Power distribution network cable joint fault monitoring method and system based on voiceprint analysis

Abstract

The invention relates to the technical field of fault monitoring equipment, in particular to a power distribution network cable joint fault monitoring method and system based on voiceprint analysis, and the method comprises the steps: carrying out the continuous sampling to obtain an original acoustic signal, and carrying out the noise reduction and time domain normalization to obtain a processing signal; performing time alignment on the processing signal, performing short-time Fourier transform to obtain a time-frequency distribution diagram, and calculating a frequency band energy ratio; applying wavelet packet decomposition based on the time-frequency distribution diagram to calculate sub-band energy, and calculating pulse density; calculating an abnormal score based on the normalized frequency band energy ratio, the sub-band energy and the pulse density; generating a trend direction based on the historical score, dividing a fault level in combination with the abnormal score, triggering a corresponding early warning notification based on the fault level, and sending a corresponding receiving object; and establishing a health file and implementing system performance evaluation and parameter dynamic adaptation. The method has the effect of improving the recognition capability and monitoring continuity of the early fault of the cable joint.

Description

Technical Field

[0001] This application relates to the technical field of fault monitoring equipment, and in particular to a method and system for monitoring faults in power distribution network cable joints based on acoustic signature analysis. Background Technology

[0002] In the operation and maintenance of power distribution networks, cable joints, as key connection points for power transmission, are subjected to high voltage, high current and complex environmental conditions for a long time. They are prone to partial discharge due to insulation aging, poor contact or mechanical damage, which can lead to overheating, breakdown or even fire.

[0003] In existing technologies, condition monitoring is typically carried out using methods such as infrared thermometry, partial discharge electrical detection, or periodic manual inspection. Among these, infrared thermometry can only reflect surface temperature changes and is not sensitive to early internal defects; electrical detection requires power outages or connection to high-voltage circuits, which is costly and difficult to implement continuously; manual inspection suffers from problems such as long cycles, strong subjectivity, and incomplete coverage.

[0004] Therefore, existing technologies are insufficient for non-contact, continuous, and sensitive identification of early-stage faults inside cable joints. In particular, the lack of effective acoustic feature capture and quantitative criteria in the early stages of faults leads to delayed detection of potential hazards and passive operation and maintenance response. Summary of the Invention

[0005] To improve the ability to identify early faults in cable joints and the continuity of monitoring, this application provides a method and system for monitoring cable joint faults in power distribution networks based on acoustic signature analysis.

[0006] Firstly, this application provides a method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis, employing the following technical solution:

[0007] A method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis includes the following steps:

[0008] The raw acoustic signal is obtained by continuous sampling based on the deployed acoustic sensors, and the raw acoustic signal is then denoised and time-domain normalized to obtain the processed signal.

[0009] After time alignment of the processed signal, a short-time Fourier transform is performed to obtain a time-frequency distribution map, and the frequency band energy ratio is calculated to determine whether partial discharge exists.

[0010] Based on the time-frequency distribution map, wavelet packet decomposition is applied to calculate the subband energy, and the pulse density is calculated based on the number of over-limit pulse events, wherein the over-limit pulse event is characterized as the subband energy exceeding a preset threshold value;

[0011] Anomaly scores are calculated based on the normalized frequency band energy ratio, the sub-band energy, and the pulse density.

[0012] Based on historical scores, a trend direction is generated, and the fault level is divided according to the abnormal scores. Based on the fault level, a corresponding early warning notification is triggered and sent to the corresponding receiving object.

[0013] Establish health records and implement system performance evaluation and dynamic parameter adaptation to verify effectiveness by statistically analyzing early warning response rates and handling results.

[0014] In some embodiments, the original acoustic signal is denoised and time-domain normalized to obtain a processed signal, including the following steps:

[0015] A reference noise channel is introduced, and filter weights are constructed based on the minimum mean square error and combined with the filter order to denoise the original acoustic signal to obtain a stage signal. The filter weights are dynamically adjusted based on the original acoustic signal and the denoised stage signal.

[0016] The root mean square values ​​of several stage signals are calculated to perform time-series normalization on the stage signals to obtain the processed signal.

[0017] In some embodiments, the processed signal is time-aligned and then subjected to a short-time Fourier transform to obtain a time-frequency distribution map, and the frequency band energy ratio is calculated to determine whether partial discharge exists, including the following steps:

[0018] The multi-channel processed signals are acquired, and the cross-correlation values ​​of several processed signals are calculated based on the cross-correlation function;

[0019] The corresponding delay is obtained based on the cross-correlation value, and the processed signal is shifted based on the delay to complete time alignment.

[0020] The processed signal is subjected to a short-time Fourier transform using the Hanning window function to obtain a two-dimensional time-frequency distribution map, where the horizontal axis represents time, the vertical axis represents frequency, and the gray level represents energy intensity.

[0021] Obtain the set of frequency indices corresponding to the target frequency band, and calculate the band energy ratio based on the ratio to all frequency indices;

[0022] If the frequency band energy ratio is greater than the historical baseline value, it is determined that partial discharge may exist.

[0023] In some embodiments, wavelet packet decomposition is applied based on the time-frequency distribution map to calculate the subband energy, and the pulse density is calculated based on the number of out-of-limit pulse events, including the following steps:

[0024] Wavelet packet decomposition is performed based on a preset number of decomposition layers to obtain several sub-bands, and high-frequency detail coefficients in key layers are selected.

[0025] The sub-band energy corresponding to each sub-band in the key layer is calculated based on the high-frequency detail coefficients, and the existence of sudden increases and periodicity is determined based on the sub-band energy.

[0026] If the sub-band energy value is greater than the preset threshold value, it is marked as an over-limit pulse event, and the pulse density is calculated based on the ratio between the number of over-limit pulse events and the total number of pulses during the acquisition time.

[0027] If the pulse density is greater than a preset value, it is marked as potentially active discharge.

[0028] In some embodiments, anomaly scores are calculated based on the normalized band energy ratio, the sub-band energy, and the pulse density, including the following steps:

[0029] A weight vector group is set, and scores corresponding to the frequency band energy ratio, the sub-band energy and the pulse density are calculated based on the weight vector group. Several scores are added together to obtain the anomaly score, wherein the anomaly score is between 0 and 1, and the closer it is to 1, the more severe the anomaly.

[0030] In some embodiments, a trend direction is generated based on historical scores, and the fault level is classified in combination with the abnormal scores. Based on the fault level, a corresponding early warning notification is triggered and sent to the corresponding receiving object, including the following steps:

[0031] Obtain the historical scores corresponding to the preset first historical duration to form a time score sequence;

[0032] The differences between adjacent historical scores in the time-based scoring sequence are summed to obtain a trend vector, and the trend direction is determined based on the magnitude of the trend vector.

[0033] The level range corresponding to the values ​​of the abnormal scores is selected to determine the basic level, and the basic level is adjusted according to the trend direction to obtain the final fault level.

[0034] Match the corresponding communication method and receiving object according to the fault level;

[0035] The degradation rate is calculated by dividing the difference between the current anomaly score and the historical score two historical periods ago by the second historical period.

[0036] Based on the fault level and the degradation rate, a response strength index is calculated using a preset allocation coefficient, and the communication method is adjusted based on the response strength index.

