Switch cabinet discharge point coordinate resolving method based on electromagnetic pulse and time difference positioning

Abstract

The invention relates to the technical field of partial discharge positioning, and particularly discloses a switch cabinet discharge point coordinate resolving method based on electromagnetic pulse and time difference positioning, and the method comprises the following steps: initializing a self-adaptive structure model and a drift compensation reference of a switch cabinet, and building a multipath matching database; and acquiring a multi-source synchronizing signal in the switch cabinet, and generating a four-dimensional hierarchical label according to the characteristic parameters of the multi-source synchronizing signal and the health degree monitoring result of the sensor. According to the electromagnetic pulse and time difference positioning-based switch cabinet discharge point position coordinate resolving method, through dynamic sampling and algorithm configuration driven by a four-dimensional hierarchical label, a system can adaptively adjust a processing strategy when various drifts or signal quality fluctuates occur in a sensor; for example, edge preserving filtering is adopted, noise is restrained, meanwhile, the pulse rising edge gradient is maintained, and the stability of the positioning precision under the non-ideal working condition is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of partial discharge positioning, in particular to a switch cabinet discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning. BACKGROUND

[0002] At present, using ultra-high frequency (UHF) or ultrasonic sensor combined with time difference positioning method (TDOA) to locate the internal partial discharge of the switch cabinet is the mainstream means of power system fault monitoring. However, the existing positioning method is usually based on an ideal static model, that is, the preset switch cabinet is a standard structure and the sensor is assumed to be in a constant ideal performance state. However, in actual operation, a large number of non-standard or transformed switch cabinets have complex and variable structures, and the pre-set signal propagation path model cannot adapt to the actual environment. At the same time, the long-term running sensor inevitably appears the sub-health state such as sensitivity decline, reference drift or installation position micro-motion. The mismatch between this static ideal model and the dynamic complex scene leads to that not only the signal time delay cannot be accurately calculated, but also a large number of false alarms are often produced due to the inability to distinguish between environmental noise and drift error, which seriously affects the robustness of the positioning system.

[0003] Moreover, when fine coordinate calculation is carried out in the complex cabinet interior, the existing technology faces a conflicting problem. On the one hand, in order to suppress the complex electromagnetic interference in the field, a high-order filtering algorithm is usually needed, but this will smooth the extremely steep rising edge of the nanosecond discharge pulse, causing the key feature point of time difference extraction to be blurred, thus causing the contradiction that the stronger the noise reduction, the lower the time domain accuracy.

[0004] On the other hand, the switch cabinet is filled with irregular components such as insulators and copper bars, resulting in a highly non-convex complex topology of the discharge feasible space. The existing geometric gravity center method or simple statistical average method often obtains a coordinate solution located in the interior of the component entity, i.e. the physical forbidden zone. This mathematically optimal solution is completely wrong in physics, directly leading to the inability to carry out subsequent operation and maintenance. SUMMARY

[0005] The present application aims to at least solve one of the technical problems in the related art. To this end, the purpose of the present application is to provide a switch cabinet discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning, so as to improve the positioning accuracy of the discharge point.

[0006] To achieve the above-mentioned purpose, the first aspect of the present application provides a switch cabinet discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning, comprising the following steps:

[0007] initializing the adaptive structure model of the switch cabinet and the drift compensation reference, and establishing a multi-path matching database;

[0008] Collect multi-source synchronization signals in the switch cabinet, and generate a four-dimensional hierarchical label based on the characteristic parameters of the multi-source synchronization signals and the sensor health monitoring results;

[0009] In response to the four-dimensional hierarchical label, the corresponding dynamic sampling frequency and algorithm configuration strategy are triggered, and the time delay and drift correction of the multi-source synchronization signal are performed in combination with the adaptive structural model to obtain the corrected time difference and distance data.

[0010] Based on the corrected time difference and distance data, a pre-optimized positioning solution is performed on the edge side to obtain an approximate optimal solution, and incremental iterative optimization is performed with the approximate optimal solution as the initial value to output the final discharge point coordinates.

[0011] The generation process of the four-dimensional hierarchical label includes: parsing the collected signal to extract signal strength features and noise level features;

[0012] The collected signals are initially matched with the multipath matching database to determine path complexity characteristics;

[0013] The drift state characteristics of the sensors are determined based on the amplitude standard deviation of continuously acquired signals, the amplitude attenuation relative to the drift compensation benchmark, and the path matching deviation among multiple sensors.

[0014] The four-dimensional hierarchical label is generated by combining the signal strength characteristics, noise level characteristics, path complexity characteristics, and drift state characteristics.

[0015] To achieve the above objectives, a second aspect of the present invention proposes a system for calculating the coordinates of a switchgear discharge point based on electromagnetic pulse and time difference positioning, the system comprising:

[0016] The initialization module is used to initialize the adaptive structural model and drift compensation benchmark of the switchgear, and to establish a multi-path matching database.

[0017] The signal sensing and grading module is used to collect multi-source synchronous signals in the switch cabinet and generate four-dimensional grading labels based on the characteristic parameters of the multi-source synchronous signals and the sensor health monitoring results.

[0018] The dynamic processing module is used to respond to the four-dimensional hierarchical label, trigger the corresponding dynamic sampling frequency and algorithm configuration strategy, and combine the adaptive structural model to perform time delay extraction and drift correction on the multi-source synchronization signal.

[0019] The collaborative solution module is used to perform edge-side pre-optimization positioning solution based on the corrected data to obtain an approximate optimal solution, and to perform incremental iterative optimization with the approximate optimal solution as the initial value, and output the final discharge point coordinates;

[0020] The generation logic of the four-dimensional hierarchical label in the signal perception and hierarchical module includes: generating the four-dimensional hierarchical label by integrating signal strength features, noise level features, path complexity features, and drift state features obtained based on sensor benchmark data analysis.

[0021] To achieve the above objectives, a third aspect of the present invention provides a readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for calculating the coordinates of the discharge point of a switchgear based on electromagnetic pulse and time difference positioning.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] The switch cabinet discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning in this embodiment of the invention, through dynamic sampling and algorithm configuration driven by four-dimensional hierarchical tags, enables the system to adaptively adjust the processing strategy when various types of sensor drift or signal quality fluctuations occur. For example, edge-preserving filtering is used to suppress noise while maintaining the steepness of the pulse rising edge, thereby achieving stability of positioning accuracy under non-ideal working conditions.

[0024] Secondly, by introducing a constraint-aware center mapping process, the problem of the centroid falling into the forbidden zone caused by the non-convex feasible region is overcome by voxel projection in the edge-side solution, ensuring that the output approximate coordinates are always within the physically feasible space and avoiding invalid mathematical solutions. Finally, a collaborative architecture of edge pre-optimization + cloud incremental iteration is adopted, which not only ensures the real-time performance of online monitoring, but also achieves high-precision final positioning through the deviation compensation of the lightweight model. Attached Figure Description

[0025] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0026] Figure 1 This is a flowchart illustrating the method for calculating the coordinates of the discharge point of a switchgear based on electromagnetic pulse and time difference positioning provided by the present invention.

[0027] Figure 2 This is a schematic diagram of the spatial distribution of four-dimensional hierarchical tags based on multi-source feature parameters in the switch cabinet discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning provided by the present invention.

[0028] Figure 3 This is a schematic diagram of the nonlinear mapping relationship between local gradient eigenvalues ​​and dynamic smoothing weights, and the signal reconstruction process in the switchgear discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning provided by this invention.

