Method for monitoring electrical connection stress of high-low voltage switch cabinet, distribution box and cable branch box

Abstract

The application discloses a kind of high-low voltage switch cabinet, distribution box and cable branch box electric connection stress monitoring method, belong to power system state monitoring technical field, comprising: the real-time stress data of cable connection is collected by laying fiber grating stress sensor array;Through temperature compensation, digital filtering and outlier rejection, data preprocessing is carried out;Extract time domain characteristic parameters and frequency domain characteristic parameters to construct stress feature vector;Adopt support vector machine classifier to carry out pattern recognition to three states of bolt loosening, bolt overtightening and normal fastening and trigger early warning;Early warning information is uploaded to background monitoring system, and incremental learning method is used to update self-adaptive threshold of classifier;Through the above technical scheme, the real-time online monitoring of high voltage switch cabinet cable connection fastening stress is realized, has the advantages that anti-electromagnetic interference ability is strong, abnormal identification accuracy is high, with adaptive learning ability, can provide intelligent guarantee for the safe operation of high voltage switch cabinet.High-low voltage switch cabinet, distribution box, cable branch box electric connection stress monitoring method.

Description

Technical Field

[0001] This invention belongs to the field of power system condition monitoring technology, specifically relating to a method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes. Background Technology

[0002] High-voltage switchgear, as a core device in the power system responsible for power distribution, control, and protection, directly affects the safety and stability of the power grid. With the rapid advancement of the State Grid construction and the in-depth development of smart grid technology, the number of high-voltage switchgear units installed in substations, distribution networks, and industrial users continues to grow. The level of intelligent and refined equipment operation and maintenance management has become a key factor in ensuring the reliability of power supply. In the overall structure of high-voltage switchgear, the cable connection, as the junction of electrical conduction and mechanical support, plays a crucial role in reliably connecting the internal electrical components to the external power grid. The connection quality and mechanical condition of this part are decisive factors affecting the overall performance of the switchgear.

[0003] The tightening stress at cable connections is a core parameter for ensuring the reliability of electrical connections. Cables are electrically connected to the internal terminals of the switchgear via bolts, and the tightening torque directly determines the contact resistance and thermal stability. Insufficient tightening torque leads to insufficient contact area, increased contact resistance, localized overheating under high current, accelerated oxidation of the contact surface, and a vicious cycle. Excessive tightening torque may cause the bolt preload to exceed the material's yield strength, resulting in plastic deformation or even breakage of the bolt, leading to loosening of the connection under cyclic stress from vibration or thermal expansion and contraction. Furthermore, the sealing performance of cable connections is also affected by tightening stress. Improper tightening torque may cause seal failure, allowing humid gases to enter the switchgear, causing insulation aging or short-circuit faults. Therefore, accurate monitoring and effective control of the tightening stress at high-voltage switchgear cable connections is a crucial technical foundation for ensuring the long-term safe operation of the switchgear.

[0004] In existing technologies, monitoring the status of cable connections in high-voltage switchgear mainly relies on two methods: periodic power outage manual inspections and offline specialized testing. Periodic power outage inspections typically combine visual checks and torque wrench spot checks, which have limitations such as low testing efficiency, high labor costs, and inability to cover all connection points. Furthermore, the power outage operation itself can impact the reliability of the power supply system. Offline specialized testing is usually conducted after equipment malfunctions or during annual maintenance. While it allows for detailed inspection of designated connection points, it is a passive, reactive response, making it difficult to detect and warn of early-stage tightening stress anomalies. In addition, while some stress monitoring devices exist for online monitoring, these devices generally suffer from high installation complexity, poor compatibility with high-voltage environments, insufficient signal acquisition accuracy, or weak electromagnetic interference resistance. These limitations make it difficult to achieve stable and reliable long-term online monitoring in the strong electromagnetic field environment inside high-voltage switchgear.

[0005] The aforementioned technical issues have resulted in the current lack of a mature technical solution capable of real-time, accurate, and continuous monitoring of the fastening stress state at cable connections in high-voltage switchgear. There is an urgent need for an innovative online monitoring method to meet the pressing requirements of intelligent operation and maintenance of power equipment. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes and cable branch boxes, which addresses the shortcomings of the prior art.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes, including the following steps: Step 1: Collect raw stress data from the fiber Bragg grating sensor array: Deploy the fiber Bragg grating stress sensor array at the bolt mounting position of the cable connection in the high-voltage switchgear, and use wavelength division multiplexing technology to achieve synchronous data acquisition from multiple measurement points to obtain real-time stress data at the cable connection. Step 2, preprocessing the raw stress data: The collected raw stress data are sequentially subjected to temperature compensation processing, digital filtering processing, and outlier removal processing. Temperature compensation processing corrects the stress data using data from a pre-embedded temperature compensation sensor, and digital filtering processing uses a finite impulse response filter to remove high-frequency noise components. Step 3, extract stress feature parameters and establish feature vector: Based on the preprocessed stress data, calculate time-domain feature parameters and frequency-domain feature parameters. The time-domain feature parameters include mean, standard deviation, peak factor and slope coefficient. The frequency-domain feature parameters include main frequency component, spectral energy distribution and power spectral density. Combine the time-domain feature parameters and frequency-domain feature parameters to construct stress feature vector. Step 4, perform abnormal pattern recognition based on stress feature vector: use a support vector machine classifier to classify and determine the stress feature vector, and identify three working states: bolt loosening, bolt overtightening and normal tightening. When the classification result is an abnormal state, trigger an early warning mechanism. Step 5: Output monitoring and early warning information and perform adaptive threshold update: Based on the classification decision result of Step 4, generate monitoring and early warning information containing the measurement point location, anomaly type and stress value, and upload the early warning information to the background monitoring system through the communication interface. At the same time, adaptively update the classification threshold of the support vector machine based on historical monitoring data and the latest decision result.

[0008] As a further preferred embodiment of the method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, the fiber optic stress sensor array in step 1 covers the key bolts at all cable connections within the high-voltage switchgear. These key bolts include the connection between the main circuit cable and the busbar, the terminals at the branch points of the three-phase cables, and the junction box location of the secondary circuit of the current transformer. At least one fiber optic stress sensor is installed at each key bolt location. The sampling frequency at each measuring point is set to a preset sampling frequency, determined based on the characteristic frequency range of bolt loosening. The center wavelength range of the fiber optic grating is within a preset wavelength range, typically configured as the C-band. The wavelength spacing between adjacent sensors meets the preset wavelength spacing requirement to ensure that there is no spectral crosstalk between sensor channels during wavelength division multiplexing. The data acquisition system adopts a synchronous sampling mode, with all measuring points starting data acquisition at the same time, and the acquisition time window length is set to a preset time window length.

[0009] As a further preferred embodiment of the electrical connection stress monitoring method for high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, in step 1, the fiber optic grating sensor adopts a metal encapsulation structure to achieve mechanical protection and stress transmission. The encapsulation material is stainless steel, and its coefficient of thermal expansion matches the bolt base material to ensure the linearity and sensitivity of stress transmission. The sensor encapsulation design adopts a tight fit structure, firmly embedding the fiber optic grating area into the stainless steel housing. The outer surface of the housing is machined with an arc-shaped groove that fits against the hexagonal head or flange face of the bolt. During installation, the sensor is fixed to the bolt surface by adhesive or mechanical clamping. The fiber optic grating strain sensitivity coefficient is a preset strain sensitivity coefficient, the strain measurement range is a preset strain measurement range, and the temperature sensitivity coefficient is a preset temperature sensitivity coefficient.

