A residual current fault judgment method and system based on multi-feature fusion

Abstract

The application provides a residual current fault judgment method and system based on multi-feature fusion, and relates to the field of low-voltage power distribution network protection. The residual current signal, the environmental temperature signal and the load current signal in the low-voltage power distribution network line are synchronously collected, the time-frequency feature, the temperature change rate feature and the load fluctuation coefficient feature of the residual current are extracted after pretreatment, and the multi-dimensional feature vector is constructed. The weighted fusion weight is dynamically adjusted through the feature-fault sensitivity mapping matrix, the dynamic fusion characteristic value is calculated and compared with the preset fault threshold, whether there is a residual current fault is judged, the fault type is identified, and the signals judged as faults are secondarily verified. The fusion feature time sequence is constructed, the adaptive ARIMA model is used for short-time prediction, the early warning signal is output and the potential fault type is marked when the prediction value reaches the early warning condition. The application improves the accuracy and reliability of fault judgment and has early warning capability.

Description

Technical Field

[0001] This invention relates to the field of low-voltage distribution network protection, and in particular to a residual current fault judgment method and system based on multi-feature fusion. Background Technology

[0002] Residual current monitoring systems are core equipment for preventing electrical fires and protecting people from electric shock in low-voltage power distribution networks. They are widely used in residential communities, industrial parks, commercial complexes, and other locations. Their basic principle is to detect residual current in the circuit using a zero-sequence current transformer. When the residual current exceeds a preset threshold, an alarm is triggered or the circuit is cut off, thereby preventing electrical fires and electric shock accidents.

[0003] With the increasing complexity of low-voltage distribution network load structures, harmonic interference, electromagnetic noise, and load fluctuations in lines have significantly increased, placing higher demands on the accuracy and reliability of residual current fault diagnosis. Traditional residual current protection devices mainly rely on single characteristic quantities for judgment. While their technical architecture is simple and their response speed is fast, they are prone to misjudgment and missed judgment under complex operating conditions. In recent years, academia and industry have begun to explore residual current fault diagnosis methods based on multi-feature fusion. By extracting the time-domain and frequency-domain features of the residual current and combining them with auxiliary information such as ambient temperature and load current, a comprehensive judgment of the fault state can be made, aiming to improve the judgment accuracy and anti-interference capability.

[0004] However, existing multi-feature fusion-based technical solutions still have several common defects, mainly reflected in the static nature of feature fusion methods, the disconnect between signal preprocessing and feature extraction, the lack of fault prediction capabilities, and the absence of system self-diagnosis mechanisms, which limit their engineering application effectiveness under complex working conditions.

[0005] Therefore, there is a need to provide a residual current fault detection method and system based on multi-feature fusion to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a residual current fault judgment method and system based on multi-feature fusion. Through dynamic weighted fusion, improved Kalman filtering to retain key frequency bands, ARIMA trend prediction and closed-loop self-diagnosis, the accuracy, reliability and initiative of residual current fault judgment under complex operating conditions are significantly improved, and the core defects of the prior art are effectively solved.

[0007] To achieve the above objectives, this invention provides a residual current fault detection method based on multi-feature fusion, comprising the following steps: S1: The real-time residual current signal in the low-voltage distribution network line is collected by the zero-sequence current transformer as the original residual current signal. The ambient temperature signal and load current signal are collected simultaneously to form a multi-source original signal set. S2: Preprocess the original multi-source signal set formed in S1 to obtain a standardized multi-source signal set; S3: Extract the core features of each signal from the standardized multi-source signal set to form a multi-dimensional feature vector; the residual current signal features include time domain features and frequency domain features, the ambient temperature signal features are the temperature change rate, and the load current signal features are the load fluctuation coefficient; S4: Construct a feature-fault sensitivity mapping matrix, dynamically adjust the weight coefficients of the weighted fusion algorithm based on the real-time values ​​of each feature parameter, and calculate the dynamic fusion feature value; S5: Fault diagnosis, using the dynamic fusion feature values ​​calculated in S4. With preset fault threshold If the dynamic fusion feature value is not less than the preset fault threshold, the line is determined to have a residual current fault. The fault type is identified based on the value range of each feature parameter and the activated sensitive feature set; otherwise, the line is determined to be normal. S6: Perform secondary verification on the fault signal identified in S5. Use S-transform to extract the time-frequency characteristics of the residual current under fault conditions and match them with preset fault feature templates. If the matching degree is ≥95%, the fault is confirmed; otherwise, it is judged as a false judgment and the process returns to S1 for signal acquisition. S7: Construct a fusion feature time series by arranging the dynamically fused feature values ​​obtained from multiple consecutive sliding windows in chronological order. An adaptive ARIMA model is used to perform short-term predictions on the fused feature time series, obtaining the predicted fused feature values ​​for the next T time windows. ; S8: When predicting fusion feature values At that time, it outputs a warning signal and marks the potential fault type, realizing early warning of faults, among which This is the early warning coefficient, with a value range of [0.6-0.8].

