Partial discharge intelligent detection method and system

Abstract

The invention discloses a partial discharge intelligent detection method and a partial discharge intelligent detection system. The method comprises the following steps: firstly, synchronously acquiring a multi-channel ultrasonic signal and a surface infrared image sequence of to-be-detected power equipment; performing pulse detection, cross-channel clustering based on a sliding time window and time difference positioning on the ultrasonic signal to obtain a preliminary discharge point; the multi-moment infrared images are intelligently selected and fused based on the ultrasonic pulse active time period, and a target surface infrared image highlighting the abnormal temperature rise area is generated; performing spatial aggregation on the ultrasonic positioning point and the infrared abnormal region to form a suspected sound-heat associated discharge source; respectively extracting waveform features and temperature features from the aggregated ultrasonic pulse set and the infrared region, fusing the waveform features and the temperature features, and inputting the fused features into a dual-channel neural network model; and finally, the model outputs a discharge type identification result and a discharge intensity evaluation value of the suspected discharge source. According to the invention, through deep space-time fusion and feature level fusion, the positioning precision and identification accuracy of partial discharge detection are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of partial discharge detection, and particularly relates to a partial discharge intelligent detection method and system. BACKGROUND

[0002] Partial discharge is an important early sign of insulation deterioration of power equipment, and effective detection and identification thereof are of great significance for ensuring safe operation of power grids and preventing equipment from sudden failure. At present, the mainstream partial discharge detection technologies mainly include pulse current method, ultra-high frequency method and ultrasonic method. Among them, the ultrasonic method is widely used due to its strong anti-electromagnetic interference ability and convenience for positioning. However, there are some inherent limitations in relying on ultrasonic signals alone: firstly, ultrasonic signals are easily affected by the internal structure of the equipment, the propagation medium and background noise, and the detection sensitivity for small discharges or deep discharges is insufficient; secondly, the single ultrasonic positioning result has certain error, and it is difficult to distinguish the discharge types (such as internal discharge, surface discharge, corona discharge, etc.).

[0003] The infrared thermal imaging technology provides intuitive morphological evidence for partial discharge by detecting abnormal temperature rise on the surface of the equipment caused by discharge. However, the infrared method usually reacts slowly and is difficult to capture instantaneous discharge pulses, and the temperature rise is easily disturbed by non-discharge factors such as environmental temperature, load current, sunlight, etc., and the single use is easy to lead to misjudgment.

[0004] In the prior art, although attempts have been made to combine ultrasonic and infrared methods, they are mostly simple information juxtaposition or manual comparison, and lack deep temporal and spatial correlation and feature-level fusion. For example, the prior art fails to solve the core problem of how to actively and efficiently obtain high-quality infrared image evidence in partial discharge detection, and extract features that can reliably represent the thermal effect of discharge and have strong correlation with ultrasonic events. SUMMARY

[0005] The present application aims to overcome the above-mentioned defects, and provides a partial discharge intelligent detection method and system, which adopts an infrared image intelligent processing strategy guided by ultrasonic signals, focused on the discharge active period and fused with multi-time information, to significantly improve the efficiency and value of infrared technology in partial discharge joint detection.

[0006] In a first aspect, the present application provides a partial discharge intelligent detection method, comprising: acquiring a plurality of ultrasonic signals of a power equipment to be detected within a preset time period, wherein the plurality of ultrasonic signals are respectively collected by a plurality of sensors; pulse analyze the plurality of ultrasonic signals to obtain ultrasonic pulse signal sequences corresponding to the plurality of ultrasonic signals, and slide a preset sliding window on each ultrasonic pulse signal sequence to cluster each ultrasonic pulse signal meeting a time synchronization condition to obtain at least one ultrasonic pulse signal set, wherein each ultrasonic pulse signal sequence contains at least one ultrasonic pulse signal corresponding to the same ultrasonic signal; determine preliminary discharge points corresponding to the at least one ultrasonic pulse signal set; determine a target surface infrared image of the power equipment under test in a preset time period based on a preset image processing strategy, and aggregate each preliminary discharge point based on at least one abnormal temperature rise area in the target surface infrared image to obtain at least one preliminary discharge point set, wherein one preliminary discharge point set corresponds to one suspected discharge source, and one preliminary discharge point set is associated with one abnormal temperature rise area; extract a certain waveform feature sub-vector corresponding to a certain preliminary discharge point set and a certain temperature feature sub-vector of a certain abnormal temperature rise area, and perform feature fusion on the certain waveform feature sub-vector and the certain temperature feature sub-vector to obtain a certain fusion feature vector; input the certain fusion feature vector into a preset partial discharge recognition model, and the partial discharge recognition model outputs a discharge type recognition result and a discharge intensity evaluation value of a certain suspected discharge source.

[0007] In a second aspect, the present application provides a partial discharge intelligent detection system, comprising: an acquisition module configured to acquire a plurality of ultrasonic signals of a power equipment under test in a preset time period, wherein the plurality of ultrasonic signals are collected by a plurality of sensors respectively; an analysis module configured to pulse analyze the plurality of ultrasonic signals according to a preset signal detection rule to obtain ultrasonic pulse signal sequences corresponding to the plurality of ultrasonic signals, and slide a preset sliding window on each ultrasonic pulse signal sequence to cluster each ultrasonic pulse signal meeting a time synchronization condition to obtain at least one ultrasonic pulse signal set, wherein each ultrasonic pulse signal sequence contains at least one ultrasonic pulse signal corresponding to the same ultrasonic signal; a determination module configured to determine preliminary discharge points corresponding to the at least one ultrasonic pulse signal set; The aggregation module is configured to determine the infrared image of the target surface of the power equipment under test within a preset time period based on a preset image processing strategy, and aggregate each preliminary discharge point based on at least one abnormal temperature rise area in the infrared image of the target surface to obtain at least one set of preliminary discharge points, wherein one set of preliminary discharge points corresponds to one suspected discharge source, and one set of preliminary discharge points is associated with one abnormal temperature rise area. The extraction module is configured to extract a waveform feature vector corresponding to a certain set of initial discharge points and a temperature feature vector of a certain abnormal temperature rise region, and to fuse the waveform feature vector and the temperature feature vector to obtain a fused feature vector. The output module is configured to input a certain fused feature vector into a preset partial discharge identification model, and the partial discharge identification model outputs the discharge type identification result and discharge intensity evaluation value of a certain suspected discharge source.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the partial discharge intelligent detection method of any embodiment of the present invention.

[0009] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the partial discharge intelligent detection method of any embodiment of the present invention.

[0010] This application presents a partial discharge intelligent detection method and system. Through strict alignment of multi-channel synchronous ultrasound and infrared data, combined with a sliding window clustering algorithm based on a physical propagation model, it achieves accurate extraction and cross-sensor correlation of discharge events. Subsequently, it intelligently guides the screening of infrared keyframes during ultrasound active periods and generates enhanced diagnostic images through regional spatiotemporal fusion, effectively locating thermal anomalies. Furthermore, it utilizes a spatial aggregation mechanism of acoustic-thermal dual-modal evidence chains to cross-validate and fuse ultrasound positioning points and infrared hot zones, filtering out false signals. Based on this, multi-dimensional features are extracted from the aggregated pulse sequences and hot zones to construct a unified fusion feature vector characterizing the physical essence of the discharge. Finally, the partial discharge identification model analyzes this vector, simultaneously outputting high-precision discharge type identification results and quantitative intensity assessments. This method significantly improves the accuracy, reliability, and intelligence level of partial discharge detection. Attached Figure Description

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0012] Figure 1 A flow chart of a partial discharge intelligent detection method provided by an embodiment of the present application is shown in Figure 2 A structural block diagram of a partial discharge intelligent detection system provided by an embodiment of the present application is shown in Figure 3 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in DETAILED DESCRIPTION

[0013] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort are within the protection scope of the present application.