[0029] Figure 4 This is a simulation comparison of the time-domain effects of edge-preserving adaptive filtering and traditional moving average filtering in the switchgear discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning provided by this invention.

[0030] Figure 5 This is a simulation diagram of the non-convex feasible region discretization and center mapping path under complex component constraints in the switch cabinet discharge point coordinate solution method based on electromagnetic pulse and time difference positioning provided by the present invention.

[0031] Figure 6 This is a schematic diagram illustrating the implementation of the switchgear discharge point coordinate calculation system based on electromagnetic pulse and time difference positioning provided by the present invention.

[0032] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0033] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0034] The following describes, with reference to the accompanying drawings, a method, system, and electronic device for calculating the discharge point coordinates of a switchgear based on electromagnetic pulse and time difference positioning, according to embodiments of the present invention.

[0035] Example 1:

[0036] This embodiment details a method for calculating the coordinates of discharge points in switchgear based on electromagnetic pulse and time-difference positioning. This method aims to address the problems of low accuracy and poor robustness in partial discharge positioning caused by the complex structure of switchgear, sensor aging and drift, and environmental interference in existing technologies. This embodiment achieves collaborative optimization positioning from the edge to the cloud by constructing an adaptive physical model and intelligent signal processing strategies.

[0037] like Figure 1 As shown, the method in this embodiment mainly includes the following steps:

[0038] S1. Before conducting any discharge monitoring, the primary task is to construct a digital benchmark that reflects the true physical state of the switchgear. This involves first initializing the switchgear's adaptive structural model and drift compensation benchmark, and then establishing a multipath matching database.

[0039] For example, the internal structure of a switchgear has a decisive influence on the propagation of electromagnetic waves and ultrasonic waves. Traditional positioning methods often assume that the cabinet is a cavity or contains only simple geometry, ignoring the blocking, reflection, and refraction effects of components such as insulators, copper busbars, and circuit breakers on signals. Therefore, this embodiment first establishes an adaptive structural model.

[0040] The process of initializing the adaptive structural model of the switchgear adopts a hierarchical processing strategy, specifically including: determining whether the current switchgear is a standard cabinet. Here, a standard cabinet refers to a cabinet that has a complete 3D CAD model and material parameters at the time of manufacture and has not undergone any subsequent modifications; a non-standard cabinet refers to an old model, a cabinet with missing data, or a cabinet that has been modified on-site. On-site modifications refer to cabinets that have had dehumidifiers installed or insulation components replaced.

[0041] If the judgment result is yes, meaning the current switchgear is a standard cabinet, the system will directly call the pre-set structured database. This structured database stores the precise three-dimensional dimensions, component layout, and material dielectric constants of mainstream switchgear models. Based on this prior data, the system can quickly generate an adapted multipath database and a three-dimensional temperature field basic model. The multipath database pre-calculates the theoretical time delay and attenuation characteristics of direct wave paths, single reflection paths, and multiple reflection paths from various typical spatial points within the cabinet to the sensor installation location; the three-dimensional temperature field basic model constructs an initial temperature gradient distribution map based on the heat source distribution within the cabinet.

[0042] If the judgment result is negative, indicating that the current switch cabinet is a non-standard cabinet, the system will initiate an active detection mechanism due to the lack of readily available structural data. Specifically, a preliminary sampling process at a first preset frequency is triggered to collect background signals from the empty cabinet. This first preset frequency is typically low, such as 100MHz to 300MHz, primarily used to acquire large-scale structural response characteristics rather than capturing high-frequency discharge pulses. The system controls a UHF sensor or a specific transmitting device to send detection signals and receives echoes. Subsequently, based on the propagation attenuation characteristics of the empty cabinet background signal and the number of ultrasonic reflections, the system uses a reverse imaging algorithm or feature matching algorithm to infer the core structural parameters of the cabinet within 500ms, such as the distance to the cabinet walls and the location of major obstacles. Based on the inference results, the system dynamically generates a simplified three-dimensional structural model and constructs an adaptive lightweight multipath database accordingly. This adaptive modeling capability ensures that the positioning algorithm can operate based on the real physical environment regardless of changes in the cabinet.

[0043] Simultaneously, to address the performance degradation caused by long-term sensor operation, the system synchronously initializes a drift compensation benchmark. This benchmark is sensor response data acquired through a high-precision standard source during initial system installation or periodic calibration, including standard sensitivity curves, inherent response delays, and baseline noise levels, providing a benchmark for subsequent health monitoring.

[0044] S2. After initialization, the system enters the real-time monitoring state. At this time, the following steps are performed: collect multi-source synchronization signals in the switch cabinet, and generate a four-dimensional hierarchical label based on the characteristic parameters of the multi-source synchronization signals and the sensor health monitoring results.

[0045] For example, the acquired multi-source synchronization signals include ultra-high frequency electromagnetic signals acquired by a UHF sensor, acoustic signals acquired by an ultrasonic sensor, transient ground voltage signals acquired by a TEV sensor, and environmental data acquired by a temperature sensor. All signals are time-aligned to the nanosecond level using a GPS or fiber optic synchronization system.

[0046] To achieve intelligent resource scheduling, this embodiment innovatively proposes the concept of four-dimensional hierarchical labels. This is not merely a simple assessment of signal quality, but also a guiding principle for subsequent algorithmic decisions. The generation process of the four-dimensional hierarchical labels includes the following detailed steps:

[0047] First, the acquired signals are analyzed to extract signal strength and noise level features. Signal strength features are determined based on the peak amplitude or energy integral of the UHF signal and are divided into three levels: high, medium, and low. Noise level features are determined by calculating the mean square error (MSE) or signal-to-noise ratio (SNR) of the signal in the non-pulse region and are also divided into three levels: low, medium, and high.

[0048] Secondly, the collected signals are initially matched with the multipath matching database to determine path complexity characteristics. The system compares the real-time collected signal waveform characteristics with the fingerprint characteristics in the multipath database. If the signal characteristics highly match the direct wave characteristics, the path complexity is determined to be simple (direct); if the signal exhibits obvious reflection and superposition characteristics, it is determined to be complex (multipath).

[0049] Furthermore, based on the amplitude standard deviation of continuously acquired signals, the amplitude attenuation relative to the drift compensation benchmark, and the path matching deviation among multiple sensors, the drift state characteristics of the sensors are determined. This is crucial to ensuring the long-term robustness of the system. Specifically:

[0050] 1. Calculate the standard deviation of the amplitude of multiple consecutive sets (e.g., 10 sets) of signals collected by the same sensor. If the standard deviation exceeds the preset threshold, it indicates that the sensor output is unstable and is marked as stability drift.

[0051] 2. Under normalization conditions, compare the current signal amplitude with the initial amplitude in the drift compensation reference. If the attenuation exceeds the preset sensitivity threshold, it is marked as sensitivity drift.

[0052] 3. The sensor position is inferred from the signal propagation path matching results between multiple sensors. If the inferred position deviates from the recorded installation position by more than a preset distance threshold, such as 3cm, it is marked as position drift.

[0053] Finally, the system integrates the signal strength characteristics (e.g., Class A / Class B / Class C), noise level characteristics, path complexity characteristics, and drift state characteristics to generate the final four-dimensional hierarchical label. Refer to "High Intensity - Low Noise - Direct Wave - No Drift" or "Low Intensity - High Noise - Multipath - Sensitivity Drift".

