A hyper-converged pole-mounted circuit breaker fault detection method and circuit breaker

Abstract

The application relates to a kind of super-fusion pole circuit breaker fault detection method and circuit breaker, comprising the following steps: S1, the electrical signal, mechanical vibration signal and thermal imaging temperature signal of pole circuit breaker are collected, denoising, normalization and time synchronization alignment are carried out, and standardization multimodal data stream is output;S2, the time domain, frequency domain and time-frequency domain features of electrical signal, mechanical vibration signal and thermal imaging temperature signal are extracted respectively, feature-level fusion is carried out based on cross-modal attention mechanism, and fusion feature vector is output;S3, based on fusion feature vector, fault type identification, fault severity assessment and fault location are carried out, and fault diagnosis result is output;S4, based on fusion feature vector and fault diagnosis result, health state assessment, residual service life prediction and hierarchical alarm are carried out, and alarm instruction is output.The beneficial effects of the application are: early accurate detection and residual life prediction of pole circuit breaker multi-class fault can be carried out, and the fault recognition accuracy is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power distribution network equipment testing technology, specifically to a method for detecting faults in a hyper-converged pole-mounted circuit breaker and the circuit breaker itself. Background Technology

[0002] Pole-mounted circuit breakers are installed on outdoor utility poles and are responsible for line segmentation, load switching, and short-circuit protection. Their operating status directly affects the reliability of the power distribution network. Pole-mounted circuit breakers are exposed to harsh outdoor environments for extended periods, enduring multiple stresses such as temperature cycling, humidity changes, mechanical shocks, and electrical load fluctuations. Therefore, fault detection methods are necessary to ensure the safe operation of the power distribution network.

[0003] In existing technologies, fault detection of pole-mounted circuit breakers mostly relies on a single signal source and periodic manual inspections. Detection using a single signal source results in a low rate of identifying latent faults and a high false alarm rate. Furthermore, it cannot provide early warning of faults. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application proposes a hyper-converged pole-mounted circuit breaker fault detection method and circuit breaker, which can perform early and accurate detection of multiple types of faults in pole-mounted circuit breakers and predict their remaining life, significantly improving the fault identification accuracy.

[0005] The following is the technical solution of the present invention: a fault detection method for a hyper-converged pole-mounted circuit breaker, comprising the following steps:

[0006] S1. Collect electrical signals, mechanical vibration signals, and thermal imaging temperature signals from the pole-mounted circuit breaker, perform noise reduction, normalization, and time synchronization alignment, and output a standardized multimodal data stream;

[0007] S2. Extract time-domain, frequency-domain, and time-frequency-domain features from electrical signals, mechanical vibration signals, and thermal imaging temperature signals respectively. Perform feature-level fusion based on cross-modal attention mechanism and output fused feature vector.

[0008] S3. Based on the fused feature vector, identify the fault type, assess the fault severity, and locate the fault location, and output the fault diagnosis results;

[0009] S4. Based on the fused feature vector and fault diagnosis results, perform health status assessment, remaining service life prediction and hierarchical alarm, and output alarm commands.

[0010] As a preferred embodiment of the present invention, S1 includes the following steps:

[0011] S101. Collect the three-phase current, zero-sequence current, three-phase voltage, and opening / closing coil current signals of the pole-mounted circuit breaker;

[0012] S102. Collect the vibration acceleration of the circuit breaker body and the stroke displacement signal of the operating mechanism;

[0013] S103. Acquire infrared thermal images of the arc-extinguishing chamber, contacts, and terminals;

[0014] S104. Perform wavelet denoising, z-score normalization, and GPS timestamp-based multimodal time synchronization alignment to output a standardized multimodal data stream.

[0015] In a preferred embodiment of the present invention, in S104, the electrical signal and the mechanical vibration signal are decomposed into 5 levels using db4 wavelet decomposition, and the high-frequency coefficients of each level are processed using an adaptive threshold for soft thresholding. The expression for the adaptive threshold is as follows:

[0016]

[0017] In the above formula, For adaptive wavelet denoising threshold, The standard deviation of the noise signal. The number of signal sampling points. The input signal-to-noise ratio. For reference signal-to-noise ratio, This is the threshold adjustment coefficient. It is a non-linear exponent.

[0018] As a preferred embodiment of the present invention, S2 includes the following steps:

[0019] S201. Extract fundamental amplitude, harmonic distortion rate, phase difference, and active power fluctuation characteristics;

[0020] S202. Extract vibration peak value, energy spectral density, envelope kurtosis and wavelet packet energy ratio features;

[0021] S203. Extract the features of the highest temperature, temperature gradient, and thermal time constant;

[0022] S204. Based on the cross-modal attention mechanism, the three-modal features are weighted and fused to output the fused feature vector.

