Power cable fault discharge sound recognition method and system based on multi-feature fusion

Abstract

The present invention relates to the technical field of power cable fault detection and location, and proposes a method and system for identifying power cable fault discharge sounds based on multi-feature fusion, comprising the following steps: acquiring the sound signal of the power cable, and performing adaptive framing and windowing processing based on the periodic peak value; calculating the envelope variance and local maximum attenuation coefficient of the sound signal; extracting the high-frequency modal energy ratio of the sound signal and calculating the normalized spectral entropy; inputting the local maximum attenuation coefficient and envelope variance data into a first classifier; inputting the high-frequency modal energy ratio and normalized spectral entropy into a trained second classifier; and fusing the two classification results. Based on the adaptive windowing method of autocorrelation analysis and the discharge sound recognition strategy of multi-dimensional feature fusion, a dual classifier structure is constructed and time-frequency information is integrated, effectively improving the accuracy and robustness of cable fault discharge sound recognition.

Description

Technical Field

[0001] The present invention relates to the technical field related to power cable fault detection and positioning, and in particular to a method and system for recognizing power cable fault discharge sound based on multi-feature fusion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Cable fault detection is a critical step in power system maintenance. Discharge sound signals, as an important basis for fault location, offer the advantages of directness and speed in actual testing. Power cable fault testing technology typically involves two key steps: cable fault ranging (coarse location) and cable fault location (precise location). During fault location, the industry often employs acoustic testing or acoustic-magnetic synchronization methods. The acoustic testing method uses a ground-based acoustic probe to collect sound signals emitted from the discharge point. After filtering and amplification by the host computer, the signal is transmitted to a listening headset, allowing the tester to determine the fault location by hearing. The acoustic-magnetic synchronization method simultaneously collects the sound and electromagnetic signals emitted from the discharge point, analyzing the time difference between the two signals to assist in determining the fault location. Regardless of the method used, both rely on the accurate identification of the discharge sound signal.

[0004] However, in actual test sites, cable burial environments are complex, and discharge signals are often drowned out by background noise (such as vehicle sounds, conversations, and wind), making discharge sound identification difficult. Furthermore, due to differences in laying methods, medium types, and burial depths, even for the same type of fault, the discharge sound signals exhibit significant diversity and uncertainty in waveform structure, frequency content, and time attenuation. Currently, mainstream discharge sound recognition still relies on subjective listening by fault testers. Recognition accuracy is heavily dependent on experience, and the probability of misjudgment is high in the presence of interfering noise. Furthermore, existing automatic recognition methods generally rely on fixed-frame-length windows for signal processing, making it difficult to account for the rhythm, period, and duration of different discharge signals. This can easily lead to feature truncation or drift, making it impossible to effectively adapt to the changing rhythm of different discharge signals, thus affecting recognition accuracy. Furthermore, most methods use only a single feature (such as energy or spectrum) for classification, making it difficult to fully characterize the multidimensional characteristics of the signal. The classification structure is often a single model, with limited ability to recognize complex signal patterns. Summary of the Invention

[0005] In order to solve the above problems, the present invention proposes a method and system for identifying power cable fault discharge sounds based on multi-feature fusion. The adaptive windowing method based on autocorrelation analysis and the discharge sound recognition strategy based on multi-dimensional feature fusion are used to construct a dual classifier structure and fuse time-frequency information, which effectively improves the accuracy and robustness of cable fault discharge sound recognition.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] One or more embodiments provide a method for identifying power cable fault discharge sound based on multi-feature fusion, comprising the following steps:

[0008] Acquire the sound signal of the power cable, identify the periodic peak of the autocorrelation coefficient in the sound signal, and perform adaptive framing and windowing processing based on the periodic peak;

[0009] Calculate the envelope variance of the sound signal based on the acquired sound signal, and calculate the local maximum attenuation coefficient of the sound signal for the time window signal after frame windowing processing;

[0010] The high-frequency modal energy ratio of the sound signal is extracted based on the acquired sound signal, and the normalized spectral entropy is calculated for the time window signal after frame and window processing;

[0011] Inputting the local maximum attenuation coefficient and envelope variance data into the trained first classifier to obtain a first classification result;

[0012] Input the high-frequency modal energy proportion and normalized spectral entropy into the trained second classifier to obtain the second classification result;

[0013] The first classification result and the second classification result are integrated to obtain the classification result of the sound signal of the power cable.

