A method and system for detecting power lines based on sound signals

Abstract

The embodiment of the present invention relates to the technical field of power line system detection, and discloses a power line detection method based on sound signals, including: obtaining sound detection information in the current detection scene through a sound acquisition array; identifying and processing the processed sound detection information through a partial discharge pulse recognition algorithm to determine the partial discharge pulse in the sound detection information, determining the amplitude parameters of each phase interval according to the pulse peak characteristics of the partial discharge pulse, and generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter; inputting the phase-amplitude distribution diagram into a pre-built partial discharge recognition model for recognition to obtain a corresponding partial discharge type recognition result; and outputting the corresponding partial discharge type recognition result and the phase-amplitude distribution diagram. It can improve the accuracy and efficiency of power line system safety detection.

Description

Technical Field

[0001] The present invention relates to the technical field of power line detection, and in particular to a power line detection method and system based on sound signals. Background Art

[0002] At present, in the power system, whether it is high-voltage or ultra-high-voltage transmission and distribution lines, or transformers and other equipment in substations, the phenomenon of partial discharge (PD) cannot be ignored. Partial discharge refers to the non-penetrating discharge between electrodes of the insulating medium in high-voltage electrical appliances under the action of high electric field strength. This phenomenon may cause serious accidents and economic losses, so power production and power transmission units are in urgent need of timely detection of partial discharge phenomena through effective means. Therefore, designing a solution that can perform efficient power line detection has become a technical problem that needs to be solved urgently by technicians in this field. Summary of the invention

[0003] In view of the above-mentioned defects, an embodiment of the present invention discloses a power line detection method based on sound signals, which can realize efficient partial discharge detection of the power line system and improve the accuracy and efficiency of the overall detection results.

[0004] A first aspect of an embodiment of the present invention discloses a power line detection method based on sound signals, comprising:

[0005] Acquiring sound detection information in a current detection scenario through a sound collection array, and filtering the sound detection information according to filter parameters to obtain processed sound detection information; wherein the filter parameters are determined according to the frequency characteristics of the partial discharge signal;

[0006] Dividing a power frequency cycle into a plurality of phase intervals, identifying and processing the processed sound detection information by a partial discharge pulse identification algorithm to determine the partial discharge pulse in the sound detection information, determining the amplitude parameter of each phase interval according to the pulse peak characteristics of the partial discharge pulse, and generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter;

[0007] Inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification, so as to obtain a corresponding partial discharge type identification result;

[0008] Output the corresponding partial discharge type identification results and phase-amplitude distribution diagram.

[0009] As an optional implementation, in the first aspect of the embodiment of the present invention, the sound collection array includes using 136 sound spiral arrays to acquire sound signals; after filtering the sound detection information according to the filter parameters to obtain the processed sound detection information, it also includes:

[0010] Analyze and process the processed sound detection information to determine the time information when the sound source arrives at each sound sensor in the sound collection array;

[0011] Calculate the time difference of the sound signal arriving at each sound sensor in the sound collection array;

[0012] The location information of the sound source in the current detection environment is determined according to the calculated time difference and the preset sound source calculation formula; and the corresponding direction information and distance information are determined according to the location information; the sound source calculation formula is: , where θ is the direction angle of the sound source, δ is the distance between the two microphones, c is the speed of sound, and Δt is the time difference between the sound source signal reaching the two microphones;

[0013] Performing data correction on the amplitude parameter according to the direction information and the distance information to obtain a corrected partial discharge pulse amplitude;

[0014] The generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter comprises:

[0015] A corresponding phase-amplitude distribution diagram is generated according to the phase interval and the corrected partial discharge pulse amplitude.

[0016] As an optional implementation manner, in the first aspect of the embodiment of the present invention, determining the amplitude parameter of each phase interval according to the pulse peak characteristic of the partial discharge pulse includes:

[0017] Data matching is performed on the local discharge pulse according to a set pulse detection threshold. When it is detected that the amplitude of the local discharge pulse exceeds the set pulse detection threshold, sequence detection is performed along the signal time from the pulse point exceeding the threshold until a point is found where the pulse amplitude no longer increases but begins to decrease. The currently detected amplitude is used as the pulse peak feature of the corresponding local discharge pulse, and the amplitude parameters of each phase interval are determined according to the pulse peak feature.

[0018] As an optional implementation manner, in the first aspect of the embodiment of the present invention, after the processed sound detection information is identified by the partial discharge pulse identification algorithm to determine the partial discharge pulse in the sound detection information, it also includes:

[0019] Extracting a pulse sequence from the pre-processed partial discharge signal based on a set detection threshold and a set pulse width, and identifying the occurrence time and amplitude of each partial discharge pulse;

[0020] According to the phase information recorded during acquisition, each pulse is associated with the corresponding phase, and a corresponding phase-time distribution diagram is generated according to the phase information and the time information, and a corresponding amplitude-time distribution diagram is generated according to the amplitude information and the time information;

[0021] The phase-time distribution diagram and the amplitude-time distribution diagram are input into a trained convolutional neural network for recognition to obtain corresponding recognition results, and whether there is partial discharge is determined based on the recognition results.

[0022] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification to obtain a corresponding partial discharge type identification result includes:

[0023] Performing region division on the phase-amplitude distribution map according to a set image division logic to divide the phase-amplitude distribution map into a first image detection region, a second image detection region, a third image detection region, and a fourth image detection region;

[0024] Performing first feature extraction on the first image detection area and the third image detection area to obtain first feature parameters associated with the first image detection area and the third image detection area, and inputting the first feature parameters into a first partial discharge recognition model for recognition to determine a corresponding first partial discharge detection result, wherein the first partial discharge recognition model is constructed based on a support vector machine or a random forest model;

[0025] performing second feature extraction on the second image detection area and the fourth image detection area to obtain second feature parameters associated with the second image detection area and the fourth image detection area, and inputting the second feature parameters into a second partial discharge recognition model for recognition to determine a corresponding second partial discharge detection result, wherein the second partial discharge recognition model is constructed based on a convolutional neural network;

[0026] A corresponding partial discharge type identification result is determined according to the first partial discharge detection result and the second partial discharge detection result.

[0027] As an optional implementation manner, in the first aspect of the embodiment of the present invention, determining a corresponding partial discharge type identification result according to the first partial discharge detection result and the second partial discharge detection result includes:

[0028] When the probability of the partial discharge type corresponding to the first partial discharge detection result exceeds a first set value, the probability of the partial discharge type corresponding to the second partial discharge detection result exceeds a second set value, and the partial discharge type corresponding to the first partial discharge detection result is different from the partial discharge type corresponding to the second partial discharge detection result, executing the next step;

[0029] Performing time domain analysis on the pre-processed sound detection information to calculate the corresponding waveform features, pulse features and energy features; wherein the waveform features include amplitude features, peak features, mean features and variance features; the pulse features include pulse width, rise time and fall time; and the energy features include signal energy distribution and change features;

[0030] Determine a first time domain characterization parameter of the corresponding waveform according to a ratio of an amplitude feature to a peak feature in the waveform feature, determine a second time domain characterization parameter of the corresponding waveform according to a ratio of a mean feature to a variance feature in the waveform feature, and determine a third time domain characterization parameter of the corresponding waveform according to a ratio of a pulse width to a sum of a rise time and a fall time in the waveform feature;

[0031] Use Fourier transform to perform spectrum analysis on the pre-processed sound detection information to obtain spectrum information, determine the frequency components of the sound detection information, calculate the energy distribution characteristics of different frequency bands, and determine the main frequency characteristics in the frequency information;

[0032] The sum of the products of the main frequency characteristics and the energy of each frequency band is calculated to obtain the first frequency domain characterization parameter, and the ratio of the high frequency band energy to the low frequency band energy is calculated to determine the second frequency domain characterization parameter; the first time domain characterization parameter, the second time domain characterization parameter, the third time domain characterization parameter, the first frequency domain characterization parameter and the second frequency domain characterization parameter are standardized, and the standardized parameters are input into the waveform recognition model to obtain the corresponding partial discharge result.