[0023] As a preferred embodiment of the present invention, in S204, the expression for the cross-modal attention fusion weights is as follows:

[0024]

[0025] In the above formula, Let be the attention fusion weight of the i-th query feature and the j-th key feature. The query vector is obtained by linear transformation of the i-th modal feature. The key vector is obtained by linear transformation of the j-th modal feature. The dimension of the key feature vector is... Let be the similarity matrix between the i-th mode and the j-th mode, calculated based on prior knowledge. This is the prior equilibrium factor.

[0026] As a preferred embodiment of the present invention, S3 includes the following steps:

[0027] S301. Based on an integrated classifier, identify fault types such as overload, short circuit, failure to operate, malfunction, and insulation aging.

[0028] S302. Evaluate the severity level of a fault based on fused feature vectors;

[0029] S303. Locate the fault location to the contacts, operating mechanism, insulation component or opening / closing coil based on the characteristic contribution degree.

[0030] In a preferred embodiment of the present invention, in S301, a fault confidence score is calculated by weighted fusion based on the historical accuracy of each base classifier, as shown in the following expression:

[0031]

[0032] In the above formula, Score the confidence level of the fault. Let be the fault probability output by the k-th base classifier. Let be the accuracy of the k-th base classifier on the historical validation set. The confidence level sharpening coefficient;

[0033] The ensemble classifier consists of five base classifiers: two random forest classifiers, two gradient boosting tree classifiers, and one one-dimensional convolutional neural network classifier.

[0034] As a preferred embodiment of the present invention, S4 includes the following steps:

[0035] S401. Calculate the health index based on multi-dimensional degradation indicators;

[0036] S402. Predicting remaining useful life based on health index time series;

[0037] S403: Based on health index, remaining service life and fault diagnosis results, make hierarchical alarm decisions and output alarm commands.

[0038] In a preferred embodiment of the present invention, in S401, the expression for the health index is as follows:

[0039]

[0040]

[0041] In the above formula, For health index, Let i be the current value of the i-th degradation index. Let i be the failure threshold for the i-th degradation index. Let be the weight of the i-th degradation index, and n be the non-linear degradation index;

[0042] In S403, based on health index and the predicted remaining useful life Implement tiered alarms, including:

[0043] when and The timing was deemed normal.

[0044] when or The timing is opportune, and it is deemed worthy of attention.

[0045] when or The timing was deemed opportune, warranting an early warning.

[0046] when or The timing was deemed urgent.

[0047] A hyper-converged pole-mounted circuit breaker, employing the aforementioned fault detection method for hyper-converged pole-mounted circuit breakers.

[0048] The beneficial effects of this invention are: this invention can perform early and accurate detection and remaining life prediction of various types of faults in pole-mounted circuit breakers, which significantly improves the accuracy of fault identification. Attached Figure Description

[0049] Figure 1 This is a diagram illustrating the steps of the method of the present invention;

[0050] Figure 2 This is a flowchart of the method of the present invention;

[0051] Figure 3 This is a schematic diagram of the multi-source heterogeneous data acquisition and preprocessing module of the present invention;

[0052] Figure 4 This is a schematic diagram of the multimodal feature extraction and fusion module of the present invention;

[0053] Figure 5 This is a schematic diagram of the fault diagnosis and classification module of the present invention;

[0054] Figure 6 This is a schematic diagram of the fault prediction and alarm module of the present invention;

[0055] In the diagram: 1. Multi-source heterogeneous data acquisition and preprocessing module; 101. Electrical signal acquisition unit; 102. Mechanical vibration signal acquisition unit; 103. Thermal imaging temperature acquisition unit; 104. Data preprocessing unit; 2. Multimodal feature extraction and fusion module; 201. Electrical feature extraction unit; 202. Mechanical feature extraction unit; 203. Thermal feature extraction unit; 204. Cross-modal fusion unit; 3. Fault diagnosis and classification module; 301. Fault type identification unit; 302. Fault severity assessment unit; 303. Fault location unit; 4. Fault prediction and alarm module; 401. Health status assessment unit; 402. Lifespan prediction unit; 403. Alarm decision unit. Detailed Implementation

[0056] To make the technical problems solved by the present invention, the technical solutions adopted, and the technical effects achieved clearer, the technical solutions of the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Example

[0058] like Figures 1 to 4 As shown, a fault detection method for a hyper-converged pole-mounted circuit breaker includes the following steps:

[0059] S1. Collect electrical signals, mechanical vibration signals, and thermal imaging temperature signals from the pole-mounted circuit breaker, perform noise reduction, normalization, and time synchronization alignment, and output a standardized multimodal data stream;

[0060] S2. Extract time-domain, frequency-domain, and time-frequency-domain features from electrical signals, mechanical vibration signals, and thermal imaging temperature signals respectively. Perform feature-level fusion based on cross-modal attention mechanism and output fused feature vector.