[0037] In some embodiments, adjusting the base level according to the trend direction to obtain the final fault level specifically includes the following steps:

[0038] If the trend is upward, upgrade the base level;

[0039] If the trend direction is not upward, maintain the current base level.

[0040] In some embodiments, establishing a health record includes the following steps:

[0041] Calculate the age of the joint;

[0042] A statistical period is set, and the variance index corresponding to the abnormal score and the average score at each time point is calculated based on the number of data points within the statistical period, wherein the number of data points represents the number of abnormal scores within the statistical period.

[0043] Calculate the mean score and standard deviation of the scores for the same batch of interfaces within the same specified time period, and calculate the relative degradation index based on the mean score and the standard deviation of the scores;

[0044] The highest score, the number of warnings after normalization, and the average rate of deterioration are summarized on a monthly basis, and a comprehensive health score is calculated by combining them with preset weighting coefficients.

[0045] The health record is generated based on the joint age, the variance index, the relative deterioration index, and the comprehensive health score.

[0046] In some embodiments, system performance evaluation and dynamic parameter adaptation are implemented to statistically verify the effectiveness of early warning response rates and handling results, including the following steps:

[0047] The effective warning rate is calculated by statistically analyzing the number of warnings in the health records over a preset historical period and comparing it with the number of on-site confirmed hazard warnings.

[0048] The weight adjustment factor for updating the weight vector group is calculated based on the contribution energy of each feature in the false alarm case, wherein the features include the band energy ratio, the subband energy, and the pulse density;

[0049] Obtain the duration of the current season, and calculate the environmental noise baseline value by combining the root mean square value of the reference noise channel under the duration of the season. The environmental noise baseline value is used to update the filter weights.

[0050] The difference between the effective early warning rate before and after the update is calculated to generate parameter iteration revenue, and the effectiveness of the action is judged based on the parameter iteration revenue.

[0051] Secondly, this application provides a power distribution network cable joint fault monitoring system based on acoustic signature analysis, which adopts the following technical solution:

[0052] A power distribution network cable joint fault monitoring system based on acoustic signature analysis, used to implement the above method, includes:

[0053] Acoustic sensors are used to perform continuous sampling to obtain raw acoustic signals;

[0054] Edge computing nodes are used to perform noise reduction, time-domain normalization, and time alignment on the original acoustic signal;

[0055] The central analysis platform is used to perform short-time Fourier transform and wavelet packet decomposition, as well as to calculate anomaly scores to classify fault levels and trigger corresponding early warning notifications based on the fault levels and send them to the corresponding receiving objects.

[0056] The user interaction page is used to receive the warning notification and view the generated health record, and supports status tracking and trend tracking.

[0057] The technical solutions provided by the embodiments of this application have the following technical effects:

[0058] By deploying acoustic sensors to continuously collect acoustic signals in the joint area, and combining adaptive filtering and time-domain normalization to improve the signal-to-noise ratio, time alignment is used to ensure multi-channel collaboration. High-frequency energy, sub-band details, and pulse density are extracted through short-time Fourier transform and wavelet packet decomposition. A weighted fusion anomaly scoring system is constructed to achieve early quantitative identification of faults. Furthermore, trend prediction and fault level classification are combined to trigger graded early warnings, generate operation and maintenance guidelines, and establish health records, forming a closed-loop management system from perception to response.

[0059] This invention can sensitively capture the acoustic characteristics generated by partial discharge, significantly improve the identification capability and monitoring continuity of early faults in cable joints, overcome the problems of slow response and insufficient sensitivity of traditional methods, and provide reliable support for the safe and stable operation of power distribution networks. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the steps of the power distribution network cable joint fault monitoring method based on acoustic analysis provided in this embodiment.

[0061] Figure 2 This is a schematic diagram of a power distribution network cable joint fault monitoring system based on acoustic signature analysis provided in an embodiment of this application. Detailed Implementation

[0062] To better understand the purpose, technical solutions, and advantages of this application, it has been described and illustrated below with reference to the accompanying drawings and embodiments. However, those skilled in the art should understand that this application can be implemented without these details. In some cases, to avoid obscuring various aspects of this application due to unnecessary description, well-known methods, processes, systems, components, and / or circuits already described at a higher level will not be elaborated upon. It will be apparent to those skilled in the art that various modifications can be made to the embodiments disclosed in this application, and the general principles defined in this application can be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the illustrated embodiments, but conforms to the broadest scope consistent with the scope of protection claimed in this application.

[0063] It should be noted that the descriptions of these embodiments are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0064] In the description of this application, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0065] In the description of this application, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples.

[0066] like Figure 1As shown in the figure, this application discloses a method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis. This method is used to monitor the health status of underground power distribution network cable joints throughout their entire lifecycle. By collecting acoustic signals generated by the cable joints during operation and combining signal processing and pattern recognition techniques, it achieves early identification and trend judgment of potential faults such as partial discharge, poor contact, and insulation aging. The system is deployed on a power distribution network operation and maintenance platform and has the capabilities for online acquisition, real-time analysis, intelligent early warning, fault level classification, and operation and maintenance suggestion push. It can effectively reduce sudden power outages caused by cable joint deterioration, improve power supply continuity and operation and maintenance response efficiency, and includes the following steps:

[0067] S100 continuously samples based on deployed acoustic sensors to obtain raw acoustic signals, and performs noise reduction and time-domain normalization on the raw acoustic signals to obtain processed signals.

[0068] The acoustic sensor is installed inside the cable joint well and is fixed to the surface of the joint housing by magnetic attraction or bolts to ensure acoustic coupling effect. The sensor is a piezoelectric acoustic sensor with wide frequency response characteristics, covering a frequency response range of 20Hz-150KHz, which can effectively capture ultrasonic signals caused by partial discharge and mid-frequency vibrations caused by loose conductor contact.

[0069] The acoustic sensor has a sampling frequency of 512 kHz to meet the Nyquist sampling theorem's requirement for high-frequency component reproduction. It triggers a 10-second acquisition cycle every 30 minutes, forming a time-series signal x(t), where... .

[0070] This sampling strategy balances data volume and information integrity, ensuring the capture of transient discharge pulses while avoiding excessive storage and transmission pressure caused by continuous 24 / 7 acquisition. The acquired signal is represented in discrete form as follows: .

[0071] in, The second is the sampling interval at a sampling frequency of 512kHz. This represents the total number of sample points in a single acquisition. This expression transforms continuous sound pressure changes at the physical level into a sequence that can be processed in the digital domain, serving as the fundamental input for all subsequent analyses.

[0072] The original acoustic signal needs to be further subjected to adaptive filtering to eliminate environmental background noise caused by traffic vibrations, water pump operation and other environmental interferences, and time-domain normalization is implemented to eliminate amplitude differences caused by temperature and humidity changes, sensor sensitivity drift or power fluctuations, so that the signal energy is unified to a similar level.

[0073] S200 performs time alignment on the processed signal and then performs a short-time Fourier transform to obtain a time-frequency distribution map, and calculates the frequency band energy ratio to determine whether partial discharge exists.

[0074] In some cable joint monitoring scenarios, multiple acoustic sensors are configured in an array layout (e.g., three acoustic sensors arranged around the joint). In this case, it's crucial to ensure that the signals from multiple channels corresponding to these sensors remain aligned on the time axis to match the starting points of the multi-channel data, facilitating subsequent sound source localization or feature fusion. This time alignment is primarily achieved by shifting sample points, ensuring the synergy of the multi-channel data.