[0010] As a further preferred embodiment of the method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, the temperature compensation process in step 2 adopts a dual fiber grating differential structure to achieve high-precision temperature compensation; two fiber grating sensors are installed simultaneously at each measuring point. The first fiber grating sensor is encapsulated in a capillary steel tube and filled with thermally conductive silicone grease, so that it only senses changes in ambient temperature and is not affected by stress, serving as a dedicated temperature compensation sensor; the second fiber grating sensor adopts a metal encapsulation structure, simultaneously sensing changes in temperature and stress; the initial set values ​​of the center wavelengths of the two sensors differ by a preset wavelength difference, so as to perform spectral separation in the demodulation system; when the ambient temperature changes, the pure stress component is calculated by the difference in wavelength drift between the two sensors, and the calculation formula is: ; in, This refers to the wavelength shift caused by stress. This represents the total wavelength shift of the stress-sensitive sensor. To compensate for the wavelength drift of a dedicated temperature-compensating sensor, is the fiber grating strain sensitivity coefficient.

[0011] As a further preferred embodiment of the method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, the digital filtering process in step 2 uses a finite impulse response filter to remove high-frequency noise components; the filter coefficients are designed using Hanning or Hamming window functions, the passband ripple is less than the preset passband ripple value, and the stopband attenuation is greater than the preset stopband attenuation value; the filter type is a low-pass filter, and the cutoff frequency is set to a preset cutoff frequency. The cutoff frequency is determined based on the characteristic frequency range of bolt loosening and is lower than half of the sampling frequency to avoid aliasing; the stopband attenuation rate is set to a preset stopband attenuation rate; the linear phase characteristic of the finite impulse response filter ensures that the filtered signal has no phase distortion and maintains the waveform integrity of the stress signal; the signal-to-noise ratio of the filtered signal meets the preset signal-to-noise ratio requirement.

[0012] As a further preferred embodiment of the method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, the outlier removal process in step 2 adopts the Grubbs criterion based on statistical principles to automatically identify and remove measurement gross errors. The core idea of ​​the Grubbs criterion is to assume that the measurement data follows a normal distribution. Under a given confidence level, when the deviation between the measured value and the sample mean exceeds a certain critical value, it is judged as an outlier. The processing procedure is as follows: first, calculate the sample mean and sample standard deviation of the filtered data sequence; then, calculate the residual of each data point, i.e., the absolute deviation of the data point from the sample mean; next, look up the critical value coefficient in the Grubbs critical value table according to the preset confidence level; finally, compare the residual of each data point with the critical value to determine and remove outliers; after outlier removal, a linear interpolation method is used to fill in the missing data to maintain the continuity of the data sequence, and the removal ratio is controlled within the preset removal ratio of the total data.

[0013] As a further preferred embodiment of the method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, the calculation method for the time-domain characteristic parameters in step 3 is as follows: The mean is obtained by summing the values ​​of all sampled points and dividing by the total number of sampled points. The formula is: This parameter reflects the average level of the stress signal. The standard deviation is calculated by taking the square root of the average of the sum of the squares of the differences between each sample point and the mean. The formula is as follows: This parameter reflects the degree of dispersion of the stress signal. The peak factor is obtained by dividing the maximum value of the sampled points by the root mean square value. The calculation formula is as follows: This parameter reflects the degree of impact of the stress signal. The slope coefficient is obtained by dividing the average of the cubes of the differences between each sampling point's value and the mean by the cube of the standard deviation. The formula is as follows: This parameter reflects the asymmetry of the probability distribution of the stress signal. in, The total number of sampling points. For the first Stress values ​​at each sampling point The mean, Standard deviation As the peak factor, This is the slope coefficient. The maximum value of the sampled points. It is the root mean square value; The feature vectors are normalized to map the feature values ​​of each dimension to a preset normalization interval in order to eliminate the influence of the difference in the dimensions of different parameters on subsequent pattern recognition.

[0014] As a further preferred embodiment of the stress monitoring method for electrical connections of high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, the calculation method of the frequency domain characteristic parameters in step 3 is as follows: The preprocessed stress data is subjected to a fast Fourier transform to obtain the spectrum data. The number of spectrum points is equal to the total number of sampling points, and the frequency resolution is equal to the sampling frequency divided by the number of spectrum points. The dominant frequency component is taken as the frequency point with the largest amplitude in the spectrum, and this parameter reflects the dominant vibration frequency of the stress signal. The spectrum energy distribution is taken as the energy proportion of the first few dominant frequency components, and the sum of the squares of the amplitudes of each dominant frequency component is calculated and divided by the sum of the squares of the amplitudes of the entire spectrum. The power spectral density is obtained by normalizing the squared amplitude of the spectrum, and the area under the power spectral density curve is equal to the total power of the signal. The time domain characteristic parameters and frequency domain characteristic parameters are arranged and combined in a fixed order to construct a multidimensional stress characteristic vector.

[0015] As a further preferred embodiment of the method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, the support vector machine classifier in step 4 adopts a radial basis function kernel function, the expression of which is: ; in, These are the kernel function parameters; set them to the preset kernel function parameters. The feature vector is set as follows; the penalty parameter is set to the preset penalty parameter; the support vector machine adopts a one-to-one strategy to achieve three classifications. For the three classification problem, three binary classifiers are constructed: a binary classifier for bolt looseness and normal tightening, a binary classifier for bolt overtightening and normal tightening, and a binary classifier for bolt looseness and bolt overtightening. A voting mechanism is used for classification decision, and the category with the most votes is taken as the final classification result; when the classification confidence is lower than the preset confidence threshold, the classifier outputs an uncertain judgment and triggers the manual review process.

[0016] As a further preferred embodiment of the method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes of the present invention, in step 4, when the support vector machine classifier determines that the bolt is loose, it further distinguishes between mild loosening and severe loosening based on the mean decrease of the stress feature vector. Mild loosening corresponds to a mean decrease within a preset first loosening range, while severe loosening corresponds to a mean decrease exceeding a preset second loosening threshold. Mild loosening indicates that the bolt tightening torque has slightly decreased but is still within a safe range, requiring increased monitoring frequency. Severe loosening indicates that the bolt tightening torque has severely decreased, which may lead to increased contact resistance and local overheating risk, requiring immediate handling. The distinction between the two loosening states helps maintenance personnel take corresponding maintenance measures according to the warning level.

[0017] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects: 1. This invention achieves continuous monitoring of the tightness of all critical bolts by deploying a fiber optic stress sensor array at the cable connection point. The sampling frequency reaches the preset sampling frequency, which can capture the stress change process in real time. This solves the problems of low detection efficiency and incomplete coverage caused by the reliance on periodic power outage inspections in the existing technology, and provides continuous and reliable data support for equipment condition maintenance. 2. This invention uses a fiber optic grating sensor as the stress sensing element. Its inherent insulating properties make it completely unaffected by electromagnetic fields. Combined with temperature compensation technology and digital filtering, it can maintain stable and reliable signal acquisition and stress monitoring accuracy in high electromagnetic interference environments, thus solving the technical defects of existing online monitoring devices with weak anti-electromagnetic interference capabilities. 3. This invention constructs feature vectors by extracting multi-dimensional stress feature parameters in the time and frequency domains, and uses a support vector machine classifier for pattern recognition. The accuracy of identifying bolt loosening and bolt overtightening both meet the preset accuracy requirements. It can distinguish between slight loosening and severe loosening, effectively solving the problem that existing technologies cannot detect and warn of abnormal fastening stress in a timely manner.

