Leakage detection and partial discharge detection method and system based on artificial intelligence

Abstract

The invention relates to the technical field of partial discharge detection, in particular to a leak detection and partial discharge detection method and system based on artificial intelligence, and the method comprises the following steps: obtaining a partial discharge pulse signal, extracting a phase mutation coordinate, analyzing an incremental feature, reconstructing a propagation waveform, constructing an amplitude gradient matrix, and screening a direction vector; and a backtracking path calculates a change rate to generate a time track, and track mapping and space positioning are completed in combination with image data. According to the method, the accuracy and the response speed of partial discharge detection are improved through multi-stage signal processing, extreme points and phase abrupt change are analyzed, more accurate discharge feature capture and equipment defect positioning are achieved, phase and amplitude data are combined, the discharge mode is recognized, a signal propagation path is reconstructed, the fault source positioning accuracy is improved, and the fault source positioning accuracy is improved. Through dynamic rendering, the intuition of a detection result is enhanced, real-time decision is supported, the method is more intelligent and efficient, the equipment fault risk is greatly reduced, and the safe operation level of power equipment is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of partial discharge detection, and in particular, to a leak detection and partial discharge detection method and system based on artificial intelligence. Background Art

[0002] The technical field of partial discharge detection includes technologies for monitoring and identifying partial discharge phenomena caused by internal insulation defects of high-voltage electrical equipment. The core content is to use various physical signals such as electromagnetic waves, ultrasonic waves, current pulses, etc. to obtain, identify, and analyze partial discharge activities generated during the operation of the equipment in real time. It generally covers links such as signal acquisition, signal recognition, discharge type classification, defect location, and risk assessment, and is widely used in the operation and maintenance and condition monitoring of power equipment to provide early warnings for potential faults and hidden dangers of the equipment through non-invasive detection means.

[0003] Among them, the leak detection and partial discharge detection method based on artificial intelligence refers to introducing methods such as pattern recognition and statistical learning into partial discharge detection and leakage fault judgment. By establishing a learning model based on historical operation state data and fault sample data, different types of partial discharge patterns and leakage characteristics are identified. For partial discharge and equipment leakage technical matters, mainly by constructing a sample library containing historical feature data, and forming discrimination criteria based on data mining and model training methods, using a digital signal conversion device to obtain the original signal data in the operating state, and after unified data standardization processing, inputting it into the model for feature extraction and category recognition to complete the technical judgment of the target state.

[0004] In the existing partial discharge detection process, it mainly relies on traditional signal processing methods, and it is difficult to deal with partial discharge and leakage problems in complex environments. Usually, only a single type of signal is concerned, and the analysis is based on static standards, resulting in poor adaptability to different discharge patterns and equipment conditions. When facing the changing operating environment of the equipment, it is impossible to flexibly adjust the processing method according to the actual situation, and it is easy to have missed detections or false alarms. Due to the limited accuracy of signal processing, the detection of partial discharge is usually relatively rough, and it is easy to miss some abnormal signals that are difficult to detect, and it is difficult to provide an accurate equipment health assessment in real time. For example, when there are slight changes in the operating state of the equipment or the signal characteristics are relatively fuzzy, the existing technology may not be able to accurately capture these changes, resulting in delays in early warnings or neglect of faults. The lack of comprehensive analysis of multi-dimensional signals makes it difficult for the existing technology to continuously provide efficient and accurate monitoring and fault diagnosis in complex environments. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art and to propose a leak detection and partial discharge detection method based on artificial intelligence.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A leakage detection and partial discharge detection method based on artificial intelligence, comprising the following steps: S1: Obtain partial discharge pulse signal data, perform sliding window division, extract local extreme points and calculate phase difference values, determine whether it exceeds the mutation threshold, record the timestamp and spatial coordinates, and generate a phase mutation recognition coordinate set; S2: Based on the phase mutation recognition coordinate set, extract the amplitude increment and phase difference values, determine the increasing trend, construct a unidirectional propagation path and perform reverse fitting, and reconstruct the waveform data to obtain an acoustic wave propagation reconstructed waveform group; S3: According to the amplitude data of each measurement point in the acoustic wave propagation reconstructed waveform group, calculate the amplitude difference and construct a gradient matrix, screen the measurement points that meet the gradient change conditions, and obtain a set of main direction vectors of the sound source; S4: Perform path backtracking according to the set of main direction vectors of the sound source, extract the amplitude value and time interval of each point, calculate the amplitude change rate, screen the trajectory segments that meet the conditions, and generate sound source time backtracking trajectory data; S5: Based on the sound source time backtracking trajectory data, obtain visible light image frames, perform trajectory mapping, set the transparency and line width, perform trajectory rendering, and obtain the spatial positioning result of partial discharge detection.

[0007] As a further solution of the present invention, the phase mutation recognition coordinate set includes the time, spatial coordinates, phase angle difference, mutation angle threshold, and difference screening result of the extreme points; the acoustic wave propagation reconstructed waveform group includes the reconstructed acoustic wave signal sequence, the fitted phase sequence, and the amplitude sequence; the set of main direction vectors of the sound source includes spatial coordinates, change direction, amplitude increment, phase difference, and gradient change direction; the sound source time backtracking trajectory data includes the path start point, the amplitude change rate of the path segment, the time interval, the ratio of the change rate to the time interval, and the trajectory segment standard; the spatial positioning result of partial discharge detection includes the layer path, the spatial coordinates of the trajectory points, the time interval, the transparency, and the line width value.

[0008] As a further solution of the present invention, the steps for obtaining the phase mutation recognition coordinate set are specifically as follows: S111: Based on the collected original waveform data of partial discharge pulse ultrasonic waves, perform sliding window division on the single-frame signal according to equal time intervals, call the signal envelope within each sliding window, track the fluctuation trend of the internal signal amplitude sequence, obtain the sampling time of all local extreme points and mark the sequence position number, and obtain an extreme time sequence index set; S112: According to the extreme time sequence index set, calculate the phase angle value between each pair of adjacent extreme points, and construct a difference sequence of adjacent extreme points based on the sequence position. Perform a one-to-one difference comparison operation on each phase difference and the set mutation angle threshold, using the formula: ; Calculate the mutation determination value between each pair of adjacent extreme points , and unidirectionally compare all the mutation determination values with the mutation angle threshold value, and screen the extreme points whose mutation determination values are greater than the mutation angle threshold value to obtain the mutation extreme value determination result set, where represents the phase angle value of the th extreme point, represents the sampling time of the th extreme point, represents the signal envelope amplitude of the th extreme point; S113: According to the extreme points determined to exceed the mutation threshold in the mutation extreme value determination result set, trace the corresponding original sampling time and combine it with the spatial coordinate value of the sensor for coordinate pairing, and use the combination of the sampling time of all extreme points that meet the screening conditions and the sensor spatial coordinates as the positioning identifier to establish the phase mutation recognition coordinate set.