[0075] After time alignment is completed, a short-time Fourier transform is performed on the signal to obtain a time-frequency distribution map. The proportion of high-frequency components is quantified by calculating the frequency band energy ratio, and the presence of partial discharge activity is determined by the high-frequency energy situation.

[0076] S300 calculates the subband energy by applying wavelet packet decomposition based on the time-frequency distribution map, and calculates the pulse density based on the number of over-limit pulse events, where an over-limit pulse event is characterized as the subband energy exceeding a preset threshold value.

[0077] To further capture the fine structure of transient pulses, wavelet packet decomposition is applied to the time-frequency distribution map to extract multi-scale detail coefficients, and the degree of discharge activity is initially determined based on a predetermined pulse density threshold. If the pulse density exceeds the preset threshold, it is marked as "suspected discharge activity".

[0078] S400 calculates anomaly scores based on normalized band energy ratio, subband energy, and pulse density.

[0079] The calculated frequency band energy ratio, sub-band energy, and pulse density are used as feature values. Based on these feature values, a voiceprint anomaly recognition scoring system is constructed. The feature values ​​are then normalized and weighted to generate an anomaly score with a timestamp and corresponding record, which is then uploaded to the central platform.

[0080] The S500 generates trend directions based on historical scores and classifies fault levels by combining abnormal scores. Based on the fault level, it triggers corresponding early warning notifications and sends them to the corresponding recipients.

[0081] Fault development trend prediction and level classification are carried out based on anomaly scoring. The fault level is classified by combining the trend direction reflected by the current score and the historical score. The status report with level label is generated based on the degree of matching between the fault level and the dynamically adjusted level range. The status report is stored in the database as the basis for subsequent operation and maintenance decision-making and for traceability analysis.

[0082] Simultaneously, the system matches corresponding early warning channels and recipients based on the fault level. Multi-level early warning pushes can be implemented through various methods such as enterprise apps / SMS, telephone voice calls, and platform pop-ups to achieve tiered notifications. Different levels of early warning notifications need to be sent to personnel with corresponding levels and permissions. Furthermore, when pushing early warning notifications, to ensure recipients quickly understand the status of the alarm, it is also necessary to include information such as voiceprint screenshots, trend charts, and the reason for the early warning, enhancing readability.

[0083] S600 establishes health records and implements system performance evaluation and dynamic parameter adaptation to verify effectiveness by statistically analyzing early warning response rates and handling results.

[0084] Health records are established based on various early warning information, dynamic monitoring data entries are added in chronological order, system performance is evaluated and parameters are dynamically adjusted and adapted to the health records, and the effectiveness of early warning response rate and handling results is verified by statistical analysis.

[0085] Using the above methods, acoustic sensors are deployed to continuously collect acoustic signals in the joint area. The signal-to-noise ratio is improved by combining adaptive filtering and time-domain normalization. Time alignment is used to ensure multi-channel coordination. High-frequency energy, sub-band details, and pulse density are extracted through short-time Fourier transform and wavelet packet decomposition. A weighted fusion anomaly scoring system is constructed to achieve early quantitative identification of faults. Furthermore, trend prediction and fault level classification trigger graded early warning, generating operation and maintenance guidelines and establishing health records to form a closed-loop management system from perception to response.

[0086] This invention can sensitively capture the acoustic characteristics generated by partial discharge, significantly improve the identification capability and monitoring continuity of early faults in cable joints, overcome the problems of slow response and insufficient sensitivity of traditional methods, and provide reliable support for the safe and stable operation of power distribution networks.

[0087] In other embodiments, the original acoustic signal is denoised and time-domain normalized to obtain a processed signal, including the following steps:

[0088] S110 introduces a reference noise channel and constructs filter weights based on the minimum mean square error, combining the filter order to denoise the original acoustic signal to obtain the stage signal. The filter weights are dynamically adjusted based on the original acoustic signal and the denoised stage signal.

[0089] Due to the presence of periodic mechanical noise (such as the start and stop of drainage pumps) and random traffic vibrations within the well, the original acoustic signal... Since interference unrelated to the fault is often mixed in, a reference noise channel is introduced in this application. The signal in this channel is acquired by an auxiliary sensor installed far from the connector, primarily to reflect environmental common-mode noise that contains as little connector acoustic signal as possible.

[0090] Furthermore, the minimum mean square error criterion is used to construct the filter weights. This is used to reduce noise in the main signal (original acoustic signal).

[0091] The processed signal output after filtering is set as follows: The calculation process is as follows:

[0092] .

[0093] Where L represents the filter order, which is usually taken as 64; The filter weights are dynamically updated based on the error signals before and after filtering. The principle of minimizing the squared expectation is used for dynamic adjustment; n represents the index of the current sample point at time, and k represents the weight corresponding to the order. It is characterized as a canceled signal that is completely opposite to the interference component of the original signal.

[0094] This process effectively reduces steady-state noise from the external environment, improves the signal-to-noise ratio, and allows weak fault-related acoustic signatures to be revealed.

[0095] S120, calculate the root mean square value of several stage signals to perform time-series normalization on the stage signals to obtain the processed signal.

[0096] Signals collected at different times may exhibit overall amplitude differences due to variations in temperature and humidity, sensor sensitivity drift, or power fluctuations, which is detrimental to subsequent cross-time period comparisons. Therefore, filtered and processed signals are necessary. Amplitude standardization is performed to keep the energy level consistent.

[0097] In this embodiment of the application, the time-domain normalization operation is defined as follows:

[0098] .

[0099] Wherein, the denominator is the root mean square of the processed signal, which represents the effective value of the energy corresponding to the processed signal; To prevent a small constant with a denominator of 0 when all signals are 0.

[0100] The root mean square (RMS) reflects the equivalent energy level of a signal. Using it as the denominator can unify signals of different time periods and amplitudes to a level of "similar energy," eliminating amplitude differences caused by sensor drift and environmental fluctuations.

[0101] After the above transformation, the energy of all the acquired segments is unified to a similar level, eliminating the amplitude shift caused by changes in external conditions and enhancing the consistency and feasibility of the data.

[0102] In other embodiments, after time alignment of the processed signal, a short-time Fourier transform is performed to obtain a time-frequency distribution map, and the frequency band energy ratio is calculated to determine whether partial discharge exists, including the following steps:

[0103] S210: Acquire multi-channel processed signals and calculate the cross-correlation values ​​of several processed signals based on the cross-correlation function.

[0104] S220 obtains the corresponding delay based on the cross-correlation value, and shifts the processed signal based on the delay to complete time alignment.

[0105] When multiple sensors are present, in order to ensure that the signals of each channel are aligned on the time axis, this embodiment of the application uses a cross-correlation function to find the maximum similarity position between each channel to achieve synchronous correction.

[0106] With two channel signals and For example, its cross-correlation function is defined as: .

[0107] Characterized as a delay (number of sample point shifts), making Get the maximum delay of the peak ,Will Overall translation Each sample point is paired for time alignment. This operation ensures the spatial synergy of multi-channel data, providing structured input for the next step of feature extraction.