[0014] One or more embodiments provide a power cable fault discharge sound recognition system based on multi-feature fusion, including: a sound collection device and a processor;

[0015] A sound collection device for collecting discharge sound signals of power cables;

[0016] The processor is configured to execute the steps of the above-mentioned method for identifying power cable fault discharge sound based on multi-feature fusion.

[0017] One or more embodiments provide a power cable fault discharge sound recognition system based on multi-feature fusion, including:

[0018] An adaptive windowing module is configured to obtain a sound signal of the power cable, identify a periodic peak value of an autocorrelation coefficient in the sound signal, and perform adaptive frame segmentation and windowing processing according to the periodic peak value;

[0019] A time domain feature extraction module is configured to calculate the envelope variance of the sound signal based on the acquired sound signal, and calculate the local maximum attenuation coefficient of the sound signal for the time window signal after the frame and window processing;

[0020] A frequency domain feature extraction module is configured to extract the high-frequency modal energy ratio of the sound signal based on the acquired sound signal, and calculate the normalized spectral entropy for the time window signal after frame and window processing;

[0021] A first classification module is configured to input the local maximum attenuation coefficient and the envelope variance data into the trained first classifier to obtain a first classification result;

[0022] a second classification module configured to input the high-frequency modal energy proportion and the normalized spectral entropy into the trained second classifier to obtain a second classification result;

[0023] The fusion module is configured to fuse the first classification result and the second classification result to obtain a classification result of the sound signal of the power cable.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The recognition method of the present invention first adopts an adaptive windowing strategy based on the periodic peak of the autocorrelation coefficient, which can dynamically adjust the analysis windowing according to the actual rhythm and attenuation law of the sound signal, solving the problems of feature truncation, energy leakage or time-frequency drift that are prone to occur in traditional fixed-length windowing when facing signal diversity. This strategy enhances the matching between feature extraction and signal structure, improves the expression stability of time domain features, and has good data-driven adaptability and field application robustness. Secondly, a multi-dimensional complementary feature system consisting of local maximum attenuation coefficient, envelope variance, high-frequency modal energy ratio and normalized spectral entropy is constructed, which can comprehensively characterize the essential characteristics of sound signals from multiple angles such as local time domain dynamics, global fluctuation characteristics, frequency domain high-frequency components and energy complexity. This feature system takes into account both physical interpretability and discriminant validity. Two classifiers based on time domain features and frequency domain features are constructed, which effectively improve the robustness and classification accuracy of the discharge sound recognition model; by utilizing the synergistic advantages of feature expression and model structure, the discriminant ability of complex signal patterns is significantly enhanced. The present invention forms complementary innovations in three aspects: windowing mechanism, feature design and classification decision, effectively improving the accuracy and reliability of power cable partial discharge sound signal recognition.

[0026] The advantages of the present invention and its additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention but do not constitute a limitation of the present invention.

[0028] Figure 1 Flowchart of a method for identifying power cable fault discharge sound according to embodiment 1 of the present invention;

[0029] Figure 2 1 is the waveform of the power cable fault discharge sound signal according to the example of embodiment 1 of the present invention;

[0030] Figure 3 1 is the waveform of the non-fault discharge sound signal of the example of embodiment 1 of the present invention;

[0031] Figure 4 1 is the waveform and envelope waveform of the power cable fault discharge sound signal according to the example of embodiment 1 of the present invention;

[0032] Figure 5 1 is a non-fault discharge sound signal waveform and envelope waveform of an example of embodiment 1 of the present invention;

[0033] Figure 6 This is the maximum value detection result of the power cable fault discharge sound signal waveform in the example of embodiment 1 of the present invention;

[0034] Figure 7 This is the maximum value detection result of the non-fault discharge sound signal waveform in the example of the first embodiment of the present invention.

[0035] Figure 8 is the modal decomposition result of the power cable fault discharge sound signal in the simulation experiment of Example 1 of the present invention;

[0036] Figure 9 This is the modal decomposition result of the non-fault discharge sound signal in the simulation experiment of Example 1 of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0038] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0039] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments of the present invention and the features in the embodiments can be combined with each other. The embodiments will be described in detail below with reference to the accompanying drawings.