[0008] Preferably, the specific process of preprocessing the multi-source raw signal set in S2 is as follows: S21: An improved Kalman filter algorithm is used to eliminate interference noise in the original residual current signal, resulting in a denoised residual current signal. ; The state equation and observation equation of the improved Kalman filter algorithm are as follows: ; ; In the formula, The system state vector at time k contains only the state variable of the amplitude of the denoised residual current signal, and is unique. This is the state transition matrix, with a fixed value of 0.98. This is the system noise driving matrix, with a value of 1; for The system noise at time t follows a mean of 0 and a variance of 0. The normal distribution; for k The observation vector at time; This is the observation matrix, with a value of 1. for k The observation noise at time t follows a mean of 0 and a variance of 0. The normal distribution; S22: The improved Kalman filter algorithm introduces an adaptive noise covariance matching mechanism in state estimation to estimate and update the system noise variance in real time. and observation noise variance ; S23: For the fundamental frequency of the residual current and its characteristic harmonic components, the preset retention frequency band is 0-500Hz, and the state transition matrix is ​​adjusted accordingly. The bandwidth parameters are set to ensure that the signal in the fault characteristic frequency band passes through without attenuation, while filtering out high-frequency switching noise and pulse interference above 500Hz. S24: The denoised residual current signal, ambient temperature signal, and load current signal are standardized using the sliding window method to eliminate the influence of dimensions, resulting in standardized multi-source signals. , , .

[0009] Preferably, the time-domain characteristics of the residual current signal in S3 include peak value. Valid value and waveform distortion rate Frequency domain characteristics include fundamental frequency amplitude Second harmonic amplitude Fifth harmonic amplitude Each feature parameter is uniquely defined and does not interfere with the others. RMS value of residual current signal In the formula The number of sampling points within a single sliding window. For the first i Each sampling time; Waveform distortion rate In the formula The value is the nth harmonic amplitude of the residual current, where n ≥ 2; Rate of temperature change In the formula The time interval is fixed and is set to 10 seconds. Load fluctuation coefficient In the formula, , , These represent the maximum, minimum, and average values ​​of the standardized load current within a single sliding window.

[0010] Preferably, the dynamically fused feature values ​​in S4 The calculation formula is: = ; In the formula, , , , , These are the dynamic weight coefficients corresponding to each feature, satisfying... , = ; This is a dynamic adjustment coefficient, with a value range of [0.5, 1.5]. Features i An indicator function for whether it belongs to the current activation sensitive feature set; it takes a value of 1 when activated and 0 otherwise. This is the effective value of the residual current. The residual current waveform distortion rate. The amplitude of the second harmonic of the residual current. The rate of change of ambient temperature. This is the load fluctuation coefficient.

[0011] A residual current fault monitoring system based on multi-feature fusion is used to implement a residual current fault judgment method based on multi-feature fusion, including a signal acquisition module, a signal preprocessing module, a feature extraction module, an adaptive feature fusion module, a fault judgment module, a trend prediction module, an alarm and control module, a self-diagnosis module, and a data storage and communication module. The signal acquisition module, signal preprocessing module, feature extraction module, and adaptive feature fusion module are connected in sequence. The adaptive feature fusion module is connected to the fault judgment module and the trend prediction module, respectively. The fault judgment module and the trend prediction module are connected to the alarm and control module. The self-diagnosis module is connected to the signal acquisition module, signal preprocessing module, feature extraction module, adaptive feature fusion module, and fault judgment module, respectively, and the communication module is connected to all of the above modules.