[0033] As an optional implementation manner, in the first aspect of the embodiment of the present invention, the detection method further includes:

[0034] Short-time Fourier transform is used to calculate the frequency variance characteristics of sound detection information in different time windows;

[0035] The wavelet transform algorithm is used to extract the wavelet coefficient energy of the sound detection information at different scales, and the sum of the wavelet coefficient energy at different scales is calculated as the wavelet energy feature;

[0036] Calculate the energy center of gravity coordinates of the sound detection information on the entire time-frequency plane according to the Wigner probability distribution, and use the time center of gravity and the frequency center of gravity as the corresponding center of gravity features; obtain the corresponding time-frequency domain comprehensive features according to the frequency variance features, wavelet energy features and center of gravity features;

[0037] Calculating the sound detection information to determine corresponding skewness characteristics, kurtosis characteristics, autocorrelation coefficients, and cross-correlation coefficients;

[0038] The time-frequency domain comprehensive features, skewness features, kurtosis features, autocorrelation coefficients and cross-correlation coefficients are input as input features into a waveform recognition model to obtain corresponding partial discharge results;

[0039] When it is determined that the partial discharge result is that there is partial discharge behavior, corresponding partial discharge level information is determined according to each acquired waveform feature, and the partial discharge level information is output;

[0040] The detection method further comprises:

[0041] Obtain corresponding vibration parameters by means of vibration sensors arranged at corresponding detection points;

[0042] The vibration parameters and sound detection information are input into a signal enhancement model for processing to obtain corresponding sound enhancement information; and the sound enhancement information is used as extrusion for subsequent pulse recognition; the signal enhancement model is a signal enhancement model constructed based on ultrasonic signals and vibration signals.

[0043] A second aspect of an embodiment of the present invention discloses a power line detection system based on sound signals, comprising:

[0044] A sound collection module is used to obtain the sound detection information in the current detection scene through the sound collection array, and filter the sound detection information according to the filter parameters to obtain the processed sound detection information; wherein the filter parameters are determined according to the frequency characteristics of the partial discharge signal;

[0045] Phase diagram generation module: used for dividing a power frequency cycle into multiple phase intervals, identifying and processing the processed sound detection information through a partial discharge pulse recognition algorithm to determine the partial discharge pulse in the sound detection information, determining the amplitude parameter of each phase interval according to the pulse peak characteristic of the partial discharge pulse, and generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter;

[0046] Identification module: used for inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification, so as to obtain a corresponding partial discharge type identification result;

[0047] Output module: used to output the corresponding partial discharge type identification results and phase-amplitude distribution diagram.

[0048] A third aspect of an embodiment of the present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the power line detection method based on sound signals disclosed in the first aspect of the embodiment of the present invention.

[0049] A fourth aspect of an embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the power line detection method based on sound signals disclosed in the first aspect of an embodiment of the present invention.

[0050] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0051] The power line detection method based on sound signals in the embodiment of the present invention uses a pre-built partial discharge identification model to identify the phase-amplitude distribution diagram, and can accurately determine the type of partial discharge, such as corona discharge, surface discharge, internal discharge, etc., based on a large amount of historical data and machine learning algorithms. Different partial discharge types correspond to different fault causes and severity, and accurate identification helps to take targeted maintenance measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 It is a flow chart of a power line detection method based on sound signals disclosed in an embodiment of the present invention;

[0054] Figure 2 It is a schematic diagram of a specific process for determining the type of partial discharge disclosed in an embodiment of the present invention;

[0055] Figure 3 It is another specific flow chart of determining the type of partial discharge disclosed in the embodiment of the present invention;

[0056] Figure 4 It is a structural schematic diagram of a power line detection system based on sound signals provided by an embodiment of the present invention;

[0057] Figure 5 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] It should be noted that the terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having" in the embodiments of the present invention and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0060] At present, in the power system, whether it is high-voltage or ultra-high-voltage transmission and distribution lines, or transformers and other equipment in substations, the phenomenon of partial discharge (PD for short) cannot be ignored. Partial discharge refers to the non-penetrating discharge between electrodes of the insulating medium in high-voltage electrical appliances under the action of high electric field strength. This phenomenon may cause serious accidents and economic losses, so power production and power transmission units are in urgent need of timely detection of partial discharge phenomena through effective means. Based on this, the embodiment of the present invention discloses a power line detection method, system, electronic device and storage medium based on sound signals, which uses a pre-built PD identification model to identify the phase-amplitude distribution diagram, and can accurately determine the type of PD based on a large amount of historical data and machine learning algorithms, such as corona discharge, surface discharge, internal discharge, etc. Different types of PD correspond to different fault causes and severity, and accurate identification helps to take targeted maintenance measures.

[0061] Embodiment 1

[0062] See also Figure 1 , Figure 1It is a flow chart of the power line detection method based on sound signals disclosed in the embodiment of the present invention. Among them, the execution subject of the method described in the embodiment of the present invention is an execution subject composed of software and / or hardware, and the execution subject can receive relevant information by wired or / and wireless means, and can send certain instructions. Of course, it can also have certain processing functions and storage functions. The execution subject can control multiple devices, such as a remote physical server or cloud server and related software, or it can be a local host or server and related software that performs related operations on a device placed somewhere. In some scenarios, multiple storage devices can also be controlled, and the storage devices can be placed in the same place or different places as the devices. For example Figure 1 As shown, the power line detection method based on sound signals includes the following steps:

[0063] S101: Acquire sound detection information in a current detection scenario through a sound collection array, and filter the sound detection information according to filter parameters to obtain processed sound detection information; wherein the filter parameters are determined according to frequency characteristics of a partial discharge signal;

[0064] S102: dividing a power frequency cycle into a plurality of phase intervals, identifying and processing the processed sound detection information by a partial discharge pulse identification algorithm to determine a partial discharge pulse in the sound detection information, determining an amplitude parameter of each phase interval according to a pulse peak characteristic of the partial discharge pulse, and generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter;

[0065] S103: inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification, so as to obtain a corresponding partial discharge type identification result;

[0066] S104: Outputting corresponding partial discharge type identification results and phase-amplitude distribution diagrams.

[0067] The sound collection array of the embodiment of the present invention can obtain sound detection information from the current detection scene comprehensively and multi-directionally. Compared with a single collection device, it can more accurately capture various sound signals emitted by power lines, including weak local discharge sound signals. The filter parameters are determined according to the frequency characteristics of the local discharge signal, and the collected sound detection information is filtered. Noise and interference signals with different frequencies from the local discharge signal can be specifically filtered out, the signal-to-noise ratio of the sound detection information can be improved, and subsequent analysis can be more accurate. For example, in a complex power environment, interference such as environmental noise and other equipment operation noise can be filtered out, and sound signals related to local discharge can be highlighted.

[0068] More specifically, dividing the power frequency cycle into multiple phase intervals can more carefully analyze the characteristics of local discharge pulses at different phases. The local discharge pulses in the sound detection information are determined by the partial discharge pulse recognition algorithm, and the amplitude parameters of each phase interval are determined according to the pulse peak characteristics, which helps to capture the dynamic changes of partial discharge and discover potential partial discharge anomalies. A phase-amplitude distribution diagram is generated based on the phase interval and amplitude parameters to display the distribution of partial discharge pulses in the power frequency cycle in an intuitive graphical manner. Operation and maintenance personnel can quickly understand the intensity, phase distribution and other key information of partial discharge through this diagram, which is convenient for judging the operating status of the power line. For example, it can be intuitively found whether the partial discharge is concentrated in certain specific phase intervals, thereby judging the type and location of the fault.

