system for power transmission equipment, the system comprising:

[0144] The first unit is used to collect multi-physical field signals of power transmission equipment through a distributed sensor array, perform cross-domain cross-correlation calculations on the multi-physical field signals to extract phase difference features and amplitude ratio features, and combine them to form a coupling feature vector.

[0145] The second unit is used to map the coupling feature vector to the feature space, and determine the discharge type identifier and severity level based on the distribution position of the coupling feature vector in the feature space;

[0146] The third unit is used to extract the signal energy intensity of each node in the distributed sensor array, calculate the spatial attenuation gradient of the energy intensity of each node, determine the main radiation direction of the discharge source based on the spatial attenuation gradient, use the main radiation direction to perform directional compensation on the signal arrival time of each node, and determine the three-dimensional spatial coordinates of the local discharge source based on the compensated signal arrival time.

[0147] The fourth unit is used to establish a discharge source archive, record the coupling feature vector and three-dimensional spatial coordinates, calculate the feature distance and spatial distance between the newly detected discharge event and the data stored in the archive, and perform weighted fusion. Based on the fusion result, the discharge source is assigned and the archive is updated.

[0148] The fifth unit is used to extract the severity level and three-dimensional spatial coordinates of the same discharge source at different times from the updated discharge source file, calculate the spatial migration rate, and generate graded early warning signals.

[0149] A third aspect of the present invention provides an electronic device, comprising:

[0150] processor;

[0151] Memory used to store processor-executable instructions;

[0152] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0153] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0154] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing and locating partial discharge characteristics in power transmission equipment, characterized in that, include: Multi-physical field signals of power transmission equipment are collected by a distributed sensor array. Cross-domain cross-correlation is performed on the multi-physical field signals to extract phase difference and amplitude ratio features, which are then combined to form a coupling feature vector. The coupling feature vector is mapped to the feature space, and the discharge type identifier and severity level are determined based on the distribution position of the coupling feature vector in the feature space. Extract the signal energy intensity of each node in the distributed sensor array, calculate the spatial attenuation gradient of the energy intensity of each node, determine the main radiation direction of the discharge source based on the spatial attenuation gradient, use the main radiation direction to perform directional compensation on the signal arrival time of each node, and determine the three-dimensional spatial coordinates of the local discharge source based on the compensated signal arrival time. Establish a discharge source archive to record the coupled feature vector and three-dimensional spatial coordinates, calculate the feature distance and spatial distance between the newly detected discharge event and the data stored in the archive, and perform weighted fusion. Based on the fusion result, assign the discharge source and update the archive. Extract the severity level and three-dimensional spatial coordinates of the same discharge source at different times from the updated discharge source file, calculate the spatial migration rate, and generate graded early warning signals.

2. The method according to claim 1, characterized in that, Multi-physics field signals from power transmission equipment are acquired using a distributed sensor array. Cross-domain cross-correlation is performed on these signals to extract phase difference and amplitude ratio features, which are then combined to form a coupling feature vector, including: The geometric delay difference of the signal propagation path is calculated based on the spatial position coordinates of each sensor node in the distributed sensor array. The geometric delay difference is then used to synchronously correct the time reference of the multi-physics field signals collected by each sensor node, thereby obtaining the time-domain aligned multi-physics field signals. The time-domain aligned multiphysics signals are grouped according to the physical field type. Cross-domain cross-correlation is performed between signals of different physical field types. The peak position of the cross-correlation function between signals of different physical field types is calculated by using a sliding time window. The time offset corresponding to the peak position of the cross-correlation function is used as the phase difference feature, and the amplitude ratio corresponding to the peak of the cross-correlation function is extracted as the amplitude ratio feature. The extracted phase difference features and amplitude ratio features are encoded using feature vectorization. The phase difference features and amplitude ratio features are then arranged in series according to a preset physical field combination order to form a coupled feature vector with a multi-dimensional structure.