[0012] Preferably, the signal acquisition module includes a zero-sequence current transformer, a temperature sensor, and a current sensor, which are used to acquire residual current signals, ambient temperature signals, and load current signals, respectively. The acquired analog signals are converted into multi-source digital signals and then transmitted to the signal preprocessing module. The signal preprocessing module receives multi-source digital signals transmitted from the signal acquisition module, performs noise reduction using an improved Kalman filter algorithm, performs standardization using a sliding window method, and transmits the processed signal to the feature extraction module. The feature extraction module receives the preprocessed multi-source digital signal, extracts the time-domain and frequency-domain features of the residual current, as well as the feature parameters of the ambient temperature and load current, and forms a multi-dimensional feature vector, which is then transmitted to the adaptive feature fusion module. The adaptive feature fusion module has a built-in adaptive weighted fusion algorithm and feature-fault sensitivity mapping matrix. It receives multi-dimensional feature vectors transmitted by the feature extraction module, dynamically adjusts the weight coefficients according to real-time feature values, calculates dynamically fused feature values, and transmits them to the fault judgment module. The fault diagnosis module has a preset unique fault threshold. By comparing the fused feature values ​​with the fault threshold, it is determined whether there is a fault in the line and the type of fault. At the same time, the fault is verified again through S-transform. The trend prediction module has a built-in adaptive ARIMA prediction model. It receives dynamic fusion feature values ​​from multiple sliding windows, forms a fusion feature time series, and performs short-term prediction. When the predicted value reaches the warning threshold, it outputs a warning signal to the alarm and control module. When the alarm and control module confirms the presence of a residual current fault, it immediately issues an audible and visual alarm signal and simultaneously outputs a control signal to the circuit breaker to disconnect the faulty line; when the trend prediction module outputs a warning signal, it issues a warning prompt but does not perform a tripping operation. The self-diagnostic module periodically injects preset self-test signal waveforms and uses closed-loop detection to check the response output of each module to determine whether each module is working properly. When any module responds abnormally or its output deviation exceeds the preset self-test threshold, a module fault code is generated and uploaded to the remote monitoring center. At the same time, the faulty module is automatically bypassed and the backup algorithm channel is activated. The data storage and communication module is used to store the acquired raw residual current signal, preprocessed multi-source digital signal, characteristic parameters, fused characteristic values ​​and fault judgment results. It also supports communication with the remote monitoring center to upload fault information in real time.

[0013] Preferably, the sampling frequency of the signal acquisition module The frequency is fixed at 1000Hz, the sampling accuracy is 16 bits, the measurement range of the zero-sequence current transformer is 0-500mA, the measurement range of the temperature sensor is -20℃~80℃, the absolute value of the measurement error is no greater than 0.5℃, the measurement range of the current sensor is 0-100A, and the absolute value of the measurement error is no greater than ±1%.

[0014] Preferably, the fault threshold in the fault judgment module The value was obtained through training with 1000 sets of sample data of different fault types, and was set to 0.72. It was then fine-tuned through a remote monitoring center based on actual line operating conditions. In the adaptive feature fusion module, the basic weight coefficients were determined through sample training, and were respectively... =0.35, , , , Dynamic adjustment coefficient The value is 1.0.

[0015] Preferably, the self-test cycle of the self-diagnostic module is set to 24 hours, and the self-test signal waveform includes an amplitude of =50Hz fundamental and second harmonic amplitudes =100Hz, third harmonic amplitude The composite waveform has an amplitude of 50% of the full scale; the absolute value of the normal response deviation threshold of each module does not exceed 5%, otherwise it is judged as a module abnormality.

[0016] Therefore, the present invention employs the above-mentioned residual current fault judgment method and system based on multi-feature fusion, and the technical effects are as follows: (1) The present invention adds an adaptive dynamic weighted fusion mechanism. By constructing a feature-fault sensitivity mapping matrix, the corresponding sensitive feature set is automatically activated according to the real-time feature value and the weight coefficient is dynamically adjusted so that the monitoring system maintains the optimal sensitivity for different fault types such as human electric shock, equipment leakage, and insulation degradation.

[0017] (2) The present invention adopts an improved Kalman filter algorithm, introduces an adaptive noise covariance matching mechanism in state estimation, and presets a 0-500Hz reserved frequency band for the residual current fault characteristic frequency band to ensure that the key fault characteristic signal passes through without attenuation, while effectively filtering out high-frequency switching noise and pulse interference.

[0018] (3) The present invention adds a trend prediction and early warning mechanism. By forming a time series by dynamically fusing feature values ​​of multiple consecutive sliding windows, an adaptive ARIMA model is used for short-term prediction. When the predicted fused feature value reaches the early warning threshold, an early warning signal is output in advance.

[0019] (4) The present invention adds a self-diagnosis module, which performs closed-loop detection on the signal acquisition module, signal preprocessing module, feature extraction module and adaptive feature fusion module by periodically injecting preset self-test signal waveforms. It can automatically locate faulty modules and generate fault codes, and at the same time enable the backup algorithm channel.

[0020] (5) All core parameters in this invention have unique definitions and standardized calculation methods, and the parameters do not overlap or interfere with each other.

