A method and system for detecting harmonic resonance fault of an electric meter relay based on harmonic analysis

Abstract

The application discloses a kind of electric meter relay resonance fault detection method and system based on harmonic analysis, it is related to the field of measurement electric variable, method includes: obtaining current signal and voltage signal and pre-processing, obtain the current cycle of current analysis signal and power grid harmonic spectrum;Based on power grid harmonic spectrum, the resonance risk of power grid is quantified, and the resonance sensitivity index of current cycle power grid is obtained;Pre-set candidate wavelet basis set, obtain the time-frequency focusing ability evaluation index and task adaptability index of each candidate wavelet basis in set;Based on three indexes, construct dynamic comprehensive cost function and select optimal wavelet basis;Based on optimal wavelet basis, wavelet packet decomposition is carried out to current analysis signal, and the current working state of relay is extracted fault feature and identified to judge.The application can realize the adaptive selection of wavelet basis, solve the problem that fault feature extraction is not accurate, false positive rate is high, and computing resource is not reasonably utilized in existing detection method.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology, and in particular to a method and system for detecting resonant faults in meter relays based on harmonic analysis. Background Technology

[0002] With the deepening deployment of smart grids and advanced measurement systems, smart meters have become fundamental nodes for power system data acquisition, two-way communication, and advanced applications. The safety relays inside smart meters, as key components for controlling power supply, directly affect power quality, user experience, and the safety of grid operations. Under long-term, frequent operation and complex grid environments, the mechanical components of relays, such as springs and contacts, can undergo irreversible deterioration due to electrolytic corrosion, wear, and material fatigue, causing random drift in the natural frequency of mechanical vibration within the relay. Harmonic pollution, primarily caused by nonlinear loads such as frequency converters and rectifiers, is prevalent in power grids, providing periodic excitation sources for relays. When a harmonic frequency approaches the relay's actual natural frequency, forced resonance of the mechanical structure occurs. This resonance fault can lead to relay malfunctions, such as uninstructed opening and closing, or even contact welding, coil overheating, and permanent damage, resulting in continuous power outages.

[0003] Because resonant faults are characterized by transient suddenness, weak characteristics, and high hazard, and traditional resonant fault monitoring methods based on fixed thresholds or steady-state currents are difficult to effectively capture such transient vibration signals related to specific frequencies, harmonic analysis methods based on signal processing, especially time-frequency analysis techniques that can simultaneously provide time-domain and frequency-domain information, have become an effective means of detecting resonant faults by analyzing the harmonic components and transient responses in current or voltage in real time.

[0004] In existing technologies, harmonic fault detection methods based on harmonic analysis mainly include the fixed threshold method and the fixed wavelet basis wavelet packet decomposition method. The fixed threshold method issues alarms by setting a threshold for harmonic amplitude or total harmonic distortion rate, but it cannot distinguish between background harmonics and resonant faults, resulting in a high false alarm rate. The fixed wavelet basis wavelet packet decomposition method utilizes methods such as... , Preset wavelets are used to perform time-frequency analysis on current signals and extract specific frequency band energy as features. However, in existing wavelet packet decomposition methods with fixed wavelet bases, the selection of wavelet bases depends on engineers' experience and lacks objective standards, making it difficult to adapt to different types of relays and the changing harmonic environment of the power grid. Secondly, the spectral response of the wavelet base selected by existing methods may not match the actual resonant frequency band of the relay, resulting in fault features being contaminated by background harmonics. Furthermore, since the wavelet base selected by existing methods is fixed, the use of high-complexity wavelet bases in low-risk scenarios can easily lead to unnecessary waste of computational resources and make it difficult to meet the stringent requirements of smart meter embedded platforms for low power consumption and high efficiency. Summary of the Invention

[0005] To improve the accuracy, robustness, and engineering practicality of meter relay resonance fault detection, and to achieve adaptive selection of wavelet bases, thereby overcoming the shortcomings of existing technologies such as subjective wavelet base selection, poor frequency band adaptability, and uneven resource utilization, this invention provides a method and system for meter relay resonance fault detection based on harmonic analysis. The technical solution is as follows:

[0006] In a first aspect, the present invention provides a method for detecting resonant faults in a meter relay based on harmonic analysis. The method includes the following steps: acquiring and preprocessing the load current signal flowing through the main contacts of the relay and the voltage signal at the meter's input port to obtain the current analysis signal and the grid harmonic spectrum for the current cycle; quantifying the grid's resonance risk based on the grid harmonic spectrum to obtain the grid's resonance sensitivity index for the current cycle; pre-setting a set of candidate wavelet bases and quantifying the ability of each candidate wavelet base to analyze transient narrowband signals to obtain a time-frequency focusing capability evaluation index for each candidate wavelet base; quantifying the energy leakage degree of each candidate wavelet base within the relay's resonant sensitive frequency band based on the pre-set set of candidate wavelet bases to obtain a task adaptability index for each candidate wavelet base; constructing a dynamic comprehensive cost function and selecting the optimal wavelet base based on the grid's resonance sensitivity index for the current cycle, the time-frequency focusing capability evaluation index for each candidate wavelet base, and the task adaptability index for each candidate wavelet base; and performing wavelet packet decomposition on the current analysis signal based on the optimal wavelet base to extract fault features and identify the current operating state of the relay.

[0007] Specifically, the natural exponential function is used to reverse map the resonance sensitivity index of the current cycle power grid. The mapped value is multiplied by the task adaptability index and then added to the time-frequency focusing capability evaluation index to form a dynamic comprehensive cost function. The candidate wavelet basis corresponding to the minimum value of the dynamic comprehensive cost function is selected from the preset candidate wavelet basis set as the optimal wavelet basis.

[0008] Preferably, with a content of not less than 20 The sampling frequency is set to periodically and in real-time acquire the load current signal flowing through the main contacts of the relay using a current transformer. At the same sampling frequency, the voltage signal at the meter's input port is periodically and in real-time acquired using a resistor divider network. The load current and voltage signals are digitized separately using an analog-to-digital converter to obtain current and voltage data sequences. A digital high-pass filter is used to remove DC offset and low-frequency drift from the current data sequence, and an adaptive notch filter is used to precisely suppress 50°C. The power frequency component is used to obtain the current analysis signal for the current cycle. The voltage data sequence is processed using the Fast Fourier Transform method to extract the frequency and amplitude information of the power grid background harmonics contained in the voltage signal, and generate the power grid harmonic spectrum for the current cycle.