[0069] The embodiment of the present invention utilizes a pre-built partial discharge identification model to identify the phase-amplitude distribution diagram, and can accurately determine the type of partial discharge, such as corona discharge, surface discharge, internal discharge, etc., based on a large amount of historical data and machine learning algorithms. Different partial discharge types correspond to different fault causes and severity levels, and accurate identification helps to take targeted maintenance measures. Manual complex analysis and judgment of the phase-amplitude distribution diagram is required, and the model can quickly give the partial discharge type identification result, which greatly improves the efficiency of power line fault diagnosis and reduces the time and workload of manual analysis. It is particularly suitable for the detection and maintenance of large-scale power lines.

[0070] At the same time, the partial discharge type identification results and phase-amplitude distribution diagram are output, providing a comprehensive and intuitive basis for the operation and maintenance decision-making of power lines. Operation and maintenance personnel can not only understand the type of partial discharge, but also view the specific distribution of partial discharge in the power frequency cycle, so as to more accurately assess the fault risk of power lines and formulate reasonable inspection and maintenance plans and strategies.

[0071] More preferably, the processed sound detection information is analyzed and processed to determine the time information when the sound source arrives at each sound sensor in the sound collection array;

[0072] Calculate the time difference of the sound signal arriving at each sound sensor in the sound collection array;

[0073] The location information of the sound source in the current detection environment is determined according to the calculated time difference and the preset sound source calculation formula; and the corresponding direction information and distance information are determined according to the location information; the sound source calculation formula is: , where θ is the direction angle of the sound source, δ is the distance between the two microphones, c is the speed of sound, and Δt is the time difference between the sound source signal reaching the two microphones;

[0074] Performing data correction on the amplitude parameter according to the direction information and the distance information to obtain a corrected partial discharge pulse amplitude;

[0075] The generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter comprises:

[0076] A corresponding phase-amplitude distribution diagram is generated according to the phase interval and the corrected partial discharge pulse amplitude.

[0077] The embodiment of the present invention uses a preset sound source calculation formula to determine the location information of the sound source in the current detection environment based on the calculated time difference, and further obtains the direction information and distance information. The formula is based on the acoustic principle and combines parameters such as time difference, microphone spacing and sound speed to accurately calculate the direction angle of the sound source, thereby determining the specific location of the sound source. This is crucial for power line detection, because accurately locating the location of the local discharge source helps operation and maintenance personnel quickly locate the fault point and reduce troubleshooting time and workload. For example, in the complex equipment environment of a large substation, the specific equipment or part that emits the local discharge sound can be quickly located.

[0078] The amplitude parameters are corrected according to the direction and distance information of the sound source to obtain the corrected PD pulse amplitude. Since the amplitude will decay with the increase of distance during the sound propagation process, and different directions may be affected differently by environmental factors, this correction can more accurately reflect the true intensity of the partial discharge. For example, the amplitude of a distant PD source may be too small due to propagation attenuation. The correction can restore its true amplitude intensity and provide more accurate data for subsequent analysis.

[0079] When generating the phase-amplitude distribution diagram, the corrected PD pulse amplitude is used to replace the original amplitude parameter. The generated phase-amplitude distribution diagram can more accurately reflect the actual intensity distribution of PD at different phases, and provide a more reliable basis for subsequent PD type identification. For example, for some PD type identification models that are sensitive to amplitude changes, the corrected phase-amplitude distribution diagram can improve the accuracy and reliability of model identification, and help to more accurately determine the type and severity of PD.

[0080] In the specific implementation process, due to the complex and changeable environment of partial discharge detection, the characteristics of partial discharge signals may also change over time. Prior information can be used for the dynamic adjustment and tracking mechanism of the algorithm. If a small drift in the frequency range of the partial discharge signal is found during the detection process, the weight vector update strategy is adjusted based on the prior frequency range information and the real-time monitoring of the signal frequency changes. For directional changes, when it is detected that the signal direction has a tendency to deviate from the prior direction, the algorithm can adjust the beam direction in advance to maintain continuous tracking of the signal. Prior information is used to provide the algorithm with a benchmark and direction for dynamic adjustment, so that the algorithm can better adapt to changes in partial discharge signals and continue to maintain good performance.

[0081] When implementing it, the weights can be updated and adjusted according to the block method: First, data block: divide the received signal into multiple data blocks according to a certain time period, and each data block contains data at multiple discrete time points. Second, calculate the autocorrelation matrix of the data block: for each data block, calculate its autocorrelation matrix; Third, update the weight vector: at the end of each data block, calculate the new weight vector according to the MVDR algorithm formula based on the autocorrelation matrix of the data block and the expected signal direction vector: the weight vector is not updated at each time point, but is updated every time point (that is, the time period of each data block). This block update method can reduce the amount of calculation, and at the same time adapt to the slow changes of the signal and interference environment to a certain extent. In practical applications, it is necessary to reasonably select the length of the data block according to the speed of change of the signal and interference to balance the calculation efficiency and adaptive performance.

[0082] Accurately determining the location of partial discharge is crucial for equipment inspection and maintenance. The adaptive beamforming algorithm can accurately calculate the time difference, phase difference and other parameters of the signal reaching each array element by processing the array received signal, and then use these parameters to accurately locate the source of the partial discharge signal. Compared with traditional detection methods, this algorithm has significantly improved positioning accuracy. In the detection of partial discharge inside large power transformers, the positioning error can be controlled within a small range, providing strong support for maintenance personnel to quickly locate the fault location, shortening the maintenance time and improving equipment maintenance efficiency.

[0083] During the specific implementation, a synchronous data acquisition system is constructed to ensure that the signals collected by each array element are precisely synchronized in time. A high-precision clock source, such as a GPS synchronized clock or an atomic clock, is used to provide a unified time reference for data collection of each array element.

[0084] More preferably, the step of determining the amplitude parameters of each phase interval according to the pulse peak characteristics of the partial discharge pulse includes:

[0085] Data matching is performed on the local discharge pulse according to a set pulse detection threshold. When it is detected that the amplitude of the local discharge pulse exceeds the set pulse detection threshold, sequence detection is performed along the signal time from the pulse point exceeding the threshold until a point is found where the pulse amplitude no longer increases but begins to decrease. The currently detected amplitude is used as the pulse peak feature of the corresponding local discharge pulse, and the amplitude parameters of each phase interval are determined according to the pulse peak feature.

[0086] The embodiment of the present invention sets a pulse detection threshold to perform data matching on the local discharge pulse, which can filter out interference signals and noise with smaller amplitudes, and only focus on local discharge pulses with amplitudes exceeding the threshold, thereby improving the recognition accuracy of the true local discharge signal, avoiding misjudging weak signals of non-local discharge as local discharge pulses, and reducing the false alarm rate. From the pulse point exceeding the threshold, sequence detection is performed along the signal time until a point is found where the pulse amplitude no longer increases but begins to decrease, which is used as the pulse peak feature, and the true peak value of the local discharge pulse can be accurately captured. This helps to accurately reflect the intensity of the local discharge, because the peak value often represents the maximum energy release or signal strength in the local discharge process, providing key accurate data for subsequent analysis.

[0087] Based on the accurately acquired pulse peak characteristics, the amplitude parameters of each phase interval are determined, so that the amplitude parameters of the phase interval can more truly reflect the actual intensity distribution of partial discharge in different phases. This is crucial for generating an accurate phase-amplitude distribution diagram, which can enable subsequent partial discharge type identification and fault analysis to be based on more reliable data, improve the accuracy of partial discharge type judgment and the reliability of power line fault diagnosis. The method of setting thresholds and sequence detection is equivalent to setting up a double screening mechanism for partial discharge pulse identification, making the entire detection method more resistant to various interferences in the environment. Even in complex electromagnetic environments or in the presence of multiple noise sources, the key features of partial discharge pulses can be extracted more stably and accurately, ensuring the stability and reliability of the detection results and reducing the impact of environmental factors on the detection results.