[0021] (6) This invention organically integrates five major innovations: adaptive dynamic weighted fusion, improved Kalman filtering, S-transform secondary verification, ARIMA trend prediction, and system self-diagnosis, forming a complete technical closed loop from signal acquisition, preprocessing, feature extraction, adaptive fusion, fault judgment, secondary verification, trend prediction to system self-diagnosis. Attached Figure Description

[0022] Figure 1 This is a flowchart of a residual current fault judgment method based on multi-feature fusion according to the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0025] Example 1 like Figure 1 As shown, this invention provides a residual current fault detection method based on multi-feature fusion, comprising the following steps: S1: The real-time residual current signal in the low-voltage distribution network line is collected using a zero-sequence current transformer as the original residual current signal. The sampling frequency is set to a fixed value. The sampling time interval is Simultaneously collect ambient temperature signals and load current signal This forms a collection of multiple source original signals; S2: Preprocess the original multi-source signal set formed in S1 to obtain a standardized multi-source signal set; The specific process of preprocessing the multi-source raw signal set in S2 is as follows: S21: An improved Kalman filter algorithm is used to eliminate interference noise in the original residual current signal, resulting in a denoised residual current signal. ; The state equation and observation equation of the improved Kalman filter algorithm are as follows: ; ; In the formula, The system state vector at time k contains only the state variable of the amplitude of the denoised residual current signal, and is unique. This is the state transition matrix, with a fixed value of 0.98. This is the system noise driving matrix, with a value of 1; for The system noise at time t follows a mean of 0 and a variance of 0. The normal distribution; for k The observation vector at time; This is the observation matrix, with a value of 1. for k The observation noise at time t follows a mean of 0 and a variance of 0. The normal distribution; S22: The improved Kalman filter algorithm introduces an adaptive noise covariance matching mechanism in state estimation to estimate and update the system noise variance in real time. and observation noise variance ; S23: For the fundamental frequency of the residual current and its characteristic harmonic components, the preset retention frequency band is 0-500Hz, and the state transition matrix is ​​adjusted accordingly. The bandwidth parameters are set to ensure that the signal in the fault characteristic frequency band passes through without attenuation, while filtering out high-frequency switching noise and pulse interference above 500Hz. S24: The denoised residual current signal, ambient temperature signal, and load current signal are standardized using the sliding window method to eliminate the influence of dimensions, resulting in standardized multi-source signals. , , .

[0026] S3: Extract the core features of each signal from the standardized multi-source signal set to form a multi-dimensional feature vector; the residual current signal features include time domain features and frequency domain features, the ambient temperature signal features are the temperature change rate, and the load current signal features are the load fluctuation coefficient; The time-domain characteristics of the residual current signal in S3 include peak value. Valid value and waveform distortion rate Frequency domain characteristics include fundamental frequency amplitude Second harmonic amplitude Fifth harmonic amplitude Each feature parameter is uniquely defined and does not interfere with the others. RMS value of residual current signal In the formula The number of sampling points within a single sliding window. For the first i Each sampling time; Waveform distortion rate In the formula The value is the nth harmonic amplitude of the residual current, where n ≥ 2; Rate of temperature change In the formula The time interval is fixed and is set to 10 seconds. Load fluctuation coefficient In the formula, , , These represent the maximum, minimum, and average values ​​of the standardized load current within a single sliding window.

[0027] S4: Construct a feature-fault sensitivity mapping matrix, dynamically adjust the weight coefficients of the weighted fusion algorithm based on the real-time values ​​of each feature parameter, and calculate the dynamic fusion feature value; Dynamic fusion eigenvalues ​​in S4 The calculation formula is: = ; In the formula, , , , , These are the dynamic weight coefficients corresponding to each feature, satisfying... , = ; This is a dynamic adjustment coefficient, with a value range of [0.5, 1.5]. Features i An indicator function for whether it belongs to the current activation sensitive feature set; it takes a value of 1 when activated and 0 otherwise. This is the effective value of the residual current. The residual current waveform distortion rate. The amplitude of the second harmonic of the residual current. The rate of change of ambient temperature. This is the load fluctuation coefficient.