[0061] S3. Based on the fused feature vector, identify the fault type, assess the fault severity, and locate the fault location, and output the fault diagnosis results;

[0062] S4. Based on the fused feature vector and fault diagnosis results, perform health status assessment, remaining service life prediction and hierarchical alarm, and output alarm commands.

[0063] In step S1, the electrical signals, mechanical vibration signals, and thermal imaging temperature signals of the pole-mounted circuit breaker are acquired through the multi-source heterogeneous data acquisition and preprocessing module. Noise reduction, normalization, and time synchronization alignment are then performed to output a standardized multimodal data stream, including the following steps:

[0064] The multi-source heterogeneous data acquisition and preprocessing module is used to acquire electrical, mechanical, and thermal multi-modal signals during the operation of the pole-mounted circuit breaker and perform standardized preprocessing. The module includes an electrical signal acquisition unit, a mechanical vibration signal acquisition unit, a thermal imaging temperature acquisition unit, and a data preprocessing unit. The outputs of the electrical signal acquisition unit, mechanical vibration signal acquisition unit, and thermal imaging temperature acquisition unit are connected to the input of the data preprocessing unit, and the output of the data preprocessing unit is connected to the input of the multi-modal feature extraction and fusion module.

[0065] S101, The electrical signal acquisition unit acquires the three-phase current, zero-sequence current, three-phase voltage, and opening and closing coil current signals of the pole-mounted circuit breaker;

[0066] The electrical signal acquisition unit is used to acquire electrical quantities from the primary circuit and control circuit of the circuit breaker. The electrical signal acquisition unit includes a Rogowski coil current transformer, a capacitive voltage divider transformer, a Hall current sensor, and a signal conditioning circuit. The Rogowski coil current transformer is connected to the three-phase outgoing cables of the circuit breaker, the capacitive voltage divider transformer is connected in parallel to the three-phase busbar, and the Hall current sensor is connected in series in the opening and closing coil circuit. The outputs of the three sensors are amplified and filtered by the signal conditioning circuit before being output to the data preprocessing unit. The electrical signal acquisition unit includes the following steps:

[0067] The S1011 Rogowski coil current transformer acquires the instantaneous values ​​of three-phase current and zero-sequence current at a sampling rate of 10kHz, and outputs a sequence of instantaneous three-phase current values. , , With the instantaneous value sequence of zero-sequence current ;

[0068] S1012, the capacitive voltage divider transformer acquires the instantaneous values ​​of the three-phase voltage at a sampling rate of 10kHz and outputs a sequence of instantaneous three-phase voltage values. , , ;

[0069] The S1013 Hall current sensor acquires the opening and closing coil current waveforms at a 50kHz sampling rate and outputs a sequence of instantaneous coil current values. ;

[0070] S1014, the signal conditioning circuit performs anti-aliasing low-pass filtering and programmable amplification on the four signals, sets the cutoff frequency to 0.4 times the sampling rate, and outputs the conditioned electrical signal sequence to the data preprocessing unit.

[0071] S102, Mechanical vibration signal acquisition unit acquires the vibration acceleration of the circuit breaker body and the stroke displacement signal of the operating mechanism;

[0072] The mechanical vibration signal acquisition unit is used to acquire mechanical vibration and displacement signals during the opening and closing operations of the circuit breaker. The mechanical vibration signal acquisition unit includes a piezoelectric accelerometer, a magnetic grating displacement sensor, and a charge amplifier. The piezoelectric accelerometer is installed on the side wall of the circuit breaker's arc-extinguishing chamber, and the magnetic grating displacement sensor is installed on the main shaft of the operating mechanism. The outputs from both sensors are converted by the charge amplifier and then output to the data preprocessing unit. The mechanical vibration signal acquisition unit includes the following steps:

[0073] The S1021 piezoelectric accelerometer collects vibration acceleration data throughout the entire opening and closing operation process at a sampling rate of 25kHz, and outputs a vibration acceleration sequence. ;

[0074] S1022, the magnetic grating displacement sensor acquires the stroke displacement of the main shaft of the operating mechanism at a sampling rate of 5kHz, and outputs the stroke displacement sequence. ;

[0075] S1023, the charge amplifier converts the piezoelectric charge signal into a voltage signal and performs impedance matching, then outputs a standardized mechanical vibration signal to the data preprocessing unit.