[0088] More preferably, after the processed sound detection information is identified by the partial discharge pulse identification algorithm to determine the partial discharge pulse in the sound detection information, the method further includes:

[0089] Extracting a pulse sequence from the pre-processed partial discharge signal based on a set detection threshold and a set pulse width, and identifying the occurrence time and amplitude of each partial discharge pulse;

[0090] According to the phase information recorded during acquisition, each pulse is associated with the corresponding phase, and a corresponding phase-time distribution diagram is generated according to the phase information and the time information, and a corresponding amplitude-time distribution diagram is generated according to the amplitude information and the time information;

[0091] The phase-time distribution diagram and the amplitude-time distribution diagram are input into a trained convolutional neural network for recognition to obtain corresponding recognition results, and whether there is partial discharge is determined based on the recognition results.

[0092] According to the phase information recorded during acquisition, the embodiment of the present invention associates each pulse with the corresponding phase and generates a phase-time distribution diagram, which can determine the distribution of local discharge pulses over time at different phases. Generating an amplitude-time distribution diagram can clearly present the change of the amplitude of the local discharge pulse over time. The phase-time distribution diagram and the amplitude-time distribution diagram are input into the trained convolutional neural network for recognition. The powerful feature extraction and pattern recognition capabilities of the convolutional neural network can be used to automatically learn and extract the feature information in the diagram, so as to more accurately identify the pattern and characteristics of the local discharge. Compared with traditional manual recognition or simple algorithm recognition, it can greatly improve the accuracy and reliability of recognition and reduce the situation of misjudgment and missed judgment.

[0093] If the two-dimensional representation of time-amplitude is used, the time series and amplitude information of the local discharge pulse are retained, which can reflect the change of the discharge pulse over time; if the two-dimensional representation of phase-amplitude is used, the two-dimensional graph may contain more detailed information about the individual pulses, because it is not just a statistical result, but may also reflect the order of pulse occurrence and time interval. It also helps to improve the comprehensiveness of the detection.

[0094] In the specific implementation, feature extraction is performed first. Feature extraction can be performed on the time-phase diagram to obtain phase distribution, phase concentration and phase symmetry, where phase distribution is the distribution of statistical pulses in different phase intervals (such as 0°-90°, 90°-180°, etc.); phase concentration is to calculate the concentration of pulses in a specific phase interval (such as standard deviation, variance, etc.), and phase symmetry is to analyze the symmetry of pulses in the positive and negative half cycles (such as corona discharge is usually symmetrical, and surface discharge is asymmetric).

[0095] The amplitude-time distribution diagram also needs to be feature extracted to obtain the amplitude distribution, amplitude change trend, pulse density, etc., among which the amplitude distribution is the amplitude distribution of the statistical pulse (such as maximum amplitude, average amplitude, etc.); the amplitude change trend is to analyze the change trend of the pulse amplitude over time (such as gradual increase, gradual decrease, etc.), and the pulse density is to calculate the number of pulses per unit time, reflecting the activity level of partial discharge.

[0096] Then, the model is trained and classified according to the above characteristics, and a training data set containing multiple types of partial discharge is constructed, and the phase-time distribution diagram and amplitude-time distribution diagram of each type of partial discharge are annotated. Model selection: Select a suitable machine learning model (such as SVM, random forest) or deep learning model (such as CNN, LSTM). Model training: Use the training data set to train the model and optimize the model parameters to ensure that the model can accurately identify the type of partial discharge. Model verification: Use the verification data set to evaluate the performance of the model and adjust the model parameters to improve the recognition accuracy.

[0097] More preferably, Figure 2 As shown, the phase-amplitude distribution diagram is input into a pre-built partial discharge identification model for identification to obtain a corresponding partial discharge type identification result, including:

[0098] S1031: performing region division on the phase-amplitude distribution map according to a set image division logic to divide the phase-amplitude distribution map into a first image detection region, a second image detection region, a third image detection region, and a fourth image detection region;

[0099] S1032: performing first feature extraction on the first image detection area and the third image detection area to obtain first feature parameters associated with the first image detection area and the third image detection area, and inputting the first feature parameters into a first partial discharge recognition model for recognition to determine a corresponding first partial discharge detection result, wherein the first partial discharge recognition model is constructed based on a support vector machine or a random forest model;

[0100] S1033: performing second feature extraction on the second image detection area and the fourth image detection area to obtain second feature parameters associated with the second image detection area and the fourth image detection area, and inputting the second feature parameters into a second partial discharge recognition model for recognition to determine a corresponding second partial discharge detection result, wherein the second partial discharge recognition model is constructed based on a convolutional neural network;

[0101] S1034: Determine a corresponding partial discharge type identification result according to the first partial discharge detection result and the second partial discharge detection result.

[0102] The embodiment of the present invention divides the phase-amplitude distribution diagram into different regions, which can highlight the features of different parts of the diagram. Different regions may correspond to different local discharge characteristics or modes. Through this division, the features of each region can be analyzed and extracted more specifically, avoiding the confusion or masking of features that may occur during overall analysis, and helping to more accurately capture the detailed information of local discharge in different phases and amplitude intervals. The complex phase-amplitude distribution diagram is decomposed into relatively small and easier-to-process sub-regions, reducing the complexity of data processing. The amount of data in each sub-region is relatively small. When performing feature extraction and model recognition, the amount of calculation and processing difficulty are reduced accordingly, which improves the efficiency and speed of the entire recognition process, allowing the model to be trained and inferred more quickly. In addition, different models can be constructed according to different regions during specific implementation to improve recognition speed and accuracy. By combining the first local discharge detection result and the second local discharge detection result to determine the final local discharge type recognition result, the advantages of different models are fully utilized, and the complementarity of features and information is achieved. Different models analyze and identify the phase-amplitude distribution diagram from different perspectives. Their results verify and complement each other, which can determine the type of partial discharge more comprehensively and accurately, reduce the misjudgment and missed judgment that may occur in a single model, and improve the overall accuracy and reliability of partial discharge type identification.

[0103] In the specific implementation, different detection logic and models can be set for each region. Different detection logic and models can be set for each phase interval to more accurately identify the type of partial discharge. The following is a specific implementation method: Region 1 (0°-90°), detection logic: This region usually corresponds to the rising edge of voltage, and surface discharge and suspension discharge may show strong signals in this region. Detection model: Support vector machine (SVM) or random forest (Random Forest) model can be used, and the pulse amplitude and density characteristics of this region are focused on during training. Region 2 (90°-180°), detection logic: This region usually corresponds to the peak of the positive half cycle of voltage, and corona discharge may show strong signals in this region; detection model: Convolutional neural network (CNN) can be used, and the pulse shape and spectrum characteristics of this region are focused on during training. Region 3 (180°-270°), detection logic: This region usually corresponds to the falling edge of voltage, and surface discharge and suspension discharge may show strong signals in this region. Detection model: The same detection logic and model as region 1 can be used, but it is necessary to combine the data of region 1 for correlation analysis. Region 4 (270°-360°), detection logic: This region usually corresponds to the peak of the negative half cycle of the voltage, and corona discharge may show a strong signal in this region. Detection model: The same detection logic and model as region 2 can be used, but the data of region 2 needs to be combined for correlation analysis.

[0104] Data association between regions: In order to improve the accuracy of detection, data between different regions can be analyzed for association. For example: Association between region 1 and region 3: Surface discharge and suspension discharge usually show similar signal characteristics at the rising and falling edges of voltage. The type of partial discharge can be further confirmed by comparing the data of region 1 and region 3. Association between region 2 and region 4: Corona discharge usually shows symmetrical signal characteristics near the positive and negative peak values ​​of voltage. The type of partial discharge can be further confirmed by comparing the data of region 2 and region 4.