[0028] S5: Fault diagnosis, using the dynamic fusion feature values ​​calculated in S4. With preset fault threshold If the dynamic fusion feature value is not less than the preset fault threshold, the line is determined to have a residual current fault. The fault type is identified based on the value range of each feature parameter and the activated sensitive feature set; otherwise, the line is determined to be normal. S6: Perform secondary verification on the fault signal identified in S5. Use S-transform to extract the time-frequency characteristics of the residual current under fault conditions and match them with preset fault feature templates. If the matching degree is ≥95%, the fault is confirmed; otherwise, it is judged as a false judgment and the process returns to S1 for signal acquisition. S7: Construct a fusion feature time series by arranging the dynamically fused feature values ​​obtained from multiple consecutive sliding windows in chronological order. An adaptive ARIMA model is used to perform short-term predictions on the fused feature time series, obtaining the predicted fused feature values ​​for the next T time windows. ; S8: When predicting fusion feature values At that time, it outputs a warning signal and marks the potential fault type, realizing early warning of faults, among which This is the early warning coefficient, with a value range of [0.6-0.8].

[0029] A residual current fault monitoring system based on multi-feature fusion is used to implement a residual current fault judgment method based on multi-feature fusion, including a signal acquisition module, a signal preprocessing module, a feature extraction module, an adaptive feature fusion module, a fault judgment module, a trend prediction module, an alarm and control module, a self-diagnosis module, and a data storage and communication module. The signal acquisition module, signal preprocessing module, feature extraction module, and adaptive feature fusion module are connected in sequence. The adaptive feature fusion module is connected to the fault judgment module and the trend prediction module, respectively. The fault judgment module and the trend prediction module are connected to the alarm and control module. The self-diagnosis module is connected to the signal acquisition module, signal preprocessing module, feature extraction module, adaptive feature fusion module, and fault judgment module, respectively, and the communication module is connected to all of the above modules.

[0030] The signal acquisition module includes a zero-sequence current transformer, a temperature sensor, and a current sensor, which are used to acquire residual current signals, ambient temperature signals, and load current signals, respectively. The acquired analog signals are converted into multi-source digital signals and then transmitted to the signal preprocessing module. The signal preprocessing module receives multi-source digital signals transmitted from the signal acquisition module, performs noise reduction using an improved Kalman filter algorithm, performs standardization using a sliding window method, and transmits the processed signal to the feature extraction module. The feature extraction module receives the preprocessed multi-source digital signal, extracts the time-domain and frequency-domain features of the residual current, as well as the feature parameters of the ambient temperature and load current, and forms a multi-dimensional feature vector, which is then transmitted to the adaptive feature fusion module. The adaptive feature fusion module has a built-in adaptive weighted fusion algorithm and feature-fault sensitivity mapping matrix. It receives multi-dimensional feature vectors transmitted by the feature extraction module, dynamically adjusts the weight coefficients according to real-time feature values, calculates dynamically fused feature values, and transmits them to the fault judgment module. The fault diagnosis module has a preset unique fault threshold. By comparing the fused feature values ​​with the fault threshold, it is determined whether there is a fault in the line and the type of fault. At the same time, the fault is verified again through S-transform. The trend prediction module has a built-in adaptive ARIMA prediction model. It receives dynamic fusion feature values ​​from multiple sliding windows, forms a fusion feature time series, and performs short-term prediction. When the predicted value reaches the warning threshold, it outputs a warning signal to the alarm and control module. When the alarm and control module confirms the presence of a residual current fault, it immediately issues an audible and visual alarm signal and simultaneously outputs a control signal to the circuit breaker to disconnect the faulty line; when the trend prediction module outputs a warning signal, it issues a warning prompt but does not perform a tripping operation. The self-diagnostic module periodically injects preset self-test signal waveforms and uses closed-loop detection to check the response output of each module to determine whether each module is working properly. When any module responds abnormally or its output deviation exceeds the preset self-test threshold, a module fault code is generated and uploaded to the remote monitoring center. At the same time, the faulty module is automatically bypassed and the backup algorithm channel is activated. The data storage and communication module is used to store the acquired raw residual current signal, preprocessed multi-source digital signal, characteristic parameters, fused characteristic values ​​and fault judgment results. It also supports communication with the remote monitoring center to upload fault information in real time.

[0031] Sampling frequency of the signal acquisition module The frequency is fixed at 1000Hz, the sampling accuracy is 16 bits, the measurement range of the zero-sequence current transformer is 0-500mA, the measurement range of the temperature sensor is -20℃~80℃, the absolute value of the measurement error is no greater than 0.5℃, the measurement range of the current sensor is 0-100A, and the absolute value of the measurement error is no greater than ±1%.

[0032] Fault threshold in the fault diagnosis module The value was obtained through training with 1000 sets of sample data of different fault types, and was set to 0.72. It was then fine-tuned through a remote monitoring center based on actual line operating conditions. In the adaptive feature fusion module, the basic weight coefficients were determined through sample training, and were respectively... =0.35, , , , Dynamic adjustment coefficient The value is 1.0.