[0054] like Figure 2 This is a visualization of the spatial distribution of four-dimensional hierarchical labels based on multi-source feature parameters. The diagram visually illustrates how four-dimensional hierarchical labels map complex multi-source signal features into separable spatial clusters, thereby driving subsequent dynamic sampling strategies. Figure 2 The three-dimensional Cartesian coordinate system established in the process corresponds to the normalized signal intensity features, noise level features, and path complexity features obtained in the signal feature extraction step, while the different shapes and grayscale colors of the data points represent the fourth dimension of sensor drift state features.

[0055] Depend on Figure 2 As can be seen, the collected historical and real-time data do not exhibit a disordered distribution in the feature space, but rather naturally form three significant clustering regions based on the operating condition characteristics. The green square data group located in the upper left of the coordinate system represents the ideal Class A operating condition region. Here, the signal strength characteristic value approaches the normalized high value, the noise level characteristic value is low, and the path complexity characteristic shows high correlation, indicating that it is mainly a direct wave component, while the sensor is in a drift-free state. Tags falling into this region will directly trigger the system to enter a low-power mode, calling the second preset frequency sampling and simplified particle swarm optimization algorithm.

[0056] Conversely, the red cross data cluster located in the lower right corner of the coordinate system represents the harsh Class C operating condition area. Here, the signal strength is weak and accompanied by high-intensity background noise. The low path complexity eigenvalue indicates that the signal has experienced severe multipath reflection, and the sensor is diagnosed with sensitivity drift or position drift. Tags falling into this area pose an extremely high computational risk; therefore, the system will forcibly trigger the highest-level fourth preset frequency sampling and start the hybrid optimization algorithm, while simultaneously activating the drift correction module to compensate for amplitude attenuation and coordinate deviation.

[0057] The orange dotted area between these two conditions corresponds to Level B operating conditions. The system detects baseline jitter, or stability drift, in the sensor and activates edge-preserving adaptive filtering. This four-dimensional visualization classification mechanism based on feature space distribution demonstrates that the invention is not based on a simple judgment of a single threshold, but rather achieves an accurate profile of the complex operating environment of the switchgear through the coupling relationship of multi-dimensional features, ensuring the scientific nature and robustness of subsequent resource scheduling strategies.

[0058] S3. After obtaining the four-dimensional hierarchical label, the system does not process the data in a static manner, but instead executes the following steps: in response to the four-dimensional hierarchical label, it triggers the corresponding dynamic sampling frequency and algorithm configuration strategy, and combines the adaptive structural model to extract the time delay and correct the drift of the multi-source synchronization signal, thereby obtaining the corrected time difference and distance data.

[0059] The response to the four-dimensional hierarchical label triggers a corresponding dynamic sampling frequency and algorithm configuration strategy, including the following three typical configuration modes:

[0060] 1. When the signal strength indicated by the four-dimensional hierarchical label is higher than the first strength threshold and there is no drift, the system determines that the current operating condition is good and no excessive computing resources are needed. At this time, sampling is performed using the second preset frequency, and a simplified particle swarm optimization algorithm is configured for solution. The simplified particle swarm optimization algorithm has a smaller population size, fewer iterations, and can output results quickly;

[0061] 2. When the signal strength indicated by the four-dimensional hierarchical label is between the first and second strength thresholds, or when there is stability drift, it indicates that the signal quality is generally poor or fluctuates. In this case, sampling is performed using a third preset frequency, and a standard particle swarm optimization algorithm is configured for solution. A higher sampling rate and a standard iteration scale help capture the true features amidst fluctuations.

[0062] 3. When the signal strength of the four-dimensional hierarchical label is lower than the second strength threshold, or when there is sensitivity drift or position drift, it indicates poor operating conditions and a high risk of missed detections or misjudgments. In this case, sampling is performed using a fourth preset frequency, and a hybrid optimization algorithm is configured for solution. The hybrid optimization algorithm typically combines algorithms with strong global search capabilities, such as genetic algorithms, with algorithms with strong local convergence capabilities, such as sequential quadratic programming (SQP). Although the computation time increases, it maximizes the success rate of solution under complex conditions.

[0063] To ensure logical consistency, the values ​​of the first, second, third, and fourth preset frequencies increase sequentially. For example, the first preset frequency is 300MHz, which can be used for background scanning; the second preset frequency is 500MHz; the third preset frequency is 800MHz; and the fourth preset frequency is 1GHz or higher.

[0064] After determining the sampling strategy and algorithm configuration, the system needs to perform refined drift correction and time delay extraction on the signal. Existing methods often directly process the original waveform, ignoring the errors caused by sensor drift. In this embodiment, the adaptive structural model is used to extract time delay and correct drift for multi-source synchronization signals, specifically including targeted treatment for different drift types:

[0065] For example, in response to the drift state feature indicating position drift, the system no longer trusts the preset sensor coordinates, but corrects the sensor coordinates based on the reversed position offset, and recalculates the theoretical path length based on the corrected coordinates.

[0066] For example, in response to the drift state characteristic indicating sensitivity drift, the system identifies that the signal amplitude has been non-linearly suppressed. Therefore, it amplifies the amplitude of the acquired signal proportionally according to the amplitude attenuation to restore the true shape of the waveform and ensure that the subsequent threshold triggering logic works normally.

[0067] For example, in response to the drift state characteristic indicating stability drift, it suggests that non-Gaussian noise or baseline jitter has been introduced into the signal. In this case, the system performs edge-preserving adaptive filtering on the acquired signal. This filtering, unlike ordinary moving averages, smooths out noise while protecting the steep rising edge of the discharge pulse by detecting the signal gradient, avoiding time delay extraction lag caused by filtering.

[0068] Furthermore, besides the sensor's own drift, environmental factors, especially temperature, have a significant impact on the propagation speed of ultrasound, thus affecting distance calculation. Therefore, the process of time delay extraction and drift correction for multi-source synchronous signals also includes temperature field correction.

[0069] The temperature field correction includes: First, acquiring real-time temperature data inside the cabinet. This data comes from temperature sensors or infrared temperature probes deployed in different areas inside the cabinet. Second, updating the temperature distribution of each signal propagation path sub-segment in the three-dimensional temperature field basic model using the real-time data. Due to the uneven temperature distribution inside the switch cabinet, for example, the temperature in the busbar compartment is much higher than that in the cable compartment, the signal propagation path is discretized into several tiny sub-segments, each with an independent temperature value.

[0070] Then, based on the correspondence between gas sound velocity, thermodynamic temperature, and specific humidity, the propagation speed of ultrasound in each sub-segment is calculated. The specific calculation formula is as follows:

[0071] ;

[0072] in, Indicates the speed of ultrasonic wave propagation; This represents the specific heat ratio of a gas; for dry air, it is usually taken as 1.4. This represents the molar gas constant, which is approximately 8.314 J / (mol·K). This represents the corrected thermodynamic temperature variable. With the actual measured thermodynamic temperature There is a functional relationship between q, which can usually be represented as W This is to correct for the slight effect of humidity on the speed of sound.

[0073] Finally, the equivalent propagation distance is corrected based on the propagation speed of each sub-segment. By integrating or summing the velocity-time products of each sub-segment, a highly accurate physical propagation distance is obtained, thereby eliminating the positioning error caused by the single temperature assumption.

[0074] S4. After the complex signal conditioning described above, the system obtains high-quality time difference and distance data. The next step is to perform pre-optimized positioning calculations on the edge side based on the corrected time difference and distance data to obtain an approximate optimal solution. This step is typically performed in an edge computing gateway deployed near the switch cabinet and requires a fast response time.