[0105] When carrying out specific implementation, there can be three implementation methods. The first is to identify the probability of all partial discharge types in the first area, the second area, the third area and the fourth area, and then assign different weights according to different areas to determine the corresponding identification results; the second is that each area identifies the first result and the second result, the first result is used as the identification result of the area, and the second result is used to assist other areas to perform comprehensive identification and determination, that is, the second result is used as an auxiliary result of other areas; the third is that each area is identified to obtain a result. If the four areas only obtain one partial discharge type identification result, then the corresponding identification result is directly determined. If the four areas obtain two partial discharge type identification results, both results are output, and when performing identification, the probability threshold can be dynamically adjusted according to the actual situation to determine the result. For example, the probability of corona discharge type is 80%, and the probability of surface discharge in another area is 90%. At this time, it is found that there will be a large conflict, so the user is reminded, and further comprehensive detection is combined with other methods.

[0106] In the specific implementation, some rules are formulated to screen and correct the test results. For example, if the types of PD detected in adjacent areas are too different and do not conform to the general development law of PD, the results should be re-evaluated, and it may be necessary to adjust the model parameters or re-extract features.

[0107] More preferably, Figure 3 As shown, the method of determining the corresponding partial discharge type identification result according to the first partial discharge detection result and the second partial discharge detection result further includes:

[0108] S1035: when the probability of the partial discharge type corresponding to the first partial discharge detection result exceeds a first set value, the probability of the partial discharge type corresponding to the second partial discharge detection result exceeds a second set value, and the partial discharge type corresponding to the first partial discharge detection result is different from the partial discharge type corresponding to the second partial discharge detection result, executing the next step;

[0109] S1036: Performing time domain analysis on the pre-processed sound detection information to calculate corresponding waveform features, pulse features and energy features; wherein the waveform features include amplitude features, peak features, mean features and variance features; the pulse features include pulse width, rise time and fall time; and the energy features include signal energy distribution and variation features;

[0110] S1037: determining a first time domain characterization parameter of the corresponding waveform according to a ratio of an amplitude feature to a peak feature in the waveform feature, determining a second time domain characterization parameter of the corresponding waveform according to a ratio of a mean feature to a variance feature in the waveform feature, and determining a third time domain characterization parameter of the corresponding waveform according to a ratio of a pulse width to a sum of a rise time and a fall time in the waveform feature;

[0111] S1038: Performing spectrum analysis on the pre-processed sound detection information using Fourier transform to obtain spectrum information, determining frequency components of the sound detection information, calculating energy distribution characteristics of different frequency bands, and determining main frequency characteristics in the frequency information;

[0112] S1039: Calculate the sum of the products of the main frequency characteristics and the energy of each frequency band to obtain a first frequency domain characterization parameter, calculate the ratio of the high frequency band energy to the low frequency band energy to determine the second frequency domain characterization parameter; standardize the first time domain characterization parameter, the second time domain characterization parameter, the third time domain characterization parameter, the first frequency domain characterization parameter and the second frequency domain characterization parameter, and input the standardized parameters into the waveform recognition model to obtain the corresponding partial discharge result.

[0113] The embodiment of the present invention triggers subsequent steps when the corresponding partial discharge type probabilities in the first partial discharge detection result and the second partial discharge detection result respectively exceed the set value and the types are different, thereby avoiding misjudgment caused by the uncertainty of a single model or noise interference. Only when both models have a high confidence level but the results are inconsistent, further in-depth analysis is performed, which ensures that subsequent analysis resources are used for fuzzy situations that really need to be solved, and improves the reliability of the entire partial discharge type identification system. In this way, complex situations that are difficult to judge directly through the results of the two models are screened out, and subsequent time domain and frequency domain analysis is focused on truly challenging cases, which improves the pertinence and efficiency of the analysis and avoids the waste of computing resources caused by comprehensive and in-depth analysis of all situations.

[0114] The first, second, and third time domain characterization parameters are calculated, and the ratio of different time domain features is calculated to highlight the relative relationship between the features. For example, the ratio of the amplitude feature to the peak feature can reflect the relative stability of the signal amplitude; the ratio of the mean feature to the variance feature can reflect the concentration and dispersion of the signal; the ratio of the pulse width to the sum of the rise time and fall time can describe the shape characteristics of the pulse. These characterization parameters further simplify and refine the time domain features, which helps to more clearly reveal the differences in the time domain between different types of partial discharge.

[0115] Using Fourier transform to perform spectrum analysis can obtain the frequency components of sound detection information, energy distribution characteristics of different frequency bands, and main frequency characteristics. Different types of partial discharges may have different performances in frequency. For example, some partial discharges may have stronger energy distribution in high frequency bands, while others may be more prominent in low frequency bands. By analyzing the frequency domain characteristics, important clues can be provided for partial discharge type identification from another perspective.

[0116] The first and second frequency domain characterization parameters are calculated to quantify the frequency domain features, further highlighting the relationship and characteristics between the frequency domain features. For example, the sum of the products of the main frequency characteristics and the energy of each frequency band can reflect the contribution of the main frequency in the entire spectrum energy distribution; the ratio of the high frequency band energy to the low frequency band energy can reflect the difference in the energy distribution of the signal in the high and low frequency bands. These characterization parameters help to more accurately describe the characteristics of different types of partial discharge in the frequency domain. The time domain and frequency domain characterization parameters are standardized so that different parameters have the same scale and range, avoiding the problem of unstable or inaccurate model training caused by differences in parameter scales. Inputting the standardized parameters into the waveform recognition model can improve the learning effect and recognition accuracy of the model, and more effectively use the comprehensive information of the time domain and frequency domain to determine the final partial discharge type recognition result.

[0117] In the embodiment of the present invention, the skewness is used to measure the asymmetry of the signal probability distribution. For partial discharge signals, the signals generated by electrical equipment under normal operating conditions usually have a relatively symmetrical distribution, and the skewness is close to 0. When partial discharge occurs, due to the complexity and asymmetry of the discharge process, the skewness of the partial discharge signal may change significantly. For example, corona discharge may produce signals with positive or negative skewness. This asymmetry can be captured by calculating the skewness, thereby providing clues to the occurrence of partial discharge. The skewness of signals generated by the same type of partial discharge may be different. For example, due to the differences in their discharge mechanisms and physical processes, surface discharge and suspended discharge may differ in the asymmetry of the signal. Analyzing the skewness can help distinguish these different types of partial discharges to a certain extent.

[0118] Kurtosis mainly describes the degree of peak or flatness of the signal probability distribution, which can reflect the probability of the occurrence of extreme values ​​in the signal. In partial discharge identification, the signal of normally operating electrical equipment usually has a relatively stable distribution, and the kurtosis value is within a certain range. However, partial discharge signals often have some pulses with large amplitudes, which leads to an increase in the kurtosis value of the signal. By monitoring the change of kurtosis, it is possible to timely detect whether there are abnormal peaks in the signal, and then determine whether partial discharge occurs. Generally speaking, the more severe the partial discharge is, the more abnormal pulses are generated and the larger the amplitude is, the higher the kurtosis value of the signal will be. Therefore, kurtosis can be used as an indicator to assess the severity of partial discharge. For example, when the kurtosis value is significantly higher than the normal level, it may mean that the partial discharge is more serious and timely measures need to be taken for inspection and maintenance. Combining skewness and kurtosis for analysis can more comprehensively describe the distribution form of partial discharge signals. Through these two high-order statistics, a richer feature space can be constructed to provide more information dimensions for partial discharge identification, which helps to improve the accuracy and reliability of identification.