[0009] As a further solution of the present invention, the steps for obtaining the acoustic wave propagation reconstruction waveform group are specifically as follows: S211: Based on the phase mutation recognition coordinate set, extract the signal envelope amplitude and phase value of each adjacent measurement point along the time axis direction, calculate the amplitude increment and phase difference between adjacent measurement points, and judge whether the amplitude increment is positive and whether the phase difference is continuously increasing. Sequentially screen the measurement point pairs that meet both conditions and connect them in the order of sampling time to generate a phase amplitude growth path group; S212: According to the phase amplitude growth path group, identify the time series corresponding to each path direction, perform reverse matching on each path, compare whether the forward and reverse paths have symmetry of the starting and ending measurement points, and use the formula: ; Calculate the direction symmetry difference between the th path and the th path. According to the difference, screen the path groups that meet the symmetry conditions, and perform corresponding matching on the bidirectional paths to obtain the diffusion direction symmetric path set, where represents the spatial position coordinate value of the th measurement point of the th path, represents the spatial position coordinate value of the th measurement point of the th path, represents the amplitude of the th measurement point of the th path, represents the amplitude of the th measurement point of the The amplitude value of the measurement point, representing the th path and the phase value of the measurement point, representing the th path and the phase value of the measurement point, representing the th path and the sampling time of the measurement point, representing the th path and the sampling time of the measurement point, is the number of measurement points in the path; S213: Based on all the paths in the diffusion direction symmetric path set, perform a phase value re - fitting operation on the peak point signals of the measurement points on each path, combine the fitted phase value sequence and the corresponding amplitude sequence of the measurement points in the order of the measurement point time, reconstruct and form continuous propagation information, and establish an acoustic wave propagation reconstruction waveform group.

[0010] As a further solution of the present invention, the acquisition steps of the sound source main direction vector set are specifically as follows: S311: Based on the amplitude value sequences of each measurement point in the acoustic wave propagation reconstruction waveform group, extract the adjacent three - frame amplitude value sequence data, perform point - to - point difference operations on the amplitude values of the first frame and the second frame, and the second frame and the third frame respectively, and arrange the obtained difference sequences in the order of the measurement point time to form a matrix structure, and construct a three - dimensional amplitude difference matrix; S312: Call the three - dimensional amplitude difference matrix, extract the gradient change trend of the front and rear measurement points in the continuous two - group difference sequences according to the spatial coordinate index, judge the consistency of the amplitude change directions between two measurement points, screen the measurement point pairs with consistent change directions, extract the spatial coordinates and mark the corresponding directions, and obtain the measurement point direction vector sequence; S313: According to the spatial coordinates and direction markings of each measurement point in the measurement point direction vector sequence, merge the continuous direction vectors in the order of the measurement point numbers, exclude the path segments with direction deviation values exceeding the direction similarity reference angle, summarize the retained vectors, and establish a sound source main direction vector set.

[0011] As a further solution of the present invention, the acquisition steps of the sound source time backtracking trajectory data are specifically as follows: S411: According to the path starting point in the sound source main direction vector set, trace back the relevant measurement point data frame by frame, extract the amplitude value and time stamp of the corresponding measurement points of each path segment, calculate the amplitude change amount and sampling time interval between the front and rear measurement points of each path segment, and obtain the path amplitude change time set; S412: Based on the path amplitude change time set, calculate the ratio of the amplitude change amount to the time interval for each path segment, compare the difference interval between the ratio and the linear increase / decrease trend standard, screen out the path segments that continuously satisfy the condition that the ratio difference is less than the change consistency threshold, and obtain a continuous trend path segment group; S413: According to the path segment numbers in the continuous trend path segment group, combine the start and end coordinates, timestamps, and amplitude values of each path segment in reverse chronological order, organize the path point sequence information in the backward tracing direction, and establish the sound source time backward tracing trajectory data.

[0012] As a further solution of the present invention, the steps for obtaining the spatial localization result of partial discharge detection are specifically as follows: S511: Based on the spatial coordinates of each trajectory point in the sound source time backward tracing trajectory data, obtain the image frame data collected by the visible light camera, extract the spatial mapping parameters of each frame of the image, perform projection conversion between the three-dimensional coordinates and the image coordinates of the trajectory points, map the trajectory points to the image frame, and establish an image mapping path group; S512: Based on the path point sequence in the image mapping path group, calculate the time interval from the start point to the end point of each path segment, and set the transparency and line width values of the path segment in the layer in combination with the time interval value. Use the formula: ; Calculate the rendering influence factor value of each path segment , set the transparency and line width values of the path segment according to the rendering influence factor value, attach the set parameter values to the layer path point attributes, and obtain the path rendering parameter matrix, where is the end point timestamp, is the start point timestamp of the path segment, is the path segment length, is the path segment speed change amount, 、 and 、 are the horizontal and vertical position values of the start and end points of the path segment in the image frame coordinates; S513: According to the transparency and line width values of each path segment in the path rendering parameter matrix, combine with the coordinate points in the corresponding image frame, perform layer path rendering, label all path segments on the image frame in the path order, and establish the spatial localization result of partial discharge detection.

[0013] An artificial intelligence-based leak detection and partial discharge detection system, comprising: The data acquisition module collects partial discharge pulse ultrasonic signals, divides them according to time intervals, extracts local extreme points, calculates the phase angle and compares it with the mutation angle threshold, screens out the extreme points and records the time and coordinates, and generates a phase mutation recognition coordinate set; The partial discharge signal processing module extracts the amplitude increment and phase difference based on the phase mutation recognition coordinate set, determines whether the conditions are met, connects the measurement points, forms a diffusion trajectory by reverse matching the path, fits the peak points and reconstructs the waveform to obtain a set of reconstructed acoustic wave propagation waveforms; The acoustic wave propagation reconstruction module performs difference operations based on the amplitude sequence of the set of reconstructed acoustic wave propagation waveforms, constructs a gradient difference matrix, determines the gradient change direction of the measurement points, extracts the spatial coordinates and directions, and generates a set of main direction vectors of the sound source; The gradient analysis module traces back the path segment data based on the set of main direction vectors of the sound source, calculates the amplitude change rate and time interval, compares the change rate with the trend standard, filters the real trajectory segments, and generates the sound source time backtracking trajectory data; The spatial positioning rendering module extracts the image coordinates and maps the spatial trajectory according to the sound source time backtracking trajectory data and the image frame data, sets the transparency and line width, completes the trajectory rendering, and obtains the spatial positioning result of the partial discharge detection.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, through the multi-level processing of the partial discharge pulse signal, from the envelope extreme value of the signal to the phase mutation analysis, more detailed discharge characteristics are captured, and the tiny partial discharge phenomena in the equipment can be discovered and located in time. By using the dual extraction of phase and amplitude, more expressive signal features are constructed, making the identification of partial discharge and leakage problems not only more accurate, but also enhancing the discrimination of fault modes. Through the acoustic wave propagation reconstruction and path analysis, not only can the diffusion trajectory of the signal be tracked, but also the spatial distribution and propagation direction of the discharge source can be inferred, improving the accurate positioning ability of the fault source. By using the visualization technology of time backtracking and layer dynamic rendering, the spatial positioning information is displayed in real time, enhancing the intuitiveness of the detection result and the decision support ability. The data-driven method has an adaptive ability, can optimize the detection strategy according to different equipment and environmental conditions, making the detection more intelligent and efficient, and can provide real-time fault monitoring in a more complex environment, significantly reducing the potential damage risk of the equipment and improving the operation safety of the power equipment. Description of the Drawings