[0108] This step completes the conversion and initial purification from physical sound field to digital signal. High sampling rate ensures that high frequency details are not lost, adaptive filtering effectively removes environmental noise, normalization processing improves data stability, and time alignment lays the foundation for multi-sensor collaborative analysis.

[0109] The processed signal, after the above steps, exhibits higher purity and consistency, significantly enhancing the reliability of subsequent feature extraction and avoiding misjudgments due to issues with the quality of the original data. The output of this stage will serve as the direct input for subsequent frequency domain and time-frequency domain analysis.

[0110] Based on the above signal preprocessing, it is further necessary to focus on extracting distinguishable acoustic features from the purified acoustic signal and to preliminarily identify whether there is partial discharge activity.

[0111] Partial discharge is a significant precursor to insulation degradation in cable joints, often accompanied by high-frequency acoustic emission, characterized by short, repetitive pulse signals. Frequency domain and time-frequency domain analysis can effectively separate these events and quantify their activity levels. Specifically:

[0112] S230 uses the Hanning window function to perform a short-time Fourier transform on the processed signal to obtain a two-dimensional time-frequency distribution map.

[0113] To reveal the time-varying frequency components of a signal, a Short-Time Fourier Transform (STFT) is performed on the normalized processed signal. The Hanning window function is selected. The window length was set to 4096 points (approximately 8 milliseconds), and the overlap rate was 50% to balance temporal and frequency resolution.

[0114] The transformation result is expressed in complex form:

[0115] .

[0116] Where m represents the frame index (that is, the 10s signal is divided into frames), R=2048 is the frame shift, k is the frequency index, and L=4096 is the window length.

[0117] After obtaining the amplitude of the transformation, square it to obtain the power spectral density: .

[0118] This generates a two-dimensional time-frequency graph, with time on the horizontal axis and frequency on the vertical axis, and grayscale representing energy intensity. This graph can visually show whether there are sudden energy clusters concentrated in a specific frequency band in the signal, which is an important basis for judging partial discharge.

[0119] S240: Obtain the set of frequency indices corresponding to the target frequency band, and calculate the band energy ratio based on the ratio to all frequency indices.

[0120] S250, if the frequency band energy ratio is greater than the historical baseline value, it is determined that partial discharge may exist.

[0121] Since acoustic emissions caused by partial discharge are mostly concentrated in the frequency band above 40kHz, while mechanical vibrations under normal operating conditions are mainly distributed in the low-frequency region (below 20kHz), the ratio of high-frequency energy to total energy is defined as a preliminary basis.

[0122] Define a frequency index set This set represents the target frequency band (high frequency) corresponding to 40-150kHz, while Given the total frequency index corresponding to all acquisition frequency bands of the sensor, then the bandwidth energy ratio is... Defined as the proportion of the total energy in the high-frequency band to the total energy in all frequency bands: .

[0123] And when When the value is significantly higher than the historical baseline, it suggests the possible presence of partial discharge activity. This indicator is simple and intuitive, and is suitable for quickly screening abnormal signal segments.

[0124] In other embodiments, wavelet packet decomposition is applied based on the time-frequency distribution map to calculate the subband energy, and the pulse density is calculated based on the number of out-of-limit pulse events, including the following steps:

[0125] S310 performs wavelet packet decomposition based on a preset number of decomposition layers to obtain several sub-bands, and selects high-frequency detail coefficients in the key layers.

[0126] To further capture the subtle structure of transient pulses, wavelet packet decomposition is used to refine the signal in multiple layers. In this embodiment, the db4 wavelet basis is selected, and the decomposition layer is set to 6 layers, generating a total of 64 sub-bands. This application focuses on the high-frequency detail coefficients in the 6th layer. , where j represents the sub-band number.

[0127] The db4 wavelet basis is a compactly supported orthogonal wavelet with good time-frequency localization characteristics. The formula for calculating the number of subbands in wavelet packet decomposition is: Number of subbands = (n is the number of decomposition layers). The choice of the number of decomposition layers needs to be combined with the sampling frequency of the signal. The more layers there are, the narrower the sub-band and the higher the frequency domain resolution, but the amount of computation will increase. Six layers is the choice that balances resolution and efficiency after experiments in the embodiments of this application.

[0128] After decomposition, the 64 sub-bands correspond to different frequency ranges, and are sorted from low to high frequency:

[0129] Sub-bands 1-32 are mainly low-frequency components, corresponding to the power frequency signals and low-frequency interference of electrical equipment;

[0130] Subbands 33-64 are dominated by high-frequency components, and the transient pulses of partial discharge happen to contain abundant high-frequency components. Therefore, this application focuses on analyzing the high-frequency subband coefficients of the 6th layer.

[0131] S320 calculates the sub-band energy corresponding to each sub-band in the key layer based on the high-frequency detail coefficients, and determines whether there are sudden increases and periodicities based on the sub-band energy.

[0132] The energy of each subband is calculated as follows: Subband energy reflects the energy intensity of the signal within that subband. When a partial discharge pulse occurs, the energy of the corresponding high-frequency subband will increase significantly, while the energy distribution of a stable interference signal is relatively uniform.

[0133] By analyzing the energy distribution of each subband, if a significant surge in energy is found in a certain high-frequency subband (such as 80-100kHz) and it exhibits a periodic pulse pattern, it is highly suspected to be a partial discharge feature. Wavelet packet decomposition provides a more detailed video focusing capability than STFT, which helps to distinguish the discharge type from the interference signal.

[0134] S330, if the sub-band energy value is greater than the preset threshold, it is marked as an over-limit pulse event, and the pulse density is calculated based on the ratio between the number of over-limit pulse events and the total number of pulses during the acquisition time.

[0135] S340, if the pulse density is greater than the preset value, it is marked as suspected active discharge.

[0136] Based on the wavelet decomposition results, the pulse times of all pulses with amplitudes exceeding a set threshold are identified, and the number of occurrences per unit time, i.e., the pulse density, is counted. .

[0137] Pulse density is defined as: .

[0138] Where T = 10 seconds is the acquisition duration, and M represents the total number of pulses detected. This is an indicator function (1 if the condition in parentheses is met, 0 if not). It is a multiple factor (usually taken as 3). It is characterized as the standard deviation of the subband coefficient (used to reflect the background noise level). This is represented by the index of the sample point corresponding to the i-th pulse. If... (Preset value, 5 times / second), then the current segment is marked as "suspected active discharge".

[0139] This step maps the signal from the time domain to a multi-dimensional feature space. STFT provides a global time-frequency view, the band energy ratio is used for macroscopic judgment, wavelet packet decomposition delves into transient details, and pulse density quantifies the frequency of discharge activity. The combination of these four elements forms a complementary judgment system, improving the robustness of the identification. The output includes a time-frequency plot and a band energy ratio. Each sub-carrier carries energy and pulse density These features will serve as the core input for subsequent anomaly scoring calculations, supporting more refined state assessments.

[0140] In other embodiments, anomaly scores are calculated based on normalized band energy ratios, sub-band energy, and pulse density, including the following steps:

[0141] S410, set a weight vector group, and calculate the scores corresponding to the frequency band energy ratio, sub-band energy and pulse density based on the weight vector group, and add several scores together to obtain an anomaly score. The anomaly score is between 0 and 1, and the closer it is to 1, the more serious the anomaly.