[0119] More preferably, the detection method further comprises:

[0120] Short-time Fourier transform is used to calculate the frequency variance characteristics of sound detection information in different time windows;

[0121] The wavelet transform algorithm is used to extract the wavelet coefficient energy of the sound detection information at different scales, and the sum of the wavelet coefficient energy at different scales is calculated as the wavelet energy feature;

[0122] Calculate the energy center of gravity coordinates of the sound detection information on the entire time-frequency plane according to the Wigner probability distribution, and use the time center of gravity and the frequency center of gravity as the corresponding center of gravity features; obtain the corresponding time-frequency domain comprehensive features according to the frequency variance features, wavelet energy features and center of gravity features;

[0123] Calculating the sound detection information to determine corresponding skewness characteristics, kurtosis characteristics, autocorrelation coefficients, and cross-correlation coefficients;

[0124] The time-frequency domain comprehensive features, skewness features, kurtosis features, autocorrelation coefficients and cross-correlation coefficients are input as input features into a waveform recognition model to obtain corresponding partial discharge results;

[0125] When it is determined that the partial discharge result is that there is partial discharge behavior, corresponding partial discharge level information is determined according to each acquired waveform feature, and the partial discharge level information is output;

[0126] The embodiment of the present invention uses short-time Fourier transform to calculate the frequency variance characteristics of the sound detection information in different time windows. The frequency variance can reflect the fluctuation of the frequency in different time periods, and different types of partial discharges may have different frequency stability. For example, a stable partial discharge may have a smaller frequency variance, while an intermittent or unstable partial discharge may have a larger frequency variance. By extracting this feature, the dynamic changes in the frequency of the partial discharge signal can be more keenly captured, providing a basis for accurately identifying the type of partial discharge.

[0127] The wavelet transform algorithm is used to extract the wavelet coefficient energy of the sound detection information at different scales, and the sum is calculated as the wavelet energy feature. Wavelet transform has the characteristics of multi-resolution analysis, which can decompose the signal at different scales, so as to capture the energy distribution of the signal at different frequency ranges and time scales. Different types of partial discharges may have different energy distributions at different scales. Wavelet energy features can fully reflect this energy distribution and help distinguish partial discharges of different types and severity.

[0128] The energy center of gravity coordinates of the sound detection information on the entire time-frequency plane are calculated according to the Wigner probability distribution, and the time center of gravity and the frequency center of gravity are used as the center of gravity features. The energy center of gravity can reflect the energy concentration position of the signal on the time-frequency plane. Different types of partial discharge may cause different distributions of signal energy on the time-frequency plane, thus having different center of gravity features. By analyzing the center of gravity features, the time-frequency distribution characteristics of the partial discharge signal can be further explored to improve the accuracy of partial discharge identification.

[0129] The frequency variance feature, wavelet energy feature and centroid feature are combined into the time-frequency domain comprehensive feature, which integrates the information of multiple time-frequency domain dimensions. A single feature may not be able to fully describe the characteristics of the partial discharge signal, while the time-frequency domain comprehensive feature can more comprehensively and accurately reflect the time-frequency characteristics of the partial discharge signal, providing richer and more effective input for the subsequent waveform recognition model, and improving the recognition performance of the model.

[0130] The skewness feature, kurtosis feature, autocorrelation coefficient and cross-correlation coefficient are calculated for the sound detection information. Skewness and kurtosis can reflect the shape of the probability distribution of the signal. Different types of partial discharge signals may have different distribution characteristics. These two features can distinguish the distribution differences of the signal. The autocorrelation coefficient and cross-correlation coefficient are used to describe the correlation of the signal, which can reflect the characteristics of the signal such as periodicity and similarity. These statistical features supplement the feature description of the sound detection information from different angles, further enrich the input feature set, and help improve the recognition ability of the waveform recognition model. For skewness and kurtosis, their difference can be calculated as a new indicator, which can reflect the complexity of the signal distribution deviation from the normal distribution. The autocorrelation function can calculate its maximum value at a specific delay time, which can reflect the degree of self-similarity of the signal at the delay time. For the cross-correlation function, the maximum value of the cross-correlation function between different sensor signals and its corresponding delay time are calculated. The maximum value reflects the strength of the correlation between the signals, and the delay time can be used to judge the time difference of signal propagation. These new indicators are combined to form a comprehensive statistical feature indicator.

[0131] The time-frequency domain comprehensive features, skewness features, kurtosis features, autocorrelation coefficients and mutual correlation coefficients are input as input features into the waveform recognition model. The model can comprehensively utilize these multi-dimensional feature information for partial discharge identification. Compared with a single or small number of feature inputs, it can more accurately determine whether there is partial discharge behavior. When judging the existence of partial discharge behavior, the corresponding partial discharge level information is determined based on the acquired waveform features. Different waveform features are closely related to the severity of partial discharge. By comprehensively analyzing these features, the degree of harm of partial discharge can be more accurately assessed, providing more targeted decision-making basis for the maintenance and inspection of power lines. When making a specific partial discharge severity assessment, the initial waveform characteristic value can also be used as the corresponding judgment basis. The initial waveform characteristic value here can be an amplitude feature, a peak feature, a main frequency feature, etc.

[0132] The embodiments of the present invention effectively process the collected ultrasonic signals, and by adopting a variety of detection means and methods, it is helpful to extract the occasional burst pulse signals from the complex background noise. Using time-frequency analysis methods such as wavelet transform, the signal can be analyzed in two dimensions of time and frequency, and the moment when the burst pulse signal appears and its frequency characteristics can be accurately located, so that even if the signal appears briefly, it can be accurately identified. Machine learning algorithms can also be used to train a large amount of ultrasonic signal data, allowing the algorithm to automatically identify the characteristic patterns of burst pulse signals and improve the accuracy and reliability of detection.

[0133] In addition to the above parameters, fractal dimension, approximate entropy, sample entropy and Lyapunov index can also be calculated. The product of fractal dimension and approximate entropy can be calculated, which can reflect the complexity and comprehensive degree of irregularity of the signal. Sample entropy and Lyapunov index can be weighted and summed. The weight can be determined based on experiments or experience. If it is found that Lyapunov index is more important for partial discharge detection, it can be given a higher weight. These newly generated indicators are combined into nonlinear feature comprehensive indicators. The fractal dimension is calculated to describe the complexity and self-similarity of the signal. The larger the fractal dimension, the more complex the signal. Entropy features such as approximate entropy and sample entropy are used to measure the irregularity of the signal. The larger the entropy value, the higher the uncertainty of the signal. The chaotic characteristics of the signal are analyzed by Lyapunov index to determine whether the signal has chaotic behavior. These nonlinear characteristics can reveal the deep characteristics of partial discharge signals and help identify complex partial discharge faults.

[0134] The comprehensive index obtained by fusing the above-mentioned feature domains is used as the input feature of the decision tree. The decision tree constructs decision rules by recursively dividing these features. For example, at the node of the decision tree, a judgment is made based on the value of a certain feature index. If the index is greater than a certain threshold, the data is divided into one branch, otherwise it is divided into another branch. This division is continued until each leaf node corresponds to a specific partial discharge detection result category, such as normal, corona discharge, surface discharge, etc.

[0135] In addition to using the decision tree model, support vector machines can also be used. Taking support vector machines (SVM) as an example, the comprehensive indicators obtained by fusing various feature domains are used as the input vector of the SVM. The input vector is mapped to a high-dimensional space through a kernel function, and an optimal classification hyperplane is found in the high-dimensional space to distinguish different types of partial discharge signals. For neural networks, such as multi-layer perceptrons (MLP), these comprehensive indicators are used as neuron inputs in the input layer, and the partial discharge detection results are finally obtained in the output layer through nonlinear transformation and weight adjustment of the hidden layer.

[0136] The detection method further comprises:

[0137] Obtain corresponding vibration parameters by means of vibration sensors arranged at corresponding detection points;

[0138] The vibration parameters and sound detection information are input into a signal enhancement model for processing to obtain corresponding sound enhancement information; and the sound enhancement information is used as extrusion for subsequent pulse recognition; the signal enhancement model is a signal enhancement model constructed based on ultrasonic signals and vibration signals.