[0014] Please refer to Figure 1 , which shows a flow chart of a partial discharge intelligent detection method.

[0015] As shown in Figure 1 , the partial discharge intelligent detection method specifically includes the following steps: Step S101, obtaining a plurality of ultrasonic signals of a power equipment to be tested within a preset time period, wherein the plurality of ultrasonic signals are respectively collected by a plurality of sensors.

[0016] In this step, a wideband or resonant ultrasonic sensor suitable for partial discharge detection of power equipment is selected, and its effective frequency range usually covers 40 kHz to 300 kHz to capture characteristic ultrasonic signals generated by different discharge types. The sensor should have high sensitivity, good signal-to-noise ratio and certain directivity.

[0017] According to the structure, size and key monitoring area of the power equipment to be tested (such as transformer, GIS, switch cabinet), a plurality of sensors are pre-planned and installed. The laying principle is to form a certain geometric distribution (such as triangle, quadrilateral or around the key parts of the equipment) in space so as to facilitate subsequent time difference positioning. The three-dimensional spatial coordinates (X, Y, Z) of each sensor need to be accurately calibrated and recorded through measurement or design drawing.

[0018] In a preset time period, the integrated multi-channel synchronous data is used to sample the ultrasonic signals from each sensor at a high speed and convert them into digital signals. Specifically, the integrated multi-channel synchronous data acquisition system comprises: a synchronous acquisition card: provided with independent analog-to-digital converters (ADC) and sample-and-hold circuits, each channel shares the same high-stability crystal clock source, ensuring that the sampling clock of all channels is strictly synchronized, and the synchronization error is less than 10 nanoseconds; a signal conditioning unit: built-in programmable gain amplifier (PGA) and anti-aliasing filter. For each channel signal, the gain is uniformly set to 60 dB, and the band-pass filter with a bandwidth of 20 kHz-400 kHz is configured; a timing and triggering module: integrated GPS / Beidou dual-mode timing unit, providing accurate UTC time synchronization signal (PPS) and absolute time information.

[0019] Step S102, pulse analysis is performed on the plurality of ultrasonic signals to obtain an ultrasonic pulse signal sequence corresponding to the plurality of ultrasonic signals, and a sliding window is slid on each ultrasonic pulse signal sequence based on a preset sliding window, and each ultrasonic pulse signal meeting a time synchronization condition is clustered to obtain at least one ultrasonic pulse signal set, wherein one ultrasonic pulse signal sequence comprises at least one ultrasonic pulse signal corresponding to the same ultrasonic signal.

[0020] Specifically, the specific implementation of step S102 comprises the following sequentially executed sub-steps: Step S1021, parallel preprocessing of multi-channel ultrasonic signals: The original digital sequence of each ultrasonic signal channel obtained in step S101 is preprocessed in parallel: Band-pass filtering: a zero-phase distortion digital band-pass filter (such as a FIR filter) is used to limit the signal of each channel within the partial discharge characteristic ultrasonic frequency band of 40 kHz to 300 kHz, so as to maximally suppress low-frequency mechanical vibration and high-frequency electromagnetic noise.

[0021] Noise baseline estimation: for each channel signal after filtering, a sliding median method is used to estimate the dynamic noise baseline level. With a window length of 1 millisecond, the absolute median of the signal amplitude in the window is calculated as the estimated value N(t) of the noise level in this period.

[0022] Preliminary noise reduction: optionally, a wavelet threshold-based noise reduction algorithm can be applied to further improve the signal-to-noise ratio.

[0023] Step S1022, pulse event detection based on an adaptive threshold: Pulse detection is independently performed for each preprocessed channel signal: Dynamic threshold calculation: the pulse detection threshold Th(t) is dynamically calculated according to the noise baseline, for example: where K is a preset signal-to-noise ratio coefficient (usually 3 to 5). This mechanism ensures that the detection sensitivity remains constant when the noise fluctuates.

[0024] Threshold-crossing identification and pulse segmentation: identify all time points where the signal amplitude exceeds Th(t). Merge the signal amplitudes corresponding to consecutive threshold-crossing time points into one ultrasound pulse signal, and determine the peak occurrence time of the ultrasound pulse signal, where the peak occurrence time is the time point corresponding to the maximum signal amplitude in the ultrasound pulse signal.

[0025] For a single channel, arrange all the ultrasound pulse signals detected by it within a preset time period in ascending order of timestamp, i.e., form the ultrasound pulse signal sequence corresponding to this channel.

[0026] Step S1023: Multi-channel pulse clustering based on sliding time window: The goal of this step is to associate the pulses generated by the same spatial discharge source in different channel sequences.

[0027] Initialization of sliding window: Set a sliding time window W, whose length is determined according to the maximum propagation delay of ultrasound waves in the device medium and the maximum spacing of sensors, for example, set to 50 microseconds. The window slides on the global time axis with a fixed step size.

[0028] In-window pulse collection and association: For each window position, perform: Collection: Find all ultrasound pulses with timestamps falling within the current window from the ultrasound pulse signal sequences of all channels.

[0029] Channel deduplication: If there are multiple pulses in the current window for the same channel, usually only the one with the largest amplitude is retained to avoid interference from multiple pulses within a single channel.

[0030] Form a candidate set: The collected ultrasound pulse signals from different channels are considered as a candidate ultrasound pulse signal set. This set is initially considered to possibly correspond to an independent discharge event.

[0031] Set validity verification and refinement: Verify each candidate set: Quantity verification: The set contains at least ultrasound pulses from 2 different spatially located sensors. Sets with less than 2 ultrasound pulses are rejected.

[0032] Time synchronization verification (core condition): Calculate the arrival time difference between all pairs of ultrasound pulses in the set. Theoretically, ultrasound pulses generated by the same source should have an arrival time difference less than the maximum theoretical time difference Tmax required for the pulse to propagate from the source to the two farthest sensors. If all pulse pairs in the set satisfy this physical constraint, it is determined to satisfy the "time synchronization condition".

[0033] Output: At least one set of ultrasonic pulse signals is obtained after the above sliding window traversal and conditional verification.

[0034] Step S103, determine a preliminary discharge point corresponding to the at least one set of ultrasonic pulse signals.

[0035] In this step, it is determined whether the number of ultrasonic pulse signals in a set of ultrasonic pulse signals is greater than a preset number threshold. If it is greater than the preset number threshold, ultrasonic pulse signals from three different spatial position sensors in a set of ultrasonic pulse signals are selected. The time stamp and peak amplitude corresponding to each of the ultrasonic pulse signals of the three different spatial position sensors are obtained, and based on the three-dimensional spatial coordinates of the three different spatial position sensors pre-calibrated, the time difference of arrival between each pair of ultrasonic pulse signals is calculated. Based on the known propagation speed of ultrasonic waves in the insulating medium of the device, the three-dimensional spatial coordinates of the three different spatial position sensors, and the time difference of arrival between each pair of ultrasonic pulse signals, a set of hyperbolic positioning equations is established. The least square estimation algorithm is used to solve the hyperbolic positioning equation set, and a three-dimensional spatial coordinate corresponding to a set of ultrasonic pulse signals is obtained as a preliminary discharge point. If it is not greater than the preset number threshold, a target sensor is selected from the sensors corresponding to each ultrasonic pulse signal, and the three-dimensional spatial coordinates of the target sensor are taken as a reference coordinate to obtain a preliminary discharge point, wherein the target sensor is the sensor corresponding to the ultrasonic pulse signal with the highest peak amplitude.