[0040] Example 1

[0041] In the technical solutions disclosed in one or more embodiments, Figures 1 to 9 As shown, a method for recognizing power cable fault discharge sound based on multi-feature fusion includes the following steps:

[0042] Step 1: Acquire the sound signal of the power cable, identify the periodic peak of the autocorrelation coefficient in the sound signal, and perform adaptive framing and windowing processing according to the periodic peak;

[0043] Step 2: Calculate the envelope variance of the sound signal based on the acquired sound signal, and calculate the local maximum attenuation coefficient of the sound signal for the time window signal after the frame windowing process;

[0044] Step 3: extract the high-frequency modal energy ratio of the sound signal based on the acquired sound signal, and calculate the normalized spectral entropy for the time window signal after frame windowing processing;

[0045] Step 4: Input the local maximum attenuation coefficient and envelope variance data into the trained first classifier to obtain a first classification result;

[0046] Step 5: Input the high-frequency modal energy proportion and the normalized spectral entropy into the trained second classifier to obtain the second classification result;

[0047] Step 6: Fusing the first classification result and the second classification result to obtain a classification result of the sound signal of the power cable; the classification result includes a fault signal and a non-fault signal;

[0048] In this embodiment, first, an adaptive windowing strategy based on the periodic peak of the autocorrelation coefficient is adopted. This strategy can dynamically adjust the analysis window according to the actual rhythm and attenuation pattern of the sound signal, solving the problems of feature truncation, energy leakage, or time-frequency drift that are prone to occur in traditional fixed-length windowing when facing signal diversity. This strategy enhances the matching between feature extraction and signal structure, improves the expression stability of time-domain features, and has good data-driven adaptability and robustness in field applications. Secondly, a multidimensional complementary feature system composed of local maximum attenuation coefficient, envelope variance, high-frequency modal energy proportion, and normalized spectral entropy is constructed. It can comprehensively characterize the essential characteristics of the sound signal from multiple perspectives, including local time-domain dynamics, global fluctuation characteristics, frequency-domain high-frequency components, and energy complexity. This feature system takes into account both physical interpretability and discriminant validity. Two classifiers based on time-domain features and frequency-domain features are constructed, effectively improving the robustness and classification accuracy of the discharge sound recognition model. By leveraging the synergistic advantages of feature expression and model structure, the discriminant ability of complex signal patterns is significantly enhanced. The method of this embodiment forms complementary innovations in three aspects: windowing mechanism, feature design, and classification decision-making. It effectively improves the accuracy and reliability of power cable partial discharge sound signal recognition, is suitable for high-confidence detection needs under various field conditions, and has good engineering application prospects.

[0049] In step 1, the method of identifying the periodic peak of the autocorrelation coefficient in the sound signal and performing adaptive framing and windowing processing according to the periodic peak includes the following steps:

[0050] Step 11: Perform Hilbert transform on the acquired sound signal to form a complex analytical signal with the sound signal and extract the envelope of the sound signal. ;

[0051] Specifically, assuming the sound signal is , the total number of sampling points of the signal is , then we get the complex analytic signal:

[0052] ;

[0053] in, represents the sampling time of the sound signal, It is the signal component obtained by performing Hilbert transform on the original sound signal.

[0054] Calculate the envelope of a sound signal as follows:

[0055] (1);

[0056] Step 12: Calculate the envelope of the sound signal The mean :

[0057] (2);

[0058] in, It's a signal The total number of sampling points.

[0059] Step 13: Based on the obtained mean, the envelope of the sound signal To perform decentralized operations:

[0060] (3);

[0061] in, Represents the envelope of the decentralized sound signal.

[0062] The example waveforms of the discharge sound signal and the non-discharge sound signal are as follows: Figure 2 and Figure 3 As shown, Figure 4 and Figure 5 That is the envelope of the corresponding extracted sound signal;

[0063] Step 13: Envelope of decentralized sound signal , calculate the autocorrelation coefficient of each signal point, identify the periodic peak of the autocorrelation coefficient, use the interval length corresponding to the starting point of the corresponding signal to the periodic peak as the window size of the current signal point, frame and window the sound signal, and divide the sound signal into different time window signals;

[0064] Specifically, the autocorrelation coefficient , the calculation formula is as follows:

[0065] (4);

[0066] in, Indicates sampling point t Envelope signal Interval with it Envelope signal of sampling points Considering that the autocorrelation coefficient needs to traverse all the m sampling points, the interval The value range is ;

[0067] By calculation of formula (4), when the interval The value varies from 1 to When , the envelope of the sound signal can be obtained The autocorrelation coefficient sequence of .