[0139] Vibration parameters are obtained by setting vibration sensors at corresponding detection points. Partial discharge may cause vibration of surrounding equipment, and vibration parameters can reflect the impact of partial discharge at the mechanical level. Combined with sound detection information, partial discharge can be monitored from different physical quantity angles, providing more comprehensive information, which helps to more accurately identify and analyze partial discharge phenomena. The vibration parameters and sound detection information are input into the signal enhancement model constructed based on ultrasonic signals and vibration signals for processing to obtain sound enhancement information. The signal enhancement model can use the correlation between vibration parameters and sound detection information to optimize and enhance the sound signal, remove noise interference, and improve the quality and clarity of the sound signal. Using the enhanced sound information as the basis for subsequent pulse recognition can improve the accuracy of pulse recognition, thereby further improving the performance of the entire partial discharge detection system.

[0140] In the specific implementation, PRPD analysis may be difficult to accurately judge some complex partial discharge types or situations where multiple partial discharge types occur simultaneously. When corona discharge and internal discharge exist simultaneously inside the equipment, the PRPD characteristics of the two discharges will overlap with each other, causing the spectrum to become complicated and difficult to clearly distinguish different types of discharges. The characteristics of weak partial discharge signals on the PRPD spectrum may not be obvious and may be easily masked by noise or other interference signals, thus affecting the judgment of the partial discharge type. In the actual operating environment of power equipment, there are a large number of electromagnetic interferences, which may produce false discharge points or interference stripes on the PRPD spectrum, interfering with the identification of the true partial discharge type. The above-mentioned complex scenarios can be identified through the above-mentioned multi-dimensional detection method.

[0141] In the embodiment of the present invention, partial discharge usually generates acoustic signals (such as ultrasound) in a specific frequency band, but different types of partial discharge (such as corona discharge, surface discharge, etc.) may show different characteristics in different frequency bands. By analyzing acoustic signals in multiple frequency bands, the type and location of partial discharge can be identified more accurately.

[0142] Implementation: Use a wideband microphone array to cover multiple frequency bands from low frequency to ultrasonic waves. Perform spectrum analysis on the collected acoustic signals to extract the characteristics of different frequency bands (such as peak frequency, spectrum energy distribution, etc.). Combined with machine learning algorithms, train a multi-band acoustic signal classification model to identify the type and location of partial discharge. Through multi-band analysis, the interference of environmental noise can be reduced and the accuracy of partial discharge detection can be improved.

[0143] The power line detection method based on sound signals in the embodiment of the present invention uses a pre-built partial discharge identification model to identify the phase-amplitude distribution diagram, and can accurately determine the type of partial discharge, such as corona discharge, surface discharge, internal discharge, etc., based on a large amount of historical data and machine learning algorithms. Different partial discharge types correspond to different fault causes and severity, and accurate identification helps to take targeted maintenance measures.

[0144] Embodiment 2

[0145] See also Figure 4 , Figure 4 FIG. 1 is a schematic diagram of the structure of a power line detection system based on sound signals disclosed in an embodiment of the present invention. Figure 4 As shown, the power line detection system based on sound signals may include:

[0146] The sound collection module 21 is used to obtain the sound detection information in the current detection scene through the sound collection array, and filter the sound detection information according to the filter parameters to obtain the processed sound detection information; wherein the filter parameters are determined according to the frequency characteristics of the partial discharge signal;

[0147] Phase diagram generation module 22: used for dividing a power frequency cycle into multiple phase intervals, identifying and processing the processed sound detection information through a partial discharge pulse recognition algorithm to determine the partial discharge pulse in the sound detection information, determining the amplitude parameter of each phase interval according to the pulse peak characteristic of the partial discharge pulse, and generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter;

[0148] Identification module 23: used for inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification, so as to obtain a corresponding partial discharge type identification result;

[0149] Output module 24: used to output the corresponding partial discharge type identification result and phase-amplitude distribution diagram.

[0150] The power line detection method based on sound signals in the embodiment of the present invention uses a pre-built partial discharge identification model to identify the phase-amplitude distribution diagram, and can accurately determine the type of partial discharge, such as corona discharge, surface discharge, internal discharge, etc., based on a large amount of historical data and machine learning algorithms. Different partial discharge types correspond to different fault causes and severity, and accurate identification helps to take targeted maintenance measures.

[0151] Embodiment 3

[0152] See also Figure 5 , Figure 5Schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. The electronic device may be a computer, a server, etc. Of course, in certain circumstances, it may also be a smart device such as a mobile phone, a tablet computer, a monitoring terminal, and an image acquisition device with processing functions. Figure 5 As shown, the electronic device may include:

[0153] A memory 510 storing executable program codes;

[0154] a processor 520 coupled to the memory 510;

[0155] The processor 520 calls the executable program code stored in the memory 510 to execute part or all of the steps in the power line detection method based on sound signals in the first embodiment.

[0156] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute part or all of the steps in the power line detection method based on sound signals in the first embodiment.

[0157] The embodiment of the present invention further discloses a computer program product, wherein when the computer program product is run on a computer, the computer is enabled to execute part or all of the steps in the power line detection method based on sound signals in the first embodiment.

[0158] An embodiment of the present invention further discloses an application publishing platform, wherein the application publishing platform is used to publish a computer program product, wherein when the computer program product runs on a computer, the computer executes part or all of the steps in the power line detection method based on sound signals in embodiment one.

[0159] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the processes does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0160] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed over multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0161] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0162] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a memory and includes several requests to enable a computer device (which can be a personal computer, server or network device, etc., specifically a processor in a computer device) to perform some or all steps of the method described in each embodiment of the present invention.

[0163] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0164] A person of ordinary skill in the art may understand that part or all of the steps in the various methods of the embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, the storage medium including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically-erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0165] The above is a detailed introduction to the power line detection method, system, electronic device and storage medium based on sound signals disclosed in the embodiments of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A power line detection method based on sound signals, characterized in that: include: Acquiring sound detection information in a current detection scenario through a sound collection array, and filtering the sound detection information according to filter parameters to obtain processed sound detection information; wherein the filter parameters are determined according to the frequency characteristics of the partial discharge signal; Dividing a power frequency cycle into a plurality of phase intervals, identifying and processing the processed sound detection information by a partial discharge pulse identification algorithm to determine the partial discharge pulse in the sound detection information, determining the amplitude parameter of each phase interval according to the pulse peak characteristics of the partial discharge pulse, and generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter; Inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification to obtain a corresponding partial discharge type identification result; inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification to obtain a corresponding partial discharge type identification result, including: Performing region division on the phase-amplitude distribution map according to a set image division logic to divide the phase-amplitude distribution map into a first image detection region, a second image detection region, a third image detection region, and a fourth image detection region; Performing first feature extraction on the first image detection area and the third image detection area to obtain first feature parameters associated with the first image detection area and the third image detection area, and inputting the first feature parameters into a first partial discharge recognition model for recognition to determine a corresponding first partial discharge detection result, wherein the first partial discharge recognition model is constructed based on a support vector machine or a random forest model; performing second feature extraction on the second image detection area and the fourth image detection area to obtain second feature parameters associated with the second image detection area and the fourth image detection area, and inputting the second feature parameters into a second partial discharge recognition model for recognition to determine a corresponding second partial discharge detection result, wherein the second partial discharge recognition model is constructed based on a convolutional neural network; Determine a corresponding partial discharge type identification result according to the first partial discharge detection result and the second partial discharge detection result; Output the corresponding partial discharge type identification results and phase-amplitude distribution diagram.