[0015] Figure 1 is the main process flow chart of the present invention; Figure 2 is the flow chart for obtaining the phase mutation recognition coordinate set of the present invention; Figure 3 is the flow chart for obtaining the set of reconstructed acoustic wave propagation waveforms of the present invention; Figure 4 is the flow chart for obtaining the set of main direction vectors of the sound source of the present invention; Figure 5 is the flow chart for obtaining the sound source time backtracking trajectory data of the present invention; Figure 6 This is the flowchart for obtaining the spatial localization result of partial discharge detection in the present invention. Specific embodiments

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0018] Please refer to Figure 1 , a leak detection and partial discharge detection method based on artificial intelligence, comprising the following steps: S1: Collect the original waveform data of partial discharge pulse ultrasonic waves, perform sliding window division on a single-frame signal according to equal time intervals, extract all local extreme points of the signal envelope within each window, calculate the phase angles of adjacent extreme points and count the differences, perform one-way difference comparison between the differences and the mutation angle threshold, screen the extreme points exceeding the mutation threshold, record the corresponding sampling time and the spatial coordinates of the sensor, and generate a phase mutation recognition coordinate set; S2: Based on the phase mutation recognition coordinate set, extract the amplitude increment and phase difference between adjacent measurement points along the time axis direction, determine whether the amplitude increment is positive and whether the phase difference is continuously increasing, connect the measurement points that meet the conditions in chronological order to form a one-way path, reverse-match the forward and reverse paths to form a diffusion trajectory, perform phase fitting operation on the wave peak points of all measurement points on the path, and combine and reconstruct the fitted phase sequence and the corresponding amplitude sequence to obtain a sound wave propagation reconstruction waveform group; S3: According to the amplitude value sequence of each measurement point in the sound wave propagation reconstruction waveform group, perform pairwise difference operations on the amplitude values of three consecutive frames and arrange them along the time axis to construct a three-dimensional amplitude gradient difference matrix, determine whether there is a consistent gradient change direction between the front and rear measurement points, extract the corresponding spatial coordinates and change directions, and obtain a set of main direction vectors of the sound source; S4: According to the path start points in the main direction vector set of the sound source, trace back the measurement point data frame by frame, calculate the amplitude change rate and time interval of adjacent path segments, compare the ratio between the change rate and the time interval with the linear increase and decrease trend standard, screen out the path segments that continuously meet the condition of close ratio as the true trajectory segments, and generate the sound source time traceback trajectory data; S5: Based on the spatial coordinates of each trajectory point in the sound source time traceback trajectory data, obtain the image frame data collected by the visible light camera and extract the image coordinate mapping data, map the spatial trajectory points to the image frame to establish a layer path, calculate the time interval from the trajectory point to the end point, and set the transparency and line width values of each path segment to complete the dynamic rendering annotation of the trajectory on the layer, and obtain the local discharge detection spatial positioning result.

[0019] The phase mutation recognition coordinate set includes the time and spatial coordinates of the extreme points, the phase angle difference, the mutation angle threshold, and the difference screening result; the acoustic wave propagation reconstructed waveform group includes the reconstructed acoustic wave signal sequence, the fitted phase sequence, and the amplitude sequence; the main direction vector set of the sound source includes the spatial coordinates, the change direction, the amplitude increment, the phase difference, and the gradient change direction; the sound source time traceback trajectory data includes the path start point, the path segment amplitude change rate, the time interval, the ratio of the change rate to the time interval, and the trajectory segment standard; the local discharge detection spatial positioning result includes the layer path, the spatial coordinates of the trajectory points, the time interval, the transparency, and the line width value.

[0020] Please refer to Figure 2 , the S1 step is as follows: S111: Based on the collected original waveform data of the partial discharge pulse ultrasonic wave, perform sliding window division on the single-frame signal according to equal time intervals, call the signal envelope within each sliding window, track the fluctuation trend of the internal signal amplitude sequence, obtain the sampling time of all local extreme points and mark the sequence position number, and obtain the extreme time sequence index set; Based on the acquisition of the original waveform data of partial discharge pulse ultrasonic waves, sliding window division is performed on a single-frame signal at equal time intervals. The entire signal time series slides forward with a window length of 10 μs as a unit, and each slide is 2 μs, forming overlapping time-domain subsequences. For each subsequence, its envelope signal is extracted as the analysis object. Specifically, the Hilbert transform can be used to perform envelope extraction on the original waveform, and then local extreme points are searched in each envelope signal sequence, that is, for each sampling point in the envelope sequence that satisfies the reverse trend of the amplitude change before and after, it is marked as a maximum or minimum value. For example, in a 20 μs window, if the sampling frequency is 2 MHz, a total of 40 points are sampled, and there are 4 points that meet the extreme value definition, and their times are 3 μs, 8 μs, 12 μs, and 18 μs respectively, and the corresponding amplitudes are 0.9 V, 1.1 V, 0.85 V, and 1.3 V. These extreme points are numbered in the order of appearance, and their sampling times in the original signal are recorded. After integration, a set of extreme point data with time stamps and sequence positions is obtained, such as: Table 1 Data Table of Extreme Point Time and Amplitude As shown in Table 1, this type of data will be used in subsequent analysis for processing procedures such as phase angle judgment and mutation identification. Therefore, after this step, an extreme value time series index set is obtained.

[0021] S112: According to the extreme value time series index set, calculate the phase angle value between each pair of adjacent extreme points, and construct a difference sequence of adjacent extreme points based on the sequence position. Perform a one-to-one difference comparison operation on each phase difference and the set mutation angle threshold, using the formula: ; Calculate the mutation determination value between each pair of adjacent extreme points , and perform a one-way comparison of all mutation determination values with the mutation angle threshold value, and screen out the extreme points whose mutation determination values are greater than the mutation angle threshold value to obtain a mutation extreme value determination result set, where, represents the phase angle value of the th extreme point, represents the sampling time of the th extreme point, represents the signal envelope amplitude of the th extreme point; According to the extreme value time series index set, extract the information of each pair of adjacent extreme points, and respectively take key features such as their phase angles, sampling times, and envelope amplitudes to calculate the phase mutation judgment value. First, perform a difference operation on the angle values and normalize them, that is, divide the phase angle difference by the square root of the corresponding time difference to form a time-normalized angle mutation rate. Secondly, introduce a waveform amplitude perturbation adjustment factor, that is, take the product of the amplitudes of the current extreme point and the next extreme point and divide it by the absolute value of their angle sum, and superimpose it into the aforementioned angle mutation rate to construct a comprehensive mutation judgment value. Collect the following data: Table 2 Data table of adjacent extreme points As shown in Table 2, substituting into the formula gives: The first item of the angle normalization value is: ; The second item of the perturbation adjustment term is: ; After comprehensively adding, the mutation determination value is obtained: ; Set the mutation threshold angle to 30°. This value is obtained from the statistical analysis of the mutation critical values in the historical waveform samples. Calculated according to the normal distribution probability density, 90% of the samples are concentrated between 20 and 28. Therefore, 30 is a reasonable limit. The current calculation result is significantly higher than the threshold value of 30°, indicating that this point constitutes a mutation behavior. Add it to the mutation recognition result, and finally obtain the mutation extreme value determination result set.