[0076] S103, The thermal imaging temperature acquisition unit acquires infrared thermal images of the arc-extinguishing chamber, contacts and terminals;

[0077] The thermal imaging temperature acquisition unit is used to acquire the temperature distribution of key parts of the circuit breaker. The unit includes an uncooled infrared thermal imager, an optical lens, and an image acquisition card. The infrared thermal imager is installed 1.5 meters below the circuit breaker, and the optical lens is aimed at the arc-extinguishing chamber and terminal area. The thermal imager output is digitized by the image acquisition card and then sent to the data preprocessing unit. The thermal imaging temperature acquisition unit includes the following steps:

[0078] S1031, the infrared thermal imager acquires infrared thermal images at a sampling rate of 1Hz and outputs an infrared thermal image sequence. ;

[0079] S1032, The image acquisition card performs non-uniformity correction and blind pixel compensation on the infrared thermal image, and outputs the corrected infrared thermal image sequence. Non-uniformity correction is performed by the two-point correction method.

[0080] S1033. Perform region of interest segmentation on the corrected infrared thermal image, extract the temperature matrix of the arc-extinguishing chamber region, contact region and terminal region, and output it to the data preprocessing unit.

[0081] S104, the data preprocessing unit performs wavelet denoising, z-score normalization, and GPS timestamp-based multimodal time synchronization alignment, outputting a standardized multimodal data stream;

[0082] The data preprocessing unit is used to perform denoising, normalization, and time synchronization on multi-source heterogeneous signals. The data preprocessing unit includes an adaptive wavelet denoising subunit, a z-score normalization subunit, and a GPS time synchronization subunit. The input of the adaptive wavelet denoising subunit is connected to the outputs of the electrical signal acquisition unit, the mechanical vibration signal acquisition unit, and the thermal imaging temperature acquisition unit. The output of the adaptive wavelet denoising subunit is connected to the input of the z-score normalization subunit. The output of the z-score normalization subunit is connected to the input of the GPS time synchronization subunit. The output of the GPS time synchronization subunit is connected to the multimodal feature extraction and fusion module. The data preprocessing unit includes the following steps:

[0083] S1041, the adaptive wavelet denoising subunit performs 5-level decomposition of the electrical signal and mechanical vibration signal using db4 wavelet. The high-frequency coefficients of each level are processed using an adaptive threshold for soft thresholding. The expression for the adaptive threshold is as follows:

[0084]

[0085] In the above formula, For adaptive wavelet denoising threshold, The standard deviation of the noise signal is estimated from the median absolute deviation of the wavelet coefficients at the finest scale. The number of signal sampling points. The input signal-to-noise ratio. For reference signal-to-noise ratio, This is the threshold adjustment coefficient. It is a non-linear exponent. In this embodiment, , .

[0086] S1042, the z-score normalization subunit, performs zero-mean unit variance normalization on each modal signal after denoising, and outputs a standardized signal sequence, as shown in the following expression:

[0087]

[0088] In the above formula, The mean of the signal. The standard deviation of the signal. The original sample signal sequence, It is a standardized signal sequence.

[0089] S1043, the GPS time synchronization subunit performs time alignment of electrical signals, mechanical vibration signals and thermal imaging temperature signals based on GPS timestamps. Using the electrical signal timestamp as a reference, it performs linear interpolation resampling on the mechanical vibration signal and nearest neighbor interpolation resampling on the thermal imaging temperature signal, and outputs a time-synchronized standardized multimodal data stream to the multimodal feature extraction and fusion module.

[0090] In step S2, time-domain, frequency-domain, and time-frequency-domain features are extracted from electrical signals, mechanical vibration signals, and thermal imaging temperature signals respectively through the multimodal feature extraction and fusion module. Feature-level fusion is performed based on the cross-modal attention mechanism to output a fused feature vector, including the following steps:

[0091] The multimodal feature extraction and fusion module is used to extract electrical, mechanical, and thermal features from a standardized multimodal data stream and perform cross-modal fusion. The module includes an electrical feature extraction unit, a mechanical feature extraction unit, a thermal feature extraction unit, and a cross-modal fusion unit. The inputs of the electrical, mechanical, and thermal feature extraction units are connected to their respective outputs of the data preprocessing unit. Their outputs are connected to the input of the cross-modal fusion unit, and the output of the cross-modal fusion unit is connected to the input of the fault diagnosis and classification module. The multimodal feature extraction and fusion module includes the following steps:

[0092] S201, the electrical feature extraction unit extracts the fundamental amplitude, harmonic distortion rate, phase difference, and active power fluctuation characteristics;

[0093] The electrical feature extraction unit extracts deep features reflecting the electrical state of the circuit breaker from the electrical signal. The electrical feature extraction unit consists of an FFT transformation subunit, a harmonic analysis subunit, and a power calculation subunit. The electrical feature extraction unit includes the following steps:

[0094] S2011, the FFT transformation subunit performs Hanning window FFT transformation on the three-phase current and three-phase voltage to extract the fundamental amplitude and phase, and outputs the fundamental amplitude feature vector and phase feature vector;

[0095] S2012, the harmonic analysis subunit calculates the harmonic content rate of the 2nd to 25th harmonics and the total harmonic distortion rate (THD), and outputs the harmonic characteristic vector;

[0096] S2013, The power calculation subunit calculates active power, reactive power and power fluctuation variance based on the instantaneous product of voltage and current, and outputs a power feature vector;

[0097] S2014. Concatenate the above features into a 32-dimensional electrical feature vector. The output is sent to the cross-modal fusion unit.

[0098] S202, the mechanical feature extraction unit extracts vibration peak value, energy spectral density, envelope kurtosis and wavelet packet energy ratio features;

[0099] The mechanical feature extraction unit is used to extract time-frequency features reflecting the state of the operating mechanism from the mechanical vibration signal. The mechanical feature extraction unit includes a time-domain statistics subunit, a frequency-domain analysis subunit, and a time-frequency analysis subunit. The mechanical feature extraction unit includes the following steps:

[0100] S2021, the time-domain statistical sub-unit calculates the peak value, root mean square value, kurtosis and margin factor of vibration acceleration, and outputs the time-domain statistical feature vector;

[0101] S2022, the frequency domain analysis subunit performs FFT transformation on the vibration signal and calculates the energy spectral density and centroid frequency, and outputs the frequency domain feature vector;

[0102] S2023, the time-frequency analysis subunit performs 4-level db4 wavelet packet decomposition on the vibration signal, calculates the energy ratio of each frequency band, and outputs the wavelet packet energy ratio feature vector.

[0103] S2024. Calculate the opening and closing times, stroke curve slope, and rebound amplitude of the travel displacement signal, and concatenate them with vibration characteristics to form a 28-dimensional mechanical feature vector. The output is sent to the cross-modal fusion unit.

[0104] S203, the thermal feature extraction unit extracts the features of the highest temperature, temperature gradient and thermal time constant;

[0105] The thermal feature extraction unit is used to extract features reflecting the thermal state of the circuit breaker from infrared thermal images. The thermal feature extraction unit includes a temperature statistics subunit and a thermodynamic analysis subunit, and includes the following steps:

[0106] S2031, the temperature statistics subunit extracts the highest temperature, average temperature and temperature standard deviation of the arc-extinguishing chamber, contact and terminal area, and outputs the temperature statistics feature vector;

[0107] S2032, the thermodynamic analysis subunit calculates the temperature rise gradient and thermal time constant based on the temperature time series, and outputs a thermodynamic feature vector;

[0108] S2033. Concatenate the above features into a 16-dimensional thermal feature vector. The output is sent to the cross-modal fusion unit.

[0109] S204. The cross-modal fusion unit performs weighted fusion of three-modal features based on a cross-modal attention mechanism and outputs a fused feature vector.

[0110] The cross-modal fusion unit is used to perform weighted fusion of electrical, mechanical, and thermal features. The cross-modal fusion unit includes a feature mapping subunit, an attention calculation subunit, and a fusion output subunit. The cross-modal fusion unit includes the following steps:

[0111] S2041, The feature mapping subunit will map the electrical feature vector Mechanical feature vectors With thermal eigenvectors Each feature vector is mapped to a 64-dimensional unified feature space via a fully connected layer. , , ;

[0112] S2042. The attention calculation subunit calculates the inter-modal attention weights. The expression for the cross-modal attention fusion weights is as follows:

[0113]

[0114] In the above formula, Let be the attention fusion weight of the i-th query feature and the j-th key feature. The query vector is obtained by linear transformation of the i-th modal feature. The key vector is obtained by linear transformation of the j-th modal feature. The dimension of the key feature vector is 64. Let be the similarity matrix between the i-th mode and the j-th mode, calculated based on prior knowledge. This is the prior equilibrium factor, with a value of 0.2.