[0009] Preferably, based on the factory test and aging test data of the electricity meter, the inherent frequency and its distribution parameters of the mechanical structure of the relay of the same model of electricity meter are statistically obtained. The distribution parameters of the inherent frequency include the statistical mean, variance and standard deviation of the inherent frequency. Based on the power grid harmonic spectrum of the current period, the number of all harmonics, the amplitude of all harmonics and the frequency of all harmonics in the power grid harmonic spectrum are statistically obtained. Based on the actual application scenario, an upper limit for the number of harmonics is set, and the harmonics exceeding the upper limit and their corresponding amplitude and frequency data in the power grid harmonic spectrum are removed to obtain the number of each harmonic, the amplitude and frequency of each harmonic in the remaining range.

[0010] Preferably, the maximum value among the amplitudes of each harmonic is extracted, and the ratio between the amplitude of a harmonic and the maximum value is taken as the relative intensity of that harmonic. Similarly, the relative intensities of each harmonic are obtained. Using a Gaussian kernel function, the square of the difference between the frequency of a harmonic and the statistical mean of the natural frequency is first calculated to obtain the frequency deviation corresponding to that harmonic. Then, the ratio between the frequency deviation and twice the variance of the natural frequency is calculated, and the natural exponential function is used to perform a reverse mapping of the comparison value. The mapped value is taken as the natural frequency fluctuation factor corresponding to that harmonic. Similarly, the natural frequency fluctuation factors corresponding to each harmonic are obtained. The product of the relative intensity of each harmonic and the corresponding natural frequency fluctuation factor is taken as the risk contribution of each harmonic. Based on the upper limit of the harmonic order, the risk contributions of each harmonic are accumulated, and the accumulated value is taken as the resonance sensitivity index of the power grid in the current cycle.

[0011] Preferably, based on the types and performance of existing wavelet bases, multiple commonly used wavelet bases suitable for relay resonance fault detection are selected as a candidate wavelet base set. The time-domain expression and frequency-domain expression, as well as the time-domain energy center and frequency-domain energy center of each candidate wavelet base in the candidate wavelet base set are obtained, and the time-domain variance and frequency-domain variance of each candidate wavelet base are calculated. The product of the time-domain variance and frequency-domain variance of a certain candidate wavelet base is used as the evaluation index of the time-frequency focusing capability of the candidate wavelet base. Similarly, the evaluation index of the time-frequency focusing capability of each candidate wavelet base in the candidate wavelet base set is obtained.

[0012] Preferably, based on the statistical mean and standard deviation of the natural frequencies, the three-sigma law is used. The difference between the statistical mean and three times the standard deviation of the natural frequencies is taken as the lower limit of the relay resonant sensitive frequency band, and the sum of the statistical mean and three times the standard deviation of the natural frequencies is taken as the upper limit of the relay resonant sensitive frequency band, thus obtaining the bandwidth of the relay resonant sensitive frequency band. Based on a preset set of candidate wavelet bases, the frequency domain expression of each candidate wavelet base is obtained. The frequency domain expression of each candidate wavelet base is integrated over the bandwidth of the relay resonant sensitive frequency band to obtain the resonant energy of each candidate wavelet base within the relay resonant sensitive frequency band. The frequency domain expression of each candidate wavelet base is integrated over the entire frequency band, and the full frequency band range is [missing information]. The total energy corresponding to each candidate wavelet basis is obtained. The ratio of the resonant energy of each candidate wavelet basis to the corresponding total energy is taken as the energy retention rate of each candidate wavelet basis. The energy leakage rate is obtained by subtracting the energy retention rate from 1. The energy leakage rate of each candidate wavelet basis is taken as the task adaptability index of each corresponding candidate wavelet basis.

[0013] Preferably, the resonance sensitivity index of the power grid in the current cycle, the time-frequency focusing capability evaluation index of each candidate wavelet basis, and the task adaptability index of each candidate wavelet basis are extracted. Based on the constructed dynamic comprehensive cost function, the function value corresponding to each candidate wavelet basis in the candidate wavelet basis set is calculated in sequence, the minimum value among the function values ​​is selected, and the candidate wavelet basis corresponding to the minimum value is taken as the optimal wavelet basis for the current cycle.

[0014] Preferably, the current analysis signal of the current period is decomposed into a set of sub-band signals covering the entire frequency band using the optimal wavelet basis of the current period. Based on the bandwidth range of the relay resonance sensitive frequency band, all sub-band signals corresponding to the bandwidth range are identified and extracted as the key sub-band set. Based on Passevar's theorem, the energy value of each sub-band signal in the key sub-band set is calculated, and the energy value of each sub-band signal is normalized by summation normalization. The normalized energy values ​​of each sub-band signal constitute a set of feature vectors for the current period.

[0015] Preferably, a classifier model is constructed using a support vector machine or a rule criterion module. The classifier model is trained using experimental data on relay resonance faults from the meter's factory testing and aging experiments. The feature vector of the current cycle is input into the trained classifier model. The classifier model identifies the characteristic pattern of the relay in the current cycle and determines the current operating state of the relay based on the feature pattern. When the feature vector of the current cycle is significantly concentrated within the bandwidth of the relay's resonance sensitive frequency band and the duration conforms to transient characteristics, the classifier model determines that the relay has a resonance fault in the current cycle and outputs a fault alarm. Otherwise, the classifier model determines that the relay is in normal operating condition in the current cycle.