[0033] The self-diagnostic module's self-test cycle is set to 24 hours, and the self-test signal waveform includes an amplitude of =50Hz fundamental and second harmonic amplitudes =100Hz, third harmonic amplitude The composite waveform has an amplitude of 50% of the full scale; the absolute value of the normal response deviation threshold of each module does not exceed 5%, otherwise it is judged as a module abnormality.

[0034] Example 2 This embodiment takes a low-voltage distribution network in a residential community as the application scenario. The lines in this scenario have complex loads, including household appliances, public lighting, charging piles, etc., and the ambient temperature fluctuates between -5℃ and 35℃. S1: Real-time residual current, ambient temperature, and load current signals at the low-voltage incoming line of the community are synchronously collected via a zero-sequence current transformer, temperature sensor, and current sensor. The collection time is continuous for 24 hours, including the original residual current signal. The fluctuation range is 0~300mA, and the ambient temperature signal is... The fluctuation range is 5℃~28℃, and the load current signal is... The fluctuation range is 10A~80A. After all the acquired signals are converted into 16-bit digital signals, they are transmitted to the signal preprocessing module.

[0035] S2: The original residual current signal is processed using an improved Kalman filter algorithm. Denoising is performed to filter out electromagnetic interference and harmonic interference, resulting in a denoised residual current signal. Testing showed that the signal-to-noise ratio (SNR) of the denoised signal improved to 38 dB, a 12 dB improvement compared to traditional Kalman filtering; further improvements were achieved using the sliding window method. , , Standardization is performed to eliminate the influence of dimensions and obtain a standardized signal. , , After standardization, the value range of each signal is [0,1].

[0036] S3: For the standardized multi-source signal, extract each core feature separately. The calculation process is as follows: ① Time-domain characteristics of residual current: According to the formula described in claim 3, the standardized effective value of residual current is calculated. Peak Waveform distortion rate ; ② Frequency domain characteristics of the normalized residual current: the fundamental amplitude is extracted by Fourier transform. 2nd harmonic amplitude 5th harmonic amplitude ; ③ Calculate the rate of temperature change using the corresponding formula. After standardization, it becomes 0.15; ④ Calculate the standardized load fluctuation coefficient according to the corresponding formula. Integrating all the above features, a multi-dimensional feature vector is formed as follows: [0.32, 0.48, 18.6%, 0.30, 0.08, 0.03, 0.15, 0.22].

[0037] S4: Substitute the extracted multi-dimensional feature vectors into the core fusion formula to calculate the fusion feature values: = =0.2115.

[0038] S5: Merge feature values With preset fault threshold In contrast, due to This indicates that the line was operating normally during that period and there was no residual current fault.

[0039] To verify the effectiveness of this solution, three common residual current faults were simulated, and the above steps were performed for each. The specific results are as follows: ① Simulated electric shock fault: A single-phase electric shock was simulated, with a current of 30mA. After acquiring and processing the signal, standardized feature parameters were extracted. =0.85、 =25.3、 =0.32、 =0.03℃ / s, standardized to 0.22. =0.25, calculate the fusion feature value: ==0.897, If the value is ≥0.72, it is initially judged as a fault. The time-frequency features are extracted by S-transform and matched with the human electric shock fault template. The matching degree is 97.2% ≥95%, confirming that the fault is human electric shock. The alarm and control module immediately issues an audible and visual alarm and outputs a control signal to cut off the faulty line. The response time is 38ms.

[0040] ② Simulate equipment leakage fault: Simulate air conditioning equipment leakage with a leakage current of 50mA, and extract standardized feature parameters: =0.78、 =22.1、 =0.28、 =0.025℃ / s, standardized to 0.19. =0.23, calculate the fusion eigenvalue. =0.813, ≥0.72, initially judged as a fault; S-transform feature matching degree is 96.5%≥95%, confirming the fault as equipment leakage, triggering an alarm and tripping, with a response time of 42ms.

[0041] ③ Simulate line insulation degradation fault: Simulate line insulation aging, with the residual current slowly rising to 80mA, and extract standardized characteristic parameters: =0.92、 =30.5、 =0.38、 =0.02℃ / s, standardized to 0.15. =0.21, calculate the fusion eigenvalue. =0.958, ≥0.72, initially judged as a fault; S-transformation characteristic matching degree is 98.1%≥95%, confirming the fault as line insulation deterioration, triggering an alarm and tripping, with a response time of 45ms.

[0042] In this embodiment, the scheme runs continuously for 72 hours. The operating effect of the traditional residual current protection device is compared with that of this scheme. The verification results are shown in Table 1. The core parameters and fused characteristic value data tables for different fault types are supplemented, and the characteristic differences of each fault are clarified in Table 2.