[0142] After acquiring a series of voiceprint features, this step is used to construct a quantifiable anomaly score by integrating various indicators, reflecting the degree to which the current cable joint deviates from its normal state. The anomaly score result is not only used to determine whether an early warning is triggered, but also provides a basis for subsequent fault level classification. In this embodiment of the application, the scoring system adopts a weighted fusion method, taking into account both the discriminative power and stability of different features.

[0143] Because of the frequency band energy ratio Each sub-carrier carries energy and pulse density Each of these features has a different physical meaning and numerical range. Directly adding them together can lead to a problem of dimensional dominance. Therefore, it is necessary to first normalize each feature to map it to the [0,1] interval to unify the dimensional range.

[0144] With frequency band energy ratio For example, let its historical maximum value be... The historical minimum value is Then the normalized score of the current value for .

[0145] Similarly, for pulse density and sub-band energy within the target range Perform the same processing to obtain the following results: and This eliminates the difference in dimensions, making different characteristics comparable.

[0146] Secondly, weighting coefficients need to be assigned based on the discriminative power of different features, as different features have varying sensitivities to faults. Long-term experiments and observations show that pulse density... It is most sensitive to early discharge, and the high-frequency energy ratio It is more representative of the mid-to-late stage of degradation. Therefore, this application assigns... Higher weight.

[0147] Specifically, this application sets a weight vector group. ,satisfy The value is set to... Furthermore, this value can be adjusted based on actual circumstances. This allocation is derived from historical case backtesting analysis, reflecting the contribution ratio of each feature in actual discrimination.

[0148] Finally, a comprehensive anomaly score is calculated to determine if any limits have been exceeded. The normalized score is combined with its corresponding weight to obtain the final anomaly score. : .

[0149] The S value ranges from [0,1], with values ​​closer to 1 indicating a more severe deviation from the normal state. Furthermore, a primary warning threshold can be set. If the abnormality score S is greater than 0.6, it is judged as "mild abnormality" and the preliminary warning process is initiated.

[0150] Under the initial warning process, a structured anomaly record is immediately generated, including a timestamp, connector number, various characteristic values, a comprehensive score S, and the judgment result. This record is uploaded to the central analysis platform via a wireless communication network for subsequent trend analysis. Simultaneously, local edge nodes retain complete acoustic data for the past 30 days for post-event traceability.

[0151] This step achieves the fusion and quantification of multi-source features, condensing complex acoustic performance into an intuitive anomaly score. This is easy for non-professionals to understand. The weighted scoring mechanism takes into account the discriminative effectiveness of different features, improving the overall rationality of the judgment. The generated anomaly records are not only used for immediate alerts but also provide data support for long-term trend modeling. The output of this step is the anomaly score. Structured records will become a key input for trend forecasting and grading in subsequent steps.

[0152] Based on the single anomaly score, the historical score sequence is used to analyze the fault evolution trend and classify the fault severity level accordingly. Trend judgment helps to distinguish between occasional disturbances and continuous degradation, while the level classification provides a priority basis for subsequent operation and maintenance response.

[0153] In other embodiments, a trend direction is generated based on historical scores, and fault levels are classified in conjunction with abnormal scores. Based on the fault level, a corresponding early warning notification is triggered and sent to the corresponding recipient, including the following steps:

[0154] S510: Obtain the historical score corresponding to the preset first historical duration to form a time score sequence.

[0155] In this embodiment of the application, the first historical duration is 7 days, so the abnormality scores generated hourly over the past 7 days for the target connector are retrieved from the central platform. This forms a historical rating time series. .

[0156] S520: Accumulate the differences between adjacent historical scores in the time score sequence to obtain a trend vector, and determine the trend direction based on the magnitude of the trend vector.

[0157] Taking 24 hours as the window length, calculate the trend vector within the current window .

[0158] The trend vector is used to characterize the change trend of historical scores within a certain period of history, and the magnitude of its value corresponds to the change direction of the trend.

[0159] Define the trend vector as the accumulation of adjacent score differences: , where represents the historical score at time t - i, t is the current time, and i is the hourly score corresponding to the past i hours.

[0160] If is greater than 0, it indicates that the recent scores show an upward trend, otherwise it is a downward trend. This indicator reflects the change direction of fault activity and is an important basis for judging whether to enter the accelerated deterioration stage. At the same time, it distinguishes "occasional anomalies (such as a single noise causing S to be high, but v < 0 and then decreasing)" from "continuous deterioration (S is high and v > 0, fault accelerating)", avoiding the lag of static thresholds.

[0161] S530. Screen the level interval corresponding to the value of the abnormal score to determine the basic level, and adjust the basic level according to the trend direction to obtain the final fault level.

[0162] Set three level intervals, namely: attention level (0.6 < S ≤ 0.75), concern level (0.75 < S ≤ 0.9), warning level (S > 0.9).

[0163] Determine the basic level according to the level interval corresponding to the abnormal score. At the same time, further, if both the score limit is exceeded and the trend is upward ( is greater than 0), it will be automatically upgraded by one level, and at most upgraded to the warning level.

[0164] For example, the abnormal score S of a certain joint is 0.72, its basic level is the attention level, but the corresponding trend vector is 0.15, then its level needs to be raised to the concern level; when the trend vector V is less than 0, the final level is still the attention level. This dynamic adjustment mechanism avoids the lag caused by static thresholds and can respond to rapidly developing hidden dangers earlier.

[0165] S540. Match the corresponding communication method and recipient according to the fault level.

[0166] After completing the fault level division, it is also necessary to trigger a warning notification at the corresponding response level according to the level result and automatically generate targeted operation and maintenance suggestions. The push content is sent to relevant personnel through multiple channels to ensure that information is timely reached and promote closed-loop processing.

[0167] Different notification methods and responsible persons are assigned to different levels.

[0168] S550 calculates the degradation rate by dividing the difference between the current anomaly score and the historical score two historical periods ago by the second historical period.

[0169] To further quantify the development speed, the degradation rate is also defined in this application embodiment. Its representation is the difference between the current anomaly score and the score from the second historical period ago (72 hours in this application) divided by the time interval: The unit is "scores / hour".

[0170] If the degradation rate r is greater than the preset value, such as 0.005, it indicates rapid degradation and requires priority handling. This indicator supplements the trend vector information, providing a more accurate description of the rate of degradation.

[0171] Comprehensive anomaly score S, trend vector Deterioration rate Based on the final fault level, a status report is generated, labeled "Attention Level," "Concern Level," or "Warning Level," and the basis for the judgment is recorded. This report is stored in the database as a basis for operation and maintenance decisions, and also used for subsequent traceability analysis.

[0172] This achieves a leap from static judgment to dynamic trend analysis. By capturing the direction of score changes through a sliding window and quantifying the speed of progress using rate indicators, it effectively distinguishes between occasional anomalies and continuous degradation. The grading mechanism provides clear guidance for the allocation of operational resources, improving response efficiency. The output status reports and grade labels will serve as the basis for generating early warning push content, ensuring the accuracy and urgency of information delivery.

[0173] The S560 calculates a response strength index based on the fault level and degradation rate, combined with a preset allocation coefficient, and adjusts the communication method based on the response strength index.