[0016] Secondly, the present invention provides a meter relay resonance fault detection system based on harmonic analysis, used to implement the above-mentioned meter relay resonance fault detection method based on harmonic analysis, comprising: a processor, a memory, a communication interface, and a data acquisition device. The processor stores computer program instructions for implementing the above-mentioned meter relay resonance fault detection method based on harmonic analysis, and the communication interface is communicatively connected to the data acquisition device.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] This invention constructs a resonance sensitivity index, combining the statistical characteristics of relay aging with the real-time power grid harmonic environment, to achieve a quantitative assessment of resonance risk. This provides a reliable scenario-aware basis for subsequent adaptive analysis, improving the targeting and environmental adaptability of fault detection. Simultaneously, by pre-setting a set of candidate wavelet bases and constructing time-frequency focusing capability evaluation indicators and task adaptability indicators for each candidate wavelet base, the invention objectively evaluates the candidate wavelet bases from two dimensions: general signal processing performance and specific physical frequency band adaptability. Combined with the resonance sensitivity index of the power grid in the current cycle, the optimal wavelet base suitable for the current cycle can be selected, achieving adaptive wavelet base selection. This overcomes the limitations of traditional methods that rely on experience to select wavelet bases or use fixed wavelet bases, and solves the defects of subjective wavelet base selection, poor frequency band adaptability, and uneven resource utilization in existing technologies. It enhances the rationality and interpretability of feature extraction, thereby effectively improving the accuracy, robustness, and engineering practicality of meter relay resonance fault detection. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an implementation method for a meter relay resonance fault detection method based on harmonic analysis, according to an embodiment of this application.

[0020] Figure 2 This is a structural block diagram of a meter relay resonance fault detection system based on harmonic analysis, according to an embodiment of this application. Detailed Implementation

[0021] The technical features of the present invention will be further described in detail below with reference to the accompanying drawings so that those skilled in the art can understand them.

[0022] A method for detecting resonant faults in meter relays based on harmonic analysis, the implementation process of which is as follows: Figure 1 As shown, the specific implementation steps are as follows:

[0023] Step S1: Obtain the load current signal flowing through the main contacts of the relay and the voltage signal at the input port of the meter, and perform preprocessing to obtain the current analysis signal and the power grid harmonic spectrum for the current cycle.

[0024] Specifically, with a minimum of 20 The sampling frequency is set to periodically and in real-time acquire the load current signal flowing through the main contacts of the relay using a current transformer. At the same sampling frequency, the voltage signal at the meter's input port is periodically and in real-time acquired using a resistor divider network. The load current and voltage signals are digitized separately using an analog-to-digital converter to obtain current and voltage data sequences. A digital high-pass filter is used to remove DC offset and low-frequency drift from the current data sequence, and an adaptive notch filter is used to precisely suppress 50°C. The power frequency component is used to obtain the current analysis signal for the current cycle. The current analysis signal is a set of high-frequency components with zero mean and high signal-to-noise ratio. It retains only the harmonics and transient information closely related to relay resonance faults and serves as the sole input for subsequent wavelet packet decomposition, effectively avoiding interference from irrelevant components in feature extraction. The voltage data sequence is processed using the Fast Fourier Transform method to extract the frequency and amplitude information of the power grid background harmonics contained in the voltage signal, generating the power grid harmonic spectrum for the current cycle.

[0025] The data acquisition period can be set to 200ms, which can fully capture the resonance decay process and meet the real-time requirements of diagnosis. The load current signal is used to capture the abnormal current oscillation during relay operation transients and resonance faults. It can detect whether the relay has actually experienced abnormal vibration. Once the relay generates a small mechanical vibration due to resonance, it will directly modulate the current flowing through it, forming a unique high-frequency disturbance. Therefore, the load current signal is the core basis for resonance fault diagnosis. The acquired voltage signal is used to determine whether there are harmonic components in the external power grid that may cause relay vibration. Since harmonics are mainly reflected in the voltage waveform, they can reflect whether there is a dangerous excitation. Therefore, the resonance risk of the power grid can be assessed by acquiring the voltage signal, and the resonance fault of the relay can be determined by the load current signal.

[0026] Furthermore, since the DC offset and low-frequency drift components in the load current signal are usually caused by sensor zero drift, temperature changes, or slow fluctuations in grid voltage, they occupy the low-frequency subband energy of wavelet decomposition and interfere with the judgment of resonance characteristics. Therefore, a digital high-pass filter is needed for detrending term processing. Preferably, this invention uses a digital high-pass filter with a cutoff frequency of 0.5Hz. Simultaneously, since the power frequency component and its energy in the load current signal are much higher than harmonics and possible resonance signals, if not filtered out, their enormous energy will dominate the wavelet packet decomposition results, leading to relatively low energy and extremely low signal-to-noise ratio in the high-frequency subband used to characterize faults, making effective detection difficult. Therefore, an adaptive notch filter is needed to accurately suppress 50% of the signal. Power frequency component; In this invention, only the load current signal is filtered because the power frequency component and slow drift contained in the load current signal will mask the weak high-frequency signal generated when the relay vibrates abnormally. It must be removed in order to effectively identify the fault. The voltage signal is used to analyze whether there are harmonics that induce vibration in the power grid. This process needs to retain the complete power frequency and its harmonic structure, so there is no need to filter it.

[0027] Step S2: Based on the power grid harmonic spectrum, quantify the resonance risk of the power grid and obtain the resonance sensitivity index of the power grid in the current cycle.

[0028] Since relay resonance is essentially a forced resonance phenomenon caused by the approach of the grid voltage harmonic excitation frequency to the natural frequency of the relay's mechanical structure, according to the failure physics model, the relay is most prone to resonance when the harmonic excitation frequency is close to the natural frequency. However, the natural frequency can drift due to factors such as spring aging and contact wear, and cannot be directly measured in real time during meter operation. Therefore, it is necessary to obtain the distribution parameters of the relay's natural frequency, namely the statistical mean, variance, and standard deviation of the natural frequency, through the meter's factory test and aging experiment data. Combined with the grid harmonic spectrum extracted from the real-time voltage signal, the overall risk level of relay resonance caused by the current grid environment can be quantified without relying on direct measurement, providing a key decision-making basis for subsequent adaptive selection of wavelet basis.