[0115] S2043. The fusion output subunit performs a weighted summation of the three-modal features based on attention weights, and then compresses them into a 128-dimensional fusion feature vector through two fully connected layers. The output is sent to the fault diagnosis and classification module.

[0116] In step S3, the fault diagnosis and classification module performs fault type identification, severity assessment, and fault location based on the fused feature vector, and outputs the fault diagnosis result, including the following steps:

[0117] The fault diagnosis and classification module performs fault type identification, severity assessment, and fault location based on fused feature vectors. The module includes a fault type identification unit, a fault severity assessment unit, and a fault location unit. The input of the fault type identification unit is connected to the output of the cross-modal fusion unit. The outputs of the fault type identification unit are connected to the inputs of the fault severity assessment unit and the fault location unit, respectively. The outputs of all three units are connected to the fault prediction and alarm module. The fault diagnosis and classification module includes the following steps:

[0118] S301, The fault type identification unit identifies overload, short circuit, failure to operate, malfunction and insulation aging fault types based on an integrated classifier;

[0119] The fault type identification unit identifies circuit breaker fault types based on fused feature vectors. The fault type identification unit includes a feature input subunit, an integrated classifier subunit, and a confidence calculation subunit. The fault type identification unit includes the following steps:

[0120] S3011, Feature Input Subunit receives 128-dimensional fused feature vector Normalization is performed, and the normalized feature vector is output.

[0121] S3012, the ensemble classifier subunit consists of 5 base classifiers, including 2 random forest classifiers, 2 gradient boosting tree classifiers and 1 one-dimensional convolutional neural network classifier. Each base classifier outputs the probability distribution of 5 types of faults.

[0122] S3013. The confidence calculation subunit performs weighted fusion based on the historical accuracy of each base classifier to calculate the fault confidence score, as shown in the following expression:

[0123]

[0124] In the above formula, Score the confidence level of the fault. Let be the fault probability output by the k-th base classifier. Let be the accuracy of the k-th base classifier on the historical validation set. This is the confidence level sharpening factor, with a value of 4.

[0125] S3014. Output the fault type label and fault confidence score to the fault severity assessment unit and fault location unit.

[0126] S302. The fault severity assessment unit assesses the fault severity level based on the fused feature vector;

[0127] The fault severity assessment unit is used to quantify the severity of a fault. The fault severity assessment unit includes a severity regression subunit and a level mapping subunit. The fault severity assessment unit includes the following steps:

[0128] S3021, Severity Regression Subunit Based on Fusion Feature Vector F_ and the fault type label are used to output a continuous severity value through a support vector regression model. ;

[0129] S3022, The severity mapping subunit maps continuous severity values ​​to three levels: mild, moderate, and severe.

[0130] in, It is mild. Moderate. It is severe.

[0131] S3023, Output the fault severity level to the fault prediction and alarm module.

[0132] S303, the fault location unit locates the fault location to the contact, operating mechanism, insulation component or opening / closing coil based on the characteristic contribution degree.

[0133] The fault location unit is used to pinpoint the specific location of a fault. The fault location unit consists of a SHAP contribution analysis subunit and a location mapping subunit, and includes the following steps:

[0134] S3031, SHAP contribution analysis subunit calculates the contribution of each modal feature in the fused feature vector to the fault type identification result based on the SHAP value, and outputs the feature contribution ranking;

[0135] S3032, The location mapping subunit locates the fault location to the contact, operating mechanism, insulating part or opening / closing coil based on the feature contribution ranking and mode and location mapping table.

[0136] The modality and location mapping table is constructed based on failure mechanisms and historical cases and is used to map modes and locations.

[0137] S3033, Output the fault location result to the fault prediction and alarm module.

[0138] In step S4, the fault prediction and alarm module performs health status assessment, remaining life prediction, and hierarchical alarm decision-making based on the fused feature vector and fault diagnosis results, and outputs alarm commands, including the following steps:

[0139] The fault prediction and alarm module performs health status assessment, remaining lifetime prediction, and hierarchical alarm decision-making based on fused feature vectors and fault diagnosis results. The module includes a health status assessment unit, a lifetime prediction unit, and an alarm decision-making unit. The input of the health status assessment unit is connected to the output of the cross-modal fusion unit, and the output of the health status assessment unit is connected to the input of the lifetime prediction unit. The output of the lifetime prediction unit and the output of the fault diagnosis and classification module are jointly connected to the input of the alarm decision-making unit. The output of the alarm decision-making unit outputs an alarm command. The fault prediction and alarm module includes the following steps:

[0140] S401, The health status assessment unit calculates the health index based on multi-dimensional degeneration indicators;

[0141] The health status assessment unit is used to quantify the overall health status of the circuit breaker. The health status assessment unit includes a degradation index extraction subunit and a health index calculation subunit, and includes the following steps:

[0142] S4011, Degradation index extraction subunit extracts contact ablation degree from fused feature vector. Mechanism wear Insulation aging degree With coil degradation Four degradation indicators are used, which are obtained by quantifying the deviation between the corresponding modal characteristics and the baseline value.

[0143] S4012, The health index calculation subunit calculates the health index based on a nonlinear weighted model. The expression for the health index is as follows:

[0144]

[0145]

[0146] In the above formula, For health index, Let i be the current value of the i-th degradation index. Let i be the failure threshold for the i-th degradation index. Let be the weight of the i-th degradation index, and n be the nonlinear degradation index. In this embodiment, n is 2. The values ​​are 0.35, 0.30, 0.20, and 0.15, respectively.

[0147] S4013, Output health index to life prediction unit.

[0148] S402, The lifespan prediction unit predicts the remaining lifespan based on the health index time series;

[0149] The life prediction unit is used to predict the remaining operating time of a circuit breaker from its current state to failure. The life prediction unit consists of an LSTM prediction subunit and a RUL calculation subunit, and includes the following steps:

[0150] The S4021 LSTM prediction subunit takes the health index time series of the past 30 days as input, predicts the trend of health index changes in the next 7 days through a two-layer LSTM network, and outputs the rate of health index degradation. The LSTM network has a hidden layer dimension of 64 and uses the Adam optimizer with a learning rate of 1e-3 during training.

[0151] S4022, the RUL calculation sub-unit calculates the remaining useful life based on the current health index and degradation rate. The expression for the remaining useful life prediction is as follows:

[0152]

[0153] In the above formula, Let t be the predicted remaining useful life at time t. Let t be the health index at time t. The failure threshold Let t be the rate of decline of the health index. These are the initial correction coefficients. This is the time decay coefficient. In this embodiment, The value is 0.3. The value is 0.1. The value is 0.005.

[0154] S4023, Output the remaining useful life prediction value to the alarm decision unit.

[0155] S403 The alarm decision unit makes hierarchical alarm decisions based on health index, remaining service life and fault diagnosis results, and outputs alarm commands.

[0156] The alarm decision unit executes tiered alarms based on multi-dimensional evaluation results. The alarm decision unit consists of a rule engine subunit and an alarm output subunit, and includes the following steps:

[0157] S4031, the rule engine subunit performs multi-condition fusion judgment based on health index, remaining service life, fault type, fault severity and fault location, and outputs four alarm levels.

[0158] when and The timing was deemed normal; when or The timing was deemed opportune; when... or The timing was deemed a warning; when or The timing was deemed critical.

[0159] S4032, the alarm output subunit encapsulates the alarm level, fault type, fault location, health index and remaining service life into an alarm message, and uploads it to the distribution network master station system through the 5G communication module. At the same time, the LED indicator on the edge side of the pole-mounted circuit breaker performs hierarchical display.

[0160] When the alarm level is emergency, the alarm output subunit synchronously triggers the circuit breaker tripping and blocking signal to prevent the fault from escalating.

[0161] The test was conducted using the scheme of this embodiment, and compared with traditional electrical testing, traditional vibration testing, and traditional infrared testing methods. The test items included fault identification accuracy, early warning lead time, and RUL prediction error. The test results are shown in Table 1.

[0162] Table 1 Comparison of Fault Detection Methods

[0163]

[0164] As shown in the table above, the fault identification accuracy of this application reaches 98.6%, which is more than 20% higher than the fault identification rate of traditional single signal source detection methods. Furthermore, it can provide early warning, with an advance warning period of up to 23.5 days. This invention can perform early and accurate detection and remaining life prediction of multiple types of faults in pole-mounted circuit breakers, significantly improving the fault identification accuracy.

[0165] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Clearly, those skilled in the art can make various alterations and variations to the invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of equivalents of the invention, the invention is also intended to include these modifications and variations.