[0068] Autocorrelation coefficient series With fluctuating changes, the time point when the autocorrelation coefficient value first appears at a peak in the sequence is , the start time of the signal is set to 0, then the window size T That is:

[0069] (5);

[0070] in, It represents the window size, which is called the fluctuation period of the autocorrelation coefficient peak.

[0071] In this embodiment, the interval corresponding to the first peak position of the autocorrelation coefficient is selected by periodic peak selection of the autocorrelation coefficient. As the window size, the frame window size is adaptively determined. Based on the frame window setting, the sound signal is windowed for subsequent calculation of relevant feature indicators.

[0072] In this embodiment, the envelope signal of the sound signal is first extracted, and based on the waveform change law of the envelope signal, the frame window size of the sound signal is adaptively determined. By introducing an adaptive windowing mechanism based on the extraction of the peak period of the autocorrelation function, the automatic determination of the window length of the sound signal is achieved, and the optimal window scale can be autonomously matched according to the differences in duration and rhythm of different types of discharge sound signals. Compared with the traditional method of fixing the window length or relying on experience, this method significantly improves the consistency of feature extraction and the correspondence with the physical meaning, and avoids the problems of signal truncation, feature loss or time-frequency drift caused by window length mismatch. This technical strategy strengthens the adaptability and robustness of the sound analysis process, and improves the adaptability and application reliability of the algorithm under complex working conditions on site.

[0073] In step 2, for the time window signal after frame windowing processing, the local maximum attenuation coefficient of the sound signal is calculated. Specifically, for the time window signal, the attenuation rate is calculated by the amplitude ratio and time interval of the maxima of adjacent windows, and the average of all attenuation rates is used as the local maximum attenuation coefficient;

[0074] Specifically, the calculation formula for the local maximum attenuation coefficient is:

[0075] (6);

[0076] in, Represents the local maximum attenuation coefficient of the sound signal, Indicates sound signal The total number of divided frame windows, Indicates the The local maximum of the frame window, Indicates the The sampling time corresponding to the local maximum value in the frame window. For example, Figure 6 and Figure 7 The local maxima shown are marked with red circles;

[0077] Step 2: Calculate the envelope variance of the sound signal based on the acquired sound signal. Extract the envelope variance of the signal in the time domain. The mean , calculate the envelope variance, the formula is:

[0078] (7);

[0079] Among them, the larger the envelope variance, the more violent the fluctuation of the signal amplitude; the smaller the envelope variance, the more stable the signal amplitude.

[0080] In step 3, for the acquired sound signal, the high-frequency modal energy ratio of the signal is extracted, including the following steps:

[0081] Step 31: Transmit the sound signal CEEMDAN is used to perform empirical mode decomposition to obtain the IMF component signals of the sound signal:

[0082] (8);

[0083] in, is the total number of components decomposed, represents the qth IMF component obtained by decomposition, Represents the residual component obtained by decomposition.

[0084] CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) is an improved empirical mode decomposition (EMD) algorithm used to decompose nonlinear and non-stationary signals (such as speech and vibration) into a set of physically meaningful modal function components (IMFs).

[0085] Step 32: Calculate each IMF component separately With sound signal The similarity of the two modal components is calculated, and the IMF component corresponding to the maximum similarity value is used as the boundary component, and the boundary component and each IMF before the boundary component are used as high-frequency modal components to calculate the high-frequency modal energy ratio. ;

[0086] Figure 8The modal decomposition results of the power cable fault discharge sound signal are shown. The original discharge sound signal is decomposed into 10 IMF components. Figure 8 The first five IMF components are shown, among which IMF3 has the highest cosine similarity with the original discharge sound signal, which is 0.96. Therefore, IMF1, IMF2 and IMF3 can be selected as high-frequency modal components and can be used to calculate the proportion of high-frequency modal energy. Figure 9 The first seven IMF components of the modal decomposition result of a non-discharge sound signal are shown. The sound signal is decomposed into 10 IMF components, among which IMF3 has the highest cosine similarity with the original sound signal, which is 0.72. Therefore, IMF1, IMF2, and IMF3 can be selected as high-frequency modal components for calculating the high-frequency modal energy ratio.