[0029] Specifically, based on the factory test and aging test data of the electricity meter, the inherent frequency and its distribution parameters of the mechanical structure of the relay of the same model of electricity meter are statistically obtained. The distribution parameters of the inherent frequency include the statistical mean, variance and standard deviation of the inherent frequency. Based on the power grid harmonic spectrum of the current period, the number of all harmonics, the amplitude of all harmonics and the frequency of all harmonics within the power grid harmonic spectrum are statistically obtained. Based on the actual application scenario, an upper limit for the number of harmonics is set, and the harmonics exceeding the upper limit and their corresponding amplitude and frequency data in the power grid harmonic spectrum are removed to obtain the number of each harmonic, the amplitude and frequency of each harmonic in the remaining range.

[0030] The process for obtaining the natural frequency and its distribution parameters of the electric meter relay mechanical structure is as follows: During the factory testing phase of the electric meter, a batch of relay samples are selected from electric meters of the same model from different batches and different aging test stages. A small vibration table or electromagnetic exciter is used to apply wideband sweep vibration to multiple relays. At the same time, an accelerometer or laser vibrometer is used to monitor the response of the housing or contacts of multiple relays. The frequency with the largest response amplitude is the natural frequency of the corresponding individual relay. Subsequently, these frequency data are statistically analyzed to obtain an average value and fluctuation range representing the overall characteristics of the relay model, that is, the statistical mean, variance, and standard deviation of the natural frequency are obtained. These statistical results are pre-stored in the electric meter system. When the electric meter is running in the field, it is not necessary to measure the resonant frequency of its own relay. Instead, the pre-stored statistical parameters are directly called, thereby combining the current harmonic situation in the power grid to assess the risk of abnormal vibration of the relay in real time.

[0031] Furthermore, the maximum value among the amplitudes of each harmonic is extracted, and the ratio between the amplitude of a harmonic and its maximum value is taken as the relative intensity of that harmonic. Similarly, the relative intensities of each harmonic are obtained. Using a Gaussian kernel function, the square of the difference between the frequency of a harmonic and the statistical mean of the natural frequency is first calculated to obtain the frequency deviation corresponding to that harmonic. Then, the ratio between the frequency deviation and twice the variance of the natural frequency is calculated, and the natural exponential function is used to perform a reverse mapping of the comparison value. The mapped value is taken as the natural frequency fluctuation factor corresponding to that harmonic. Similarly, the natural frequency fluctuation factors corresponding to each harmonic are obtained. The product of the relative intensity of each harmonic and the corresponding natural frequency fluctuation factor is taken as the risk contribution of each harmonic. Based on the upper limit of the harmonic order, the risk contributions of each harmonic are accumulated, and the accumulated value is taken as the resonance sensitivity index of the power grid in the current cycle.

[0032] The resonance sensitivity index of the power grid in the current cycle is denoted as A, and its calculation formula is as follows:

[0033]

[0034] In the formula, k represents the harmonic order, and K represents the upper limit of the harmonic order. This represents the amplitude of the kth harmonic in the power grid harmonic spectrum. This represents the maximum amplitude of each harmonic within the harmonic order limit of the power grid harmonic spectrum. This represents the frequency of the kth harmonic. The statistical mean of the relay's natural frequency. The variance of the relay's natural frequency is represented by , and n represents the number of relay samples. This represents the natural exponential function. This represents the relative intensity of the kth harmonic. This represents the natural frequency fluctuation factor corresponding to the k-th harmonic. Since harmonic analysis usually focuses on odd harmonics, therefore... K is an odd number and is usually set to 15.

[0035] Relative intensity is used to measure the excitation intensity of harmonics. If the relative intensity of a certain harmonic is close to 1, it indicates that this harmonic is the dominant harmonic in the current power grid, with strong excitation capability and a substantial contribution to the resonance risk. The natural frequency fluctuation factor is used to determine whether the frequency of the corresponding harmonic is close to the resonant sensitive range of the relay. Since the natural frequency of the relay is affected by manufacturing and aging factors, it is not a single fixed value, but fluctuates within a certain range according to a normal distribution, centered on the statistical mean of the natural frequency. Its dispersion is characterized by the variance of the natural frequency. Therefore, it can be quantified by using the calculation formula of the Gaussian kernel function. The possibility that the frequency of a subharmonic falls within a high-density region of a normal distribution. The closer to The closer the Gaussian kernel function value is to 1, the higher the matching degree. The closer the frequency of the subharmonic is to the resonant sensitive range of the relay.

[0036] Because multiple harmonics typically exist simultaneously in a power grid, even if none of the harmonics has a frequency exactly equal to... However, if the frequencies of multiple harmonics are distributed in Nearby, the combined excitation of these harmonics can still induce a significant resonant response, whereas if only a few harmonics have frequencies close to [the target frequency], [the response may be negative]. Furthermore, the amplitude is very small, and the overall risk of the power grid in the current cycle is also low. Therefore, it is necessary to comprehensively consider the relative intensity and inherent frequency fluctuation factor to measure the risk contribution of each harmonic. Then, the risk contributions of each harmonic are added together to form an overall risk assessment value, which yields the resonance sensitivity index of the power grid in the current cycle. The resonance sensitivity index is used to reflect the comprehensive risk level of the current power grid harmonics to cause resonance faults in relays. It provides a reasonable, robust, and engineering-achievable risk quantification basis for adaptive selection of wavelet bases. The larger the value, the more likely it is that there are harmonics with strong amplitudes and frequencies close to the resonance sensitivity range of relays in the current power grid, and the higher the risk of resonance. The smaller the value, the weaker the amplitude of the harmonics or the frequency of the harmonics is far from the resonance sensitivity range of relays, and the safer the power grid environment.

[0037] Step S3: Preset a set of candidate wavelet bases, quantify the ability of each candidate wavelet base in the set to analyze transient narrowband signals, and obtain the time-frequency focusing capability evaluation index of each candidate wavelet base.