[0036] In one specific embodiment, the total number N of ultrasonic pulse signals contained in a set of ultrasonic pulse signals is counted, and it is determined whether the total number N of ultrasonic pulse signals is greater than a preset number threshold (with a value of 3). If it is greater than the preset number threshold, the three ultrasonic pulse signals with the highest peak signal-to-noise ratio (SNR) are selected from all the ultrasonic pulse signals in the set of ultrasonic pulse signals, and the SNR calculation formula is: (peak amplitude / noise baseline of the sensor).

[0037] If the SNR requirement cannot be met, the three different sensor pulses are selected in descending order of peak amplitude.

[0038] The accurate time stamps t1, t2, and t3 (unit: microseconds, accuracy 0.1 μs) of the three ultrasonic pulse signals are obtained.

[0039] Read the three-dimensional coordinates of the three sensors corresponding to the three ultrasonic pulse signals from the calibration database: (x1, y1, z1), (x2, y2, z2), (x3, y3, z3).

[0040] Obtain the propagation speed v of ultrasonic waves in the current insulation medium of the device (unit: m / s). This speed value is real-time compensated and corrected according to the type of medium (oil, epoxy resin, etc.) and temperature.

[0041] Calculate the time difference of arrival: Take sensor 1 as the reference, calculate the relative time difference: Δt21 = t2 - t1, Δt31 = t3 - t1.

[0042] Establish the positioning equation set: Assume the discharge point coordinates are (x, y, z), and establish the equation set according to the hyperbolic positioning principle: , , Iterative solution: Use Taylor series expansion iterative method to solve, the specific steps are as follows: Set the initial iteration value , usually take the geometric center of the three sensor coordinates.

[0043] In the kth iteration, the above nonlinear equation set is expanded to the first order Taylor at , and the linearized error equation is obtained: , is the Jacobian matrix, the elements are the partial derivatives of the distance difference with respect to the coordinates, is the difference vector between the observed value and the calculated value, is the coordinate correction amount; Solve the linear least squares problem: , is the transpose of the Jacobian matrix; Update the coordinates: ; Repeat the iteration until the norm of the coordinate correction amount is less than the preset convergence threshold (such as 1mm) or reaches the maximum number of iterations; The final convergence P is the preliminary discharge point coordinates.

[0044] If it is not greater than the preset number threshold, from all the ultrasonic pulse signals in the ultrasonic pulse signal set, select the sensor corresponding to the ultrasonic pulse signal with the largest peak amplitude as the target sensor.

[0045] ​If there are multiple ultrasonic pulse signals with similar amplitudes (difference < 3dB), the sensor corresponding to the pulse with the highest signal-to-noise ratio is preferentially selected.

[0046] Directly use the three-dimensional space coordinates (xs, ys, zs) of the target sensor pre-calibrated as the preliminary discharge point coordinates.

[0047] In summary, first, according to the data abundance of the pulse set (such as the number of ultrasonic pulse signals), the positioning strategy is automatically judged, the high-quality signal is solved by the time difference arrival method (TDOA) based nonlinear optimization, and high-precision positioning is realized; for weak signals, it is degraded to fast positioning with the strongest signal source as the reference, so as to ensure the reliability and robustness under all working conditions while ensuring the high-precision potential. In the positioning process, through the fusion of peak signal-to-noise ratio sensor optimization, medium temperature compensation sound speed model and iterative optimization algorithm based on Taylor expansion and least squares, the accuracy and stability of the positioning result are improved in multiple dimensions; each preliminary discharge point generated carries complete metadata such as coordinates, positioning type and quality index, which not only provides a unified benchmark for accurate spatial correlation with the infrared abnormal temperature rise area, but also the statistical characteristics and positioning reliability of the pulse become the key dimension for subsequent multi-modal feature fusion. In summary, this step efficiently converts the "time accuracy" advantage of ultrasonic detection into a "spatial accuracy" advantage through intelligent algorithms, becoming the core bridge connecting front-end signal processing and back-end intelligent diagnosis, significantly improving the spatial resolution, environmental adaptability and decision support capability of the entire detection system.

[0048] Step S104, determine the target surface infrared image of the power equipment to be tested within a preset time period based on a preset image processing strategy, and aggregate each preliminary discharge point based on at least one abnormal temperature rise area in the target surface infrared image, to obtain at least one preliminary discharge point set, wherein one preliminary discharge point set corresponds to one suspected discharge source, and one preliminary discharge point set is associated with one abnormal temperature rise area.

[0049] In this step, the surface infrared image sequence of the power equipment to be tested within a preset time period is obtained; In each ultrasonic pulse signal sequence selection, the target ultrasonic pulse signal sequence with the most signal quantity of ultrasonic pulse signals is selected, and the candidate surface infrared image corresponding to the target peak occurrence time is selected in the surface infrared image sequence, wherein the target peak occurrence time is the peak occurrence time of each target ultrasonic pulse signal in the target ultrasonic pulse signal sequence; Sort each candidate surface infrared image based on the time sequence to obtain a candidate surface infrared image sequence, and calculate the image similarity between adjacent two candidate surface infrared images in the candidate surface infrared image sequence; select the first candidate surface infrared image and the second candidate surface infrared image corresponding to the maximum image similarity in each image similarity, and identify a first candidate temperature rise region corresponding to the first candidate surface infrared image and a second candidate temperature rise region corresponding to the second candidate surface infrared image based on the temperature field distribution in the first candidate surface infrared image and the second candidate surface infrared image; perform spatial registration on the first candidate temperature rise region and the second candidate temperature rise region, calculate the overlapping area ratio of the first candidate temperature rise region and the second candidate temperature rise region after spatial registration, and determine whether the overlapping area ratio is greater than a preset contour similarity threshold; If the overlapping area ratio is greater than the preset contour similarity threshold, the first candidate temperature rise region and the second candidate temperature rise region are merged to obtain a fused candidate temperature rise region. Calculate the first average temperature of the fused candidate temperature rise region in the first candidate surface infrared image and the second average temperature of the fused candidate temperature rise region in the second candidate surface infrared image, and take the maximum value of the first average temperature and the second average temperature as the reference temperature rise value of the fused candidate temperature rise region. According to the position and contour of the fused candidate temperature rise region in the first candidate surface infrared image and the second candidate surface infrared image, the reference temperature rise value is mapped to a unified background infrared image by using a preset pseudo-color mapping table to obtain a target surface infrared image, wherein the background infrared image is an average temperature field image of the surface of the power equipment to be measured in a preset time period. If the overlapping area ratio is not greater than the preset contour similarity threshold, the first candidate temperature rise region and the second candidate temperature rise region are determined as two independent temperature rise regions to be analyzed; calculate the first target average temperature in the first candidate temperature rise region and the second target average temperature in the second candidate temperature rise region; according to the position and contour of the first candidate temperature rise region in the first candidate surface infrared image and the position and contour of the second candidate surface infrared image in the first candidate surface infrared image, the first target average temperature and the second target average temperature are mapped to a unified background infrared image by using a preset pseudo-color mapping table to obtain a target surface infrared image.