[0087] Optionally, calculate each IMF component With sound signal The similarity is calculated by cosine similarity;

[0088] Traditionally, high-frequency energy fraction is calculated by performing a Fourier transform on the sound signal to obtain a spectrum. After further observation of the spectrum, a frequency value is manually determined as a high-frequency threshold. Based on this high-frequency threshold, the energy fraction of high-frequency components above the threshold is calculated. Obviously, this traditional method faces the challenge of difficult to determine the high-frequency threshold and the selected high-frequency threshold is not applicable to all signals. However, the CEEMDAN decomposition strategy and adaptive boundary component determination method adopted in this step can effectively overcome the shortcomings of traditional methods of calculating high-frequency energy fraction through spectral analysis. In this step, CEEMDAN decomposition achieves a complete decomposition of the original sound signal, obtaining the component signals of the original sound signal at different frequency levels, which provides a fully refined description of the original sound signal. Adaptive boundary component determination enables quantitative calculation of the similarity between each component signal and the original sound signal. Using the IMF component corresponding to the highest similarity as the boundary, high-frequency modal components are adaptively screened out, and the high-frequency modal energy fraction is calculated based on these modal components. This method has strong flexibility and adaptability.

[0089] After screening, when the boundary component is , then the high-frequency modal component is:

[0090] ;

[0091] Calculate the proportion of high-frequency modal energy The formula is:

[0092] (9);

[0093] In this embodiment, the high-frequency characteristic strength of the signal can be effectively quantified by the high-frequency modal energy ratio.

[0094] In step 3, the method for calculating the normalized spectral entropy for the time window signal after the frame windowing process includes the following steps:

[0095] Step 301: Perform fast Fourier transform on each frame time window signal to obtain the frame time window signal spectrum; assuming that The spectrum of the frame time window is .

[0096] Step 302: Calculate the frame time window signal spectrum The energy spectrum of is:

[0097] (10);

[0098] Step 303: Normalize the energy spectrum of each frame window signal into a probability distribution:

[0099] (11)

[0100] in, Indicates the Framed windows, Indicates the The probability corresponding to the energy spectrum of the frame window signal, Indicates sound signal The total number of divided frame windows.

[0101] Step 304: Calculate the information entropy values ​​of all frame windows:

[0102] (12);

[0103] Maximum entropy corresponding to the total number of framing windows , then the normalized spectral entropy is:

[0104] (13);

[0105] The normalized spectral entropy calculated in this step is for each adaptive frame time window, rather than directly for all sampling points in the sound signal sequence. This calculation method can focus on both the local complexity of the signal spectrum and the overall spectral changes of the signal. It can more effectively quantify the concentration / randomness of the spectral energy distribution and effectively distinguish between ordered and random components in the signal. It is an effective way to describe the frequency domain characteristics of sound signals. The normalized spectral entropy is efficient and robust to amplitude changes, making it suitable for characterizing state changes in signals such as speech, mechanical vibration, and biological signals.

[0106] Optionally, the first classifier may be a support vector machine, a random forest, or a multilayer perceptron; the second classifier may be a support vector machine, a random forest, or a multilayer perceptron;

[0107] In step 6, the first classification result is obtained and the second classification results , through the integration of logic and operation, the final classification result is obtained :

[0108] (14);

[0109] A further technical solution also includes a process of training the first classifier and the second classifier, comprising the following steps:

[0110] Step S1, collecting discharge sound signals and non-discharge sound signals near the cable fault point, and constructing discharge sound and non-discharge sound data sets respectively;

[0111] Step S2, dividing the discharge sound and non-discharge sound datasets into a training dataset and a test dataset according to a set ratio;

[0112] Step S3: for the sound signals of the training set and the test set, identify the periodic peaks of the autocorrelation coefficients in the sound signals, perform adaptive frame segmentation and windowing processing based on the periodic peaks; and extract the local maximum attenuation coefficient and envelope variance of the sound signals respectively;

[0113] Then extract the high-frequency modal energy ratio and normalized spectral entropy of the sound signal respectively; the processing method of this step is the same as the above steps 1 to 3, and will not be repeated here;

[0114] Step S4, transmitting the local maximum attenuation coefficients and envelope variances of the discharge sound and non-discharge sound in the training set to the first classifier for training, obtaining the optimal model parameters of the first classifier, and obtaining the trained first classifier;

[0115] Step S5, based on the high-frequency modal energy proportion and normalized spectral entropy features of the discharge sound and non-discharge sound in the training set, transmitting them to the second classifier for training, obtaining the optimal model parameters of the second classifier, and obtaining the trained second classifier;

[0116] Step S6: Using the test data set, the trained first classifier and the second classifier are tuned to obtain the final trained first classifier and the second classifier.