[0018] 4. This invention continuously updates the classifier through incremental learning, enabling the system to adapt to changes in equipment operating conditions and new abnormal patterns, maintaining long-term stable monitoring performance. The data transmission delay of the background monitoring system meets the preset delay requirements, and can promptly push early warning information to maintenance personnel, providing intelligent protection for the safe operation of high-voltage switchgear. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall technical solution architecture of the electrical connection stress monitoring method for high and low voltage switchgear, distribution boxes and cable branch boxes proposed in this invention. Figure 2 This is a schematic diagram of the core principle framework of the fiber optic stress sensor array in this invention, showing the layout of key bolts at cable connections and the synchronous data acquisition from multiple measurement points. Figure 3 This is a flowchart of the original stress data preprocessing process in this invention, including temperature compensation processing, digital filtering processing, and outlier removal processing, and the process of using normal monitoring cycle data for benchmark interval calculation is marked. Figure 4 This is a flowchart illustrating the time-domain and frequency-domain multidimensional stress feature parameter extraction and feature vector construction, as well as the logical framework for judging two-level early warning (maintenance information and power outage protection information) based on the normal reference interval in this invention. Figure 5This is a schematic diagram of the core principle framework of the support vector machine classifier in this invention to identify three working states: loose bolts, overtight bolts, and normal tightening, and to trigger a first-level early warning (APP maintenance push) or a second-level early warning (power failure protection + emergency push) based on the stress change amplitude. Figure 6 This is a schematic diagram of the multi-level interaction and data flow of the monitoring and early warning information output (back-end monitoring and mobile APP linkage), the adaptive update of the support vector machine classification threshold, and the configuration and execution of the secondary power failure protection output logic in this invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings: The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0021] like Figures 1 to 6 As shown in Example 1: To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0022] The online monitoring method for fastening stress at cable connections in high and low voltage switchgear provided by this invention involves five core processing stages, which form a complete signal processing link according to the data flow direction. From the data acquisition perspective, the raw stress signal is acquired by a fiber optic grating sensor array and then undergoes four processing stages: preprocessing, feature extraction, pattern recognition, and information output, ultimately achieving real-time online monitoring and early warning of the bolt tightness status at the cable connection. Each step is described in detail below.

[0023] In step 1, the deployment of the fiber Bragg grating stress sensor array and the acquisition of raw stress data are completed first. The fiber Bragg grating stress sensor array is deployed to cover all critical bolts at cable connections within the high-voltage switchgear, including connections between main circuit cables and busbars, terminals at three-phase cable branches, and junction boxes in the secondary circuits of current transformers. At least one fiber Bragg grating stress sensor is installed at each critical bolt location to achieve comprehensive sensing of fastening stress.

[0024] The sensor array is deployed according to the following technical specifications: the sampling frequency of each measuring point is set to a preset sampling frequency, which is determined based on the characteristic frequency range of bolt loosening, typically set between 100Hz and 500Hz to meet the Nyquist sampling requirements for bolt micro-motion signals; the center wavelength range of the fiber grating is within a preset wavelength range, typically configured as the C-band range of 1525nm to 1565nm; the wavelength spacing between adjacent sensors meets the preset wavelength spacing requirements, typically set to 3nm to 5nm, to ensure that there is no spectral crosstalk between sensor channels during wavelength division multiplexing. Wavelength division multiplexing technology is based on the wavelength selectivity of fiber gratings. Broadband optical signals from a broadband light source are transmitted to each measuring point via incident optical fibers. Fiber gratings with different center wavelengths reflect their specific wavelength optical signals. The reflected optical signals are converged by the same optical circulator and sent to a demodulator. The wavelength shift of each sensor is demodulated using a scanning laser or interferometer, and then converted into a stress value. The data acquisition system adopts a synchronous sampling mode to ensure that all measuring points start collecting data at the same time. The acquisition time window length is set to a preset time window length, which is usually between 10 seconds and 60 seconds. The number of data points collected in a single session is determined according to the sampling frequency and time window.

[0025] After the cable connection installation is completed and the power supply is accepted, the routine monitoring cycle is immediately initiated. In this embodiment, the routine monitoring cycle is set to 72 hours (3 days). During this cycle, the system continuously records the stress data of each measuring point at a sampling frequency of 200Hz, and simultaneously records the ambient temperature data and cable load current data at a sampling rate of 1Hz. After 72 hours, the stress data of each measuring point is statistically analyzed: the time series curve of stress is calculated, and the stress fluctuation range (±5MPa) caused by diurnal temperature variation (typical temperature difference 15℃) and the stress change (±3MPa) caused by load fluctuation are identified. The baseline range of stress change value at the measuring point under normal operating conditions is obtained as the baseline center value ±8MPa (corresponding to a deviation percentage of approximately ±6%). This baseline range is stored in the system database as a reference for subsequent two-level early warning judgments.

[0026] The fiber Bragg grating sensor employs a metal encapsulation structure for mechanical protection and stress transfer. The encapsulation material is stainless steel, whose coefficient of thermal expansion matches that of the bolt substrate material, ensuring linearity and sensitivity in stress transfer. The sensor encapsulation design utilizes a tight-fit structure, firmly embedding the fiber Bragg grating area within the stainless steel housing. The outer surface of the housing is machined with arc-shaped grooves to mate with the hexagonal head of the bolt or flange face. During installation, the sensor is fixed to the bolt surface using adhesive or mechanical clamping. The fiber Bragg grating strain sensitivity coefficient is a preset value, typically 1.2 pm / με, meaning each micro-strain corresponds to a 1.2 picometer grating wavelength shift. The strain measurement range is a preset range, typically set to 0 to 2000 με, corresponding to a bolt tightening stress range of 0 to 500 MPa. The temperature sensitivity coefficient is a preset value, typically 10 pm / ℃, used for parameter settings in subsequent temperature compensation calculations.

[0027] In step 2, the raw stress data acquired in step 1 is preprocessed. This preprocessing process includes three sub-steps: temperature compensation, digital filtering, and outlier removal. The purpose of temperature compensation is to eliminate the influence of ambient temperature changes on the stress measurement results. This invention uses a dual fiber grating differential structure to achieve high-precision temperature compensation. The specific scheme is as follows: Two fiber grating sensors are installed simultaneously at each measurement point. The first fiber grating sensor is encapsulated in a capillary steel tube and filled with thermally conductive silicone grease, so that it only senses changes in ambient temperature and is not affected by stress. This sensor serves as a dedicated temperature compensation sensor. The second fiber grating sensor uses a metal encapsulation structure and senses both temperature and stress changes. The initial set values ​​of the center wavelengths of the two sensors differ by a preset wavelength difference, typically between 10 nm and 20 nm, to facilitate spectral separation in the demodulation system. When the ambient temperature changes, the wavelength drift of the two fiber gratings is the product of their respective temperature sensitivity coefficients and the temperature change. The wavelength drift of the temperature compensation sensor is entirely caused by the temperature change, while the wavelength drift of the stress-sensitive sensor is caused by both temperature and stress. The pure stress component is calculated by the difference in wavelength drift between the two sensors. The calculation formula is: the wavelength drift caused by stress equals the total wavelength drift of the stress-sensitive sensor minus the wavelength drift of the temperature-compensated sensor, divided by the strain sensitivity coefficient of the fiber optic grating. The compensation accuracy meets the preset compensation accuracy requirements, with a typical temperature drift of less than 0.5%FS after compensation.