[0174] To quantify the required response intensity for different levels, embodiments of this application further define a warning response intensity index. It is used to guide the allocation of communication resources, specifically:

[0175] .

[0176] Among them, L This indicates the fault level (1 = Attention level, 2 = Concern level, 3 = Warning level). Characterized by the rate of degradation, These are empirical allocation coefficients.

[0177] Response strength index This value combines the severity and speed of the event; a higher value indicates that a higher level of communication channels and response software needs to be used. For example... When the value is greater than 2.0, the voice call and work order system will be automatically activated to ensure that high-risk events are not missed.

[0178] Furthermore, embodiments of this application also include generating structured operation and maintenance suggestions by calling a preset template, specifically:

[0179] Based on the fault level and historical handling experience, a corresponding handling suggestion template is matched. For example, the suggestion for the "Attention" level is "Schedule to check the appearance and temperature of the connector during the next inspection"; the suggestion for the "Concern" level is "Schedule a special inspection within one week and re-inspect using an infrared thermal imager"; and the suggestion for the "Warning" level is "Inspect with power off within 48 hours and prepare to replace the connector."

[0180] To improve the adaptability of the suggestions, a suggestion matching score M is introduced to measure the similarity between the current state and historical cases: .

[0181] in, Characterized by the current k-th feature (such as the band energy ratio) Each sub-carrier carries energy and pulse density The normalized value of ) The normalized value of the k-th feature corresponding to a certain historical case is represented by its representation. For feature weights, This represents the total number of features. The formula uses the minimum ratio method to measure similarity, avoiding the dominance of extreme values. , The closer the value is to 1, the higher the matching degree. The system prioritizes recommending historical solutions with high matching degrees to improve the practicality of the suggestions.

[0182] Furthermore, additional voiceprint feature screenshots and trend charts enhance readability:

[0183] Embed key information visualizations in push notifications, including screenshots of abnormal areas in the current time-frequency distribution map, score change curves over the past 7 days, pulse density trend charts, etc. This combination of text and graphics helps recipients quickly understand the essence of the problem and improves decision-making efficiency.

[0184] To highlight abnormal periods, a visual saliency weighted map is defined. This is used to generate trend charts with highlighted annotations. .

[0185] in, For the anomaly score at time t, This represents the degradation rate during that period. =0.5 is the emphasis factor (controls the degree of highlight). For indicator functions, This is the degradation rate threshold. When the degradation rate exceeds the threshold, the corresponding time period will be automatically bolded or changed color in the image, guiding users to focus on the accelerated degradation zone and improving information delivery efficiency.

[0186] Furthermore, recording push events and feedback status enables closed-loop tracking:

[0187] The system automatically records the sending time, channel, recipient, and read status of each alert. If no processing feedback is received within the specified time limit, the alert is escalated according to preset rules to adjust the notification channel. This mechanism ensures the effective delivery of alert information and prevents omissions.

[0188] To assess information transmission efficiency, the early warning closure rate C is calculated: .

[0189] in, This refers to the total number of warnings issued within the period. This represents the number of warnings marked as "processed." This indicator reflects the timeliness of the operation and maintenance response. If C continues to fall below the set target (e.g., 85%), subsequent system optimization procedures will be triggered (see steps S660-S690).

[0190] This step introduces a response intensity index. Recommended matching degree Visual salience map With closed loop rate Four quantitative expressions enable the early warning push process to shift from experience-driven to data-driven. Response Intensity Index Dynamic allocation of communication resources was achieved, and a matching degree was suggested. The recommendations were made more targeted, and the visual salience diagram was improved. Enhanced information readability and closed-loop rate These indices provide a basis for system performance evaluation. Together, they support an intelligent, efficient, and traceable early warning mechanism, and their outputs (such as push logs and feedback status) will serve as the core data source for subsequent health record updates.

[0191] To support full lifecycle management, an independent electronic health record is established for each cable joint, continuously recording its acoustic characteristics, anomaly scores, early warning history, and handling records. The record supports multi-dimensional queries and trend backtracking, providing a basis for root cause analysis of failures and optimization of operation and maintenance strategies.

[0192] In other embodiments, establishing a health record includes the following steps:

[0193] S610, calculate the age of the joint.

[0194] Before calculating the age of the cable joints, each monitored cable joint is assigned a globally unique number, and a corresponding file is created in the database. The header of the file records static information such as installation date, location coordinates, cable type, rated voltage, and construction unit, serving as background material for subsequent analysis.

[0195] To facilitate subsequent statistical analysis, the age of the joint is defined. .

[0196] in, Represented as the current time, The time of installation of the cable joint is represented in days. This variable will serve as the basis for analyzing the relationship between failure occurrence and service life, and will be used to identify whether a "concentration period of aging" exists.

[0197] S620 sets the statistical period and calculates the variance index corresponding to the abnormal scores and average scores at each time point based on the number of data points within the statistical period, where the number of data points represents the number of abnormal scores within the statistical period.

[0198] After each voiceprint collection and analysis is completed, anomaly scores will be assigned. Data such as various characteristic values, fault levels, and early warning records are appended to the corresponding files in chronological order. The data is stored at the minute or hour level, forming a continuous time series, which facilitates the creation of long-term trend charts.

[0199] To quantify state volatility, the score variance index is calculated. .

[0200] Where T represents the number of data points within the statistical period. The abnormal score at time t is represented by the score. This represents the average score over the statistical period. (Variance index) The score is used to calculate the dispersion of the score. A high score indicates that the state is unstable and may be affected by external loads or periodic environmental factors. Further analysis based on the operating data is required.

[0201] S630 calculates the mean score and standard deviation of the scores for the same batch of interfaces within the same specified time period, and calculates the relative degradation index based on the mean score and standard deviation of the scores.

[0202] In this embodiment, a query interface is provided, allowing users to specify start and end times to extract abnormal score change curves for a specific connector or batch of connectors. Trend lines for different connectors can be overlaid to identify whether there are regional common problems (such as a batch of products generally deteriorating faster).

[0203] To enable cross-connection comparison, a relative degradation index is defined in this embodiment. .

[0204] in, This is represented by the average score of the same batch of connectors within the same time period. This is represented by the standard deviation of the scores for the same batch of connectors within the same time period. When the value is greater than the preset value (e.g., greater than 2), it indicates that the condition of the joint is significantly worse than the average level of the same batch, and manufacturing or installation defects should be investigated first.

[0205] S640 summarizes the highest score, normalized number of warnings, and average degradation rate for each month, and calculates the overall health score by combining them with preset weighting coefficients.

[0206] A health status report is generated monthly, summarizing indicators such as the highest score of each connector, the number of normalized warnings, and the average rate of degradation, and sorted by level to form a "key concern list".

[0207] This includes a comprehensive health score. : .

[0208] in, The highest rating within the month is represented by the highest score. This represents the number of warnings within the current month after normalization. This represents the average degradation rate within the current month. , , These are the corresponding weights (summing to 1). The H value represents the health status of the joint; a lower H value indicates a worse health status, and vice versa, facilitating fast sorting and resource allocation.

[0209] S650 generates health profiles based on joint age, variance index, relative deterioration index, and overall health score.