[0117] In step S1, optionally, a cable fault locator can be used to collect discharge sound signals and non-discharge sound signals near the cable fault point. For example, 300 groups of both can be collected to construct discharge sound and non-discharge sound data sets respectively; the cable fault discharge sound signal is marked as 1, and the non-discharge sound signal is marked as 0.

[0118] In step S2, the discharge sound and non-discharge sound data sets can be divided into a training data set and a test data set in a ratio of 7:3. In this embodiment, the training set includes 420 groups of sound data; the test data set includes 180 groups of sound data.

[0119] In step S3, the collected data is processed, and the features extracted from an example sound signal are shown in Table 1. It can be seen from the data in Table 1 that the local maximum attenuation coefficient of the discharge sound signal is 4.8922, while the local maximum attenuation coefficient of the non-discharge sound signal is 0.1576. The features of the two different signals differ by 30 times, and the envelope variance of the two different signals differs even more. Therefore, the time domain feature discrimination capability adopted in this embodiment is very significant. For the proportion of high-frequency modal energy, the characteristic value of the discharge sound signal is 5 times that of the non-discharge sound signal; for the normalized spectral entropy, the characteristic value of the non-discharge sound signal is 2.3 times that of the discharge sound signal. Obviously, the frequency domain feature discrimination defined and adopted in this embodiment is also more significant. In summary, the time domain features and frequency domain features defined and adopted in this embodiment have clear boundaries, there is no close feature value or boundary aliasing, and have good separability and reliability.

[0120] Table 1 Normalized feature extraction results of discharge sound and non-discharge sound;

[0121]

[0122] In step S6, the accuracy index for evaluating the model performance can be calculated for the entire test data set. The calculation formula is as follows:

[0123] (15);

[0124] Among them, TP represents the number of samples correctly judged as fault discharge sound signals, TN represents the number of samples correctly judged as non-discharge sound signals, FP represents the number of samples incorrectly judged as non-discharge sound signals as fault discharge sound signals, and FN represents the number of samples incorrectly judged as discharge sound signals as non-discharge sound signals.

[0125] In order to illustrate the effect of the method of this embodiment, a test experiment was carried out, which is described in detail as follows:

[0126] 1) In the test experiment, there are 90 groups of discharge sound and non-discharge sound signals in the test data set. The confusion matrix of the sound signal recognition of the test data set is shown in Table 2, which verifies the recognition ability of the obtained optimal training model. According to the data shown in Table 2, the accuracy of the test experiment can be calculated based on formula (15): , so the accuracy of sound signal recognition in this experiment is 95%.

[0127] Table 2 Confusion matrix of sound signal recognition of test dataset;

[0128]

[0129] 2) To verify the effectiveness of multi-feature fusion, a comparison test was conducted between multi-feature fusion and single-feature recognition in this embodiment. Table 3 shows the sound recognition performance achieved by training a support vector machine using a single feature description.

[0130] Table 3 Comparison of sound recognition results using single feature recognition and multi-feature fusion;

[0131]

[0132] 2) To verify the effect of dual classifier decision fusion, a comparative experiment was conducted;

[0133] Among them, the time domain decision is to use the first classifier alone to perform sound recognition and determine whether it is a fault signal; the frequency domain decision is to use the second classifier alone to perform sound recognition; the dual classifier decision fusion is to use two classifiers to recognize and fuse the first classification results with the second classification results. As can be seen from Table 4, the dual classifier decision greatly improves the recognition accuracy.

[0134] As shown in Table 3, the sound recognition results of the test dataset with and without decision fusion are shown. The recognition accuracy is the average of ten cross-experiments.