[0022] S113: According to the extreme points determined to exceed the mutation threshold in the mutation extreme value determination result set, trace the corresponding original sampling time and combine it with the spatial coordinate value of the sensor for coordinate pairing. Combine the sampling time of all extreme points that meet the screening conditions with the sensor spatial coordinates as the positioning identifier to establish a phase mutation recognition coordinate set; Read the extreme points determined to be mutated in the mutation extreme value determination result set, mark their positions, map their original sampling times to the actual space-time coordinate system, and obtain the fixed spatial coordinates of the sensor where the corresponding extreme points are located according to the sensor layout corresponding to the collected signals. Use the sampling time identifier as the abscissa and the sensor spatial position as the ordinate to construct a two-dimensional recognition coordinate data set of extreme points. For example, if the sampling time of the mutation point is 15 μs and the spatial coordinates corresponding to the sensor number are (120 mm, 250 mm), then mark it as the point (15, 120, 250). Perform pairing on multiple mutation points in sequence and integrate them to form a set of multi-dimensional coordinate sets. This coordinate set will be used in the subsequent construction of the discharge path and spatial offset analysis input, and finally generate a phase mutation recognition coordinate set.

[0023] Please refer to Figure 3 , step S2 is as follows: S211: Based on the coordinate set identified by phase mutation, extract the signal envelope amplitude and phase value of each adjacent measurement point along the time axis direction, calculate the amplitude increment and phase difference between adjacent measurement points, determine whether the amplitude increment is positive and whether the phase difference is continuously increasing, sequentially screen the measurement point pairs that meet both conditions, and connect them in the order of sampling time to generate a phase-amplitude growth path group; Based on the coordinate set identified by phase mutation, sequentially extract the signal envelope amplitude and phase value between adjacent measurement points along the time axis direction. First, pair the sampling times of adjacent measurement points to ensure that the time interval is controlled between 0.1 ms and 0.5 ms. If the sampling time of a certain measurement point is 3.0 ms and the next measurement point is 3.3 ms, then this measurement point pair meets the pairing condition. Subsequently, extract the maximum amplitude of the signal envelopes of the two measurement points respectively. For example, if the amplitude of the first measurement point is 1.2 V and the amplitude of the second measurement point is 1.5 V, then the amplitude increment of this pair of measurement points is , if > 0, this condition holds. Further extract the phase values of the two measurement points. For example, the phase of measurement point 1 is 80° and the phase of measurement point 2 is 110°, calculate the phase difference , and record its change trend in the continuous path. If the phase in subsequent measurement points continues to be 130° and 150°, it indicates that the phase difference shows an increasing state, which conforms to the characteristic of unidirectional phase propagation. Connect the measurement point groups that meet the conditions of positive amplitude increment and continuously increasing phase difference in time series. In practical applications, the measurement point sequences of multiple sensors in a certain area can be selected, such as point A, point B, point C, and point D, corresponding to the sampling times of 2.0 ms, 2.3 ms, 2.6 ms, and 2.9 ms. If the amplitude sequence is 0.9 V, 1.1 V, 1.3 V, 1.4 V, and the phase sequence is 60°, 85°, 105°, 130°, then an effective unidirectional path can be connected and formed. Subsequently, starting from point A, connect B, C, and D in sequence to generate path A → B → C → D, and record its path number and measurement point information to form a phase-amplitude growth path group.

[0024] S212: According to the phase-amplitude growth path group, identify the time series corresponding to each path direction, perform reverse matching on each path, and compare whether the forward and reverse paths have symmetry of starting and ending measurement points. Use the formula: ; Calculate the direction symmetry difference between the rd path and the th path. According to the difference, screen the path groups that meet the symmetry conditions, perform corresponding matching on the bidirectional paths, and obtain the diffusion direction symmetry path set. Among them, represents the th path and the Spatial position coordinate values of the measurement points representing the th path, the spatial position coordinate values of the measurement points representing the th path, the amplitude value of the measurement point representing the th path, the amplitude value of the measurement point representing the th path, the phase value of the measurement point representing the th path, the phase value of the measurement point representing the th path, the sampling time of the measurement point representing the th path, the sampling time of the measurement point is the number of measurement points in the path; Based on each path identified in the phase - amplitude growth path group, by reading the measurement point index sequence of each path to obtain its corresponding time - series information, respectively extract the sampling time and spatial coordinate values of the first and last measurement points in the path. Denote the time of the first measurement point of the forward path as , and the time of the last measurement point of the reverse path as , and then judge whether the path direction has a time - symmetric feature, and make a one - to - one correspondence between the two - direction paths. On this basis, further extract the spatial position coordinates of each measurement point in the path , construct the first dimension of direction matching by using the square of the spatial coordinate difference as the distance measure, and record the amplitude of each measurement point in the path , the acquisition method is the maximum value of the envelope extracted from the original signal of the measurement point, and this value is obtained by taking the amplitude after Hilbert transform. If the amplitude of the signal envelope of a certain measurement point is 1.2V after processing, and the measurement point at the same sequence position in another path is 1.4V, then the corresponding calculation denominator is , and use this value as the normalization term for difference judgment. In addition, perform an absolute - value operation on the phase difference between each pair of measurement points and divide it by the product of the corresponding amplitude and phase to form the second - dimension feature of the direction difference. For example, if , , and , then the value of this part is . Finally, use the square - root of the absolute value of the sampling time difference between all measurement points in the path as the time stability index. For example, , , then this item is Three parts are weighted to construct a complete directional symmetry difference index Path length Take a fixed value such as 6 to unify and standardize the path scale. For all path pairs, construct such differences in sequence and determine whether they are lower than the symmetry threshold. The directional symmetry threshold is set to 1.5. This threshold is set based on the joint floating range of the three constituent indicators. Specifically, the sum of the squares of the coordinate differences is usually between 0.05 and 0.4, the normalized amplitude combination fluctuates between 1.3 and 1.7, the normalized phase difference is stable between 0.06 and 0.1, and the root mean square of the time difference is mainly concentrated between 0.4 and 0.6. Considering the maximum combined superposition situation of each item in the actual measurement point layout and signal response characteristics, the total floating range usually does not exceed 2.2. By selecting a boundary slightly lower than the average value within this range as the basis for determining effective paths, the directional symmetry threshold is set to 1.5. This value may fluctuate slightly with changes in the spatial density of measurement points, the amplitude of signal phase changes, and the discreteness of time intervals, but still uses 1.5 as the common critical criterion. By calculating the values of all path pairs, include path pairs with values lower than 1.5 into the two-way matching path set to generate a diffusion direction symmetry path set.

[0025] Table 3 Calculation Table of Diffusion Path Direction Symmetry (Unit: ms, m, V, degree) As shown in Table 3, through the comparison difference operation of the actual path group, it can be seen that the difference corresponding to paths P3 and P4 , is lower than the preset threshold boundary of 1.5. Although it is slightly higher but in a critical position, it can be regarded as part of the path group that can be included in the discussion range. This result shows that by calculating the matching degree of paths in terms of coordinate difference, amplitude and normalized phase, and sampling time stability, path pairs with strong direction consistency can be effectively distinguished, and a diffusion direction symmetry path set can be established based on this.