3. The method according to claim 1, characterized in that, The coupling feature vectors are mapped to a feature space, and the discharge type and severity level are determined based on the distribution of the coupling feature vectors in the feature space, including: The cross-coupling term between the phase difference feature and the amplitude ratio feature in the coupling feature vector is extracted. The cross-coupling feature matrix is ​​constructed by calculating the product of the phase difference feature and the amplitude ratio feature corresponding to different combinations of physical field types. The cross-coupling feature matrix is ​​decomposed by tensor to extract multi-order coupling modes. The coupling feature vector and the multi-order coupling modes are combined and projected into the feature space. The nonlinear distribution surface formed by historical discharge events in the feature space is fitted to generate a discharge type discrimination manifold. At the same time, the temporal trajectory of the same discharge source in the feature space is tracked and the evolution direction vector field is extracted to construct the severity level evolution trajectory. Calculate the geodesic distance from the projection point of the coupled feature vector in the feature space to each discharge type discrimination manifold, extract the discharge type identifier with the smallest geodesic distance, calculate the tangential component of the severity level evolution trajectory of the projection point along the corresponding discharge type discrimination manifold, and determine the severity level based on the normalized position of the tangential component on the evolution trajectory.

4. The method according to claim 1, characterized in that, Extracting the signal energy intensity of each node in the distributed sensor array, calculating the spatial attenuation gradient of the energy intensity of each node, and determining the main radiation direction of the power source based on the spatial attenuation gradient include: Spectral decomposition is performed on the multi-physics field signals collected by each node of the distributed sensor array, and the cumulative value of spectral energy in the discharge characteristic frequency band is extracted as the signal energy intensity of each node. Calculate the ratio of the difference in signal energy intensity between adjacent nodes to the distance between nodes, obtain the energy attenuation rate of each node in three orthogonal directions in three-dimensional space, and combine the energy attenuation rates in the three orthogonal directions to form the spatial attenuation gradient of the energy intensity of each node. Extend the reverse vector of the spatial attenuation gradient of each node in three-dimensional space, calculate the spatial intersection density distribution of all reverse extension vectors, extract the peak position of the intersection density as the estimated discharge source position, calculate the line vector connecting the estimated discharge source position to each node, and sum the line vectors according to the signal energy intensity of each node to obtain the main radiation direction of the discharge source.

5. The method according to claim 1, characterized in that, Directional compensation is performed on the signal arrival time of each node using the main radiation direction. The three-dimensional spatial coordinates of the partial discharge source are determined based on the compensated signal arrival time, including: The main radiation direction is projected onto the line connecting each node and the estimated discharge source location, and the projection component of the main radiation direction in each line direction is calculated as the direction consistency coefficient of each node. Based on the directional consistency coefficient, a time correction amount is constructed for the signal arrival time of each node, and the time correction amount is superimposed on the signal arrival time of each node to obtain the compensated signal arrival time. Using the compensated signal arrival time as the time reference, establish isochronous surfaces for each node, extract the set of intersection lines of the isochronous surfaces corresponding to different nodes in three-dimensional space, and perform spatial curvature analysis on the set of intersection lines to identify the curvature abrupt change locations. The spatial coordinates corresponding to the curvature change location are used as candidate points. The direction of the candidate line connecting each candidate point and each node is calculated. The main radiation direction is projected onto the direction of each candidate line to obtain the candidate projection component. Calculate the deviation between the candidate projection component and the direction consistency coefficient corresponding to each candidate point in three-dimensional spatial coordinates, and select the candidate point with the smallest deviation value as the three-dimensional spatial coordinates of the local discharge source.