[0028] Digital filtering employs a finite impulse response (FIR) filter to remove high-frequency noise components. The design parameters of the FIR filter are as follows: the filter order is set to a preset order, typically ranging from 32 to 64; a higher order results in better frequency selectivity but also greater computational delay. The filter coefficients are designed using Hanning or Hamming window functions, with passband ripple less than 0.1 dB and stopband attenuation greater than 40 dB. During filtering, the temperature-compensated stress data sequence is convolved through the FIR filter. The filter type is a low-pass filter, and the cutoff frequency is set to a preset cutoff frequency. The cutoff frequency is determined based on the characteristic frequency range of bolt loosening, typically between 20 Hz and 50 Hz, lower than half the sampling frequency to avoid aliasing. The stopband attenuation rate is set to a preset stopband attenuation rate, typically ranging from 40 dB to 60 dB per decade. The filtered signal-to-noise ratio (SNR) meets the preset SNR requirement, typically improving to over 40 dB. The linear phase characteristic of the finite impulse response filter ensures that the filtered signal has no phase distortion, thus maintaining the waveform integrity of the stress signal.

[0029] Outlier removal employs the Grubbs criterion, based on statistical principles, to automatically identify and remove measurement gross errors. The core idea of ​​the Grubbs criterion is that, assuming the measurement data follows a normal distribution, a measurement value is considered an outlier if its deviation from the sample mean exceeds a certain critical value at a given confidence level. The process is as follows: First, the sample mean and sample standard deviation of the filtered data sequence are calculated. Then, the residual for each data point is calculated, representing the absolute deviation of that data point from the sample mean. Next, the critical value coefficient is obtained from the Grubbs critical value table based on a preset confidence level; commonly used confidence levels are 95% or 99%. Finally, the residual for each data point is compared with the critical value. When the residual for a data point exceeds the critical value at that confidence level, that data point is identified as an outlier and removed from the data sequence. After outlier removal, linear interpolation is used to impute missing data to maintain the continuity of the data sequence. The outlier removal ratio is controlled within 5% of the total data volume; if the outlier ratio exceeds this threshold, a data quality alarm is issued.

[0030] In step 3, based on the stress data preprocessed in step 2, time-domain and frequency-domain feature parameters are extracted, and the two types of feature parameters are combined to construct a stress feature vector. The calculation of the time-domain feature parameters analyzes the statistical characteristics of the preprocessed data sequence, including the following four parameters: the mean is obtained by summing the values ​​of all sampling points and dividing by the total number of sampling points; this parameter reflects the average level of the stress signal. The standard deviation is obtained by taking the square root of the average of the sum of the squares of the differences between the values ​​of each sampling point and the mean; this parameter reflects the dispersion of the stress signal. The peak factor is obtained by dividing the maximum value of the sampling point by the root mean square; this parameter reflects the impact of the stress signal. The slope coefficient is obtained by dividing the average of the cubes of the differences between the values ​​of each sampling point and the mean by the cube of the standard deviation; this parameter reflects the asymmetry of the probability distribution of the stress signal. The specific calculation formulas for each parameter are as follows:

[0031] ; ; ; ; in, The total number of sampling points. For the first Stress values ​​at each sampling point The mean, Standard deviation As the peak factor, This is the slope coefficient. The maximum value of the sampled points. This is the root mean square value.

[0032] The calculation of frequency domain characteristic parameters is based on the spectral data obtained by performing a Fast Fourier Transform (FFT) on the preprocessed stress data. The specific calculation method is as follows: A Hanning window function is applied to the preprocessed stress data sequence to reduce spectral leakage, and then a FFT is performed to obtain complex spectral data. The number of spectral points equals the total number of sampling points, and the frequency resolution is the sampling frequency divided by the number of spectral points. The dominant frequency component is the frequency point with the largest amplitude in the spectrum; this parameter reflects the dominant vibration frequency of the stress signal. The spectral energy distribution is the energy proportion of the first few dominant frequency components, typically the first 5 to 10 frequency components with the largest amplitudes. The calculation formula is the sum of the squares of the amplitudes of each dominant frequency component divided by the sum of the squares of the amplitudes of the entire spectrum. The power spectral density is obtained by normalizing the squared spectral amplitudes; the area under the power spectral density curve equals the total power of the signal.

[0033] A multidimensional stress feature vector is constructed by arranging and combining time-domain and frequency-domain feature parameters in a fixed order. The dimension of the feature vector equals the sum of the number of time-domain parameters and the number of frequency-domain parameters, typically ranging from 7 to 15 dimensions. The feature vector is represented as follows:

[0034] ; in, to For time-domain feature parameters, to These are frequency domain feature parameters. To eliminate the impact of dimensional differences between different parameters on subsequent pattern recognition, the feature vector is normalized, mapping each feature value to the interval between 0 and 1.

[0035] In step 4, a support vector machine (SVM) classifier is used to classify the stress feature vectors, identifying three working states: loose bolts, overtight bolts, and properly tightened bolts. The configuration parameters of the SVM classifier are as follows: the kernel function is a radial basis function (RBF) kernel, and its expression is...

[0036] ; in, The kernel function parameters are set to preset kernel function parameters, typically taking the inverse of the feature vector dimension; the penalty parameter is also set to a preset penalty parameter, typically taking a value between 1 and 100, used to balance the classification margin and classification error; the support vector machine uses a one-to-one strategy to achieve three-class classification, that is, constructing three binary classifiers for the three-class classification problem: a binary classifier for bolt looseness versus normal tightness, a binary classifier for bolt overtightness versus normal tightness, and a binary classifier for bolt looseness versus bolt overtightness. A voting mechanism is used for classification decisions, with the category receiving the most votes becoming the final classification result. The accuracy rates for identifying bolt looseness and bolt overtightness both meet the preset accuracy requirements, with a typical accuracy rate of over 95%.

[0037] The training process of the Support Vector Machine (SVM) classifier combines offline training with online updates. In the offline training phase, a training sample set is constructed using historical monitoring data. After expert annotation, the sample data is categorized into three classes: loose bolts, overtight bolts, and normally tightened bolts. Cross-validation is used to optimize the kernel function and penalty parameters during training. In the online classification phase, the real-time extracted stress feature vectors are input into the trained SVM classifier for judgment. The classifier outputs the class label of the sample and the classification confidence score. When the classification confidence score falls below a preset confidence threshold, the classifier outputs an uncertain judgment and triggers a manual review process.

[0038] When the support vector machine classifier determines that the bolts are loose, a two-level early warning judgment is executed based on the stress change value benchmark range determined in step 1 during the normal monitoring cycle: (1) Level 1 Early Warning (Maintenance Information) Conditions: Calculate the decrease in the current average stress value relative to the center value of the benchmark interval. For example, if the benchmark center value determined during the normal monitoring cycle is 165 MPa, the Level 1 early warning threshold is set to a deviation of 5% to 15% (i.e., the stress value drops to the range of 140 MPa to 157 MPa). If the current average stress value is 150 MPa, the decrease is 9.1%, and the classification confidence level is 87%, then a Level 1 early warning is triggered. The system generates power maintenance information, including the location of the measuring point (e.g., "the upper connection of phase A cable of switch cabinet No. 3"), the current stress value of 150 MPa, the benchmark value of 165 MPa, the deviation ratio of -9.1%, and the recommended maintenance measures ("It is recommended to arrange an inspection within 7 days to check the bolt tightening torque").