[0038] In wavelet analysis, different wavelet bases, such as Xiaobo Xiaobo Wavelets, among others, possess different shapes and spectral characteristics, and these differences directly affect their ability to characterize signals. The typical characteristics of relay resonant fault signals are: brief occurrence in the time domain, typically lasting from a few milliseconds to tens of milliseconds, classifying them as transient events; and concentration in the frequency domain within a narrow frequency band, determined by the relay's mechanical structure, for example... An ideal analytical tool should be able to concentrate such a short, sharp signal into a small region in the joint time-frequency representation, rather than dispersing its energy across a large time-frequency area. If the energy is dispersed, it becomes difficult to accurately identify whether a fault has occurred, and it is easily confused with background noise. However, traditional methods mostly rely on experience to select a commonly used wavelet basis as the analytical tool, lacking objective standards to determine which wavelet basis is more suitable for the current task. Therefore, it is necessary to propose a quantitative index based entirely on the mathematical properties of the wavelet basis itself, to objectively compare the ability of different wavelet bases to analyze transient narrowband signals, thereby measuring the applicability of different wavelet bases and overcoming the limitations of subjective selection.

[0039] Specifically, based on the existing types and performance of wavelet bases, several commonly used wavelet bases suitable for relay resonance fault detection are selected as a candidate wavelet base set. The time-domain and frequency-domain expressions, as well as the time-domain and frequency-domain energy centers, of each candidate wavelet base in the candidate wavelet base set are obtained, and the time-domain and frequency-domain variances of each candidate wavelet base are calculated. The product of the time-domain and frequency-domain variances of a candidate wavelet base is used as the evaluation index of the time-frequency focusing capability of that candidate wavelet base. Similarly, the evaluation index of the time-frequency focusing capability of each candidate wavelet base in the candidate wavelet base set is obtained.

[0040] The evaluation index for the time-frequency focusing capability of the m-th candidate wavelet basis is: The calculation formula is as follows:

[0041]

[0042]

[0043]

[0044] In the formula, This represents the m-th candidate wavelet basis within the set of candidate wavelet basis bases. Describes the m-th candidate wavelet basis The time-domain variance, Describes the m-th candidate wavelet basis The frequency domain variance, where t represents the time variable and its value ranges from 1 to 2. , Describes the m-th candidate wavelet basis The time-domain energy center, , Let m represent the time-domain expression of the m-th candidate wavelet basis. Represents a frequency variable and its value range is , Describes the m-th candidate wavelet basis The frequency domain energy center, Let represent the frequency domain expression of the m-th candidate wavelet basis, which can be obtained by Fourier transforming the corresponding time domain expression.

[0045] Time-domain variance measures the concentration of a wavelet basis along the time axis. A smaller time-domain variance indicates that the corresponding wavelet basis is compact in the time domain, with energy concentrated within a very short time window, resulting in high time resolution. A larger time-domain variance indicates that the corresponding wavelet basis has a longer time tail, resulting in ambiguous time positioning. Frequency-domain variance measures the concentration of a wavelet basis along the frequency axis. A smaller frequency-domain variance indicates that the corresponding wavelet basis is sharp in the frequency domain, mainly responding to a narrow frequency band, resulting in high frequency resolution. A larger frequency-domain variance indicates that the corresponding wavelet basis has a broad spectrum, responding to multiple frequencies, resulting in poor frequency selectivity.

[0046] Because there is usually an inverse relationship between the concentration of wavelet basis functions in the time domain and the concentration in the frequency domain, specifically, the time-domain variance reflects the degree of energy diffusion of the wavelet basis along the time axis. The smaller the value, the stronger the time locality, which is more conducive to accurately locating the occurrence time of transient events, but its spectrum is often wider and the frequency selectivity is poorer. The frequency-domain variance reflects the degree of energy diffusion of the wavelet basis along the frequency axis. The smaller the value, the stronger the frequency selectivity, which is more conducive to distinguishing narrowband resonant components, but it usually shows a longer tail in the time domain, reducing the time positioning capability. Therefore, for the resonant fault signal of a relay, an ideal wavelet basis should have both small time-domain variance and small frequency-domain variance. Thus, the product of the two is used as a comprehensive evaluation index to measure the overall focusing capability of the wavelet basis in the joint time-frequency domain. The smaller the value of the time-frequency focusing capability evaluation index, the more concentrated the corresponding candidate wavelet basis is in both time and frequency dimensions. It can tightly concentrate the energy of the transient narrowband resonant fault signal in the local area of ​​the time-frequency map, thereby improving feature clarity and detection accuracy. Conversely, the larger the value, the more significant the dispersion of the corresponding candidate wavelet basis is in the time or frequency domain dimensions, resulting in signal energy diffusion, feature ambiguity, and susceptibility to noise interference, which may lead to misjudgment or missed judgment.

[0047] Step S4: Based on the preset set of candidate wavelet bases, quantify the energy leakage degree of each candidate wavelet base in the set within the relay resonant sensitive frequency band to obtain the task adaptability index of each candidate wavelet base.

[0048] In wavelet packet decomposition, it is crucial to consider not only the general time-frequency characteristics of the wavelet basis but also the degree of matching between the wavelet basis and the specific fault frequency band. Even if two wavelet bases have similar time-frequency focusing capabilities, their energy response characteristics for a specific target frequency band may still differ significantly. For relay resonant fault signals, the energy is mainly concentrated within the resonant sensitive frequency band determined by the relay's mechanical structure, such as the relay resonant sensitive frequency band with a bandwidth of 750Hz~950Hz. If the energy response of the selected wavelet basis is excessively diffused within this frequency band, even with good global focusing performance, a large amount of background harmonics or noise energy will be introduced into the critical sub-frequency band, leading to contamination of fault characteristics and reduced diagnostic reliability. Therefore, a task suitability index is proposed to quantify the degree of energy leakage of the wavelet basis within the relay resonant sensitive frequency band, thereby selecting the most targeted analytical tool for meter relay resonant faults from the candidate wavelet basis set.