[0043] Table 1

[0044] Table 2

[0045] In this embodiment, a total of 100 faults were simulated, with approximately 33 of each of the three types of faults mentioned above. The judgment was accurate 98 times, with 1 false positive (caused by momentary interference, failure of secondary verification, and failure to trigger tripping) and 1 false negative (caused by slight insulation degradation, which was identified after subsequent characteristic changes). The judgment accuracy was 98%, the false positive rate was 1%, and the false negative rate was 1%. In contrast, the traditional device was accurate 75 times, false positive 18 times, and false negative 7 times, with an accuracy of only 75%. The false positive rate and false negative rate are much higher than those of this solution.

[0046] During periods of strong electromagnetic interference, the signal-to-noise ratio of the denoised signal in this solution remains stable at over 35dB, with characteristic parameter errors ≤3%. Traditional devices frequently produce false alarms, with a false judgment rate exceeding 25%, indicating that the anti-interference capability of this solution is significantly superior to that of traditional technologies.

[0047] The fault response time of this solution is within 50ms, with an average response time of 42ms, which can quickly disconnect the faulty line and prevent the fault from escalating. The average response time of traditional devices is 120ms, which is slow and poses a safety hazard.

[0048] Under operating conditions with ambient temperature variations of 5℃ to 28℃ and load fluctuations of 10A to 80A, the accuracy of this solution remains above 97%, demonstrating good adaptability. In contrast, traditional devices show a significant increase in misjudgment rate and poor adaptability when load fluctuations are large.

[0049] In summary, this embodiment fully verifies the feasibility and effectiveness of the residual current fault judgment method and monitoring system based on multi-feature fusion. It can effectively solve the defects of the prior art, improve the accuracy and reliability of fault judgment in low-voltage distribution networks, and meet the needs of practical engineering applications.

[0050] Therefore, the present invention adopts the above-mentioned residual current fault judgment method and system based on multi-feature fusion. Through dynamic weighted fusion, improved Kalman filtering to retain key frequency bands, ARIMA trend prediction and closed-loop self-diagnosis, it achieves a judgment accuracy of over 95%, significantly improves the accuracy, reliability and initiative of residual current fault judgment under complex working conditions, and effectively solves the core defects of the prior art.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A residual current fault detection method based on multi-feature fusion, characterized in that, Includes the following steps: S1: The real-time residual current signal in the low-voltage distribution network line is collected by the zero-sequence current transformer as the original residual current signal. The ambient temperature signal and load current signal are collected simultaneously to form a multi-source original signal set. S2: Preprocess the original multi-source signal set formed in S1 to obtain a standardized multi-source signal set; S3: Extract the core features of each signal from the standardized multi-source signal set to form a multi-dimensional feature vector; the residual current signal features include time domain features and frequency domain features, the ambient temperature signal features are the temperature change rate, and the load current signal features are the load fluctuation coefficient; S4: Construct a feature-fault sensitivity mapping matrix, dynamically adjust the weight coefficients of the weighted fusion algorithm based on the real-time values ​​of each feature parameter, and calculate the dynamic fusion feature value; S5: Fault diagnosis, using the dynamic fusion feature values ​​calculated in S4. With preset fault threshold By comparison, if the dynamic fusion feature value is not less than the preset fault threshold, it is determined that there is a residual current fault in the line. Based on the numerical range of each feature parameter and the activated sensitive feature set, the fault type is identified. Otherwise, the line is considered normal; S6: Perform secondary verification on the fault signal identified in S5. Use S-transform to extract the time-frequency characteristics of the residual current under fault conditions and match them with preset fault feature templates. If the matching degree is ≥95%, the fault is confirmed; otherwise, it is judged as a false judgment and the process returns to S1 for signal acquisition. S7: Construct a fusion feature time series by arranging the dynamically fused feature values ​​obtained from multiple consecutive sliding windows in chronological order. An adaptive ARIMA model is used to perform short-term predictions on the fused feature time series, obtaining the predicted fused feature values ​​for the next T time windows. ; S8: When predicting fusion feature values At that time, it outputs a warning signal and marks the potential fault type, realizing early warning of faults, among which This is the early warning coefficient, with a value range of [0.6-0.8].

2. The residual current fault judgment method based on multi-feature fusion according to claim 1, characterized in that, The specific process of preprocessing the multi-source raw signal set in S2 is as follows: S21: An improved Kalman filter algorithm is used to eliminate interference noise in the original residual current signal, resulting in a denoised residual current signal. ; The state equation and observation equation of the improved Kalman filter algorithm are as follows: ； ； In the formula, The system state vector at time k contains only the state variable of the amplitude of the denoised residual current signal, and is unique. This is the state transition matrix, with a fixed value of 0.