[0075] The pre-optimization localization calculation at the edge side to obtain an approximate optimal solution includes a dual localization fusion mechanism:

[0076] The first step involves performing three-dimensional geometric calculations based on ultra-high frequency electromagnetic signals to obtain the first relocation region. UHF signals propagate extremely fast and are relatively less affected by multipath effects, making them suitable for quickly defining a large, approximate area.

[0077] The second step involves constructing a spherical space centered on the point of strongest signal, based on the energy attenuation characteristics of the transient ground voltage signal, to obtain the second relocation region. The TEV signal mainly propagates along the cabinet surface, and its amplitude attenuation is strongly correlated with distance, providing another dimension of spatial constraint.

[0078] The third step is to determine the overlapping portion between the first relocation region and the second relocation region. This overlapping portion is the area where discharge is most likely to occur.

[0079] However, the internal structure of switchgear is complex, and the overlapping area may contain entities such as insulators and conductors. Discharge can only occur in the air gap or on the insulating surface, and not inside the conductor. Simply taking the geometric center of the overlapping area may result in an invalid coordinate located inside the copper busbar. Therefore, this embodiment performs a constraint-aware center mapping process in the overlapping part to obtain the approximate optimal solution.

[0080] For example, the constraint-aware center mapping process includes the following sub-steps: First, the overlapping portion after removing the component exclusion zones is discretized into a set of effective voxel points with a preset resolution. The system calls an adaptive structural model to mark the space occupied by metal conductors and insulating entities within the overlapping area as exclusion zones. The remaining space, including air and insulating surfaces, is divided into tiny cubic units to form the effective point set. Second, the geometric mean coordinates of the effective voxel point set are calculated to obtain the theoretical centroid. This is a center point obtained through purely mathematical calculation. Subsequently, using the component distribution data in the adaptive structural model, it is determined whether the theoretical centroid is located within the component exclusion zones.

[0081] If the theoretical centroid is not located within the restricted area of ​​the component, it indicates that the center point is physically feasible, and the theoretical centroid is directly used as the approximate optimal solution.

[0082] If the theoretical centroid is located within the restricted area of ​​the component, for example, inside the insulator, it indicates that the feasible region exhibits a non-convex shape, such as a ring.

[0083] At this point, the valid voxel set is traversed, and the voxel with the smallest Euclidean distance to the theoretical centroid is selected as the approximate optimal solution. This method is equivalent to projecting the invalid theoretical centroid onto the nearest legal point in space, ensuring the physical rationality of the approximate solution and providing reliable initial values ​​for subsequent fine-tuning.

[0084] S5. Although the edge side provides an approximate solution, powerful computing power from the cloud is still needed to achieve centimeter-level positioning accuracy. The system then executes the following steps: using the approximate optimal solution as the initial value, iteratively optimizes the solution incrementally, and outputs the final coordinates of the discharge point.

[0085] This embodiment introduces a deep learning model to assist in solving the physical equations. The incremental iterative optimization using the approximate optimal solution as the initial value includes: calling the corresponding pre-trained Long Short-Term Memory (LSTM) sub-model based on the four-dimensional hierarchical labels. Multiple lightweight LSTM models are pre-trained in the cloud for different scenarios. Electromagnetic signal time difference features, ultrasonic wave shape features, and environmental parameters are input into the pre-trained LSTM sub-model, which outputs multi-dimensional compensation values ​​including time difference deviation, temperature deviation, path deviation, and drift deviation. This model learns the nonlinear mapping relationship between measured values ​​and true values ​​from a large amount of historical data, enabling it to predict residual errors that are difficult for the current physical model to resolve. The multi-dimensional compensation values ​​are integrated into the incremental iterative process to correct the approximate optimal solution to obtain the final discharge point coordinates. The incremental iteration here typically uses the Newton-Gaussian method or the Levenberg-Marquardt algorithm. Due to the accurate initial value (marginal pre-optimized solution) and accurate error compensation (LSTM output), the iteration process converges extremely quickly, usually requiring only 3-5 iterations to achieve very high accuracy.

[0086] S6. In order to adapt to different monitoring needs, such as the need for speed in emergency fault diagnosis and accuracy in daily inspection, the system performs a dynamic balance judgment before outputting the final discharge point coordinates in the final execution step.

[0087] Before outputting the final discharge point coordinates, the process includes: obtaining the type of the current monitoring scenario and its corresponding preset accuracy threshold and preset real-time threshold. For example, a live-line detection scenario requires a response time of less than 50ms and an accuracy requirement of ±10cm; while an offline diagnostic scenario requires a response time of less than 500ms but an accuracy requirement of ±1cm. The system then determines whether the positioning accuracy and calculation time under the current configuration meet the preset accuracy threshold and the preset real-time threshold. Positioning accuracy can be estimated by the residual size, while the calculation time is obtained by the system timer. If not, the sampling frequency or algorithm configuration strategy is adjusted based on a preset accuracy and real-time trade-off model until the output result meets the priority weight requirements of the current monitoring scenario. For example, if insufficient accuracy is detected, the system may instruct the next round of acquisition to increase the sampling frequency (possibly from the second frequency to the third frequency) or switch to a more complex hybrid optimization algorithm, forming a closed-loop feedback control system.

[0088] In summary, this embodiment constructs a high-precision and robust partial discharge positioning system for switchgear by deeply integrating physical models with artificial intelligence, working collaboratively with edge computing and cloud computing, and adaptively compensating for sensor drift and environmental changes.

[0089] Example 2:

[0090] This embodiment addresses the most challenging issue of stability drift in partial discharge monitoring of switchgear, and describes a specific implementation of edge-preserving adaptive filtering. As mentioned in Embodiment 1, when the four-dimensional hierarchical tag indicating sensor generated by the system exhibits stability drift, it means that non-Gaussian random noise or baseline fluctuations have been introduced into the acquired signal. In this case, if traditional linear filtering methods, such as moving average or Gaussian filtering, are directly used, although they can smooth the noise, they can easily obliterate the crucial rising edge characteristics of the partial discharge pulse signal, leading to a catastrophic decrease in the accuracy of Time Difference of Origin (TDOA).

[0091] The edge-preserving adaptive filtering process described in this embodiment is not a single algorithmic step, but a closed-loop processing system that includes feature extraction, weight mapping, and signal reconstruction. This system runs in an edge computing gateway or a high-performance FPGA module, aiming to intelligently distinguish between noise fluctuations and pulse transitions on a nanosecond-level timescale, thereby powerfully suppressing drift noise while perfectly preserving the steepness of the pulse rising edge.

[0092] Optionally, the first stage of this processing flow is the calculation of local gradient feature values. The system first receives the original discrete signal sequence acquired by the analog-to-digital converter (ADC) and after removing the DC component. In order to accurately capture the subtle changes in the signal in the time domain, the system does not directly analyze the absolute amplitude of the signal, but focuses on its rate of change. Specifically, the following steps are performed: calculating the local gradient feature value of each sampling point in the acquired signal sequence within a preset local window. Here, the preset local window is a time sliding window centered on the current sampling point, and the choice of its physical length is crucial. If the window is too small, it is susceptible to single-point noise interference; if the window is too large, it will reduce the sensitivity to rapidly changing signals. In this embodiment, the window covers several adjacent sampling points.

[0093] The mathematical definition of the local gradient eigenvalue strictly follows the principle of using the statistical value of the absolute value of the first-order difference between the amplitude of the current sampling point and the amplitudes of the adjacent sampling points. Assume the sampling point index at the current time is... The original acquired signal sequence is Then at time The amplitude of the sampled signal is The system sets the radius of the preset local window to be [value missing]. For example, take The window size is then 5 data points. For any point within the window... (in The system first calculates the absolute value of its first-order difference. This is to more robustly characterize the current center point. To determine the intensity of the signal at a given point, this embodiment uses the arithmetic mean of the absolute values ​​of the first-order differences of all points within the window as the local gradient feature value for that point. The calculation process can be expressed by the following formula.