[0049] Specifically, based on the statistical mean and standard deviation of the natural frequencies, the three-sigma law is used. The difference between the statistical mean and three times the standard deviation of the natural frequencies is taken as the lower limit of the relay resonant sensitive frequency band, and the sum of the statistical mean and three times the standard deviation of the natural frequencies is taken as the upper limit of the relay resonant sensitive frequency band, thus obtaining the bandwidth of the relay resonant sensitive frequency band. Based on a preset set of candidate wavelet bases, the frequency domain expression of each candidate wavelet base is obtained. The frequency domain expression of each candidate wavelet base is integrated over the bandwidth of the relay resonant sensitive frequency band to obtain the resonant energy of each candidate wavelet base within the relay resonant sensitive frequency band. The frequency domain expression of each candidate wavelet base is integrated over the entire frequency band, and the full frequency band range is... The total energy corresponding to each candidate wavelet basis is obtained. The ratio of the resonant energy of each candidate wavelet basis to the corresponding total energy is taken as the energy retention rate of each candidate wavelet basis. The energy leakage rate is obtained by subtracting the energy retention rate from 1. The energy leakage rate of each candidate wavelet basis is taken as the task adaptability index of each corresponding candidate wavelet basis.

[0050] Wherein, the task fit index of the m-th candidate wavelet basis is The calculation formula is as follows:

[0051]

[0052] In the formula, This represents the m-th candidate wavelet basis within the set of candidate wavelet basis bases. Let m be the frequency domain expression of the m-th candidate wavelet basis, and let the bandwidth of the relay resonant sensitive frequency band be... , The statistical mean of the relay's natural frequency. The standard deviation of the relay's natural frequency is represented by n, and n represents the number of relay samples.

[0053] Task suitability index reflects the concentration of energy response to the target fault frequency band when the corresponding candidate wavelet basis is used as an analysis tool, directly determining the purity and anti-interference ability of feature extraction; among them, the numerator term The task adaptability index is used to reflect the energy concentration of the m-th candidate wavelet basis within the relay resonance sensitive frequency band. The larger the value, the stronger the response of the wavelet basis to the target fault frequency, and the more conducive it is to highlighting the resonance characteristics. The denominator is the total energy of the wavelet basis, which is used for normalization to make the index dimensionless and comparable. Therefore, the smaller the value of the task adaptability index, the more concentrated the wavelet basis energy is within the relay resonance sensitive frequency band, the stronger the selectivity to the target resonance frequency band, and the more conducive it is to extracting pure fault characteristics. The larger the value of the task adaptability index, the more dispersed the wavelet basis energy is in non-sensitive frequency bands, which makes it easier to couple the energy of the power grid background harmonics into the key sub-frequency band, causing feature confusion and increasing the risk of false alarms.

[0054] Step S5: Based on the three indicators, construct a dynamic comprehensive cost function and select the optimal wavelet basis.

[0055] After obtaining the resonance sensitivity index of the current power grid, the time-frequency focusing capability evaluation index of each candidate wavelet basis, and the task adaptability index, a scenario-driven adaptive wavelet basis selection method needs to be implemented to dynamically determine the analysis tool most suitable for the current operating conditions. The core idea of ​​this method is: in high-risk scenarios, priority is given to ensuring analysis accuracy, and the wavelet basis with the strongest time-frequency focusing capability and the most concentrated response to the resonance frequency band is selected; in low-risk scenarios, priority is given to suppressing energy leakage, and the wavelet basis with the most concentrated energy in the resonance sensitive frequency band is selected, thereby improving feature purity and indirectly optimizing computational efficiency.

[0056] Specifically, the natural exponential function is used to reverse-map the resonance sensitivity index of the current period's power grid. The mapped value is multiplied by the task adaptability index and then added to the time-frequency focusing capability evaluation index to form a dynamic comprehensive cost function. The resonance sensitivity index of the current period's power grid, the time-frequency focusing capability evaluation index of each candidate wavelet basis, and the task adaptability index of each candidate wavelet basis are extracted. Based on the constructed dynamic comprehensive cost function, the function value corresponding to each candidate wavelet basis in the candidate wavelet basis set is calculated sequentially. The minimum value among the function values ​​is selected, and the candidate wavelet basis corresponding to the minimum value is taken as the optimal wavelet basis for the current period.

[0057] The expression for obtaining the optimal wavelet basis based on the dynamic synthesis cost function is as follows:

[0058]

[0059] In the formula, This represents the optimal wavelet basis for the current period. Denotes the set of candidate wavelet bases. This represents the m-th candidate wavelet basis within the set of candidate wavelet basis bases. This indicates the resonance sensitivity index of the power grid during the current cycle. Let represent the time-frequency focusing capability evaluation index of the m-th candidate wavelet basis. Let represent the task fit index of the m-th candidate wavelet basis. This represents the natural exponential function. An index function for the independent variable representing the minimum value.

[0060] for In terms of the calculation formula, A reflects the possibility of coupling resonance between the current power grid harmonic excitation and the relay's natural frequency; the higher the value of A, the greater the possibility that the system is in a high-risk state, and the higher the accuracy requirement for signal analysis. Reflects the m-th candidate wavelet basis Overall focusing capability in the time-frequency joint domain The smaller the value, the better the candidate wavelet basis is. The stronger the ability to focus the energy of transient narrowband signals in the time-frequency joint domain, the better it is for extracting weak resonance characteristics; Reflects the m-th candidate wavelet basis Adaptability to the relay resonance sensitive frequency band is the key to achieving high signal-to-noise ratio feature extraction under low risk. Let m be the dynamic synthesis cost function of the m-th candidate wavelet basis, used to reflect the selection of the wavelet basis. As for the overall cost of the analysis tool in the current scenario, when A is high... Approaching 0, the cost is mainly due to The dominant approach is to select the wavelet basis with the best global focusing performance; when A is low, Approaching 1, High wavelet gene penalty terms are increased and suppressed, tending to select wavelet bases with more concentrated energy in the resonant sensitive frequency band; the candidate wavelet base set is traversed to find the wavelet base that minimizes the synthesis cost as the optimal wavelet base for subsequent wavelet packet decomposition.

[0061] Step S6: Perform wavelet packet decomposition on the current analysis signal based on the optimal wavelet basis, extract fault features, and identify and determine the current working state of the relay.