[0135] Table 4 Comparison of recognition effects of different decision strategies;

[0136]

[0137] In summary, the experimental results show that the power cable fault discharge sound recognition method proposed in the present invention can realize the comprehensive expression of the power cable discharge sound signal characteristics, and can highlight the key recognition features from the time domain and frequency domain perspectives respectively, overcoming the deficiency that a single feature cannot effectively describe diversified data, and improving the generalization of feature expression; through the decision fusion of the dual classification results, the accuracy and intelligence level of power cable fault discharge sound recognition are further improved, avoiding the subjective influence brought by relying on the experience of the testers.

[0138] Example 2

[0139] Based on Example 1, this embodiment provides a power cable fault discharge sound recognition system based on multi-feature fusion, including: a sound collection device and a processor;

[0140] A sound collection device for collecting discharge sound signals of power cables;

[0141] The processor is configured to execute the steps of the power cable fault discharge sound recognition method based on multi-feature fusion described in Example 1.

[0142] Specifically, the sound collection device may be a power cable fault locator, the model of which may be the T-506 intelligent power cable fault locator.

[0143] Example 3

[0144] Based on Example 1, this embodiment provides a power cable fault discharge sound recognition system based on multi-feature fusion, including:

[0145] An adaptive windowing module is configured to obtain a sound signal of the power cable, identify a periodic peak value of an autocorrelation coefficient in the sound signal, and perform adaptive frame segmentation and windowing processing according to the periodic peak value;

[0146] A time domain feature extraction module is configured to calculate the envelope variance of the sound signal based on the acquired sound signal, and calculate the local maximum attenuation coefficient of the sound signal for the time window signal after the frame and window processing;

[0147] A frequency domain feature extraction module is configured to extract the high-frequency modal energy ratio of the sound signal based on the acquired sound signal, and calculate the normalized spectral entropy for the time window signal after frame and window processing;

[0148] A first classification module is configured to input the local maximum attenuation coefficient and the envelope variance data into the trained first classifier to obtain a first classification result;

[0149] a second classification module configured to input the high-frequency modal energy proportion and the normalized spectral entropy into the trained second classifier to obtain a second classification result;

[0150] The fusion module is configured to fuse the first classification result and the second classification result to obtain a classification result of the sound signal of the power cable; the classification result includes a fault signal and a non-fault signal.

[0151] Furthermore, the adaptive windowing module includes:

[0152] An envelope extraction module is configured to perform a Hilbert transform on the acquired sound signal, form a complex analytical signal with the sound signal, and extract the envelope of the sound signal;

[0153] a mean value calculation module, configured to calculate the mean value of the envelope of the sound signal;

[0154] a decentralization module configured to perform a decentralization operation on the envelope of the sound signal based on the obtained mean value;

[0155] The window size determination module is configured to calculate the autocorrelation coefficient of each signal point for the envelope of the decentralized sound signal, identify the periodic peak of the autocorrelation coefficient, use the period corresponding to the periodic peak as the window size of the current signal point, frame and window the sound signal, and divide the sound signal into different time window signals.

[0156] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation processes are the same, which will not be repeated here.

[0157] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

[0158] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for identifying power cable fault discharge sound based on multi-feature fusion, characterized in that: The steps include: Acquire the sound signal of the power cable, identify the periodic peak of the autocorrelation coefficient in the sound signal, and perform adaptive framing and windowing processing based on the periodic peak; Calculate the envelope variance of the sound signal based on the acquired sound signal, and calculate the local maximum attenuation coefficient of the sound signal for the time window signal after frame windowing processing; The high-frequency modal energy ratio of the sound signal is extracted based on the acquired sound signal, and the normalized spectral entropy is calculated for the time window signal after frame and window processing; Inputting the local maximum attenuation coefficient and envelope variance data into the trained first classifier to obtain a first classification result; Input the high-frequency modal energy proportion and normalized spectral entropy into the trained second classifier to obtain the second classification result; The first classification result and the second classification result are integrated to obtain the classification result of the sound signal of the power cable.

2. The method for identifying power cable fault discharge sound based on multi-feature fusion according to claim 1, characterized in that: The method for identifying the periodic peak of the autocorrelation coefficient in the sound signal and performing adaptive framing and windowing processing according to the periodic peak comprises the following steps: Performing Hilbert transform on the acquired sound signal to form a complex analytical signal with the sound signal and extracting the envelope of the sound signal; Calculate the mean of the envelope of the sound signal; Based on the obtained mean, the envelope of the sound signal is decentralized; For the envelope of the decentralized sound signal, the autocorrelation coefficient of each signal point is calculated, the periodic peak of the autocorrelation coefficient is identified, and the interval length from the corresponding signal starting point to the periodic peak is used as the window size of the current signal point. The sound signal is framed and windowed to divide the sound signal into different time window signals.