[0094] ;

[0095] in, Indicates the first Local gradient feature values ​​at each sampling point; Indicates the first in the window The original signal amplitude at each location; Indicates the first The original signal amplitude at each location; This indicates the number of times the gradient is involved in the difference calculation. By introducing this statistically based gradient calculation, It can effectively smooth out instantaneous jumps caused by individual random noises, thus more realistically reflecting the overall activity or steepness of the signal within that local area. When When the value is small, it indicates that the signal in this area is flat, and the main components are baseline drift or low-frequency noise; when... A significant increase in the value indicates a drastic potential change in the region, which is highly likely to be the wavefront of a discharge pulse.

[0096] Optionally, after obtaining the local gradient feature values ​​of the entire sequence, the second stage begins: the generation of dynamic smoothing weights. The core of this step lies in using a pre-defined nonlinear mapping function to convert the local gradient feature values ​​into corresponding dynamic smoothing weights. The physical meaning of these weights is to determine whether the current sampling point retains more of its original value (trust observation) or adopts more of the neighborhood average (trust smoothing) in the final output. To achieve this, the nonlinear mapping function must possess a high degree of selectivity.

[0097] For example, the configuration logic of the nonlinear mapping function is as follows: when the local gradient eigenvalue indicates a non-pulse-jump region, a high smoothing weight is output; when the local gradient eigenvalue indicates a pulse-jump region, a low smoothing weight is output. To mathematically represent this logic, this embodiment constructs a mapping relationship based on an inverse S-shaped curve (Sigmoid-like) or a piecewise decay function. Two key decision thresholds are set: a noise gradient threshold... and pulse gradient threshold ,and These two thresholds are dynamically set based on the noise level features in the aforementioned four-dimensional hierarchical labels. The dynamic smoothing weight is defined as follows: Its value range is strictly limited to the interval [0, 1]. The nonlinear mapping function... The specific form is as follows:

[0098] ;

[0099] in, As a transitional median, it is usually taken as... ; This is an adjustment coefficient used to control the sensitivity of the weights to changes in the gradient, i.e., the steepness of the curve.

[0100] Analysis of the above formula shows that: 1. When the local gradient eigenvalues Very small ( When the system determines that it is currently in a stable background noise region, i.e., a non-pulse transition region, the smoothing weights of the function output are then determined. Approaching or equal to 1. This means the system will use maximum filtering to suppress baseline fluctuations caused by stability drift.

[0101] 2. Local gradient eigenvalues Very large ( When the system determines that it is currently in the rising edge or peak region of the discharge pulse, i.e., the pulse transition region, the smoothing weight of the function output is then determined. Approaching or equal to 0. This means the system will almost completely disable filtering and enter pass-through mode, thus ensuring that the rising edge of the pulse is not rounded or widened by the smoothing algorithm.

[0102] 3. When When the weights are in between, the transition is smooth, avoiding signal discontinuity caused by hard switching.

[0103] like Figure 3 This diagram illustrates the nonlinear mapping relationship between local gradient eigenvalues ​​and dynamic smoothing weights, as well as the signal reconstruction process. It vividly reveals the core operating mechanism and technical effect of the edge-preserving adaptive filtering algorithm in this embodiment.

[0104] Figure 3 The upper half (a) shows the preset nonlinear mapping function curve, with the horizontal axis representing the calculated local gradient eigenvalues ​​and the vertical axis representing the output dynamic smoothing weights. The curve shape shows that when the local gradient eigenvalues ​​are less than the set noise gradient threshold, the weights output by the mapping function remain stably at their maximum value, indicating that the system is in a fully smooth mode to maximize the suppression of background noise. However, as the gradient value increases and approaches the impulse gradient threshold, the weight curve exhibits a steep downward trend and rapidly approaches zero. This inverse S-shaped nonlinear response characteristic ensures that the algorithm can sensitively identify abrupt changes in the signal state.

[0105] Figure 3The lower half (b) demonstrates the actual reconstruction effect of this mapping mechanism in nanosecond-level time-domain signal processing. The gray curve represents the noisy original signal affected by stability drift, showing obvious spikes in its baseline. The blue dashed line represents the dynamically smoothing weights calculated in real time. At the pulse occurrence time of approximately forty nanoseconds, with the instantaneous burst of the signal gradient, the weight values ​​drop from high to low in a very short time, causing the system to instantly switch to pass-through mode. The red solid line represents the final edge-preserving filtered signal. As can be seen, this signal smooths most of the baseline noise in flat regions, while perfectly aligning with the steep trajectory of the original signal at the pulse rising edge, without the wavefront broadening or time delay phenomena common in traditional filtering. This dynamic time-domain weighting adjustment process strongly demonstrates that this invention can accurately preserve the key time-frequency characteristics of partial discharge pulses in strong noise environments, thus ensuring high-precision calculation of subsequent time difference positioning.

[0106] Optionally, the third stage of the processing flow is weighted reconstruction of the signal, namely, the step of: based on the dynamic smoothing weights, performing weighted summation and calculation on the acquired signal sequence to obtain a signal sequence that retains the steepness of the pulse rising edge. In this step, the system utilizes the calculated dynamic weights. Dynamic weighted fusion is performed between the original signal and the smoothed reference value. The smoothed reference value mentioned here is typically the arithmetic mean of all sampling points within the current window, denoted as... .

[0107] For example, the final output is the amplitude of the filtered signal. The calculation formula is as follows:

[0108] ;

[0109] in, The definition of is:

[0110] ;

[0111] The technical effect of this embodiment can be clearly seen through the above weighted summation formula: in the non-pulse region, because... The formula degenerates into That is, the output value equals the mean of the local window. In this case, the algorithm is equivalent to a standard moving average filter, which can extremely effectively filter out random glitches and low-frequency disturbances caused by sensor instability drift, significantly improving the signal-to-noise ratio. In the pulse rising edge region, because... The formula degenerates into This means the output value equals the original sampled value. At this point, the algorithm automatically stops the smoothing process, and the extremely steep nanosecond-level rising edges in the original signal are preserved intact in the output sequence.

[0112] Optionally, to further eliminate the slight phase delay that may be introduced during the filtering process, this embodiment can also employ a bidirectional zero-phase filtering strategy in practical engineering applications. That is, the gradient calculation and weighted summation described above are first performed in forward chronological order to obtain an intermediate sequence; then, the same operation is performed again on this intermediate sequence in reverse chronological order. Since the phase lag introduced by forward filtering and the phase lead introduced by reverse filtering mathematically cancel each other out, the final output signal sequence achieves zero drift on the time axis. This is crucial for Time Difference-of-Origin (TDOA) based algorithms because it ensures that the time references of signals acquired by different sensors remain strictly aligned after different levels of filtering, preventing artificial time difference errors caused by varying filtering intensities.

[0113] The signal sequence after the aforementioned edge-preserving adaptive filtering process exhibits a flat baseline and sharp pulses in its waveform display. Compared to the unprocessed drift-containing signal, its baseline noise standard deviation is significantly reduced; compared to the signal after traditional low-pass filtering, its pulse rise time broadening error is less than 1 ns, i.e., one sampling period at a 1 GHz sampling rate. This high-quality signal lays a solid data foundation for the accurate extraction of corrected time difference and distance data in subsequent steps, enabling the system to recover usable high-precision positioning information from the C+ stability drift condition identified by the four-dimensional hierarchical label even under conditions of sensor aging or harsh environments.