[0062] Specifically, the current analysis signal of the current period is decomposed using the optimal wavelet basis of the current period to obtain a set of sub-band signals covering the entire frequency band. Based on the bandwidth range of the relay resonance sensitive frequency band, all sub-band signals within the bandwidth range are identified and extracted as the key sub-band set. Based on Passevar's theorem, the energy value of each sub-band signal in the key sub-band set is calculated, and the energy value of each sub-band signal is normalized by summation normalization. The normalized energy values ​​of each sub-band signal constitute a set of feature vectors for the current period. Since the feature vectors of the current period are decomposed using the optimal wavelet basis of the current period, their energy distribution has a higher signal-to-noise ratio and fault sensitivity, which can effectively distinguish between the normal state caused by power grid background harmonics and the resonance fault caused by relay parameter degradation.

[0063] Furthermore, a classifier model is constructed using support vector machines or rule-based criterion modules. The model is trained using experimental data on relay resonance faults from factory testing and aging experiments of the electricity meter. The feature vector of the current cycle is input into the trained classifier model, which identifies the relay's characteristic patterns in the current cycle and determines the relay's current operating state based on these patterns. When the feature vector of the current cycle is significantly concentrated within the bandwidth of the relay's resonance-sensitive frequency band and its duration conforms to transient characteristics, the classifier model determines that the relay has a resonance fault in the current cycle and outputs a fault alarm. Otherwise, the classifier model determines that the relay is in normal working condition in the current cycle. Simultaneously, the diagnostic results of the relay's current operating state can be reported to the electricity information collection master station or edge agent device via the smart meter's communication module for triggering early warnings, remote maintenance, or data archiving, achieving closed-loop monitoring and proactive management of the relay's health status.

[0064] This invention also discloses a meter relay resonance fault detection system based on harmonic analysis, used to implement the aforementioned meter relay resonance fault detection method based on harmonic analysis. The system structure is as follows: Figure 2 As shown, the device includes: a processor, a memory, a communication interface, and a data acquisition device. The processor stores computer program instructions for implementing the aforementioned method for detecting resonant faults in a meter relay based on harmonic analysis. The communication interface is connected to the data acquisition device. The data acquisition device includes a current transformer, a resistor divider network, and an analog-to-digital converter. The current transformer is used to acquire the load current signal flowing through the main contacts of the relay, the resistor divider network is used to acquire the voltage signal at the meter's input port, and the analog-to-digital converter is used to convert the acquired analog signal into a digital signal. The data acquisition device requires a high-precision, wide-bandwidth current transformer, and the sampling frequency of the current transformer and the resistor divider network is not less than 20 Hz. The resolution of the analog-to-digital converter should be no less than 16 bits.

[0065] The embodiments included in this invention are merely descriptions of preferred embodiments of the invention and are not limited to the precise structures described above and shown in the accompanying drawings. Various modifications and changes can be made without departing from the scope of protection. Any variations and improvements made by those skilled in the art to the technical solutions of this invention without departing from the design concept of this invention should fall within the scope of protection of this invention.

Claims

1. A method for detecting resonant faults in meter relays based on harmonic analysis, characterized in that: The load current signal flowing through the main contacts of the relay and the voltage signal at the input port of the meter are acquired and preprocessed to obtain the current analysis signal and the power grid harmonic spectrum for the current cycle. Based on the power grid harmonic spectrum, the resonance risk of the power grid is quantified to obtain the resonance sensitivity index of the power grid for the current cycle. A pre-defined set of candidate wavelet bases is used to quantify the ability of each candidate wavelet base to analyze transient narrowband signals, resulting in a time-frequency focusing capability evaluation index for each candidate wavelet base. Based on the pre-defined set of candidate wavelet bases, the energy leakage degree of each candidate wavelet base within the relay resonance sensitive frequency band is quantified, resulting in a task adaptability index for each candidate wavelet base. Based on the resonance sensitivity index of the current cycle power grid, the time-frequency focusing capability evaluation index of each candidate wavelet base, and the task adaptability index of each candidate wavelet base, a dynamic comprehensive cost function is constructed, and the optimal wavelet base is selected. Based on the optimal wavelet base, wavelet packet decomposition is performed on the current analysis signal to extract fault features and identify the current operating state of the relay. Among them, the natural exponential function is used to reverse map the resonance sensitivity index of the current cycle power grid. The mapped value is multiplied by the task adaptability index and added to the time-frequency focusing capability evaluation index to form a dynamic comprehensive cost function. The candidate wavelet basis corresponding to the minimum value of the dynamic comprehensive cost function is selected from the preset candidate wavelet basis set as the optimal wavelet basis. Based on the existing types and performance of wavelet bases, several commonly used wavelet bases suitable for relay resonance fault detection are selected as a candidate wavelet base set. The time-domain and frequency-domain expressions, as well as the time-domain and frequency-domain energy centers, of each candidate wavelet base in the candidate wavelet base set are obtained, and the time-domain and frequency-domain variances of each candidate wavelet base are calculated. The product of the time-domain and frequency-domain variances of a candidate wavelet base is used as the evaluation index of the time-frequency focusing capability of that candidate wavelet base. Similarly, the evaluation index of the time-frequency focusing capability of each candidate wavelet base in the candidate wavelet base set is obtained.

2. The method for detecting resonant faults in a meter relay based on harmonic analysis according to claim 1, characterized in that, The process of acquiring and preprocessing the load current signal flowing through the main contacts of the relay and the voltage signal at the meter's input port to obtain the current analysis signal and the power grid harmonic spectrum for the current cycle includes: using a frequency of not less than 20... The sampling frequency is set to periodically and in real-time acquire the load current signal flowing through the main contacts of the relay using a current transformer. At the same sampling frequency, the voltage signal at the meter's input port is periodically and in real-time acquired using a resistor divider network. The load current and voltage signals are digitized separately using an analog-to-digital converter to obtain current and voltage data sequences. A digital high-pass filter is used to remove DC offset and low-frequency drift from the current data sequence, and an adaptive notch filter is used to precisely suppress 50°C. The power frequency component is used to obtain the current analysis signal for the current cycle. The voltage data sequence is processed using the Fast Fourier Transform method to extract the frequency and amplitude information of the power grid background harmonics contained in the voltage signal, and generate the power grid harmonic spectrum for the current cycle.