[0026] S213: Based on all paths in the diffusion direction symmetry path set, perform a phase value re-fitting operation on the peak point signals of the measurement points on each path. Combine the sequence of phase values obtained by fitting with the corresponding amplitude sequence of the measurement points in the time order of the measurement points to reconstruct and form continuous propagation information, and establish an acoustic wave propagation reconstruction waveform group; Symmetrize all the paths in the path set according to the diffusion direction, and perform phase reconstruction operations on the peak point signals of the measurement points on each path. First, read the envelope peak position index point of each measurement point, extract the corresponding original signal segment as the fitting interval, and form a phase fitting interval with 5 sampling points before and after it. For example, if the peak position index of the i-th measurement point is 3.5 ms, then the interval is taken as 3.45 ms to 3.55 ms. Select the signal phase values within this time range, use the adjacent point difference method to establish a fitting sequence, perform linear or polynomial regression on the phase change trends of all points in the sequence to extract the phase change rate, and obtain the phase fitting result sequence. For example, if the phase values of 5 points on a measurement point are 70°, 85°, 90°, 100°, 115°, the fitting result can be , where the slope part represents the phase change rate and the constant part is the intercept term. Subsequently, according to the fitting phase sequence of each measurement point, perform one-to-one combination processing on its amplitude sequence. The amplitude is selected as the signal envelope value at the same time point in the fitting phase sequence. For example, the amplitude corresponding to 3.5 ms in the phase fitting sequence is 1.6 V, 3.52 ms corresponds to 1.8 V, and 3.54 ms is 2.0 V. Combine the phase value with its corresponding amplitude to generate a combined waveform point sequence. Finally, connect this type of waveform point sequence in the time order of the measurement points to reconstruct a continuous propagation trajectory signal sequence and establish an acoustic wave propagation reconstruction waveform group.

[0027] Please refer to Figure 4 , and the S3 step is as follows: S311: Based on the amplitude value sequences of each measurement point in the acoustic wave propagation reconstruction waveform group, extract the adjacent three-frame amplitude value sequence data, perform point-to-point difference operations on the amplitude values of the first frame and the second frame, and the second frame and the third frame respectively, and arrange the obtained difference sequences in the time order of the measurement points to form a matrix structure, and construct a three-dimensional amplitude difference matrix; Based on the amplitude value sequences of each measurement point in the acoustic wave propagation reconstruction waveform group, first extract the amplitude values of each measurement point in three consecutive time frames. For example, let a measurement point number be P5, and its amplitude values at three moments t1, t2, and t3 are 1.22, 1.49, and 1.65 mV respectively. Then, pairwise frame difference operations can be performed, that is, ΔA1 = 1.49 - 1.22 = 0.27 mV, ΔA2 = 1.65 - 1.49 = 0.16 mV. The above differences represent the amplitude change conditions of the measurement point between consecutive time frames. This operation is repeated on the entire measurement point sequence to form a two-dimensional array matrix composed of two differences of each measurement point within three frames. After binding each group of differences with its time label to form a three-dimensional structure, the measurement point number and spatial coordinates also need to be incorporated into the matrix index dimension during the execution. The constructed matrix index structure is [measurement point number][difference sequence][time axis position]. For example, for five measurement points P1 to P5, their corresponding amplitude values within three frames can be referred to the following table: Table 4 Three-frame amplitude value sequence table As shown in Table 4, the amplitude change rate and direction of the measurement point in the time axis direction can be obtained through the amplitude difference between two adjacent frames. When the difference results form a three-dimensional matrix, it is necessary to further define the difference order and time tags. For example, the first-layer matrix is the ΔA1 sequence, and the second layer is the ΔA2 sequence. The corresponding time points are [t1→t2] and [t2→t3] respectively. This matrix structure is convenient for calling the amplitude gradient information of any time period of the corresponding measurement point in the follow-up. In actual operation, if the total number of measurement points is N, the total number of differences within three frames is N×2, and the size of the constructed matrix is [N×2×1]. For example, if N = 128, the matrix size is [128×2×1]. The value range of the difference data is determined by the change of the amplitude value, generally fluctuating between 0.05 mV and 0.5 mV. This range is the response interval of the equipment to the envelope change during the sound wave propagation at a sampling frequency of 10 kHz. Finally, a three-dimensional amplitude difference matrix is obtained.

[0028] S312: Call the three-dimensional amplitude difference matrix, extract the gradient change trend of the front and rear measurement points in two consecutive difference sequences according to the spatial coordinate index, judge the consistency of the amplitude change direction between two measurement points, screen out the measurement point pairs with the same change direction, extract the spatial coordinates and mark the corresponding directions to obtain the measurement point direction vector sequence; Call the difference data in the three-dimensional amplitude difference matrix, and sequentially extract the amplitude change trend between each pair of adjacent front and rear measurement points in the same time frame, and judge whether their change directions are the same, that is, if the amplitude differences of two measurement points in the same time frame are both positive or both negative, they are considered to have the same direction. In specific operation, assume that the amplitude differences between P4 and P5 between t1→t2 are 0.24 mV and 0.27 mV respectively, and the directions are the same. Then continue to judge whether the differences between them are in the same direction between t2→t3, for example, 0.22 mV and 0.16 mV respectively, which are also in the same direction. Judge that it is a measurement point pair that meets the conditions, record its number and corresponding coordinates (such as P4: 12.3, 18.6, 5.1; P5: 12.8, 18.9, 5.3), and at the same time mark its change direction as amplitude increasing. If the amplitude change direction of a certain measurement point pair is opposite in any time period, it is not included in the screening range. This judgment process is completed through two direction consistency screenings. Repeat this operation for all measurement point pairs. Finally, when the total number of measurement points is 128, about 80 - 100 groups of measurement point pairs meet this screening condition. This ratio is determined according to the measurement point density and waveform response stability at the site. Such measurement point pairs form a directional change sequence, that is, the measurement point direction vector sequence, which provides basic data for the subsequent main direction determination. Finally, the measurement point direction vector sequence is obtained.

[0029] S313: According to the spatial coordinates and direction annotations of each measurement point in the measurement point direction vector sequence, merge consecutive direction vectors in the order of the measurement point numbers, exclude the path segments where the direction deviation value exceeds the direction similarity reference angle, summarize the retained vectors, and establish the main direction vector set of the sound source; According to the selected measurement point pairs and their spatial coordinate data in the measurement point direction vector sequence, construct a continuous path vector sequence by merging point by point in the order of their sampling numbers. For example, if the difference between each direction vector in P1→P2→P3→P4 is less than 5° of the direction similarity reference angle, a continuous path is formed. If the angular difference between the directions of P3→P4 and P4→P5 is 9°, exceeding the reference angle of 5°, the path generation is terminated at the breakpoint P4. The setting basis of the direction similarity reference angle of 5° is that the sampling direction difference between adjacent measurement points in the sound wave propagation path is jointly affected by the measurement point layout density and the wave peak fitting variance. The measurement point layout density is to arrange a set of three-axis sensors every 10 cm. Combining the average value of the phase fitting residuals in the process of three-axis vector fitting, it is found that the included angle between adjacent path points is most stable between 3° and 7°. Among them, 5° is the critical value with the lowest comprehensive performance of the fitting deviation and spatial error. If the layout density is increased to a set of measurement points every 5 cm, the direction reference angle can be appropriately lowered to 3.5°. If the measurement point spacing is increased to 15 cm, the angle reference can be adjusted to 6°. Therefore, this angle value is adjusted with the combined mean square error of the layout spacing and the phase waveform fitting residuals. This angle reference value is the judgment reference obtained from the comparison test of the acquisition stability of the device in each channel. The judgment method is: based on the cosine value of the included angle between any two consecutive vectors in the three-dimensional coordinate. If cosθ>cos(5°)≈0.996, it is considered that the directions are the same. For example, if the included angle between the directions of the two vectors of P2→P3 and P3→P4 is 3.5°, the condition is satisfied, and this segment is retained in the path reconstruction. Repeat this operation to perform path generation and screening operations on all measurement point direction vectors. After the path is formed, summarize the vector information corresponding to the start and end nodes in all paths to obtain the path vector set representing the sound wave propagation direction in the three-dimensional space. This set is the result main direction vector set of the sound source in this step.