[0050] Specifically, according to the position and contour of the fused candidate temperature rise region in the first candidate surface infrared image and the second candidate surface infrared image, the reference temperature rise value is mapped to a unified background infrared image by using a preset pseudo-color mapping table to obtain a target surface infrared image, which includes: The pixel-level average temperature of all surface infrared images in the surface infrared image sequence within the preset time period is calculated to generate a time-averaged temperature field image as a unified background infrared image; for the fusion candidate temperature rise region, the contour pixel coordinate set of the fusion candidate temperature rise region in the first selected surface infrared image and the second selected surface infrared image is obtained; a pseudo-color mapping table from a reference temperature rise value to a display color is established; the contour pixel coordinate set corresponding to the fusion candidate temperature rise region is directly mapped to the corresponding coordinate position of the background infrared image, and the filling color is determined through the pseudo-color mapping table according to the reference temperature rise value of the fusion candidate temperature rise region, so as to be drawn in the background infrared image in a semi-transparent superimposed manner, so as to obtain an image in which the color identification of the fusion candidate temperature rise region and the background temperature field are fused, that is, a target surface infrared image.

[0051] In one specific embodiment, an infrared thermal imager is synchronized with an ultrasonic acquisition system in advance, and infrared image sequences of the surface of the to-be-tested equipment are acquired within a preset time period. Each image has a microsecond-level timestamp aligned with the ultrasonic data. All ultrasonic pulse signal sequences are analyzed to find a target ultrasonic pulse signal sequence with the largest number of ultrasonic pulse signals, which corresponds to the most active period of discharge. The peak occurrence time of all pulses in the target ultrasonic pulse signal sequence is extracted. The selected surface infrared image corresponding to the time stamp is extracted from the infrared image sequence. If the time stamp does not match completely, the frame with the smallest time difference is selected. The initially selected selected images are sorted by time to form a selected surface infrared image sequence. The image similarity between adjacent frames is calculated using the structural similarity index (SSIM) combined with temperature histogram correlation, and the pair of images with the highest similarity is selected as the first selected surface infrared image and the second selected surface infrared image.

[0052] The first selected surface infrared image and the second selected surface infrared image are applied to an adaptive threshold segmentation algorithm based on local temperature statistics to identify a pixel connected domain significantly higher than the surrounding background (e.g., more than 3 times the standard deviation of the background temperature) as a first candidate temperature rise region and a second candidate temperature rise region.

[0053] The first candidate temperature rise region and the second candidate temperature rise region are subjected to sub-pixel level spatial registration using existing feature matching algorithms using the inherent invariant feature points (such as bolt corners and sign corners) on the surface of the to-be-tested power equipment. The intersection over union (IoU) of the two candidate regions is calculated in the unified coordinate system as the overlap area ratio.

[0054] If the overlap area ratio is greater than a preset contour similarity threshold (e.g., 0.6), it is determined that the same heat source is present at two times, and the two region contours are merged into a fusion candidate temperature rise region.

[0055] For the fusion candidate temperature rise region, the average temperature of the fusion candidate temperature rise region in the two original images is calculated respectively, and the maximum value is taken as the reference temperature rise value of the region to capture the peak effect of the heat source. For the unmerged independent region, the first target average temperature and the second target average temperature are calculated respectively.

[0056] The pixel-level time average of all infrared image sequences in the entire preset time period is calculated to generate a time-averaged temperature field image reflecting the overall steady-state heat generation of the device as a background.

[0057] The contours of all determined regions (fusion candidate temperature rise regions or unmerged independent regions) are mapped to the same coordinate system of the background reference image.

[0058] According to the reference temperature rise value, or the first target average temperature and the second target average temperature of each region, a preset rainbow color or iron red pseudo-color mapping table is used to convert it to the corresponding color value.

[0059] The color region is rendered to the background reference image in a semi-transparent manner (such as transparency 50%) using Alpha blending (transparency superposition) technology to generate the final target surface infrared image. This image retains both the background temperature field information and the highlighted abnormal temperature rise region.

[0060] In this embodiment, first, the activity of the ultrasonic signal is used as a guide to intelligently select key frames highly related to discharge events from a large amount of infrared images, achieving data dimension reduction and target focusing, greatly improving processing efficiency; second, through cross-time frame candidate region comparison and fusion, the real and continuous discharge heat source and the transient interference are effectively distinguished, enhancing the robustness of the temperature rise region identification; finally, the generated target surface infrared image is not a simple presentation of the original data, but a "special diagnostic map" that integrates time dimension information (multi-time verification) and spatial enhancement features (highlight display, pseudo-color quantization). It not only intuitively and clearly locates the position and relative severity of abnormal temperature rise, but also provides high-quality and target-oriented input for subsequent cross-modal spatial fusion through the inherent time association with ultrasonic data, thereby solving the technical problems of complex background, heat source confusion, and difficulty in direct correlation with transient electrical signals in traditional infrared inspection.

[0061] Further, after temperature calibration and noise filtering of the target surface infrared image, at least one abnormal temperature rise region is identified using an adaptive threshold segmentation and region growing algorithm, and the contour polygon of each abnormal temperature rise region is extracted. Establish a spatial mapping relationship between the ultrasonic sensor coordinate system and the infrared image pixel coordinate system, and map the three-dimensional coordinates of each initial discharge point to the infrared image pixel coordinate system to obtain the projected coordinates of each initial discharge point. Determine whether a certain projected coordinate of a certain initial discharge point is located inside the outline polygon of any abnormal temperature rise region; If a certain projected coordinate is located inside the outline polygon of a certain abnormal temperature rise region, then a certain initial discharge point is assigned to the set of a certain initial discharge point corresponding to a certain abnormal temperature rise region. If a certain projected coordinate is not located inside the contour polygon of any abnormal temperature rise area, then calculate the shortest distance from the certain projected coordinate to the boundary of each contour polygon, and determine whether the shortest distance is less than the preset spatial tolerance threshold. If the shortest distance is less than the preset spatial tolerance threshold, then a certain initial discharge point will be assigned to the set of initial discharge points corresponding to the nearest abnormal temperature rise area. If the shortest distance is not less than the preset spatial tolerance threshold, a certain initial discharge point will be recorded as an isolated discharge event and output to the abnormal event log for manual review.

[0062] In another specific embodiment, Otsu's method is used for adaptive threshold segmentation, and combined with existing region growing algorithms, at least one abnormal temperature rise region is accurately extracted, and a list of vertex coordinates of its contour polygon is obtained. A mapping model is established from the three-dimensional coordinate system of the ultrasonic sensor to the two-dimensional coordinate system of the infrared image pixels through pre-completed joint calibration. This mapping model is typically a perspective transformation matrix or polynomial model that includes rotation, translation, and scaling. Using this mapping model, the three-dimensional coordinates (x, y, z) of each preliminary discharge point obtained in step S103 are converted into projected coordinates (u, v) on the infrared image; Use the ray method or polygon filling algorithm to determine whether each projected coordinate (u, v) is located inside the contour polygon of any abnormal temperature rise region; For points outside the contour, calculate their Euclidean shortest distance to the boundary of each contour polygon (which consists of a series of line segments).