2. The power line detection method based on sound signals according to claim 1, characterized in that: The sound collection array includes 136 sound spiral arrays for acquiring sound signals; After filtering the sound detection information according to the filter parameters to obtain processed sound detection information, the method further includes: Analyze and process the processed sound detection information to determine the time information when the sound source arrives at each sound sensor in the sound collection array; Calculate the time difference of the sound signal arriving at each sound sensor in the sound collection array; The location information of the sound source in the current detection environment is determined according to the calculated time difference and the preset sound source calculation formula; and the corresponding direction information and distance information are determined according to the location information; the sound source calculation formula is: ,in, is the direction angle of the sound source, is the distance between the two microphones, is the speed of sound, It is the time difference between the sound source signal arriving at the two microphones; Performing data correction on the amplitude parameter according to the direction information and the distance information to obtain a corrected partial discharge pulse amplitude; The generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter comprises: A corresponding phase-amplitude distribution diagram is generated according to the phase interval and the corrected partial discharge pulse amplitude.

3. The power line detection method based on sound signals according to claim 2, characterized in that: The step of determining the amplitude parameters of each phase interval according to the pulse peak characteristics of the partial discharge pulse comprises: Data matching is performed on the local discharge pulse according to a set pulse detection threshold. When it is detected that the amplitude of the local discharge pulse exceeds the set pulse detection threshold, sequence detection is performed along the signal time from the pulse point exceeding the threshold until a point is found where the pulse amplitude no longer increases but begins to decrease. The currently detected amplitude is used as the pulse peak feature of the corresponding local discharge pulse, and the amplitude parameters of each phase interval are determined according to the pulse peak feature.

4. The power line detection method based on sound signals according to claim 1, characterized in that: After the processed sound detection information is identified by the partial discharge pulse identification algorithm to determine the partial discharge pulse in the sound detection information, the method further includes: Extracting a pulse sequence from the pre-processed partial discharge signal based on a set detection threshold and a set pulse width, and identifying the occurrence time and amplitude of each partial discharge pulse; According to the phase information recorded during acquisition, each pulse is associated with the corresponding phase, and a corresponding phase-time distribution diagram is generated according to the phase information and the time information, and a corresponding amplitude-time distribution diagram is generated according to the amplitude information and the time information; The phase-time distribution diagram and the amplitude-time distribution diagram are input into a trained convolutional neural network for recognition to obtain corresponding recognition results, and whether there is partial discharge is determined based on the recognition results.

5. The power line detection method based on sound signals according to claim 1, characterized in that: The determining of the corresponding partial discharge type identification result according to the first partial discharge detection result and the second partial discharge detection result also includes: When the probability of the partial discharge type corresponding to the first partial discharge detection result exceeds a first set value, the probability of the partial discharge type corresponding to the second partial discharge detection result exceeds a second set value, and the partial discharge type corresponding to the first partial discharge detection result is different from the partial discharge type corresponding to the second partial discharge detection result, executing the next step; Performing time domain analysis on the pre-processed sound detection information to calculate the corresponding waveform features, pulse features and energy features; wherein the waveform features include amplitude features, peak features, mean features and variance features; the pulse features include pulse width, rise time and fall time; and the energy features include signal energy distribution and change features; Determine a first time domain characterization parameter of the corresponding waveform according to a ratio of an amplitude feature to a peak feature in the waveform feature, determine a second time domain characterization parameter of the corresponding waveform according to a ratio of a mean feature to a variance feature in the waveform feature, and determine a third time domain characterization parameter of the corresponding waveform according to a ratio of a pulse width to a sum of a rise time and a fall time in the waveform feature; Use Fourier transform to perform spectrum analysis on the pre-processed sound detection information to obtain spectrum information, determine the frequency components of the sound detection information, calculate the energy distribution characteristics of different frequency bands, and determine the main frequency characteristics in the frequency information; The sum of the products of the main frequency characteristics and the energy of each frequency band is calculated to obtain the first frequency domain characterization parameter, and the ratio of the high frequency band energy to the low frequency band energy is calculated to determine the second frequency domain characterization parameter; the first time domain characterization parameter, the second time domain characterization parameter, the third time domain characterization parameter, the first frequency domain characterization parameter and the second frequency domain characterization parameter are standardized, and the standardized parameters are input into the waveform recognition model to obtain the corresponding partial discharge result.

6. The power line detection method based on sound signals according to claim 5, characterized in that: The detection method further comprises: Short-time Fourier transform is used to calculate the frequency variance characteristics of sound detection information in different time windows; The wavelet transform algorithm is used to extract the wavelet coefficient energy of the sound detection information at different scales, and the sum of the wavelet coefficient energy at different scales is calculated as the wavelet energy feature; Calculate the energy center of gravity coordinates of the sound detection information on the entire time-frequency plane according to the Wigner probability distribution, and use the time center of gravity and the frequency center of gravity as the corresponding center of gravity features; obtain the corresponding time-frequency domain comprehensive features according to the frequency variance features, wavelet energy features and center of gravity features; Calculating the sound detection information to determine corresponding skewness characteristics, kurtosis characteristics, autocorrelation coefficients, and cross-correlation coefficients; The time-frequency domain comprehensive features, skewness features, kurtosis features, autocorrelation coefficients and cross-correlation coefficients are input as input features into a waveform recognition model to obtain corresponding partial discharge results; When it is determined that the partial discharge result is that there is partial discharge behavior, corresponding partial discharge level information is determined according to each acquired waveform feature, and the partial discharge level information is output; The detection method further comprises: Obtain corresponding vibration parameters by means of vibration sensors arranged at corresponding detection points; The vibration parameters and sound detection information are input into a signal enhancement model for processing to obtain corresponding sound enhancement information; and the sound enhancement information is used as extrusion for subsequent pulse recognition; the signal enhancement model is a signal enhancement model constructed based on ultrasonic signals and vibration signals.

7. A power line detection system based on sound signals, characterized in that: include: A sound collection module is used to obtain the sound detection information in the current detection scene through the sound collection array, and filter the sound detection information according to the filter parameters to obtain the processed sound detection information; wherein the filter parameters are determined according to the frequency characteristics of the partial discharge signal; Phase diagram generation module: used for dividing a power frequency cycle into multiple phase intervals, identifying and processing the processed sound detection information through a partial discharge pulse recognition algorithm to determine the partial discharge pulse in the sound detection information, determining the amplitude parameter of each phase interval according to the pulse peak characteristic of the partial discharge pulse, and generating a corresponding phase-amplitude distribution diagram according to the phase interval and the amplitude parameter; Identification module: used for inputting the phase-amplitude distribution diagram into a pre-built partial discharge identification model for identification, so as to obtain a corresponding partial discharge type identification result; the phase-amplitude distribution diagram is input into a pre-built partial discharge identification model for identification, so as to obtain a corresponding partial discharge type identification result, including: Performing region division on the phase-amplitude distribution map according to a set image division logic to divide the phase-amplitude distribution map into a first image detection region, a second image detection region, a third image detection region, and a fourth image detection region; Performing first feature extraction on the first image detection area and the third image detection area to obtain first feature parameters associated with the first image detection area and the third image detection area, and inputting the first feature parameters into a first partial discharge recognition model for recognition to determine a corresponding first partial discharge detection result, wherein the first partial discharge recognition model is constructed based on a support vector machine or a random forest model; performing second feature extraction on the second image detection area and the fourth image detection area to obtain second feature parameters associated with the second image detection area and the fourth image detection area, and inputting the second feature parameters into a second partial discharge recognition model for recognition to determine a corresponding second partial discharge detection result, wherein the second partial discharge recognition model is constructed based on a convolutional neural network; Determine a corresponding partial discharge type identification result according to the first partial discharge detection result and the second partial discharge detection result; Output module: used to output the corresponding partial discharge type identification results and phase-amplitude distribution diagram.

8. An electronic device, characterized in that: include: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the power line detection method based on sound signals according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the power line detection method based on sound signals according to any one of claims 1 to 7.

Images