[0030] Please refer to Figure 5 , step S4 is as follows: S411: According to the path start points in the main direction vector set of the sound source, trace back the relevant measurement point data frame by frame, extract the amplitude values and timestamps of the measurement points corresponding to each path segment, calculate the amplitude change amount and sampling time interval between the measurement points before and after each path segment, and obtain the path amplitude change time set; According to the path starting points in the set of main direction vectors of the sound source, select the first measurement point marked on each vector path as the starting node, trace back the measurement point information passed by the path frame by frame, extract the amplitude value and sampling time of the measurement point as basic participation items. During the implementation, assume the path number is P1, the corresponding starting point number is M1, and in reverse time order are M2, M3... in sequence. Obtain the amplitude values of every two adjacent measurement points. For example, the amplitude values of M1 and M2 are 1.42 and 1.30 respectively, and the sampling times on the time axis are 12.00 μs and 11.75 μs respectively, obtaining an amplitude change amount of 0.12 and a time interval of 0.25 μs. Continue to process M2 and M3, and obtain the amplitude change amounts and time intervals of all path segments in sequence. Record the change data of each segment to form a path segment amplitude change time set. At the same time, preliminarily identify the amplitude change direction and judge whether it is a positive change. Mark and exclude negative amplitude segments. Such path tracing is applicable to the early diffusion determination area in the judgment of the sound source propagation trajectory. For example, the continuous reflection path formed on the surface of the equipment housing. The data acquisition frequency of the traced back data is set to 40 MHz, each time frame corresponds to 25 ns, and the storage precision of the measurement point data is six significant digits in floating point type. All sampling times are uniformly adjusted relatively with the starting point of the first frame as the zero point. The processed traced back path amplitude change time data set is shown in Table 5.

[0031] Table 5 Path Segment Amplitude Change Time Table As shown in Table 5, the amplitude change amount and time interval are respectively maintained within the conventional ranges of 0.08 V - 0.12 V and 0.20 μs - 0.30 μs in different path segments, providing basic data support for the subsequent judgment of the amplitude change rate ratio. As shown in Table 1, the distributions of the amplitude change amount and time interval are maintained within the same order of magnitude range, facilitating subsequent ratio calculation and trend identification to obtain the path amplitude change time set.

[0032] S412: Based on the path amplitude change time set, perform ratio calculation on the amplitude change amount and time interval in each path segment, compare the difference interval between the ratio and the linear increase and decrease trend standard, screen the path segments that continuously meet the condition that the ratio difference is less than the change consistency threshold, and obtain a set of continuous trend path segments; Based on all the path segment data concentrated on the path amplitude change time, calculate the ratio of the amplitude change amount to the corresponding time interval for each segment one by one. Suppose a certain segment is numbered M1 - M2, its amplitude change amount is 0.12V, the time interval is 0.25μs, and the ratio is 0.48. Continue to calculate for the M2 - M3 segment. Its amplitude change amount is 0.08V, the time interval is 0.20μs, and the ratio is 0.40. After sorting the ratios of all path segments according to the path number, calculate the ratio difference between adjacent segments and determine whether it is lower than the linear trend change consistency threshold. This threshold is set by comparing the standard deviation range of the ratios of the entire path segments and is set using the consistency interval within 1σ. Suppose the average value of the ratios of the entire path segments is 0.46 and the standard deviation is 0.05, then the consistency threshold is set to 0.05. If the ratio difference between a certain segment and its previous segment is 0.06, it is not included in the judgment of the continuous trend path segment. For example, the ratio of the M2 - M3 segment is 0.40, and the M1 - M2 segment is 0.48, and their difference is 0.08, exceeding the threshold range, so it is determined as an interrupted segment; on the contrary, the ratio of the M3 - M4 segment is 0.44, and its difference from the M2 - M3 segment is 0.04, so it is retained as a continuous segment. Connect multiple path segments that meet this condition into a continuous trend path segment group to form a continuous structure for subsequent trajectory extraction, and obtain the continuous trend path segment group.

[0033] S413: According to the path segment numbers in the continuous trend path segment group, combine the start and end coordinates, timestamps, and amplitude values of each segment of the path in reverse chronological order, organize the path point sequence information in the backward tracing direction, and establish the sound source time backward tracing trajectory data; According to the order of the path segment numbers in the continuous trend path segment group, connect the start and end coordinates of each segment of the path in sequence. For example, if the path segments M3 - M4 and M4 - M5 form a continuous segment, then use the coordinates of M3 as the starting point and the coordinates of M5 as the ending point, and integrate the sampling time, amplitude value, and relative position of each measurement point in the middle in order. During the construction process, if the direction angle deviation of a certain segment exceeds the set value of the direction similarity reference angle, then this segment of the path is excluded. The direction similarity reference angle is based on the included angle between the direction vectors of each segment in the continuous segment. If the included angle difference is less than 5 degrees, it is determined that the directions are consistent, otherwise it is regarded as a direction mutation segment. The angle difference between the direction angles of the M4 - M5 segment and the M3 - M4 segment is 3.7 degrees, meeting the direction consistency condition, so the path segment is retained. After combining all the path segments, arrange the time information in reverse order, trace back from the measurement point of the first segment as the ending point to the starting point of the last segment, record the measurement point number, timestamp, amplitude value, and spatial coordinates item by item, and finally integrate them into a structured path trajectory array to establish the sound source time backward tracing trajectory data.

[0034] Please refer to Figure 6 For step S5: S511: Based on the spatial coordinates of each trajectory point in the sound source time backtracking trajectory data, obtain the image frame data collected by the visible light camera, extract the spatial mapping parameters of each frame of the image, perform projection conversion between the three-dimensional coordinates and the image coordinates of the trajectory points, map the trajectory points to the image frame, and establish an image mapping path group; Based on the spatial coordinates of each trajectory point in the sound source time backtracking trajectory data, first sort the trajectory points in chronological order and determine the three-dimensional coordinate values of the starting point of each trajectory segment. In this embodiment, there are four trajectory segments, and their starting point coordinates are (1.2, 0.5, 3.1), (1.4, 0.6, 3.3), (1.6, 0.7, 3.5), and (1.8, 0.9, 3.7) respectively. Subsequently, obtain the image frame data collected by the visible light camera, and perform spatial projection mapping according to the calibration parameters between the sensor coordinate system and the camera image coordinate system. This mapping relationship uses a perspective projection model. Given the camera internal parameter matrix , rotation matrix , and translation vector , map the three-dimensional coordinates to two-dimensional image coordinates , and the corresponding formula is . In practical applications, for example, a certain trajectory point (1.4, 0.6, 3.3) is mapped to (124, 203) in the image frame. Record the coordinates of each trajectory point in the image frame and connect them in the trajectory order to form continuous image trajectory segments. Then, use the time stamp to number each segment of the path for subsequent parameter calculation and dynamic rendering. Finally, establish an image mapping path group.