[0063] If a certain projected coordinate is located inside the contour polygon of a certain abnormal temperature rise region, it is directly assigned to the set of preliminary discharge points corresponding to that contour. If a certain projected coordinate is not located inside the contour polygon of any abnormal temperature rise region, and the shortest distance is less than the spatial tolerance threshold (for example, the image pixel distance corresponding to an actual spatial distance of 5 cm), then it is assigned to the set of preliminary discharge points corresponding to the nearest region.

[0064] If a projection coordinate is not located inside the contour polygon of any abnormal temperature rise region, and the shortest distance is not less than the spatial tolerance threshold, the preliminary discharge point is marked as an isolated discharge event, all information is recorded and stored in the abnormal event log, and further manual analysis is awaited.

[0065] Finally, a number of preliminary discharge point sets are output, each set being uniquely associated with an abnormal temperature rise region and collectively pointing to a suspected discharge source.

[0066] In this embodiment, through coordinate system mapping, the ultrasonic positioning points calculated based on time difference are accurately projected onto the infrared feature image. Through the dual criteria of "internal inclusion" and "proximity constraint", this strategy intelligently aggregates multiple ultrasonic positioning points that may be generated by the same discharge source and are slightly dispersed in space under the same abnormal heat region, thereby effectively filtering out false points or error points in ultrasonic positioning due to sound wave reflection, refraction or noise, significantly improving the confidence and clarity of discharge source positioning. Each preliminary discharge point set output is a highly reliable suspected discharge source verified by sound-heat evidence cross-validation, providing a pure, structured and physically meaningful input for subsequent feature extraction and pattern recognition.

[0067] In step S105, a waveform feature sub-vector corresponding to a preliminary discharge point set and a temperature feature sub-vector of an abnormal temperature rise region are extracted, and the waveform feature sub-vector and the temperature feature sub-vector are fused to obtain a fusion feature vector.

[0068] In this step, multidimensional waveform analysis is performed on all ultrasonic pulse signals corresponding to a preliminary discharge point set, a group of time domain waveform features and frequency domain waveform features are extracted to form a waveform feature sub-vector, wherein the time domain waveform features include pulse amplitude statistical features, pulse time interval statistical features and pulse energy features, and the frequency domain waveform features include spectral centroid, spectral width and specific frequency band energy proportion extracted by fast Fourier transform; Temperature field analysis is performed on the infrared image region corresponding to an abnormal temperature rise region, a group of geometric features and temperature field features are extracted to form a temperature feature sub-vector, wherein the geometric features include the contour area, contour perimeter and contour aspect ratio of the abnormal temperature rise region, and the temperature field features include temperature gradient features and temperature rise statistical features; The waveform feature sub-vector and the temperature feature sub-vector are standardized respectively to eliminate the dimensional influence, and the standardized waveform feature sub-vector and the temperature feature sub-vector are spliced to obtain a fusion feature vector.

[0069] In one embodiment, for a preliminary discharge point set (corresponding to a suspected discharge source), the set is aggregated from multiple sets of ultrasonic pulse signals, and each set of ultrasonic pulse signals contains multiple ultrasonic pulse signals from different sensors, which are implemented as follows: Step S1051, obtain all associated pulse signals: From the preliminary discharge point set, obtain all sets of ultrasonic pulse signals contained therein; From these sets of pulse signals, extract all ultrasonic pulse signals (including the timestamp, peak amplitude, energy, complete time-domain waveform segment, etc. of each ultrasonic pulse signal); Statistical total ultrasonic pulse signal quantity , total sensor quantity and pulse signal set quantity .

[0070] Step S1052, overall statistical analysis of time-domain waveform features: Pulse amplitude statistical features: calculate the mean , standard deviation , maximum value , kurtosis and skewness of amplitude distribution of the peak amplitudes of all ultrasonic pulse signals; Pulse interval statistical features: Sort the pulses from all pulse signal sets by global timestamp; Calculate the time interval sequence of adjacent pulses, and then calculate the mean , standard deviation ISI std and coefficient of variation of the sequence; Pulse energy statistical features: Calculate the energy of each ultrasonic pulse signal by waveform square integration; Calculate the total energy , average energy and energy standard deviation of all ultrasonic pulse signals; Calculate the average energy of each set of ultrasonic pulse signals, and then calculate the standard deviation of these set average energies to represent the energy volatility of different discharge events.

[0071] Step S1053, comprehensive extraction of frequency-domain waveform features: Representative sample selection: select the top M pulses (e.g. ) from all ultrasonic pulse signals as representative ultrasonic pulse signals according to peak amplitude ranking.

[0072] Spectrum analysis: Fast Fourier Transform (FFT) is applied to each time-domain waveform segment of representative ultrasonic pulse signals to obtain their amplitude spectrum.

[0073] Spectrum feature calculation: Spectrum centroid SC: The spectrum centroid of each ultrasonic pulse signal is calculated, and then the average of these centroids is taken and standard deviation ; Spectrum width BW: The -3dB bandwidth of each pulse spectrum is calculated, and the average is taken ; Specific band energy ratio: The energy of two key frequency bands (such as 40-80 kHz and 150-300 kHz) is calculated relative to the total spectrum energy. The average of the energy ratios of these pulses in the two frequency bands is calculated respectively ; Dominant frequency distribution: Identify the frequency component with the largest amplitude (dominant frequency) in each pulse spectrum, and count the proportion of pulses whose dominant frequencies fall into the above two key frequency bands; Spectrum consistency feature: Calculate the average correlation coefficient between all representative pulse spectra to represent the spectrum similarity of different discharge events; Arrange the above features in order to form a waveform feature sub-vector.

[0074] Step S1054, for a certain abnormal temperature rise region (from the target surface infrared image) associated with the same preliminary discharge point set, the following analysis is implemented: Region and background definition: According to the contour polygon of the region, the region range is accurately defined in the target surface infrared image and the original temperature data; Define a ring-shaped background region (such as a ring with an outer extension of 10-20 pixels) outside the region, and calculate its average temperature as the environmental background temperature ; Geometric feature calculation: Contour area: The total number of pixels inside the contour polygon, converted to the actual physical area; Contour perimeter: The sum of all side lengths of the contour polygon; Contour aspect ratio: The ratio of the long side to the short side of the minimum bounding rectangle of the region; Contour complexity: defined as (twice the contour perimeter) / ( Contour area), used to describe the deviation of the contour from a circle; Temperature field feature calculation: Temperature gradient feature: Use the Sobel operator to calculate the temperature gradient amplitude and direction of each pixel in the region, and take the maximum value and average value ; Temperature rise statistical characteristics: Region average temperature rise: , is the arithmetic mean of the temperature values of all pixel points inside the abnormal temperature rise region contour polygon; Region maximum temperature rise: , is the maximum value among the temperature values of all pixel points inside the abnormal temperature rise region contour polygon; Temperature rise standard deviation: the standard deviation of the temperature rise of each pixel in the region; The above characteristics are arranged in order to form a temperature characteristic sub-vector.

[0075] Step S1055, respectively, the waveform feature sub-vector and the temperature feature sub-vector are subjected to Z-score standardization, and the standardized waveform feature sub-vector and the temperature feature sub-vector are directly spliced at the beginning and the end to form a high-dimensional fusion feature vector: If the dimension of the fusion feature vector is too high, principal component analysis (PCA) or automatic encoder (Autoencoder) can be used for dimension reduction to retain the main information while reducing the calculation complexity.