[0210] The joint age introduced above Variance index Relative deterioration index Comprehensive health score Building health records elevates them from mere data storage to analytical tools. Through the aforementioned parameter characteristics, it supports longitudinal (time dimension) and horizontal (inter-connection) comparisons, revealing hidden patterns and assisting in the development of differentiated operation and maintenance strategies.

[0211] To ensure the long-term effective operation of the monitoring system, this step regularly evaluates its discrimination effectiveness and dynamically adjusts key parameters (such as early warning thresholds and weighting coefficients) based on field feedback and environmental changes to maintain system adaptability.

[0212] In other embodiments, system performance evaluation and dynamic parameter adaptation are performed to verify the effectiveness of the early warning response rate and handling results, including the following steps:

[0213] S660 calculates the effective warning rate by combining the number of warnings in the historical preset time period based on health records with the number of hazard warnings confirmed on site.

[0214] Collect feedback on all warnings received over a period of time, such as within a month, and calculate the ratio of "confirmed potential risks" to "no abnormalities found." If the latter percentage is too high, it suggests a possible oversensitivity issue, and the threshold settings need to be re-evaluated.

[0215] Specifically, define the effective early warning rate. : .

[0216] in, This is represented by the number of warnings issued after on-site confirmation of potential hazards. Represented by the total number of warnings within a month. Effective warning rate. Reflecting the accuracy of the system's judgment, if If the value is less than a preset threshold (e.g., 0.7), the parameter optimization process will be initiated.

[0217] S670 calculates a weight adjustment factor for updating the weight vector group based on the contribution energy of each feature in false alarm cases, wherein the features include band energy ratio, subband energy, and pulse density.

[0218] Furthermore, feature weights are optimized in reverse by analyzing false positive cases. Specifically,

[0219] Retrospective analysis was conducted on cases identified as false positives to examine which feature was responsible for the high scores. If it was found that a certain feature (such as the bandwidth energy ratio) was prone to producing false positives under specific environmental conditions (such as frequent start-stop cycles of water pumps during the rainy season), the weight corresponding to the bandwidth energy ratio was appropriately reduced. This enhances the robustness of the system.

[0220] Using weighting adjustment factors Update the weight of the k-th feature: .

[0221] in, The old weights are represented by the k-th feature. The new weights, after weight adjustment, are represented by the k-th feature. This is represented by the learning rate (e.g., 0.1). The contribution energy of the k-th feature in false alarm cases is represented by this energy. The total energy of the k-th feature across all cases is represented. By using the above method, the weight of features that contribute significantly to false positives is reduced (e.g., if the frequency band energy contributes more than the false positive contribution, the weight is reduced from 0.3 to 0.25), thus reducing the dominant role of this feature in the scoring.

[0222] It should also be noted that the above is based on the weight adjustment factor. Updates are only made based on the effective warning rate. Triggered when the value is below the threshold to avoid frequent fluctuations.

[0223] S680 obtains the seasonal duration corresponding to the current season, and calculates the environmental noise baseline value by combining the root mean square value of the reference noise channel under the seasonal duration. The environmental noise baseline value is used to update the filter weights.

[0224] The noise level of underground manholes varies with the seasons (e.g., frequent drainage in summer, vibrations due to icing in winter), affecting the signal-to-noise ratio. The reference noise statistics (reference noise channel) in the adaptive filter are updated quarterly to ensure that the noise reduction effect is not affected by seasonal environmental changes.

[0225] Specifically, define environmental noise benchmark values. ,in The reference noise channel is represented at time t, where T is the seasonal duration, and RMS is represented by the root mean square calculation. This is achieved by using a new noise reference value. As a reference input for updating the weights of the filter, it improves noise reduction accuracy, especially during the rainy season or peak construction period.

[0226] In this way, the root mean square average of the reference channel noise is calculated every quarter, and the reference noise model of the adaptive filter is updated to adapt to "seasonal noise changes" (such as high noise caused by frequent drainage pumps in summer and low noise due to icing vibration in winter), ensuring stable noise reduction effect throughout the year.

[0227] Furthermore, after dynamically adjusting the weights and thresholds, the adjusted parameters are packaged into a configuration file and distributed to each edge node via a secure channel to complete the local parameter update. The update process is logged to ensure auditability.

[0228] S690 calculates the difference in effective early warning rates before and after the update to generate parameter iteration revenue, and judges the effectiveness of the action based on the parameter iteration revenue.

[0229] Furthermore, to evaluate the update effect, the parameter iteration gain is further calculated in this embodiment of the application. : .

[0230] in, Characterized by the effective early warning rate within the updated period. This is represented by the effective early warning rate in the previous period. Compare the effective early warning rates before and after parameter adjustment; if the parameter iteration benefit... If the value is greater than 0, it indicates that the adjustment is effective, the newly updated parameters are retained, and the system enters a new stable period; if the parameter iteration benefit is... If the value is less than 0, roll back to the old parameters and re-analyze the cause.

[0231] Through the above steps, an effective early warning rate is introduced. Based on weight adjustment factor Weight update rules, environmental noise benchmark values With parameter iteration gains It has constructed a complete system self-optimization closed loop, forming a dynamic adaptation capability that avoids "blind optimization" and ensures that each parameter adjustment (such as weight and threshold) can improve the system's judgment accuracy, forming a closed loop of "evaluation → optimization → verification", which is a key guarantee for realizing intelligent monitoring of the entire life cycle of cable joints.

[0232] This application also discloses a fault monitoring system for cable joints in power distribution networks based on acoustic signature analysis. The overall system framework includes acoustic sensors, edge computing nodes, a central analysis platform, a communication network, and a user interface. The acoustic sensors are fixedly installed inside the cable joint well or near the joint body to continuously capture acoustic signals generated during operation. The edge computing nodes are responsible for the initial processing and feature extraction of the raw data. The central analysis platform performs in-depth analysis, historical data comparison, and trend prediction, and generates early warning information and maintenance guidelines according to preset rules. Users can receive push notifications through mobile terminals or a dispatch system, achieving closed-loop management.

[0233] Specifically, including:

[0234] Acoustic sensors are used to perform continuous sampling to obtain raw acoustic signals.

[0235] Edge computing nodes are used to perform noise reduction, time-domain normalization, and time alignment on the raw acoustic signal.

[0236] The central analysis platform is used to perform short-time Fourier transform and wavelet packet decomposition. It is also used to calculate anomaly scores to classify fault levels and trigger corresponding early warning notifications based on the fault levels and send them to the corresponding receiving objects.

[0237] The user interaction page is used to receive early warning notifications and view generated health records, and supports status tracking and trend tracking.

[0238] The implementation principle is as follows:

[0239] By deploying acoustic sensors to continuously collect acoustic signals in the joint area, and combining adaptive filtering and time-domain normalization to improve the signal-to-noise ratio, time alignment is used to ensure multi-channel collaboration. High-frequency energy, sub-band details, and pulse density are extracted through short-time Fourier transform and wavelet packet decomposition. A weighted fusion anomaly scoring system is constructed to achieve early quantitative identification of faults. Furthermore, trend prediction and fault level classification are combined to trigger graded early warnings, generate operation and maintenance guidelines, and establish health records, forming a closed-loop management system from perception to response.

[0240] This invention can sensitively capture the acoustic characteristics generated by partial discharge, significantly improve the identification capability and monitoring continuity of early faults in cable joints, overcome the problems of slow response and insufficient sensitivity of traditional methods, and provide reliable support for the safe and stable operation of power distribution networks.