[0039] (2) Level 2 Warning (Power Outage Protection) Conditions: If the current average stress value decreases by more than 15% (i.e., the stress value is below 140MPa), or the stress value decreases by more than 20% within 1 hour, and the classification confidence level is not less than 90%, a Level 2 warning is triggered. The system generates power outage protection information, including the measurement point location, current stress value, reference value, deviation ratio, and emergency handling suggestions ("Immediately arrange power outage maintenance and check the status of the connection"). At the same time, the system issues a trip command according to the power outage protection output logic pre-configured by the user. For example, if the user has configured a corresponding feeder circuit breaker trip output for the A-phase cable of switch cabinet No. 3, the circuit breaker will be disconnected after a 1-second delay to disconnect the faulty feeder.

[0040] When the support vector machine classifier determines that a bolt is loose, it further distinguishes between slight and severe loosening based on the decrease in the mean of the stress feature vector. The distinction thresholds are set based on statistical analysis of a large amount of experimental data: slight loosening corresponds to a mean decrease within a preset first loosening range, typically 5% to 15%; severe loosening corresponds to a mean decrease exceeding a preset second loosening threshold, typically 15%. Slight loosening indicates that the bolt tightening torque has slightly decreased but is still within a safe range, requiring increased monitoring frequency; severe loosening indicates that the bolt tightening torque has severely decreased, which may lead to increased contact resistance and the risk of localized overheating, requiring immediate action. Distinguishing between the two loosening states helps maintenance personnel take appropriate maintenance measures based on the warning level.

[0041] In step 5, based on the classification judgment results of step 4, monitoring and early warning information is generated and adaptive threshold updates are performed. The generation of monitoring and early warning information includes the following elements: measurement point location information, obtained by mapping the physical layout coordinates and logical numbers of the sensor array; anomaly type information, including subcategories such as bolt loosening, bolt overtightening, and slight or severe loosening; stress value information, including the currently measured stress value, historical average stress value, and percentage deviation relative to the benchmark value; timestamp information, recording the time when the anomaly judgment occurred; and classification confidence information, reflecting the reliability of the judgment result. The early warning information is encapsulated according to a preset format and uploaded to the backend monitoring system via a communication interface. The communication interface uses the standard Modbus protocol or the IEC61850 protocol. The Modbus protocol is suitable for communication with local data acquisition units, while the IEC61850 protocol is suitable for integration with substation automation systems. The data transmission delay of the monitoring and early warning information meets the preset delay requirements, with a typical delay index of less than 1 second, ensuring that the backend monitoring system can obtain the latest monitoring status in real time.

[0042] The warning information is simultaneously pushed to the bound maintenance personnel's mobile APP via a wireless communication module (4G / 5G or Wi-Fi). The mobile APP has the following functions: real-time display of stress status (values, trend curves, normal range) at each measuring point; receiving and displaying warning information, with level 1 warnings displayed as ordinary notifications and level 2 warnings displayed as emergency alarms (flashing red and sound alerts); displaying the power outage protection execution status (whether it has tripped, tripping time, and tripping feeder number); and supporting remote alarm confirmation and viewing of historical records. The background monitoring system records all warning events and protection action logs, supporting multi-condition queries and report export by time, measuring point, warning level, etc.

[0043] The background monitoring system includes the following functions: real-time display of stress status at each measuring point, including current stress values, historical trend curves, and stress distribution heatmaps; display of a list of early warning information, including warning time, measuring point location, anomaly type, and handling status; provision of historical data query functionality, supporting filtering by time range, measuring point location, and anomaly type; and provision of data export functionality, supporting the export of monitoring data to standard format files for further analysis. The background monitoring system also forwards received early warning information to the operation and maintenance management platform, allowing operation and maintenance personnel to arrange on-site verification and repair work based on the warning information.

[0044] Adaptive threshold updates employ an incremental learning method to continuously optimize the support vector machine (SVM) classifier. The trigger condition for incremental learning is when the amount of newly added valid monitoring data reaches a preset update sample size (typically 50 to 100 samples). The SVM classifier is then retrained using these new samples. When updating classifier parameters, the incremental learning algorithm retains existing support vectors while incorporating newly added ones, avoiding the reuse of historical training data. The updated classifier maintains its ability to recognize known abnormal patterns while improving its generalization ability to identify newly emerging abnormal operating conditions. The incremental learning cycle is determined based on the rate of change in equipment operating conditions; the update cycle is shorter for rapidly changing environments and longer for stable environments.

[0045] The monitoring method of the present invention will be described in detail below with a specific application example, focusing on demonstrating the effect of the two-level early warning and the linkage with the mobile APP.

[0046] Suppose that in a high-voltage switchgear of a 110kV substation, there are 12 critical bolt installation points at cable connections that need to be monitored. In step 1, technicians install a fiber optic stress sensor at each bolt point. The sensor is encapsulated in stainless steel and has center wavelengths of 1525nm, 1530nm, and 1535nm, with adjacent wavelengths spaced 5nm apart. The sensor array is connected to a demodulator via wavelength division multiplexing (WDM), with a sampling frequency set to 200Hz, meaning data is collected every 5 milliseconds. The data acquisition system automatically performs three timed acquisition tasks daily, with each acquisition window lasting 30 seconds, obtaining 6000 data points per acquisition. After successful power-on acceptance, the system automatically enters a 72-hour routine monitoring cycle. During this cycle, stress data at all measuring points is continuously recorded. Taking the upper connection of phase A cable in switchgear No. 3 as an example, the 72-hour monitoring data shows that the stress value fluctuates between 158MPa and 172MPa, with a baseline center value of 165MPa, and a normal fluctuation range of ±7MPa (corresponding to ±4.2%). The system stores the baseline interval in the database.

[0047] In steps 2 and 3, preprocessing and feature extraction are performed, which will not be described in detail here.

[0048] In step 4, after 30 days of operation, the system monitored in real time that the average stress at measuring point 3 had decreased to 150 MPa. The classifier output that the bolt was loose, with a confidence level of 92%. The deviation was calculated as (165-150) / 165 = 9.1%, which falls within the first-level warning range of 5% to 15%. The system triggered a first-level warning and generated a maintenance message: "[Maintenance Reminder] The stress of the bolt at the upper connection of phase A cable in switchgear 3 has decreased by 9.1%, with a current value of 150 MPa and a baseline of 165 MPa. It is recommended to arrange an inspection within 7 days." This message was pushed to the maintenance personnel's mobile APP via the 4G module and displayed as a regular notification. After receiving the message, the maintenance personnel arranged for an on-site inspection the following week and found that the bolt was slightly loose. After tightening it with a torque wrench, the stress returned to 163 MPa, and the warning was cleared.

[0049] After some time, the stress at the lower connection point of the B-phase cable in the same switchgear (measuring point 8) suddenly dropped from the benchmark 170MPa to 135MPa (a deviation of 20.6%), with a classification confidence level of 95%. The system triggered a level-two warning and generated a power outage protection message: "[Emergency Alarm] The stress on the bolt at the lower connection point of the B-phase cable in switchgear 8 has dropped significantly by 20.6%, with a current value of 135MPa and a benchmark of 170MPa. There is a risk of contact failure! Please handle this immediately!" The APP alerted the user with a red flashing light and a continuous sound. Simultaneously, based on the user-configured power outage protection logic (measuring point 8 is associated with feeder circuit breaker QF8, which trips after a 1.5-second delay), the system automatically issued a trip command, disconnecting the corresponding feeder. The background monitoring system recorded the tripping event and displayed "Power outage protection executed." Upon receiving the alarm, maintenance personnel immediately went to the site and found that the cable terminals were overheated and discolored, and the bolts were almost completely loose. They promptly replaced the damaged parts, preventing an accident.