[0114] like Figure 4 The simulation diagram comparing the time-domain performance of edge-preserving adaptive filtering and traditional moving average filtering visually demonstrates the significant advantages of the method described in this embodiment in resolving the conflict between denoising and fidelity preservation.

[0115] Figure 4 The solid gray line represents the raw signal acquired by the sensor under conditions of stability drift, showing obvious random noise and baseline fluctuations superimposed on its waveform. The dashed blue line represents the waveform after processing with a traditional moving average filter with a window size of 5. Although this method effectively suppresses noise in flat regions, in the pulse initiation phase at approximately 40 nanoseconds, its linear averaging characteristics cause the originally steep rising edge to be forcibly softened, resulting in a phase lag of about two to three sampling points. This directly introduces a positioning error of tens of centimeters.

[0116] In contrast, the red solid line represents the waveform processed by the edge-preserving adaptive filtering algorithm described in this invention. It can be seen that in the non-pulse region, the red curve highly overlaps with the blue curve, demonstrating excellent smoothing and noise reduction capabilities, effectively correcting baseline instability caused by sensor drift. At the critical moment of the pulse rising edge, the red curve almost perfectly matches the transition trajectory of the original gray signal, without significant wavefront broadening or delay. This adaptive switching effect on a nanosecond timescale is due to the algorithm's ability to keenly perceive signal changes through local gradient features and dynamically reduce the smoothing weight to an extremely low level, thereby ensuring that the starting moment features required for time difference positioning are not erased by the filtering algorithm, perfectly meeting the stringent requirements of high-precision positioning of partial discharge in switchgear.

[0117] In summary, this embodiment successfully resolves the seemingly irreconcilable contradiction between noise suppression and feature preservation in traditional signal processing through statistical calculation of local gradient eigenvalues, gradient-based nonlinear dynamic weight mapping, and adaptive weighted reconstruction. This fully demonstrates the significant advantages of this technical solution in solving practical engineering problems related to partial discharge location in switchgear.

[0118] Example 3:

[0119] This embodiment focuses on the core mechanism of edge-side pre-optimization positioning solution in this invention, especially the constraint-aware center mapping process proposed for the complex non-convex spatial structure inside the switch cabinet.

[0120] For example, the edge-side pre-optimized positioning solution described in this embodiment first relies on the complementary utilization of the different propagation characteristics of multi-source physical signals. The system performs two coarse positioning operations in parallel: the first positioning is based on three-dimensional geometric calculation using ultra-high frequency electromagnetic signals. Utilizing the characteristics of electromagnetic wave propagation speed and steep wavefronts, a large first positioning region is quickly locked using the traditional time-difference hyperboloid intersection algorithm. The second positioning is based on the energy attenuation characteristics of transient ground voltage signals. Utilizing the physical law that the amplitude of TEV signals propagates along the metal surface of the cabinet and attenuates significantly with distance, a spherical space is constructed with the strongest detected signal point as the center and a preset attenuation distance as the radius, thus obtaining the second positioning region. Subsequently, the system determines the overlapping part of the first and second positioning regions through Boolean operations. This overlapping part represents the spatial set that is determined to be a high-probability discharge from both electromagnetic and voltage physical dimensions.

[0121] However, the interior of the switchgear is not an empty free space, but rather filled with physical components such as copper busbars, insulators, transformers, and circuit breaker contacts. Partial discharge typically occurs on the surface of the insulating medium, in internal air gaps, or at the tips of conductors, but absolutely not inside the metal conductor or at the center of a perfectly intact insulating entity. Traditional positioning algorithms often ignore this physical constraint, directly calculating the geometric center of the overlapping area as the result, leading to output coordinates that may be located in physically forbidden zones. Therefore, this embodiment performs a constraint-aware center mapping process in the overlapping portion to obtain the physically legal approximate optimal solution. This process relies on the component distribution data provided by the adaptive structural model established in the preceding steps, tightly integrating geometric calculations with physical constraints.

[0122] The first step in the constraint-aware center mapping process is spatial discretization. Specifically, the system executes the following steps: discretizing the overlapping portion after removing the component exclusion zones into a set of effective voxel points with a preset resolution. In this process, the system first calls the adaptive structural model of the switchgear to obtain the three-dimensional geometric boundary information of all internal components. The system defines the three-dimensional space occupied by these component entities as the component exclusion zones. Next, the system performs a mesh scan on the aforementioned determined overlapping portion. To balance the real-time performance of edge computing with the precision of positioning, a reasonable preset resolution needs to be set. This resolution defines the side length of the discretized mesh, for example, set to 10 mm or 5 mm. The system traverses each mesh node within the overlapping portion, determining whether the node is located within the component exclusion zone. Nodes located in air gaps, insulating surfaces, or other legal areas where discharge may occur are retained and added to a set, which is the set of effective voxel points.

[0123] For example, let the effective voxel set be... This set contains The set contains discrete three-dimensional coordinate points. The coordinate vector of an individual point is denoted as ,in The value range is from 1 to Through this discretization process, a continuous space that might originally contain complex curved boundaries and internal holes is transformed into a finite set of points that is easy for a computer to process, and this set of points has naturally eliminated all locations where discharge is physically impossible.

[0124] The second step in the process is to calculate the centroid in a statistical sense. The system executes the following steps: calculating the geometric mean coordinates of the effective voxel set to obtain the theoretical centroid. The theoretical centroid reflects the centroidal position of the current high-probability discharge region in spatial distribution and is the mathematically optimal estimate. The calculation formula is as follows:

[0125] ;

[0126] in, A three-dimensional coordinate vector representing the theoretical centroid; This represents the total number of voxels in the effective voxel set. Represents the first in the effective voxel set The three-dimensional coordinate vector of an individual point. This calculation process is highly efficient and can quickly provide a preliminary reference position.

[0127] The third step in the process is the core legality verification and correction. Because the internal structure of the switchgear may exhibit a C-shaped, O-shaped, or irregular non-convex topology (such as an annular air gap around the busbar), the theoretical centroid is calculated using the above formula. It is highly likely that the centroid will fall within the hollow, non-convex shape, specifically inside the busbar or insulator. Therefore, the system must perform the following step: using the component distribution data in the adaptive structural model, determine whether the theoretical centroid is located within the component's restricted area. This determination process is implemented using a point-polyhedron geometric inclusion relationship algorithm.

[0128] If the judgment result is negative, meaning the theoretical centroid is not located within the component's restricted area, this indicates that the theoretical centroid not only mathematically represents the center of the probability distribution but also physically represents a legitimate location where discharge may occur. In this ideal situation, the system directly outputs the theoretical centroid as the approximate optimal solution. This situation typically occurs in scenarios where the discharge area is relatively concentrated and there are no complex obstructions around it.

[0129] If the judgment result is yes, that is, if the theoretical centroid is located within the component's restricted area, this indicates that the mathematical center is a pseudo-solution. In this case, to obtain a solution that is both close to the mathematical center and conforms to physical facts, the system cannot simply discard the result; instead, it needs to perform a nearest-neighbor projection operation. Specifically, the system executes the following steps: traverse the set of valid voxels and select the voxel with the smallest Euclidean distance to the theoretical centroid as the approximate optimal solution.