3. The method for recognizing power cable fault discharge sound based on multi-feature fusion according to claim 1, characterized in that: For the time window signal after frame windowing processing, the local maximum attenuation coefficient of the sound signal is calculated. Specifically: for the time window signal, the attenuation rate is calculated by the amplitude ratio and time interval of the adjacent window maxima, and the average of all attenuation rates is used as the local maximum attenuation coefficient.

4. The method for identifying power cable fault discharge sound based on multi-feature fusion according to claim 1, characterized in that: Calculate the envelope variance based on the mean of the envelope of the sound signal.

5. The method for identifying power cable fault discharge sound based on multi-feature fusion according to claim 1, characterized in that: For the acquired sound signal, extract the high-frequency modal energy ratio of the signal, including the following steps: The sound signal is subjected to empirical mode decomposition using CEEMDAN to obtain the IMF component signals of the sound signal; The similarity between each IMF component and the sound signal is calculated respectively. The IMF component corresponding to the maximum similarity value is taken as the boundary component, and the IMFs before the boundary component are taken as high-frequency modal components. The proportion of high-frequency modal components is calculated.

6. The method for identifying power cable fault discharge sound based on multi-feature fusion according to claim 1, characterized in that: The method for calculating the normalized spectral entropy of a time window signal after frame windowing processing includes the following steps: Perform fast Fourier transform on each time window signal to obtain the spectrum of the windowed signal; Calculate the energy spectrum of the signal spectrum in each time window; Normalize the energy spectrum of the signal in each time window into a probability distribution; Calculate the information entropy value and maximum entropy value of the frame window; The normalized spectral entropy is calculated based on the information entropy and maximum entropy of all frame windows.

7. The method for recognizing power cable fault discharge sound based on multi-feature fusion according to claim 1, characterized in that: The first classification result and the second classification result are merged through a logical AND operation to obtain the final classification result.

8. A power cable fault discharge sound recognition system based on multi-feature fusion, characterized by: include: Sound collection device and processor; A sound collection device for collecting discharge sound signals of power cables; The processor is configured to execute the steps of the power cable fault discharge sound recognition method based on multi-feature fusion according to any one of claims 1 to 7.

9. The power cable fault discharge sound recognition system based on multi-feature fusion is characterized by: include: An adaptive windowing module is configured to obtain a sound signal of the power cable, identify a periodic peak value of an autocorrelation coefficient in the sound signal, and perform adaptive frame segmentation and windowing processing according to the periodic peak value; A time domain feature extraction module is configured to calculate the envelope variance of the sound signal based on the acquired sound signal, and calculate the local maximum attenuation coefficient of the sound signal for the time window signal after the frame and window processing; A frequency domain feature extraction module is configured to extract the high-frequency modal energy ratio of the sound signal based on the acquired sound signal, and calculate the normalized spectral entropy for the time window signal after frame and window processing; A first classification module is configured to input the local maximum attenuation coefficient and the envelope variance data into the trained first classifier to obtain a first classification result; a second classification module configured to input the high-frequency modal energy proportion and the normalized spectral entropy into the trained second classifier to obtain a second classification result; The fusion module is configured to fuse the first classification result and the second classification result to obtain a classification result of the sound signal of the power cable; the classification result includes a fault signal and a non-fault signal.

10. The power cable fault discharge sound recognition system based on multi-feature fusion according to claim 9, characterized in that: Adaptive windowing module, including: An envelope extraction module is configured to perform a Hilbert transform on the acquired sound signal, form a complex analytical signal with the sound signal, and extract the envelope of the sound signal; a mean value calculation module, configured to calculate the mean value of the envelope of the sound signal; a decentralization module configured to perform a decentralization operation on the envelope of the sound signal based on the obtained mean value; The window size determination module is configured to calculate the autocorrelation coefficient of each signal point for the envelope of the decentralized sound signal, identify the periodic peak of the autocorrelation coefficient, use the period corresponding to the periodic peak as the window size of the current signal point, frame and window the sound signal, and divide the sound signal into different time window signals.

Images