6. The method according to claim 1, characterized in that, Establishing a discharge source archive records the coupling feature vector and three-dimensional spatial coordinates. Calculating and weighting the feature distance and spatial distance between newly detected discharge events and the data stored in the archive, and then assigning the discharge source based on the fusion result and updating the archive, includes: The coupling feature vector is bound to the three-dimensional spatial coordinates of the local discharge source as a discharge source file entry and stored in the discharge source file formation file for data storage. Extract the coupling feature vectors of the data stored in the archive at different detection times, fit them in the feature space to generate feature evolution curves, extract the tangent vectors of the feature evolution curves to construct feature evolution trend vectors, and calculate the deviation distance between the coupling feature vectors of the newly detected discharge event and the feature evolution trend vectors at the extended prediction points in the feature space as the feature distance. Extract the three-dimensional spatial coordinates of the partial discharge source at different detection times from the data stored in the archive, fit the three-dimensional space to generate a spatial migration curve, extract the tangent vector of the spatial migration curve to construct a spatial migration trend vector, and calculate the deviation distance between the three-dimensional spatial coordinates of the partial discharge source of the newly detected discharge event and the extended prediction point of the spatial migration trend vector in the three-dimensional space as the spatial distance. The fusion distance is obtained by weighted summation of the feature distance and the spatial distance. The assigned discharge source is determined based on the fusion distance. The discharge source file entries of newly detected discharge events are appended to the file corresponding to the assigned discharge source to store the data and form an updated discharge source file.

7. The method according to claim 1, characterized in that, The severity level and three-dimensional spatial coordinates of the same discharge source at different times are extracted from the updated discharge source archive. The spatial migration rate is calculated and graded early warning signals are generated, including: The severity levels of the same discharge source at different times are extracted from the updated discharge source archive, and the changes in severity levels between adjacent times are calculated to construct a severity level evolution sequence. The three-dimensional spatial coordinates of the same discharge source at different times are extracted from the updated discharge source file. The instantaneous migration rate is obtained by calculating the ratio of the displacement of the three-dimensional spatial coordinates at adjacent times to the time interval. The instantaneous migration rate is then filtered in the time dimension to obtain the spatial migration rate. A severity level evolution sequence is time-series fitted to generate a severity level evolution curve. The curvature change features of the severity level evolution curve are extracted to construct a severity aggravation factor. A spatial migration rate evolution curve is time-series fitted to generate a migration rate evolution curve. The curvature change features of the migration rate evolution curve are extracted to construct a spatial diffusion acceleration factor. The cross-correlation coefficient between the severity level evolution sequence and the spatial migration rate in the time dimension is calculated. Based on the cross-correlation coefficient, a two-domain coupling strength index is constructed. The severity aggravation factor, spatial diffusion acceleration factor and two-domain coupling strength index are nonlinearly combined to obtain the early warning hazard measurement value. Warning levels are classified according to the numerical range of the warning hazard measurement value, and graded warning signals are generated based on the warning level and the discharge type identifier of the same discharge source.

8. A partial discharge characteristic diagnosis and location system for power transmission equipment, used to implement the method of any one of claims 1-7, characterized in that, include: The first unit is used to collect multi-physical field signals of power transmission equipment through a distributed sensor array, perform cross-domain cross-correlation calculations on the multi-physical field signals to extract phase difference features and amplitude ratio features, and combine them to form a coupling feature vector. The second unit is used to map the coupling feature vector to the feature space, and determine the discharge type identifier and severity level based on the distribution position of the coupling feature vector in the feature space; The third unit is used to extract the signal energy intensity of each node in the distributed sensor array, calculate the spatial attenuation gradient of the energy intensity of each node, determine the main radiation direction of the discharge source based on the spatial attenuation gradient, use the main radiation direction to perform directional compensation on the signal arrival time of each node, and determine the three-dimensional spatial coordinates of the local discharge source based on the compensated signal arrival time. The fourth unit is used to establish a discharge source archive, record the coupling feature vector and three-dimensional spatial coordinates, calculate the feature distance and spatial distance between the newly detected discharge event and the data stored in the archive, and perform weighted fusion. Based on the fusion result, the discharge source is assigned and the archive is updated. The fifth unit is used to extract the severity level and three-dimensional spatial coordinates of the same discharge source at different times from the updated discharge source file, calculate the spatial migration rate, and generate graded early warning signals.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images