3. The method for detecting resonant faults in a meter relay based on harmonic analysis according to claim 1, characterized in that, The method of quantifying the resonance risk of the power grid based on the power grid harmonic spectrum and obtaining the resonance sensitivity index of the power grid in the current cycle includes: statistically obtaining the inherent frequency and its distribution parameters of the mechanical structure of the relay of the same model of electricity meter based on the factory test and aging test data of the meter. The distribution parameters of the inherent frequency include the statistical mean, variance and standard deviation of the inherent frequency; statistically obtaining the number of all harmonics, the amplitude of all harmonics and the frequency of all harmonics in the power grid harmonic spectrum based on the current cycle; setting an upper limit for the number of harmonics based on the actual application scenario, and removing the harmonics exceeding the upper limit in the power grid harmonic spectrum and their corresponding amplitude and frequency data to obtain the number of each harmonic, the amplitude and frequency of each harmonic in the remaining range.

4. The method for detecting resonant faults in a meter relay based on harmonic analysis according to claim 3, characterized in that, The method of quantifying the resonance risk of the power grid based on the power grid harmonic spectrum and obtaining the resonance sensitivity index of the power grid in the current cycle further includes: extracting the maximum value among the amplitudes of each harmonic, taking the ratio between the amplitude of a certain harmonic and the maximum value as the relative intensity of that harmonic, and similarly obtaining the relative intensity of each harmonic; using a Gaussian kernel function, first calculating the square of the difference between the frequency of a certain harmonic and the statistical mean of the natural frequency to obtain the frequency deviation corresponding to that harmonic, then calculating the ratio between the frequency deviation and twice the variance of the natural frequency, using the natural exponential function to perform inverse mapping of the comparison value, and taking the mapped value as the natural frequency fluctuation factor corresponding to that harmonic, and similarly obtaining the natural frequency fluctuation factor corresponding to each harmonic; taking the product between the relative intensity of each harmonic and the corresponding natural frequency fluctuation factor as the risk contribution of each harmonic, and accumulating the risk contributions of each harmonic based on the upper limit of the harmonic order, and taking the accumulated value as the resonance sensitivity index of the power grid in the current cycle.

5. The method for detecting resonant faults in a meter relay based on harmonic analysis according to claim 3, characterized in that, The method, based on a pre-defined set of candidate wavelet bases, quantifies the energy leakage degree of each candidate wavelet base within the relay resonant sensitive frequency band, obtaining the task adaptability index of each candidate wavelet base. This index includes: based on the statistical mean and standard deviation of the natural frequencies, using the three-sigma law, the difference between the statistical mean and three times the standard deviation of the natural frequencies is taken as the lower limit of the relay resonant sensitive frequency band, and the sum of the statistical mean and three times the standard deviation of the natural frequencies is taken as the upper limit of the relay resonant sensitive frequency band, thus obtaining the bandwidth of the relay resonant sensitive frequency band; based on the pre-defined set of candidate wavelet bases, the frequency domain expression of each candidate wavelet base is obtained; the frequency domain expression of each candidate wavelet base is integrated over the bandwidth of the relay resonant sensitive frequency band to obtain the resonant energy of each candidate wavelet base within the relay resonant sensitive frequency band; and the frequency domain expression of each candidate wavelet base is integrated over the entire frequency band, where the full frequency band range is [missing information]. The total energy corresponding to each candidate wavelet basis is obtained. The ratio of the resonant energy of each candidate wavelet basis to the corresponding total energy is taken as the energy retention rate of each candidate wavelet basis. The energy leakage rate is obtained by subtracting the energy retention rate from 1. The energy leakage rate of each candidate wavelet basis is taken as the task adaptability index of each corresponding candidate wavelet basis.

6. A method for detecting resonant faults in a meter relay based on harmonic analysis according to any one of claims 1 to 5, characterized in that, The process of constructing a dynamic integrated cost function and selecting the optimal wavelet basis includes: extracting the resonance sensitivity index of the power grid in the current period, the time-frequency focusing capability evaluation index of each candidate wavelet basis, and the task adaptability index of each candidate wavelet basis; based on the constructed dynamic integrated cost function, sequentially calculating the function value corresponding to each candidate wavelet basis in the candidate wavelet basis set; selecting the minimum value among the function values; and taking the candidate wavelet basis corresponding to the minimum value as the optimal wavelet basis for the current period.

7. The method for detecting resonant faults in a meter relay based on harmonic analysis according to claim 5, characterized in that, The step of performing wavelet packet decomposition on the current analysis signal based on the optimal wavelet basis to extract fault features and identify the current operating state of the relay includes: performing wavelet packet decomposition on the current analysis signal of the current period using the optimal wavelet basis of the current period to obtain a set of sub-frequency band signals covering the entire frequency band; identifying and extracting all sub-frequency band signals corresponding to the bandwidth range of the relay's resonant sensitive frequency band as a set of key sub-frequency bands based on the Passevar theorem; calculating the energy value of each sub-frequency band signal in the set of key sub-frequency bands based on Passevar's theorem; normalizing the energy value of each sub-frequency band signal by summation normalization; and constructing a set of feature vectors for the current period using the normalized energy values ​​of each sub-frequency band signal.

8. The method for detecting resonant faults in a meter relay based on harmonic analysis according to claim 7, characterized in that, The step of extracting fault features and identifying the current operating state of the relay further includes: constructing a classifier model using a support vector machine or rule criterion module; training the classifier model using experimental data on relay resonance faults from the meter's factory testing and aging experiments; inputting the feature vector of the current cycle into the trained classifier model; identifying the relay's feature pattern in the current cycle through the classifier model; and determining the relay's current operating state based on the feature pattern. When the feature vector of the current cycle is significantly concentrated within the bandwidth of the relay's resonance sensitive frequency band and its duration conforms to transient characteristics, the classifier model determines that the relay has a resonance fault in the current cycle and outputs a fault alarm. Otherwise, the classifier model determines that the relay is in a normal operating state in the current cycle.

9. A meter relay resonance fault detection system based on harmonic analysis, characterized in that: It includes a processor, a memory, a communication interface, and a data acquisition device. The processor stores computer program instructions for implementing the method for detecting the resonant fault of a meter relay based on harmonic analysis as described in any one of claims 1 to 8. The communication interface is communicatively connected to the data acquisition device.

Images