[0076] In summary, the method of the embodiment, by aggregating and analyzing all pulse signals of the multiple ultrasonic pulse signal sets corresponding to the preliminary discharge point set, waveform features that can comprehensively reflect the overall discharge mode of the discharge source are extracted; at the same time, geometric and temperature field features of the abnormal temperature rise region associated therewith are extracted, providing supplementary information from the perspective of thermal effect. This cross-modal feature fusion strategy realizes the joint digital representation of acoustic and thermal physical characteristics of the same discharge source for the first time in partial discharge detection. The fusion feature vector generated thereby not only contains the time-frequency characteristics and energy release law of the discharge, but also fuses the spatial distribution and intensity information of the thermal effect, providing a high-dimensional feature input with rich information and strong complementarity for the subsequent intelligent recognition model. This significantly improves the ability of the model to distinguish different discharge types (such as corona discharge, surface discharge, and internal discharge), and makes the assessment of discharge intensity more accurate and reliable, thereby providing a key decision basis for the accurate diagnosis and risk assessment of power equipment.

[0077] Step S106, inputting the certain fusion feature vector into a preset partial discharge recognition model, and outputting a discharge type recognition result and a discharge intensity evaluation value of a certain suspected discharge source by the partial discharge recognition model.

[0078] In this step, a dual-channel deep neural network is used as the preset partial discharge recognition model. The network includes: Feature preprocessing layer: receiving the fusion feature vector output by step S105, first performing dimension adaptation and preliminary feature interaction through a fully connected layer; Dual-branch feature extraction module: Ultrasonic feature branch: a multi-layer perceptron (MLP) that is specialized in processing the part of the fusion feature vector that belongs to the waveform feature sub-vector, learning the acoustic pattern of the discharge; Infrared feature branch: another multi-layer perceptron (MLP) that is specialized in processing the part of the fusion feature vector that belongs to the temperature feature sub-vector, learning the thermal pattern of the discharge; Feature fusion and decision module: concatenating the high-level abstract features output by the two branches, then passing them through a shared network containing multiple fully connected layers and Dropout layers, and finally connecting to two parallel output heads: Type recognition head: a Softmax classification layer that outputs the probability distribution of belonging to each of the preset discharge types where K is the number of discharge types (e.g., K = 4, corresponding to "corona discharge", "surface discharge", "internal discharge", and "other discharge").

[0079] Intensity assessment head: a linear regression layer (or an ordinal classification layer) that outputs a continuous discharge intensity assessment value I (e.g., ranging from 0 to 100) or a discrete intensity level (e.g., "weak", "moderate", "intense", "critical").

[0080] In summary, the method of the present application achieves high-sensitivity capture and cross-channel intelligent association of discrete discharge events from complex noise background through multi-channel ultrasonic acquisition and sliding window clustering based on physical propagation model, providing pure and structured discharge event units for subsequent analysis. Secondly, it generates a "thematic diagnosis map" that can highlight abnormal temperature rise related to discharge through the infrared image selection and fusion strategy guided by the ultrasonic active period, solving the problem of dispersed infrared detection targets and difficulty in correlating transient events. Further, through spatial aggregation of the acoustic and thermal dual-modality evidence chain, the ultrasonic positioning point and the infrared thermal region are accurately associated and cross-verified, effectively filtering out false signals and positioning errors in single-modality detection, significantly improving the confidence of discharge source positioning. Then, by extracting multi-level and multi-dimensional statistical and physical features from the aggregated multi-event ultrasonic pulse sequence and associated infrared region, and fusing them into a unified feature vector, a fusion feature that can fully represent the nature of discharge is constructed. Through the intelligent analysis of the fusion feature by the partial discharge recognition model, high-precision automatic recognition of discharge types (such as corona, surface, and internal discharge) and quantitative assessment of discharge intensity are achieved. Overall, the present application not only overcomes the limitations of single detection means, but also through deep information fusion and intelligent analysis, achieves earlier, more accurate, and more intuitive perception and diagnosis of the insulation state of power equipment, providing strong technical support for predictive maintenance and safe operation of power systems.

[0081] Please refer to Figure 2 , which shows a structural block diagram of a partial discharge intelligent detection system of the present application.

[0082] As Figure 2 shown, the partial discharge intelligent detection system 200 includes an acquisition module 210, an analysis module 220, a determination module 230, an aggregation module 240, an extraction module 250, and an output module 260.

[0083] The acquisition module 210 is configured to acquire a plurality of ultrasonic signals of a power equipment to be tested within a preset time period, wherein the plurality of ultrasonic signals are respectively collected by a plurality of sensors; the analysis module 220 is configured to perform pulse analysis on the plurality of ultrasonic signals according to a preset signal detection rule to obtain an ultrasonic pulse signal sequence corresponding to the plurality of ultrasonic signals, and slide a preset sliding window on each ultrasonic pulse signal sequence to cluster each ultrasonic pulse signal meeting a time synchronization condition to obtain at least one ultrasonic pulse signal set, wherein one ultrasonic pulse signal sequence contains at least one ultrasonic pulse signal corresponding to the same ultrasonic signal; the determination module 230 is configured to determine a preliminary discharge point corresponding to the at least one ultrasonic pulse signal set; the aggregation module 240 is configured to determine a target surface infrared image of the power equipment to be tested within the preset time period based on a preset image processing strategy, and aggregate each preliminary discharge point based on at least one abnormal temperature rise area in the target surface infrared image to obtain at least one preliminary discharge point set, wherein one preliminary discharge point set corresponds to one suspected discharge source, and one preliminary discharge point set is associated with one abnormal temperature rise area; the extraction module 250 is configured to extract a certain waveform feature sub-vector corresponding to a certain preliminary discharge point set and a certain temperature feature sub-vector of a certain abnormal temperature rise area, and perform feature fusion on the certain waveform feature sub-vector and the certain temperature feature sub-vector to obtain a certain fusion feature vector; and the output module 260 is configured to input the certain fusion feature vector into a preset partial discharge identification model, and the partial discharge identification model outputs a discharge type identification result and a discharge intensity evaluation value of a certain suspected discharge source.

[0084] It should be understood that Figure 2 the modules described in the specification correspond to the respective steps in the methods described with reference to Figure 1 . Thus, the operations and features described above for the methods and the corresponding technical effects apply equally to the modules in Figure 2 , and will not be described again here.

[0085] In some embodiments, the present application also provides a computer readable storage medium having stored thereon a computer program, wherein the program instructs a processor to execute the partial discharge intelligent detection method in any of the above method embodiments when executed by the processor. As an implementation form, the computer readable storage medium of the present application stores computer executable instructions, which are configured to: acquire a plurality of ultrasonic signals of the to-be-tested power equipment in a preset time period, wherein the plurality of ultrasonic signals are respectively acquired by a plurality of sensors; perform pulse analysis on the plurality of ultrasonic signals to obtain a plurality of ultrasonic pulse signal sequences corresponding to the plurality of ultrasonic signals, and slide a preset sliding window on each ultrasonic pulse signal sequence to cluster each ultrasonic pulse signal meeting a time synchronization condition to obtain at least one ultrasonic pulse signal set, wherein one ultrasonic pulse signal sequence comprises at least one ultrasonic pulse signal corresponding to the same ultrasonic signal; determine preliminary discharge points corresponding to the at least one ultrasonic pulse signal set; determine a target surface infrared image of the to-be-tested power equipment in the preset time period based on a preset image processing strategy, and aggregate each preliminary discharge point based on at least one abnormal temperature rise area in the target surface infrared image to obtain at least one preliminary discharge point set, wherein one preliminary discharge point set corresponds to one suspected discharge source, and one preliminary discharge point set is associated with one abnormal temperature rise area; extract a certain waveform feature sub-vector corresponding to a certain preliminary discharge point set and a certain temperature feature sub-vector of a certain abnormal temperature rise area, and perform feature fusion on the certain waveform feature sub-vector and the certain temperature feature sub-vector to obtain a certain fusion feature vector; input the certain fusion feature vector into a preset partial discharge recognition model, and output a discharge type recognition result and a discharge intensity evaluation value of a certain suspected discharge source by the partial discharge recognition model.