[0035] S512: Based on the sequence of path points in the image mapping path group, calculate the time interval from the starting point to the end point of each path segment, and set the transparency and line width values of the path segment in the layer in combination with the time interval value. Use the formula: ; Calculate the rendering influence factor value of each path segment. Set the transparency and line width values of the path segment according to the rendering influence factor value, attach the set parameter values to the layer path point attributes, and obtain the path rendering parameter matrix. Among them, is the end point time stamp, is the path segment starting point time stamp, is the path segment length, is the path segment speed change amount, , and , are the horizontal and vertical position values of the starting and ending points of the path segment in the image frame coordinates; Call the sequence of path points in the image mapping path group, compare the start time and end time of each path segment, calculate the time interval from the start point to the end point, and further calculate the layer rendering parameters based on the path segment length and speed change amount, combined with the image coordinate change. In a specific implementation, the start times of each path segment are 1s, 2s, 3s, and 4s respectively, the end time is 5s, the path segment lengths are 1.2m, 1.45m, 1.1m, and 1.3m respectively, the speed change amounts are 0.8m / s², 0.95m / s², 0.7m / s², and 0.85m / s² respectively, and the horizontal and vertical coordinate changes of the image are shown in Table 5 respectively. Calculate according to the path segment rendering influence factor formula.

[0036] Substitute and calculate for the first path segment. , , , , , , then its rendering influence factor: ; ; Calculate the remaining path segments in the same way. The calculation results are used to set the transparency and line width values of this path segment. The rendering transparency is inversely proportional to the influence factor, and the line width is directly proportional to the influence factor. Finally, generate a path rendering parameter matrix. See Table 6 for specific data; Table 6 Rendering path segment parameter table As shown in Table 6, substituting the parameters of each path segment into the formula gives , , , and when setting the transparency based on this, set the transparency to , and the line width to pixels. Attach the rendering parameters to the layer path, and finally obtain the path rendering parameter matrix S513: According to the transparency and line width values of each path segment in the path rendering parameter matrix, combined with the coordinate points in the corresponding image frame, perform layer path rendering, label all path segments on the image frame in the path order, and establish the local discharge detection spatial positioning result; According to the transparency and line width values of each path segment in the path rendering parameter matrix, combined with their corresponding image coordinate points, the path segments are drawn through the layer rendering engine. After connecting the trajectory points in the image frame, they are marked segment by segment. In practice, if the starting image coordinates of a path segment are (120, 205) and the ending coordinates are (135, 215), and its rendering parameters are a transparency of 0.70 and a line width of 3.0 pixels, then this path segment is rendered as a 3.0 - pixel - wide line segment with a 70% transparency onto the image frame. After all path segments are rendered, the complete path trajectory is shown in the image frame, and thus the spatial position of the partial discharge can be clearly determined, and finally the spatial positioning result of the partial discharge detection is established.

[0037] An artificial - intelligence - based leak detection and partial discharge detection system, comprising: The data acquisition module collects partial discharge pulse ultrasonic signals, divides them by time intervals, extracts local extreme points, calculates the phase angle and compares it with the mutation angle threshold, filters the extreme points and records the time and coordinates, and generates a phase mutation recognition coordinate set. The partial discharge signal processing module, based on the phase mutation recognition coordinate set, extracts the amplitude increment and phase difference, determines whether the conditions are met, connects the measurement points, forms a diffusion trajectory by reverse - matching the path, fits the peak points and reconstructs the waveform, and obtains a set of acoustic wave propagation reconstructed waveforms. The acoustic wave propagation reconstruction module, based on the amplitude sequence of the set of acoustic wave propagation reconstructed waveforms, performs difference operations, constructs a gradient difference matrix, determines the gradient change direction of the measurement points, extracts the spatial coordinates and directions, and generates a set of main direction vectors of the sound source. The gradient analysis module, based on the set of main direction vectors of the sound source, traces back the path segment data, calculates the amplitude change rate and time interval, compares the change rate with the trend standard, filters the real trajectory segments, and generates the sound source time - backtracking trajectory data. The spatial positioning rendering module, according to the sound source time - backtracking trajectory data and the image frame data, extracts the image coordinates and maps the spatial trajectory, sets the transparency and line width, completes the trajectory rendering, and obtains the spatial positioning result of the partial discharge detection.

[0038] The above is only a preferred embodiment of the present invention, and it does not impose other forms of limitations on the present invention. Any person skilled in the relevant art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above - mentioned embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. An artificial intelligence-based leak detection and partial discharge detection method, characterized in that, It includes the following steps: S1: Obtain partial discharge pulse signal data, perform sliding window partitioning, extract local extreme points and calculate phase differences, determine whether it exceeds the mutation threshold, record the time stamp and spatial coordinates, and generate a phase mutation recognition coordinate set; S2: Based on the phase mutation recognition coordinate set, extract the amplitude increment and phase difference, determine the increasing trend, construct a one-way propagation path and perform reverse fitting, and reconstruct the waveform data to obtain a set of acoustic wave propagation reconstructed waveforms; S3: According to the amplitude data of each measurement point in the set of acoustic wave propagation reconstructed waveforms, calculate the amplitude difference and construct a gradient matrix, screen the measurement points that meet the gradient change conditions, and obtain the set of main direction vectors of the sound source; S4: Perform path backtracking according to the set of main direction vectors of the sound source, extract the amplitude value and time interval of each point, calculate the amplitude change rate, screen the trajectory segments that meet the conditions, and generate the sound source time backtracking trajectory data; S5: Based on the sound source time backtracking trajectory data, obtain visible light image frames, perform trajectory mapping, set the transparency and line width, perform trajectory rendering, and obtain the local discharge detection spatial positioning result.

2. The leak detection and partial discharge detection method based on artificial intelligence according to claim 1, characterized in that, The phase mutation recognition coordinate set includes the time, spatial coordinates, phase angle difference, mutation angle threshold, and difference screening result of the extreme points; the set of acoustic wave propagation reconstructed waveforms includes the reconstructed acoustic wave signal sequence, the fitted phase sequence, and the amplitude sequence; the set of main direction vectors of the sound source includes the spatial coordinates, change direction, amplitude increment, phase difference, and gradient change direction; the sound source time backtracking trajectory data includes the path start point, the amplitude change rate of the path segment, the time interval, the ratio of the change rate to the time interval, and the trajectory segment standard; the local discharge detection spatial positioning result includes the layer path, the spatial coordinates of the trajectory points, the time interval, the transparency, and the line width value.

3. The leak detection and partial discharge detection method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the phase mutation recognition coordinate set are as follows: S111: Based on the acquisition of the original waveform data of partial discharge pulse ultrasonic waves, perform sliding window partitioning on the single-frame signal according to equal time intervals, call the signal envelope within each sliding window, track the fluctuation trend of the internal signal amplitude sequence, obtain the sampling times of all local extreme points and mark the sequence position numbers to obtain the extreme time sequence index set; S112: According to the extreme time sequence index set, calculate the phase angle values between each pair of adjacent extreme points, and construct a difference sequence of adjacent extreme points according to the sequence position. Perform a one-to-one difference comparison operation between each phase difference and the set mutation angle threshold, using the formula: ； Calculate the mutation determination value between each pair of adjacent extreme points , unidirectionally compare all mutation determination values with the mutation angle threshold value, and screen out the extreme points whose mutation determination values are greater than the mutation angle threshold value to obtain the mutation extreme value determination result set, where represents the phase angle value of the th extreme point, represents the sampling time of the th extreme point, represents the signal envelope amplitude of the th extreme point; S113: According to the extreme points determined to exceed the mutation threshold in the mutation extreme point determination result set, track the corresponding original sampling time and combine it with the spatial coordinate value of the sensor for coordinate pairing. Combine the sampling times of all extreme points that meet the screening conditions and the spatial coordinates of the sensor as the positioning identifier to establish a phase mutation recognition coordinate set.