[0130] This search process aims to find an effective set of voxels. a specific point in This makes the point an invalid theoretical centroid. The geometric distance is shortest. This optimal selection process can be described by the following mathematical formula:

[0131] ;

[0132] in, Represents the coordinate vector of the final determined approximate optimal solution; Represents any voxel in the effective voxel set; This represents the Euclidean norm, which is the straight-line distance in space. Physically, this means that the system pulls back or projects the theoretical center of mass, which is suspended inside the solid component, along the direction of the shortest spatial distance onto the nearest legal discharge surface or air gap.

[0133] This constraint-aware center mapping process has significant technical advantages. It avoids the iterative non-convexity or convergence to illegal solutions that may occur in traditional algorithms when dealing with non-convex feasible regions. This is achieved by restricting the solution to... Within the set, the system ensures that each output near-optimal solution is generated based on the actual internal structure of the switchgear. For example, when a discharge occurs at the insulator skirt, the overlapping area may surround the insulator, and the directly calculated centroid is located inside the insulator porcelain insulator. However, using the method in this embodiment, this centroid is automatically mapped onto the surface of the insulator. This not only provides extremely accurate and legitimate initial values ​​for cloud-based incremental iterative optimization in subsequent embodiments, avoiding the waste of computing power caused by invalid searches in restricted areas by the iterative algorithm, but also provides maintenance personnel with preliminary diagnostic information with clear physical meaning. Even when the network is interrupted and high-precision calculations cannot be performed by connecting to the cloud, this near-optimal solution directly output by the edge terminal has considerable reference value, sufficient to guide on-site personnel to narrow down the scope of troubleshooting.

[0134] like Figure 5 The figure shown is a simulation diagram of the discretization of the non-convex feasible region and the center mapping path under the constraints of complex components. This figure vividly illustrates the core solution logic of this embodiment when dealing with the complex non-convex spatial structure inside the switch cabinet.

[0135] Figure 5 The gray circular area in the center represents the physical components inside the switchgear, such as high-voltage insulators or metal busbars, which are physically restricted areas where partial discharge is impossible. The blue scattering around the components represents the effective voxel set generated by discretization after double coarse positioning and removal of restricted areas. These voxel sets constitute the practically possible discharge feasible domain, exhibiting a clear non-convex C-shaped topology.

[0136] Figure 5 The red cross marks the theoretical centroid obtained by directly calculating the geometric mean of the valid point set. It's clear that this centroid falls inside the gray component entity, meaning the mathematically optimal solution is physically invalid. To correct this error, this embodiment performs constraint-aware mapping, through... Figure 5The projection path, indicated by the black dashed line, forces the theoretical centroid located within the restricted area to be mapped to the solid green point in the effective voxel set that is closest to it in terms of Euclidean distance. This green point is the corrected approximate optimal solution, which preserves the centrality of the probability distribution to the maximum extent while strictly satisfying the physical constraint that it must be located in an air gap or on an insulating surface. This effectively avoids the technical defect of traditional algorithms that output illegal coordinates in complex structures.

[0137] Example 4:

[0138] like Figure 6 As shown in the illustration, corresponding to the above method embodiments, this embodiment details a switchgear discharge point coordinate calculation system based on electromagnetic pulse and time-difference positioning. This system forms the hardware and software architecture foundation for the methods described in Embodiments 1 to 3, aiming to address the technical challenges mentioned in the background art, such as the difficulty of existing positioning systems adapting to the complex and ever-changing internal structure of switchgear, the inability to effectively cope with sensor aging and drift, and the difficulty in balancing noise suppression and temporal feature extraction. Through modular design and an edge-cloud collaborative architecture, this system achieves intelligent processing throughout the entire process, from environmental perception to high-precision coordinate output.

[0139] The switchgear discharge point coordinate calculation system based on electromagnetic pulse and time difference positioning proposed in this embodiment mainly includes four core functional modules in its logical structure: an initialization module, a signal sensing and classification module, a dynamic processing module, and a collaborative calculation module. These modules are connected via a high-speed data bus and work collaboratively to achieve the high robustness and high accuracy mentioned in the beneficial effects.

[0140] First, the system includes an initialization module, which initializes the adaptive structural model and drift compensation benchmark of the switchgear and establishes a multipath matching database. Addressing the positioning errors caused by the use of ideal static models in existing technologies, the initialization module is configured as the system's digital base. This module incorporates a non-standard cabinet reverse engineering engine, capable of dynamically reconstructing the three-dimensional geometric distribution of insulators, copper busbars, and other components inside the switchgear based on the background signal characteristics of the empty cabinet described in the previous embodiment, thereby constructing an adaptive structural model highly consistent with the physical object. Simultaneously, this module also manages the drift compensation benchmark, records the standard response curves of sensors in the initial state, and establishes a multipath matching database containing the time delay characteristics of direct and reflected waves, providing accurate prior data support for subsequent signal processing.

[0141] Secondly, the system is equipped with a signal sensing and grading module. This module is used to collect multi-source synchronous signals within the switchgear and generate four-dimensional grading tags based on the characteristic parameters of the multi-source synchronous signals and the sensor health monitoring results. As the system's front-end, this module integrates a multi-channel high-speed analog-to-digital converter interface, capable of simultaneously accessing data from UHF electromagnetic sensors, ultrasonic sensors, and TEV sensors. To overcome the sub-health state of sensors after long-term operation, this module incorporates a feature extraction unit and a health assessment unit. The generation logic of the four-dimensional grading tags in the signal sensing and grading module specifically includes: comprehensively considering signal strength characteristics, noise level characteristics, path complexity characteristics, and drift state characteristics obtained based on sensor baseline data analysis to generate the four-dimensional grading tags. For example, when a low signal strength is detected accompanied by baseline fluctuations, the module automatically generates a tag indicating a low signal-to-noise ratio and stability drift, thereby guiding subsequent modules to perform targeted processing and avoiding missed detections or false alarms caused by the one-size-fits-all processing of traditional systems.

[0142] Furthermore, the system includes a dynamic processing module. This module responds to the four-dimensional hierarchical tags, triggering corresponding dynamic sampling frequencies and algorithm configuration strategies. It then combines this with the adaptive structural model to extract time delays and correct drift in the multi-source synchronization signals. This module acts as the system's adaptive engine, dynamically adjusting the ADC's sampling rate and the complexity of the filtering algorithm based on the four-dimensional hierarchical tags generated by the preceding modules. Specifically, when the tag indicates stability drift, this module invokes the edge-preserving adaptive filtering algorithm detailed in Example 2 to smooth noise while protecting the pulse rising edge. Additionally, this module is responsible for eliminating environmental interference by updating the sound velocity model using the real-time temperature distribution within the cabinet.

[0143] Finally, the core computational task of the system is undertaken by the collaborative solution module. This module performs edge-side pre-optimization localization calculations based on the corrected data to obtain an approximate optimal solution, and uses this approximate optimal solution as the initial value to perform incremental iterative optimization, outputting the final discharge point coordinates. This module adopts a heterogeneous computing architecture of edge + cloud. On the edge side, the module utilizes the constraint-aware center mapping process described in Example 3, combined with component restricted area data in the adaptive structural model, to calculate the theoretical centroid of the effective voxel point set. When the centroid falls into the restricted area, a search algorithm is used to find the nearest legal voxel point as the approximate optimal solution, fundamentally solving the problem that mathematical solutions are physically invalid due to non-convex feasible regions. Subsequently, the module uploads this physically valid approximate optimal solution to the cloud as the initial value for incremental iterative optimization, calls a pre-trained lightweight model for residual compensation, and finally outputs sub-centimeter-level discharge point coordinates.