[0050] The above examples verify the effectiveness of the two-level early warning mechanism and the linkage with the mobile APP of this invention.

[0051] The monitoring method of this invention will be described in detail below with a specific application example. Assume that in a high-voltage switchgear of a 110kV substation, there are 12 critical bolt installation points at cable connections that need to be monitored. In step 1, technicians install a fiber optic stress sensor at each bolt point. The sensor is encapsulated in stainless steel and has center wavelengths of 1525nm, 1530nm, and 1535nm, with adjacent wavelengths spaced 5nm apart. The sensor array is connected to a demodulator via wavelength division multiplexing (WDM), and the sampling frequency is set to 200Hz, meaning data is collected every 5 milliseconds. The data acquisition system automatically performs three timed acquisition tasks per day, with each acquisition window lasting 30 seconds, obtaining 6000 data points per acquisition.

[0052] In step 2, the preprocessing process is illustrated using data from a specific measurement point as an example. The raw data first undergoes temperature compensation. This measurement point is equipped with one temperature-compensated fiber Bragg grating and one stress-sensitive fiber Bragg grating. Assuming the ambient temperature rises by 2°C during a certain acquisition period, the wavelength drift of the temperature-compensated sensor is 20 pm, and the wavelength drift of the stress-sensitive sensor is 30 pm. Therefore, the wavelength drift caused by pure stress after temperature compensation is 10 pm, which translates to a stress value of approximately 8.3 MPa. The temperature-compensated data sequence is then filtered using a finite impulse response low-pass filter. The filter order is 64, the cutoff frequency is 30 Hz, and the stopband attenuation is 50 dB / decibels. The filtered data then enters the outlier removal stage, where it is tested according to the Grubbs criterion with a confidence level set at 95%. If the residual of a data point exceeds the critical value, it is removed and filled with linear interpolation. After preprocessing, the signal-to-noise ratio of the data sequence increases from the original 25 dB to 45 dB.

[0053] In step 3, feature parameters are extracted from the preprocessed data. The calculated results of the time-domain feature parameters are: mean 150.3 MPa, standard deviation 12.7 MPa, peak factor 1.35, and slope coefficient 0.18. The calculation process of the frequency-domain feature parameters is as follows: the data sequence is subjected to a fast Fourier transform to obtain the spectrum data, the main frequency component is 8.2 Hz, the energy proportion of the first 5 main frequency components in the spectrum energy distribution is 78.3%, and the power spectral density is normalized to obtain the power spectral density curve. The above 7 feature parameters are combined to obtain the stress feature vector [150.3, 12.7, 1.35, 0.18, 8.2, 78.3, 0.783]. The feature vector is normalized to map the values ​​of each dimension to the interval between 0 and 1.

[0054] In step 4, the normalized feature vectors are input into the support vector machine classifier for decision-making. The classifier uses a radial basis function kernel, with kernel function parameters... The penalty parameter was set to 10 and the classifier output a decision of bolt loosening with a classification confidence level of 92%. Since the decision was bolt loosening, the degree of loosening was further determined based on the decrease in the mean value. The baseline mean stress at this measuring point was set at 165.2 MPa. The current mean stress of 150.3 MPa corresponds to a decrease of 9.02%, which falls within the 5% to 15% range of slight loosening. Therefore, it was determined to be slight loosening, triggering a yellow alert.

[0055] In step 5, the generated monitoring and early warning information is as follows: the measuring point is the upper connection of phase A cable in switchgear No. 3; the anomaly type is slight bolt loosening; the stress value is currently 150.3 MPa, the baseline is 165.2 MPa, the deviation is -9.02%, and the classification confidence level is 92%. The early warning information is uploaded to the background monitoring system via Modbus TCP protocol, with a transmission delay of 0.8 seconds. After receiving the early warning information, the background monitoring system pops up an alarm window on the monitoring interface and issues an audible and visual alert, while simultaneously forwarding the early warning information to the operation and maintenance management platform. Based on the early warning information, the operation and maintenance personnel arrange on-site verification the following day. The on-site inspection confirms that the bolt at the upper connection of phase A cable in switchgear No. 3 is slightly loose. After tightening with a torque wrench, the stress value is re-measured and returns to near the baseline value, thus clearing the early warning.

[0056] The incremental learning process begins after the warning event ends. The system labels the valid monitoring data collected during this event as slightly loose and includes it in the incremental learning sample library. When the number of new samples in the sample library reaches the preset update sample size of 80, the system uses the new samples to perform secondary training on the support vector machine classifier. After training, the classifier's accuracy in recognizing slightly loose states increases from 92% to 95%, while maintaining the same accuracy in recognizing normally tight and overtightened bolt states.

[0057] Example 2: This example describes an alternative technical solution for monitoring the electrical connection stress of high and low voltage switchgear, distribution boxes and cable branch boxes. This solution uses different technical implementation methods in the feature extraction stage in step 3 and the pattern recognition stage in step 4.

[0058] In step 3, the calculation method for the time-domain feature parameters remains consistent with that in Example 1. The calculation method for the frequency-domain feature parameters is improved as follows: wavelet transform is used instead of fast Fourier transform for frequency domain analysis. Specifically, the preprocessed stress data sequence is decomposed using multi-scale wavelet decomposition, with a decomposition level of 5 layers. The wavelet basis function is the db4 wavelet. After decomposition, approximation coefficients and detail coefficients for each layer are obtained. The energy of the detail coefficients for each layer is calculated as the frequency-domain feature parameters. The detail coefficients of the first layer correspond to high-frequency components, and the detail coefficients of the fifth layer correspond to low-frequency components. The dominant frequency component is determined by analyzing the energy distribution of the detail coefficients at each layer, selecting the characteristic frequency range corresponding to the layer with the highest energy as the dominant frequency range. The calculation method for the spectral energy distribution is the same as in Example 1, i.e., calculating the energy proportion of the first few dominant frequency components. The method for constructing the feature vector is the same as in Example 1, where the time-domain and frequency-domain feature parameters are arranged in a fixed order and then normalized.

[0059] In step 4, the pattern recognition stage uses a deep learning classifier instead of a support vector machine classifier. The deep learning classifier uses a convolutional neural network with the following configuration: the input layer dimension matches the feature vector dimension; the hidden layer consists of two convolutional layers and two pooling layers, with the convolutional layers using the ReLU activation function, a kernel size of 3×3, and 32 and 64 kernels respectively; the pooling layers use max pooling with a pooling window size of 2×2; the fully connected layer contains 128 neurons, using the ReLU activation function; the output layer uses the Softmax function, with 3 output nodes corresponding to the three states of loose bolts, overtight bolts, and properly tightened bolts. The deep learning classifier is trained using the backpropagation algorithm, with the Adam optimizer, a learning rate of 0.001, a batch size of 32, and 500 training iterations. Dropout is used during training to prevent overfitting, with a Dropout ratio of 0.5.