[0086] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the partial discharge intelligent detection system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the partial discharge intelligent detection system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0087] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the partial discharge intelligent detection method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the partial discharge intelligent detection system. The output device 340 may include a display screen or other display device.

[0088] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0089] In one implementation, the above-described electronic device is applied to a partial discharge intelligent detection system as a client, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Multiple ultrasonic signals from the electrical equipment under test within a preset time period are acquired, wherein the multiple ultrasonic signals are respectively acquired by multiple sensors; pulse analysis is performed on the plurality of ultrasonic signals to obtain a sequence of ultrasonic pulse signals corresponding to the plurality of ultrasonic signals, and each ultrasonic pulse signal meeting a time synchronization condition is clustered based on a preset sliding window sliding on each sequence of ultrasonic pulse signals to obtain at least one set of ultrasonic pulse signals, wherein each sequence of ultrasonic pulse signals contains at least one ultrasonic pulse signal corresponding to the same ultrasonic signal; A preliminary discharge point corresponding to the at least one set of ultrasonic pulse signals is determined. A target surface infrared image of the power equipment under test in a preset time period is determined based on a preset image processing strategy, and each preliminary discharge point is aggregated based on at least one abnormal temperature rise area in the target surface infrared image to obtain at least one set of preliminary discharge points, wherein one set of preliminary discharge points corresponds to one suspected discharge source, and one set of preliminary discharge points is associated with one abnormal temperature rise area. A certain waveform feature sub-vector corresponding to a certain set of preliminary discharge points and a certain temperature feature sub-vector of a certain abnormal temperature rise area are extracted, and the certain waveform feature sub-vector and the certain temperature feature sub-vector are fused to obtain a certain fusion feature vector. The certain fusion feature vector is input into a preset partial discharge recognition model, and the partial discharge recognition model outputs a discharge type recognition result and a discharge intensity evaluation value of a certain suspected discharge source.

[0090] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus necessary universal hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in terms of the contribution to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of the embodiments or some parts of the embodiments.

[0091] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A partial discharge intelligent detection method, characterized in that, The method comprises the following steps: acquiring a plurality of ultrasonic signals of a to-be-tested power equipment within a preset time period, wherein the plurality of ultrasonic signals are respectively collected by a plurality of sensors; performing pulse analysis on the plurality of ultrasonic signals to obtain a sequence of ultrasonic pulse signals corresponding to the plurality of ultrasonic signals, and sliding a preset sliding window on each sequence of ultrasonic pulse signals to cluster ultrasonic pulse signals meeting a time synchronization condition to obtain at least one set of ultrasonic pulse signals, wherein one sequence of ultrasonic pulse signals comprises at least one ultrasonic pulse signal corresponding to the same ultrasonic signal; determining a preliminary discharge point corresponding to the at least one set of ultrasonic pulse signals; determining a target surface infrared image of the to-be-tested power equipment within the preset time period based on a preset image processing strategy, and aggregating each preliminary discharge point based on at least one abnormal temperature rise area in the target surface infrared image to obtain at least one set of preliminary discharge points, wherein one set of preliminary discharge points corresponds to one suspected discharge source, and one set of preliminary discharge points is associated with one abnormal temperature rise area; extracting a certain waveform feature sub-vector corresponding to a certain set of preliminary discharge points and a certain temperature feature sub-vector of a certain abnormal temperature rise area, and performing feature fusion on the certain waveform feature sub-vector and the certain temperature feature sub-vector to obtain a certain fusion feature vector; inputting the certain fusion feature vector into a preset partial discharge identification model, and outputting a discharge type identification result and a discharge intensity evaluation value of a certain suspected discharge source by the partial discharge identification model.

2. The partial discharge intelligent detection method according to claim 1, characterized in that, The determination of the preliminary discharge point corresponding to the at least one set of ultrasonic pulse signals comprises: determining whether the number of ultrasonic pulse signals in a certain set of ultrasonic pulse signals is greater than a preset number threshold; if greater than the preset number threshold, selecting ultrasonic pulse signals from three different spatial position sensors in the certain set of ultrasonic pulse signals; acquiring time stamps and peak amplitudes corresponding to the ultrasonic pulse signals of the three different spatial position sensors respectively, and calculating a time difference between each two ultrasonic pulse signals based on three-dimensional space coordinates of the three different spatial position sensors pre-calibrated; based on a known propagation speed of ultrasonic waves in the equipment insulation medium, the three-dimensional space coordinates of the three different spatial position sensors, and the time difference between each two ultrasonic pulse signals, a set of hyperbolic positioning equation groups is established; solving the hyperbolic positioning equation groups by using a least square estimation algorithm to obtain a three-dimensional space coordinate corresponding to the certain set of ultrasonic pulse signals as a certain preliminary discharge point; if not greater than the preset number threshold, selecting a target sensor from the sensors corresponding to each ultrasonic pulse signal, and taking the three-dimensional space coordinate of the target sensor as a reference coordinate to obtain a certain preliminary discharge point, wherein the target sensor is the sensor corresponding to the ultrasonic pulse signal with the maximum peak amplitude.

3. The partial discharge intelligent detection method according to claim 1, characterized in that, The determination of the target surface infrared image of the to-be-tested power equipment within the preset time period based on the preset image processing strategy comprises: acquire a surface infrared image sequence of the to-be-tested power equipment in a preset time period; select a target ultrasonic pulse signal sequence in which the number of signal of each ultrasonic pulse signal is the most in each ultrasonic pulse signal sequence, and select a to-be-selected surface infrared image corresponding to a target peak occurrence time in the surface infrared image sequence, where the target peak occurrence time is a peak occurrence time of each target ultrasonic pulse signal in the target ultrasonic pulse signal sequence; sort each to-be-selected surface infrared image based on a time sequence to obtain a to-be-selected surface infrared image sequence, and calculate an image similarity between two adjacent to-be-selected surface infrared images in the to-be-selected surface infrared image sequence; select a first to-be-selected surface infrared image and a second to-be-selected surface infrared image corresponding to a maximum image similarity in each image similarity, and identify a first candidate temperature rise region corresponding to the first to-be-selected surface infrared image and a second candidate temperature rise region corresponding to the second to-be-selected surface infrared image based on a temperature field distribution in the first to-be-selected surface infrared image and the second to-be-selected surface infrared image; perform spatial registration on the first candidate temperature rise region and the second candidate temperature rise region, calculate an overlapping area ratio of the first candidate temperature rise region and the second candidate temperature rise region after the spatial registration, and determine whether the overlapping area ratio is greater than a preset contour similarity threshold; if the overlapping area ratio is greater than the preset contour similarity threshold, merge the first candidate temperature rise region and the second candidate temperature rise region to obtain a fused candidate temperature rise region; calculate a first average temperature of the fused candidate temperature rise region in the first to-be-selected surface infrared image and a second average temperature of the fused candidate temperature rise region in the second to-be-selected surface infrared image, and take a maximum value of the first average temperature and the second average temperature as a reference temperature rise value of the fused candidate temperature rise region; map the reference temperature rise value to a unified background infrared image by using a preset pseudo-color mapping table according to a position and a contour of the fused candidate temperature rise region in the first to-be-selected surface infrared image and the second to-be-selected surface infrared image, to obtain the target surface infrared image, where the background infrared image is an average temperature field image of a surface of the to-be-tested power equipment in the preset time period.