4. The leak detection and partial discharge detection method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the set of acoustic wave propagation reconstructed waveforms are as follows: S211: Based on the phase mutation recognition coordinate set, extract the signal envelope amplitude and phase value of each adjacent measurement point along the time axis direction, calculate the amplitude increment and phase difference between adjacent measurement points, determine whether the amplitude increment is positive and whether the phase difference increases continuously, sequentially screen the measurement point pairs that meet both conditions, and connect them in the order of sampling time to generate a phase amplitude growth path group; S212: According to the phase amplitude growth path group, identify the time series corresponding to each path direction, perform reverse matching on each path, compare whether the forward and reverse paths have symmetry of start and end measurement points, and use the formula: ； Calculate the directional symmetry difference between the th path and the th path. Screen the path groups that meet the symmetry condition according to the difference, and perform corresponding matching on the two-way paths to obtain the diffusion direction symmetric path set. Among them, represents the spatial position coordinate value of the th measurement point on the th path; represents the spatial position coordinate value of the th measurement point on the th path; represents the amplitude value of the th measurement point on the th path; represents the amplitude value of the th measurement point on the th path; represents the phase value of the th measurement point on the th path; represents the phase value of the th measurement point on the th path; represents the sampling time of the th measurement point on the th path; represents the sampling time of the th measurement point on the th path; is the number of measurement points in the path; S213: Based on all the paths in the diffusion direction symmetric path set, perform a phase value re-fitting operation on the peak point signals of the measurement points on each path, combine the fitted phase value sequence and the corresponding amplitude sequence of the measurement points in the order of the measurement point time, reconstruct and form continuous propagation information, and establish an acoustic wave propagation reconstruction waveform group.

5. The leak detection and partial discharge detection method based on artificial intelligence according to claim 1, characterized in that The specific steps for obtaining the set of main direction vectors of the sound source are as follows: S311: Based on the amplitude value sequences of each measurement point in the acoustic wave propagation reconstruction waveform group, extract the amplitude value sequence data of adjacent three frames, perform point-to-point difference operations on the amplitude values of the first frame and the second frame, and the second frame and the third frame respectively, and arrange the obtained difference sequences in the order of the measurement point time to form a matrix structure, and construct a three-dimensional amplitude difference matrix; S312: Call the three-dimensional amplitude difference matrix, extract the gradient change trend of the front and rear measurement points in the continuous two groups of difference sequences according to the spatial coordinate index, judge the consistency of the amplitude change direction between two measurement points, screen the measurement point pairs with the same change direction, extract the spatial coordinates and mark the corresponding directions, and obtain the measurement point direction vector sequence; S313: According to the spatial coordinates and direction markings of each measurement point in the measurement point direction vector sequence, merge the continuous direction vectors in the order of the measurement point numbers, exclude the path segments with direction deviation values exceeding the direction similarity reference angle, summarize the retained vectors, and establish a set of main direction vectors of the sound source.

6. The leak detection and partial discharge detection method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the sound source time backtracking trajectory data are as follows: S411: According to the path start points in the set of main direction vectors of the sound source, trace back the relevant measurement point data frame by frame, extract the amplitude value and time stamp of the measurement points corresponding to each path segment, calculate the amplitude change amount and sampling time interval between the front and rear measurement points of each path segment, and obtain the path amplitude change time set; S412: Based on the path amplitude change time set, perform a ratio calculation on the amplitude change amount and time interval in each path segment, compare the difference interval between the ratio and the linear increase and decrease trend standard, screen the path segments that continuously meet the condition that the ratio difference is less than the change consistency threshold, and obtain a continuous trend path segment group; S413: According to the path segment numbers in the continuous trend path segment group, combine the start and end coordinates, time stamps and amplitude values of each path segment in reverse time order, organize the path point sequence information in the backtracking direction, and establish the sound source time backtracking trajectory data.

7. The leak detection and partial discharge detection method based on artificial intelligence according to claim 1, characterized in that, The specific steps for obtaining the spatial positioning result of partial discharge detection are as follows: S511: Based on the spatial coordinates of each trajectory point in the sound source time backtracking trajectory data, obtain the image frame data collected by the visible light camera, extract the spatial mapping parameters of each frame of the image, perform projection conversion between the three-dimensional coordinates and the image coordinates of the trajectory points, map the trajectory points into the image frame, and establish an image mapping path group; S512: Based on the sequence of path points in the image mapping path group, calculate the time interval from the starting point of each path to the ending point, and combine the time interval value to set the transparency and line width values of the path segment in the layer. Use the formula: ； Calculate the rendering influence factor value for each path segment , set the transparency and line width values of the path segment according to the rendering influence factor value, attach the set parameter values to the layer path point attributes, and obtain the path rendering parameter matrix, where is the end point timestamp, is the start point timestamp of the path segment, is the length of the path segment, is the speed change amount of the path segment, , and , are the horizontal and vertical position values of the start and end points of the path segment in the image frame coordinates; S513: According to the transparency and line width values of each path segment in the path rendering parameter matrix, combined with the coordinate points in the corresponding image frame, perform layer path rendering, label all path segments on the image frame in the path order, and establish the local discharge detection spatial positioning result.

8. An artificial intelligence-based leak detection and partial discharge detection system, characterized in that, The system is used to execute the method described in any one of claims 1-7, and includes: The data acquisition module collects the partial discharge pulse ultrasonic signals, divides them according to time intervals, extracts the local extreme points, calculates the phase angle and compares it with the mutation angle threshold, filters the extreme points and records the time and coordinates, and generates a phase mutation recognition coordinate set; The partial discharge signal processing module, based on the phase mutation recognition coordinate set, extracts the amplitude increment and the phase difference, determines whether the conditions are met, connects the measurement points, forms a diffusion trajectory by reverse matching the path, fits the wave peak points and reconstructs the waveform, and obtains a set of acoustic wave propagation reconstructed waveforms; The acoustic wave propagation reconstruction module, based on the amplitude sequence of the set of acoustic wave propagation reconstructed waveforms, performs a difference operation, constructs a gradient difference matrix, determines the gradient change direction of the measurement points, extracts the spatial coordinates and directions, and generates a set of main direction vectors of the sound source; The gradient analysis module, based on the set of main direction vectors of the sound source, backtracks the path segment data, calculates the amplitude change rate and the time interval, compares the change rate with the trend standard, filters the real trajectory segments, and generates the sound source time backtracking trajectory data; The spatial positioning rendering module, according to the sound source time backtracking trajectory data and the image frame data, extracts the image coordinates and maps the spatial trajectory, sets the transparency and line width, completes the trajectory rendering, and obtains the local discharge detection spatial positioning result.

Images