Claims

1. A method for fault detection of a hyper-converged pole-mounted circuit breaker, characterized in that, Includes the following steps: S1. Collect electrical signals, mechanical vibration signals, and thermal imaging temperature signals from the pole-mounted circuit breaker, perform noise reduction, normalization, and time synchronization alignment, and output a standardized multimodal data stream; S2. Extract time-domain, frequency-domain, and time-frequency-domain features from electrical signals, mechanical vibration signals, and thermal imaging temperature signals, respectively. Perform feature-level fusion based on cross-modal attention mechanism and output fused feature vector. S3. Based on the fused feature vector, identify the fault type, assess the fault severity, and locate the fault location, and output the fault diagnosis results; S4. Based on the fused feature vector and fault diagnosis results, perform health status assessment, remaining service life prediction and hierarchical alarm, and output alarm commands.

2. The method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 1, characterized in that, S1 includes the following steps: S101. Collect the three-phase current, zero-sequence current, three-phase voltage, and opening / closing coil current signals of the pole-mounted circuit breaker; S102. Collect the vibration acceleration of the circuit breaker body and the stroke displacement signal of the operating mechanism; S103. Acquire infrared thermal images of the arc-extinguishing chamber, contacts, and terminals; S104. Perform wavelet denoising, z-score normalization, and GPS timestamp-based multimodal time synchronization alignment to output a standardized multimodal data stream.

3. The method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 2, characterized in that, In S104, the electrical signal and mechanical vibration signal are decomposed into 5 levels using db4 wavelet decomposition. Adaptive thresholding is applied to the high-frequency coefficients of each level for soft thresholding. The expression for the adaptive threshold is as follows: In the above formula, For adaptive wavelet denoising threshold, The standard deviation of the noise signal. The number of signal sampling points. The input signal-to-noise ratio. For reference signal-to-noise ratio, This is the threshold adjustment coefficient. It is a non-linear exponent.

4. The method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 1, characterized in that, S2 includes the following steps: S201. Extract fundamental amplitude, harmonic distortion rate, phase difference, and active power fluctuation characteristics; S202. Extract vibration peak value, energy spectral density, envelope kurtosis and wavelet packet energy ratio features; S203. Extract the features of the highest temperature, temperature gradient, and thermal time constant; S204. Based on the cross-modal attention mechanism, the three-modal features are weighted and fused to output the fused feature vector.

5. The method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 4, characterized in that, In S204, the expression for the cross-modal attention fusion weights is as follows: In the above formula, Let be the attention fusion weight of the i-th query feature and the j-th key feature. The query vector is obtained by linear transformation of the i-th modal feature. The key vector is obtained by linear transformation of the j-th modal feature. The dimension of the key feature vector is... Let be the similarity matrix between the i-th mode and the j-th mode, calculated based on prior knowledge. This is the prior equilibrium factor.

6. The method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 1, characterized in that, S3 includes the following steps: S301. Based on an integrated classifier, identify fault types such as overload, short circuit, failure to operate, malfunction, and insulation aging. S302. Evaluate the severity level of a fault based on fused feature vectors; S303. Locate the fault location to the contacts, operating mechanism, insulation component, or opening / closing coil based on the characteristic contribution.

7. The method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 6, characterized in that, In S301, a weighted fusion is performed based on the historical accuracy of each base classifier to calculate the fault confidence score, as shown in the following expression: In the above formula, Score the confidence level of the fault. Let be the fault probability output by the k-th base classifier. Let be the accuracy of the k-th base classifier on the historical validation set. The confidence level sharpening coefficient; The ensemble classifier consists of five base classifiers: two random forest classifiers, two gradient boosting tree classifiers, and one one-dimensional convolutional neural network classifier.

8. The method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 1, characterized in that, S4 includes the following steps: S401. Calculate the health index based on multi-dimensional degradation indicators; S402. Predicting remaining useful life based on health index time series; S403: Based on health index, remaining service life and fault diagnosis results, make hierarchical alarm decisions and output alarm commands.

9. A method for fault detection of a hyper-converged pole-mounted circuit breaker according to claim 8, characterized in that, In S401, the expression for the health index is as follows: In the above formula, For health index, Let i be the current value of the i-th degradation index. Let i be the failure threshold for the i-th degradation index. Let be the weight of the i-th degradation index, and n be the non-linear degradation index; In S403, based on health index and the predicted remaining useful life Implement tiered alarms, including: when and The timing was deemed normal. when or The timing is opportune, and it is deemed worthy of attention. when or The timing was deemed opportune, warranting an early warning. when or The timing was deemed critical.

10. A hyper-converged pole-mounted circuit breaker, characterized in that, The method for fault detection of a hyper-converged pole-mounted circuit breaker as described in any one of claims 1-9 is adopted.

Images