[0144] In summary, the system in this embodiment, through the close collaboration of its various modules, not only achieves adaptive perception of the complex environment of the switchgear, but also effectively solves the conflict-related technical problems proposed in the background art through hierarchical processing and constraint mapping mechanisms, significantly improving the intelligence level and operation and maintenance efficiency of partial discharge monitoring of power equipment.

[0145] Example 5:

[0146] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0147] like Figure 7 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0148] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0149] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0150] The memory 103 stores a computer program corresponding to the switchgear discharge point coordinate calculation method based on electromagnetic pulse and time difference positioning in the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0151] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 7 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0152] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A switch cabinet discharge point position coordinate solving method based on electromagnetic pulse and time difference positioning, characterized in that, The method comprises the following steps: initializing an adaptive structure model of the switch cabinet and a drift compensation reference, and establishing a multi-path matching database; collecting multi-source synchronous signals in the switch cabinet, and generating a four-dimensional hierarchical tag according to characteristic parameters of the multi-source synchronous signals and monitoring results of sensor health degrees; in response to the four-dimensional hierarchical tag, triggering corresponding dynamic sampling frequencies and algorithm configuration strategies, and combining the adaptive structure model to perform time delay extraction and drift correction on the multi-source synchronous signals to obtain corrected time difference and distance data; based on the corrected time difference and distance data, performing pre-optimization positioning calculation on the edge side to obtain an approximate optimal solution, and performing incremental iterative optimization with the approximate optimal solution as an initial value to output a final discharge point coordinate; wherein, the generation process of the four-dimensional hierarchical tag comprises: analyzing the collected signals to extract signal strength characteristics and noise level characteristics; preliminarily matching the collected signals with the multi-path matching database to determine path complexity characteristics; based on the amplitude standard deviation of the continuously collected signals, the amplitude attenuation amount relative to the drift compensation reference, and the path matching deviation among the multiple sensors, determining the drift state characteristics of the sensors; comprehensively generating the four-dimensional hierarchical tag based on the signal strength characteristics, noise level characteristics, path complexity characteristics, and drift state characteristics.

2. The method of claim 1, wherein, The initialization of the adaptive structure model of the switch cabinet comprises: determining whether the current switch cabinet is a standard cabinet body; if yes, calling a pre-set structured database to generate an adaptive multi-path database and a three-dimensional temperature field basic model; 3. The method of claim 1, wherein, if no, triggering a preliminary sampling process at a first preset frequency to collect an empty cabinet background signal, inversely deducing cabinet structure parameters according to the propagation attenuation characteristics and the number of ultrasonic reflection waves of the empty cabinet background signal, and generating a simplified three-dimensional structure model and an adaptive multi-path database. The response to the four-dimensional hierarchical tag to trigger corresponding dynamic sampling frequencies and algorithm configuration strategies comprises: when the four-dimensional hierarchical tag indicates that the signal strength is higher than a first strength threshold and there is no drift, a second preset frequency is used for sampling, and a simplified particle swarm optimization algorithm is configured for calculation; when the four-dimensional hierarchical tag indicates that the signal strength is between the first strength threshold and a second strength threshold, or there is a stable drift, a third preset frequency is used for sampling, and a standard particle swarm optimization algorithm is configured for calculation; when the four-dimensional hierarchical tag indicates that the signal strength is lower than the second strength threshold, or there is a sensitivity drift or a position drift, a fourth preset frequency is used for sampling, and a hybrid optimization algorithm is configured for calculation; 4. The method of claim 1, wherein, wherein, the values of the first, second, third and fourth preset frequencies increase in turn. The combination of the adaptive structure model for time delay extraction and drift correction on the multi-source synchronous signals comprises: in response to the drift state characteristics indicating a position drift, correcting the sensor coordinates according to the inversely deduced position offset, and recalculating the theoretical path length; in response to the drift state characteristics indicating a sensitivity drift, amplifying the amplitude of the collected signal in proportion to the amplitude attenuation amount; In response to the drift state feature indicating stability drift, edge-preserving adaptive filtering processing is performed on the collected signal.

5. The method of claim 4, wherein, The edge-preserving adaptive filtering processing performed on the collected signal comprises: calculating a local gradient feature value of each sampling point in the collected signal sequence within a preset local window, the local gradient feature value being a statistical value of a first-order difference absolute value of amplitudes of the current sampling point and adjacent sampling points before and after the current sampling point; using a preset nonlinear mapping function to convert the local gradient feature value into a corresponding dynamic smoothing weight; the nonlinear mapping function is configured to: output a high smoothing weight when the local gradient feature value indicates a non-pulse jump region, and output a low smoothing weight when the local gradient feature value indicates a pulse jump region; based on the dynamic smoothing weight, performing weighted summation calculation on the collected signal sequence to obtain a signal sequence that retains the steepness of the pulse rising edge.

6. The method of claim 1, wherein, The process of performing time delay extraction and drift correction on the multi-source synchronous signal further comprises temperature field correction, and the temperature field correction comprises: acquiring real-time temperature data in the cabinet to update the temperature distribution of each signal propagation path sub-segment in the three-dimensional temperature field base model; calculating the propagation speed of ultrasonic waves in each sub-segment according to the corresponding relationship between gas sound speed and thermodynamic temperature and specific humidity; correcting the equivalent propagation distance based on the propagation speed of each sub-segment.

7. The method of claim 1, wherein, The execution of the pre-optimization positioning solution on the edge side to obtain an approximate optimal solution comprises: performing three-dimensional geometric solution based on the ultra-high frequency electromagnetic signal to obtain a first repositioning area; constructing a spherical space with the point of the strongest signal as the center of the sphere based on the energy attenuation characteristics of the transient ground voltage signal to obtain a second repositioning area; determining the overlapping part of the first repositioning area and the second repositioning area, and performing a constraint-aware center mapping process in the overlapping part based on the component distribution data in the adaptive structure model to obtain the approximate optimal solution.

8. The method of claim 7, wherein, The execution of the constraint-aware center mapping process comprises: discretizing the overlapping part after excluding the component forbidden area into a set of effective voxel points with a preset resolution; calculating the geometric mean coordinates of the set of effective voxel points to obtain a theoretical centroid; using the component distribution data in the adaptive structure model to determine whether the theoretical centroid is located in the component forbidden area; if the theoretical centroid is not located in the component forbidden area, directly taking the theoretical centroid as the approximate optimal solution; if the theoretical centroid is located in the component forbidden area, traversing the set of effective voxel points and selecting a voxel point with the smallest Euclidean distance from the theoretical centroid as the approximate optimal solution.

9. The method of claim 1, wherein, The execution of the incremental iterative optimization with the approximate optimal solution as the initial value comprises: calling a corresponding pre-trained long short-term memory network sub-model according to the four-dimensional hierarchical label; inputting electromagnetic signal time difference features, ultrasonic waveform features, and environmental parameters into the pre-trained long short-term memory network sub-model to output multi-dimensional compensation values including time difference deviation, temperature deviation, path deviation, and drift deviation; integrating the multi-dimensional compensation values into the incremental iterative process to correct the approximate optimal solution to obtain the final discharge point coordinate.

10. The method of claim 1, wherein, The output final discharge point coordinate further comprises the following steps: Obtain the type of the current monitoring scene and the corresponding preset accuracy threshold and preset real-time threshold; Determine whether the positioning accuracy and the calculation time under the current configuration meet the preset accuracy threshold and the preset real-time threshold; If not, adjust the sampling frequency or the algorithm configuration strategy based on the preset accuracy and real-time compromise model until the output result meets the priority weight requirement of the current monitoring scene.

Images