[0241] It should be understood that although the steps in the flowcharts in the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise expressly stated herein, there is no strict order in which these steps are performed, and they may be performed in other orders.

[0242] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis, characterized in that, Includes the following steps: The raw acoustic signal is obtained by continuous sampling based on the deployed acoustic sensors, and the raw acoustic signal is then denoised and time-domain normalized to obtain the processed signal. After time alignment of the processed signal, a short-time Fourier transform is performed to obtain a time-frequency distribution map, and the frequency band energy ratio is calculated to determine whether partial discharge exists. Based on the time-frequency distribution map, wavelet packet decomposition is applied to calculate the subband energy, and the pulse density is calculated based on the number of over-limit pulse events, wherein the over-limit pulse event is characterized as the subband energy exceeding a preset threshold value; Anomaly scores are calculated based on the normalized frequency band energy ratio, the sub-band energy, and the pulse density. Based on historical scores, a trend direction is generated, and the fault level is divided according to the abnormal scores. Based on the fault level, a corresponding early warning notification is triggered and sent to the corresponding receiving object. Establish health records and implement system performance evaluation and dynamic parameter adaptation to verify effectiveness by statistically analyzing early warning response rates and handling results.

2. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 1, characterized in that, The original acoustic signal is denoised and time-domain normalized to obtain a processed signal, including the following steps: A reference noise channel is introduced, and filter weights are constructed based on the minimum mean square error and combined with the filter order to denoise the original acoustic signal to obtain a stage signal. The filter weights are dynamically adjusted based on the original acoustic signal and the denoised stage signal. The root mean square values ​​of several stage signals are calculated to perform time-series normalization on the stage signals to obtain the processed signal.

3. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 2, characterized in that, After time alignment of the processed signal, a short-time Fourier transform is performed to obtain a time-frequency distribution map, and the frequency band energy ratio is calculated to determine whether partial discharge exists. This includes the following steps: The multi-channel processed signals are acquired, and the cross-correlation values ​​of several processed signals are calculated based on the cross-correlation function; The corresponding delay is obtained based on the cross-correlation value, and the processed signal is shifted based on the delay to complete time alignment. The processed signal is subjected to a short-time Fourier transform using the Hanning window function to obtain a two-dimensional time-frequency distribution map, where the horizontal axis represents time, the vertical axis represents frequency, and the gray level represents energy intensity. Obtain the set of frequency indices corresponding to the target frequency band, and calculate the band energy ratio based on the ratio to all frequency indices; If the frequency band energy ratio is greater than the historical baseline value, it is determined that partial discharge may exist.

4. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 3, characterized in that, Based on the time-frequency distribution map, wavelet packet decomposition is applied to calculate the sub-band energy, and the pulse density is calculated based on the number of out-of-limit pulse events, including the following steps: Wavelet packet decomposition is performed based on a preset number of decomposition layers to obtain several sub-bands, and high-frequency detail coefficients in key layers are selected. The sub-band energy corresponding to each sub-band in the key layer is calculated based on the high-frequency detail coefficients, and the existence of sudden increases and periodicity is determined based on the sub-band energy. If the sub-band energy value is greater than the preset threshold value, it is marked as an over-limit pulse event, and the pulse density is calculated based on the ratio between the number of over-limit pulse events and the total number of pulses during the acquisition time. If the pulse density is greater than a preset value, it is marked as potentially active discharge.

5. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 4, characterized in that, Anomaly scores are calculated based on the normalized frequency band energy ratio, the sub-band energy, and the pulse density, including the following steps: A weight vector group is set, and scores corresponding to the frequency band energy ratio, the sub-band energy and the pulse density are calculated based on the weight vector group. Several scores are added together to obtain the anomaly score, wherein the anomaly score is between 0 and 1, and the closer it is to 1, the more severe the anomaly.

6. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 1, characterized in that, Based on historical scores, a trend direction is generated. Combined with the abnormal scores, fault levels are categorized. Based on the fault level, a corresponding early warning notification is triggered and sent to the corresponding recipient. This includes the following steps: Obtain the historical scores corresponding to the preset first historical duration to form a time score sequence; The differences between adjacent historical scores in the time-based scoring sequence are summed to obtain a trend vector, and the trend direction is determined based on the magnitude of the trend vector. The level range corresponding to the values ​​of the abnormal scores is selected to determine the basic level, and the basic level is adjusted according to the trend direction to obtain the final fault level. Match the corresponding communication method and receiving object according to the fault level; The degradation rate is calculated by dividing the difference between the current anomaly score and the historical score two historical periods ago by the second historical period. Based on the fault level and the degradation rate, a response strength index is calculated using a preset allocation coefficient, and the communication method is adjusted based on the response strength index.

7. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 6, characterized in that, The base level is adjusted according to the trend direction to obtain the final fault level, specifically including the following steps: If the trend is upward, upgrade the base level; If the trend direction is not upward, maintain the current base level.

8. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 1, characterized in that, Establishing a health record includes the following steps: Calculate the age of the joint; A statistical period is set, and the variance index corresponding to the abnormal score and the average score at each time point is calculated based on the number of data points within the statistical period, wherein the number of data points represents the number of abnormal scores within the statistical period. Calculate the mean score and standard deviation of the scores for the same batch of interfaces within the same specified time period, and calculate the relative degradation index based on the mean score and the standard deviation of the scores; The highest score, the number of warnings after normalization, and the average rate of deterioration are summarized on a monthly basis, and a comprehensive health score is calculated by combining them with preset weighting coefficients. The health record is generated based on the joint age, the variance index, the relative deterioration index, and the comprehensive health score.

9. The method for monitoring cable joint faults in power distribution networks based on acoustic signature analysis according to claim 5, characterized in that, Implement system performance evaluation and dynamic parameter adaptation to verify effectiveness by statistically analyzing early warning response rates and handling results, including the following steps: The effective warning rate is calculated by statistically analyzing the number of warnings in the health records over a preset historical period and comparing it with the number of on-site confirmed hazard warnings. The weight adjustment factor for updating the weight vector group is calculated based on the contribution energy of each feature in the false alarm case, wherein the features include the band energy ratio, the subband energy, and the pulse density; Obtain the duration of the current season, and calculate the environmental noise baseline value by combining the root mean square value of the reference noise channel under the duration of the season. The environmental noise baseline value is used to update the filter weights. The difference between the effective early warning rate before and after the update is calculated to generate parameter iteration revenue, and the effectiveness of the action is judged based on the parameter iteration revenue.

10. A power distribution network cable joint fault monitoring system based on acoustic signature analysis, characterized in that, For implementing the method as described in any one of claims 1-9, comprising: Acoustic sensors are used to perform continuous sampling to obtain raw acoustic signals; Edge computing nodes are used to perform noise reduction, time-domain normalization, and time alignment on the original acoustic signal; The central analysis platform is used to perform short-time Fourier transform and wavelet packet decomposition, as well as to calculate anomaly scores to classify fault levels and trigger corresponding early warning notifications based on the fault levels and send them to the corresponding receiving objects. The user interaction page is used to receive the warning notification and view the generated health record, and supports status tracking and trend tracking.

Images