4. The partial discharge intelligent detection method according to claim 3, characterized in that, After determining whether the overlapping area ratio is greater than the preset contour similarity threshold, the method further includes: if the overlapping area ratio is not greater than the preset contour similarity threshold, determine the first candidate temperature rise region and the second candidate temperature rise region as two independent to-be-analyzed temperature rise regions; calculate a first target average temperature in the first candidate temperature rise region and a second target average temperature in the second candidate temperature rise region; and According to the position and contour of the first candidate temperature rise region in the first candidate surface infrared image and the position and contour of the second candidate surface infrared image in the first candidate surface infrared image, the first target average temperature and the second target average temperature are mapped into a unified background infrared image by using a preset pseudo-color mapping table, to obtain the target surface infrared image.

5. The partial discharge intelligent detection method according to claim 3, characterized in that, According to the position and contour of the fusion candidate temperature rise region in the first candidate surface infrared image and the second candidate surface infrared image, the reference temperature rise value is mapped into a unified background infrared image by using a preset pseudo-color mapping table, to obtain the target surface infrared image. The pixel-level average temperature of all surface infrared images in the surface infrared image sequence in the preset time period is calculated, to generate a time-averaged temperature field image as a unified background infrared image. For the fusion candidate temperature rise region, a set of contour pixel coordinates of the fusion candidate temperature rise region in the first candidate surface infrared image and the second candidate surface infrared image are obtained. A pseudo-color mapping table from the reference temperature rise value to the display color is established. The set of contour pixel coordinates corresponding to the fusion candidate temperature rise region is directly mapped to the corresponding coordinate position of the background infrared image, and a filling color is determined by the pseudo-color mapping table according to the reference temperature rise value of the fusion candidate temperature rise region, to be drawn in the background infrared image in a semi-transparent superimposed manner, to obtain an image in which the color identification of the fusion candidate temperature rise region is fused with the background temperature field, i.e. the target surface infrared image.

6. The partial discharge intelligent detection method according to claim 1, characterized in that, The aggregation of each preliminary discharge point based on at least one abnormal temperature rise region in the target surface infrared image includes: After temperature calibration and noise filtering of the target surface infrared image, at least one abnormal temperature rise region is identified by using adaptive threshold segmentation and region growing algorithm, and a contour polygon of each abnormal temperature rise region is extracted. A spatial mapping relationship between the ultrasonic sensor coordinate system and the infrared image pixel coordinate system is established, to map the three-dimensional coordinates of each preliminary discharge point to the infrared image pixel coordinate system, to obtain the projection coordinates of each preliminary discharge point. It is judged whether a projection coordinate of a preliminary discharge point is located inside a contour polygon of any abnormal temperature rise region. If a projection coordinate is located inside a contour polygon of an abnormal temperature rise region, the preliminary discharge point is classified into a preliminary discharge point set corresponding to the abnormal temperature rise region. If a projection coordinate is not located inside a contour polygon of any abnormal temperature rise region, the shortest distance from the projection coordinate to the boundary of each contour polygon is calculated, and it is judged whether the shortest distance is less than a preset spatial tolerance threshold. If the shortest distance is less than the preset spatial tolerance threshold, the preliminary discharge point is classified into the preliminary discharge point set corresponding to the nearest abnormal temperature rise region. If the shortest distance is not less than the preset spatial tolerance threshold, the preliminary discharge point is recorded as an isolated discharge event and output to an abnormal event log for manual review.

7. The partial discharge intelligent detection method according to claim 1, characterized in that, The extraction and a certain temperature feature sub-vector of a certain abnormal temperature rise region corresponding to a certain preliminary discharge point set are fused, and a certain fusion feature vector is obtained. A multi-dimensional waveform analysis is performed on all ultrasonic pulse signals corresponding to the certain preliminary discharge point set, a group of time domain waveform features and frequency domain waveform features are extracted, and a certain waveform feature sub-vector is constructed, wherein the time domain waveform features include pulse amplitude statistical features, pulse time interval statistical features, and pulse energy features, and the frequency domain waveform features include a frequency spectrum centroid, a frequency spectrum width, and a specific frequency band energy proportion extracted by fast Fourier transform. A temperature field analysis is performed on an infrared image region corresponding to the certain abnormal temperature rise region, a group of geometric features and temperature field features are extracted, and a certain temperature feature sub-vector is constructed, wherein the geometric features include an abnormal temperature rise region contour area, a contour perimeter, and a contour aspect ratio, and the temperature field features include temperature gradient features and temperature rise statistical features. The certain waveform feature sub-vector and the certain temperature feature sub-vector are standardized respectively to eliminate dimensional influence, and a certain fusion feature vector is obtained by splicing the standardized certain waveform feature sub-vector and the certain temperature feature sub-vector.

8. A partial discharge intelligent detection system, characterized in that, The acquisition module is configured to acquire a plurality of ultrasonic signals of the power equipment under test within a preset time period, wherein the plurality of ultrasonic signals are collected by a plurality of sensors. The analysis module is configured to perform pulse analysis on the plurality of ultrasonic signals according to a preset signal detection rule to obtain an ultrasonic pulse signal sequence corresponding to the plurality of ultrasonic signals, and to cluster ultrasonic pulse signals meeting a time synchronization condition by sliding a preset sliding window on each ultrasonic pulse signal sequence to obtain at least one ultrasonic pulse signal set, wherein one ultrasonic pulse signal sequence includes at least one ultrasonic pulse signal corresponding to the same ultrasonic signal. The determination module is configured to determine a preliminary discharge point corresponding to the at least one ultrasonic pulse signal set. The aggregation module is configured to determine a target surface infrared image of the power equipment under test within a preset time period based on a preset image processing strategy, and to aggregate each preliminary discharge point based on at least one abnormal temperature rise region in the target surface infrared image to obtain at least one preliminary discharge point set, wherein one preliminary discharge point set corresponds to one suspected discharge source, and one preliminary discharge point set is associated with one abnormal temperature rise region. The extraction module is configured to extract a certain waveform feature corresponding to a certain preliminary discharge point set and a certain temperature feature of a certain abnormal temperature rise region, and to fuse the certain waveform feature and the certain temperature feature to obtain a certain fusion feature vector. The output module is configured to input the certain fusion feature vector into a preset partial discharge identification model, and to output a discharge type identification result and a discharge intensity evaluation value of a certain suspected discharge source from the partial discharge identification model. ​ 9. An electronic device, comprising: ​ at least one processor, and a memory connected with the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program, when executed by a processor, implements the method of any one of claims 1 to 7. The program, when executed by a processor, implements the method of any one of claims 1 to 7.

Images