[0060] The training process of the deep learning classifier also employs a combination of offline training and online updates. In the offline training phase, a large amount of historical monitoring data is used to construct a training sample set. After expert annotation, the sample data is categorized into three classes: loose bolts, overtight bolts, and normally tightened bolts. During training, the model performance is verified using a validation set, and network structure parameters are adjusted. In the online classification phase, the stress feature vectors extracted in real time are input into the trained deep learning classifier for judgment. The classifier outputs the probability distribution of each class and the maximum probability value. When the maximum probability value is lower than a preset confidence threshold, the classifier outputs an uncertain judgment and triggers a manual review process.

[0061] When the deep learning classifier determines that the bolts are loose, it further distinguishes between slight and severe looseness based on the magnitude of the decrease in the mean value, with the distinction threshold remaining consistent with Example 1. The advantage of the deep learning classifier lies in its ability to automatically learn feature representations, its stronger fitting ability to complex nonlinear relationships, and its better generalization performance when facing new abnormal working conditions.

[0062] In step 5, the method for generating monitoring and early warning information and updating adaptive thresholds remains consistent with that in Example 1. Early warning information is uploaded to the backend monitoring system via a communication interface, and the functionality of the backend monitoring system remains unchanged. The adaptive threshold update employs an incremental learning method to continuously optimize the deep learning classifier; the triggering conditions and update method for incremental learning are the same as in Example 1.

[0063] The following detailed explanation of the alternative technical solution of the present invention is illustrated with a specific application example. Assume that in a high-voltage switchgear of a 220kV substation, there are 24 critical bolt installation points at cable connections that require monitoring. In step 1, technicians install a fiber optic stress sensor at each bolt point. The sensor is encapsulated in stainless steel, with a center wavelength range of 1525nm to 1565nm and an adjacent wavelength interval of 5nm. The sensor array is connected to a demodulator via wavelength division multiplexing, with a sampling frequency set to 250Hz, meaning data is collected every 4 milliseconds. The data acquisition system automatically performs four timed acquisition tasks daily, with each acquisition window lasting 20 seconds, obtaining 5000 data points per acquisition.

[0064] In step 2, the preprocessing procedure is consistent with that in Example 1, including temperature compensation, digital filtering, and outlier removal. Temperature compensation uses a dual fiber grating differential structure, digital filtering uses a 64th-order finite impulse response low-pass filter with a cutoff frequency set to 35Hz, and outlier removal uses the Grubbs criterion with a confidence level set to 95%. After preprocessing, the signal-to-noise ratio of the data sequence is improved from the original 22dB to 43dB.

[0065] In step 3, feature parameters are extracted from the preprocessed data. The calculated results of the time-domain feature parameters are: mean 158.6 MPa, standard deviation 14.2 MPa, peak factor 1.28, and slope coefficient 0.15. The frequency-domain feature parameters are calculated using wavelet transform: the data sequence is decomposed into 5-level db4 wavelet fractions, and the energy of the detail coefficients at each level is calculated to obtain 5 frequency-domain feature parameters. The dominant frequency components are determined by analyzing the energy distribution of the detail coefficients at each level. The energy of the detail coefficients at the 3rd level is the largest, and the corresponding dominant frequency range is 20Hz to 40Hz. The calculated spectral energy distribution shows that the energy proportion of the first 5 dominant frequency components is 81.2%. The above 9 feature parameters are combined to obtain a stress feature vector, which is then normalized.

[0066] In step 4, the normalized feature vector is input into a deep learning classifier for judgment. The classifier uses a convolutional neural network structure with an input layer dimension of 9, hidden layers consisting of two convolutional layers and two pooling layers, a fully connected layer containing 128 neurons, and an output layer containing 3 nodes. The classifier outputs a judgment result of bolt loosening, with probabilities of 0.87 for bolt loosening, 0.05 for bolt overtightening, and 0.08 for normal tightening. Since the judgment result is bolt loosening, the degree of loosening is further determined based on the decrease in the mean value. The baseline mean stress at this measuring point is set at 172.4 MPa, and the current mean stress of 158.6 MPa corresponds to a decrease of 8.01%, which is within the range of 5% to 15% for slight loosening. Therefore, it is judged as slight loosening, triggering a yellow warning.

[0067] In step 5, the generated monitoring and early warning information is as follows: the measuring point is the lower connection of phase B cable in switchgear No. 7; the anomaly type is slight bolt loosening; the stress value is currently 158.6 MPa, the reference is 172.4 MPa, the deviation is -8.01%, and the classification confidence level is 87%. The early warning information is uploaded to the background monitoring system via the IEC61850 protocol, with a transmission delay of 0.6 seconds. After receiving the early warning information, the background monitoring system pops up an alarm window on the monitoring interface and issues an audible and visual alert, while simultaneously forwarding the early warning information to the operation and maintenance management platform. Based on the early warning information, the operation and maintenance personnel arrange on-site verification for the same day. The on-site inspection confirms that the bolt at the lower connection of phase B cable in switchgear No. 7 is slightly loose. After tightening with a torque wrench, the stress value is re-measured and returns to near the reference value, thus clearing the early warning.

[0068] The incremental learning process begins after the warning event ends. The system labels the valid monitoring data collected during this event as slightly loose and includes it in the incremental learning sample library. When the number of new samples in the sample library reaches the preset update sample size of 60, the system uses the new samples to perform secondary training on the deep learning classifier. After training, the classifier's accuracy in recognizing slightly loose states increases from 87% to 93%, while maintaining the same accuracy in recognizing normal tightening and overtightened bolt states.

[0069] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention. All technical features in this embodiment can be freely combined according to actual needs.

[0070] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes, characterized in that, Includes the following steps: Step 1: Collect raw stress data from the fiber Bragg grating sensor array: Deploy the fiber Bragg grating stress sensor array at the bolt mounting position of the cable connection in the high-voltage switchgear, and use wavelength division multiplexing technology to achieve synchronous data acquisition from multiple measurement points to obtain real-time stress data at the cable connection. Step 2, preprocessing the raw stress data: The collected raw stress data are sequentially subjected to temperature compensation processing, digital filtering processing, and outlier removal processing. Temperature compensation processing corrects the stress data using data from a pre-embedded temperature compensation sensor, and digital filtering processing uses a finite impulse response filter to remove high-frequency noise components. Step 3, extract stress feature parameters and establish feature vector: Based on the preprocessed stress data, calculate time-domain feature parameters and frequency-domain feature parameters. The time-domain feature parameters include mean, standard deviation, peak factor and slope coefficient. The frequency-domain feature parameters include main frequency component, spectral energy distribution and power spectral density. Combine the time-domain feature parameters and frequency-domain feature parameters to construct stress feature vector. Step 4, perform abnormal pattern recognition based on stress feature vector: use a support vector machine classifier to classify and determine the stress feature vector, and identify three working states: bolt loosening, bolt overtightening and normal tightening. When the classification result is an abnormal state, trigger an early warning mechanism. Step 5: Output monitoring and early warning information and perform adaptive threshold update: Based on the classification decision result of Step 4, generate monitoring and early warning information containing the measurement point location, anomaly type and stress value, and upload the early warning information to the background monitoring system through the communication interface. At the same time, adaptively update the classification threshold of the support vector machine based on historical monitoring data and the latest decision result.

2. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: In step 1, the fiber optic stress sensor array is deployed to cover all critical bolts at cable connections within the high-voltage switchgear. These critical bolts include the connections between the main circuit cable and the busbar, the terminals at three-phase cable branches, and the junction boxes in the secondary circuit of the current transformer. At least one fiber optic stress sensor is installed at each critical bolt location. The sampling frequency for each measurement point is set to a preset sampling frequency, determined based on the characteristic frequency range of bolt loosening. The center wavelength range of the fiber optic grating is within a preset wavelength range, typically configured as the C-band. The wavelength spacing between adjacent sensors meets the preset wavelength spacing requirement to ensure no spectral crosstalk between sensor channels during wavelength division multiplexing. The data acquisition system uses a synchronous sampling mode, with all measurement points starting data acquisition at the same time, and the acquisition time window length is set to a preset time window length.

3. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: In step 1, the fiber Bragg grating sensor employs a metal encapsulation structure to achieve mechanical protection and stress transmission. The encapsulation material is stainless steel, whose coefficient of thermal expansion matches that of the bolt base material to ensure the linearity and sensitivity of stress transmission. The sensor encapsulation design adopts a tight-fit structure, firmly embedding the fiber Bragg grating area into the stainless steel housing. The outer surface of the housing is machined with an arc-shaped groove that fits against the hexagonal head or flange face of the bolt. During installation, the sensor is fixed to the bolt surface by adhesive or mechanical clamping. The fiber Bragg grating strain sensitivity coefficient is a preset strain sensitivity coefficient, the strain measurement range is a preset strain measurement range, and the temperature sensitivity coefficient is a preset temperature sensitivity coefficient.

4. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: In step 2, the temperature compensation process employs a dual fiber grating differential structure to achieve high-precision temperature compensation. Two fiber grating sensors are installed simultaneously at each measuring point. The first fiber grating sensor is encapsulated in a capillary tube and filled with thermally conductive silicone grease, ensuring it senses only ambient temperature changes and is unaffected by stress; it serves as a dedicated temperature compensation sensor. The second fiber grating sensor uses a metal encapsulation structure and senses both temperature and stress changes. The initial set values ​​of the center wavelengths of the two sensors differ by a preset wavelength difference to facilitate spectral separation in the demodulation system. When the ambient temperature changes, the pure stress component is calculated from the difference in wavelength drift between the two sensors using the following formula: ； in, This refers to the wavelength shift caused by stress. This represents the total wavelength shift of the stress-sensitive sensor. To compensate for the wavelength drift of a dedicated temperature-compensating sensor, is the fiber grating strain sensitivity coefficient.

5. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: In step 2, the digital filtering process uses a finite impulse response (FIR) filter to remove high-frequency noise components. The filter coefficients are designed using Hanning or Hamming window functions, with passband ripple less than a preset passband ripple value and stopband attenuation greater than a preset stopband attenuation value. The filter type is a low-pass filter, and the cutoff frequency is set to a preset cutoff frequency. The cutoff frequency is determined based on the characteristic frequency range of bolt loosening and is lower than half of the sampling frequency to avoid aliasing. The stopband attenuation rate is set to a preset stopband attenuation rate. The linear phase characteristics of the FIR filter ensure that the filtered signal has no phase distortion and maintains the waveform integrity of the stress signal. The signal-to-noise ratio of the filtered signal meets the preset signal-to-noise ratio requirement.

6. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: In step 2, outlier removal employs the Grubbs criterion, based on statistical principles, to automatically identify and remove measurement gross errors. The core idea of ​​the Grubbs criterion is to assume that the measurement data follows a normal distribution. At a given confidence level, if the deviation of the measured value from the sample mean exceeds a certain critical value, it is considered an outlier. The process involves first calculating the sample mean and sample standard deviation of the filtered data sequence, then calculating the residual of each data point, i.e., the absolute deviation of that data point from the sample mean. Next, the critical value coefficient is obtained by consulting the Grubbs critical value table according to the preset confidence level. Finally, the residual of each data point is compared with the critical value to determine and remove outliers. After outlier removal, linear interpolation is used to fill in missing data to maintain the continuity of the data sequence, and the removal ratio is controlled within a preset removal ratio of the total data volume.

7. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: The method for calculating the time-domain feature parameters in step 3 is as follows: The mean is obtained by summing the values ​​of all sampled points and dividing by the total number of sampled points. The formula is: This parameter reflects the average level of the stress signal. The standard deviation is calculated by taking the square root of the average of the sum of the squares of the differences between each sample point and the mean. The formula is as follows: This parameter reflects the degree of dispersion of the stress signal. The peak factor is obtained by dividing the maximum value of the sampled points by the root mean square value. The calculation formula is as follows: This parameter reflects the degree of impact of the stress signal. The slope coefficient is obtained by dividing the average of the cubes of the differences between each sampling point's value and the mean by the cube of the standard deviation. The formula is as follows: This parameter reflects the asymmetry of the probability distribution of the stress signal. in, The total number of sampling points. For the first Stress values ​​at each sampling point The mean, Standard deviation, As the peak factor, This is the slope coefficient. The maximum value of the sampled points. It is the root mean square value; The feature vectors are normalized to map the feature values ​​of each dimension to a preset normalization interval in order to eliminate the influence of the difference in the dimensions of different parameters on subsequent pattern recognition.

8. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: The calculation method for the frequency domain feature parameters in step 3 is as follows: The preprocessed stress data is subjected to a fast Fourier transform to obtain the spectral data. The number of spectral points equals the total number of sampling points, and the frequency resolution equals the sampling frequency divided by the number of spectral points. The dominant frequency component is taken as the frequency point with the largest amplitude in the spectrum, reflecting the dominant vibration frequency of the stress signal. The spectral energy distribution is taken as the energy proportion of the first few dominant frequency components, and the sum of the squares of the amplitudes of each dominant frequency component is calculated and divided by the sum of the squares of the amplitudes of the entire spectrum. The power spectral density is obtained by normalizing the squared amplitudes of the spectrum, and the area under the power spectral density curve equals the total power of the signal. The time domain feature parameters and frequency domain feature parameters are arranged and combined in a fixed order to construct a multidimensional stress feature vector.

9. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: In step 4, the support vector machine classifier uses a radial basis function kernel, the expression of which is: ； in, These are kernel function parameters; set them to the preset kernel function parameters. The feature vector is set as follows; the penalty parameter is set to the preset penalty parameter; the support vector machine adopts a one-to-one strategy to achieve three classifications. For the three classification problem, three binary classifiers are constructed: a binary classifier for bolt looseness and normal tightening, a binary classifier for bolt overtightening and normal tightening, and a binary classifier for bolt looseness and bolt overtightening. A voting mechanism is used for classification decision, and the category with the most votes is taken as the final classification result; when the classification confidence is lower than the preset confidence threshold, the classifier outputs an uncertain judgment and triggers the manual review process.

10. The method for monitoring electrical connection stress in high and low voltage switchgear, distribution boxes, and cable branch boxes according to claim 1, characterized in that: In step 4, when the support vector machine classifier determines that the bolt is loose, it further distinguishes between mild and severe loosening based on the mean decrease of the stress feature vector. Mild loosening corresponds to a mean decrease within a preset first loosening range, while severe loosening corresponds to a mean decrease exceeding a preset second loosening threshold. Mild loosening indicates that the bolt tightening torque has slightly decreased but is still within a safe range, requiring increased monitoring frequency. Severe loosening indicates that the bolt tightening torque has severely decreased, which may lead to increased contact resistance and local overheating risk, requiring immediate action. The distinction between the two loosening states helps maintenance personnel take corresponding maintenance measures based on the warning level.

Images