98. This is the system noise driving matrix, with a value of 1; for The system noise at time t follows a mean of 0 and a variance of 0. The normal distribution; for k The observation vector at time; This is the observation matrix, with a value of 1. for k The observation noise at time t follows a mean of 0 and a variance of 0. The normal distribution; S22: The improved Kalman filter algorithm introduces an adaptive noise covariance matching mechanism in state estimation to estimate and update the system noise variance in real time. and observation noise variance ; S23: For the fundamental frequency of the residual current and its characteristic harmonic components, the preset retention frequency band is 0-500Hz, and the state transition matrix is ​​adjusted accordingly. The bandwidth parameters are set to ensure that the signal in the fault characteristic frequency band passes through without attenuation, while filtering out high-frequency switching noise and pulse interference above 500Hz. S24: The denoised residual current signal, ambient temperature signal, and load current signal are standardized using the sliding window method to eliminate the influence of dimensions, resulting in standardized multi-source signals. , , .

3. The residual current fault judgment method based on multi-feature fusion according to claim 2, characterized in that, The time-domain characteristics of the residual current signal in S3 include peak value. Valid value and waveform distortion rate Frequency domain characteristics include fundamental frequency amplitude Second harmonic amplitude Fifth harmonic amplitude ; Each feature parameter is uniquely defined and does not interfere with the others; RMS value of residual current signal In the formula The number of sampling points within a single sliding window. For the first i Each sampling time; Waveform distortion rate In the formula The value is the nth harmonic amplitude of the residual current, where n ≥ 2; Rate of temperature change In the formula The time interval is fixed and is set to 10 seconds. Load fluctuation coefficient In the formula, , , These represent the maximum, minimum, and average values ​​of the standardized load current within a single sliding window.

4. The residual current fault judgment method based on multi-feature fusion according to claim 1, characterized in that, Dynamic fusion eigenvalues ​​in S4 The calculation formula is: = ； In the formula, , , , , These are the dynamic weight coefficients corresponding to each feature, satisfying... , = ; This is a dynamic adjustment coefficient, with a value range of [0.5, 1.5]. Features i An indicator function for whether it belongs to the current activation sensitive feature set; it takes a value of 1 when activated and 0 otherwise. This is the effective value of the residual current. The residual current waveform distortion rate. The amplitude of the second harmonic of the residual current. The rate of change of ambient temperature. This is the load fluctuation coefficient.

5. A residual current fault monitoring system based on multi-feature fusion, used to implement the steps of the residual current fault judgment method based on multi-feature fusion as described in any one of claims 1-4, characterized in that, It includes a signal acquisition module, a signal preprocessing module, a feature extraction module, an adaptive feature fusion module, a fault diagnosis module, a trend prediction module, an alarm and control module, a self-diagnosis module, and a data storage and communication module; The signal acquisition module, signal preprocessing module, feature extraction module, and adaptive feature fusion module are connected in sequence. The adaptive feature fusion module is connected to the fault judgment module and the trend prediction module, respectively. The fault judgment module and the trend prediction module are connected to the alarm and control module. The self-diagnosis module is connected to the signal acquisition module, signal preprocessing module, feature extraction module, adaptive feature fusion module, and fault judgment module, respectively, and the communication module is connected to all of the above modules.

6. The residual current fault monitoring system based on multi-feature fusion as described in claim 5, characterized in that: The signal acquisition module includes a zero-sequence current transformer, a temperature sensor, and a current sensor, which are used to acquire residual current signals, ambient temperature signals, and load current signals, respectively. The acquired analog signals are converted into multi-source digital signals and then transmitted to the signal preprocessing module. The signal preprocessing module receives multi-source digital signals transmitted from the signal acquisition module, performs noise reduction using an improved Kalman filter algorithm, performs standardization using a sliding window method, and transmits the processed signal to the feature extraction module. The feature extraction module receives the preprocessed multi-source digital signal, extracts the time-domain and frequency-domain features of the residual current, as well as the feature parameters of the ambient temperature and load current, and forms a multi-dimensional feature vector, which is then transmitted to the adaptive feature fusion module. The adaptive feature fusion module has a built-in adaptive weighted fusion algorithm and feature-fault sensitivity mapping matrix. It receives multi-dimensional feature vectors transmitted by the feature extraction module, dynamically adjusts the weight coefficients according to real-time feature values, calculates dynamically fused feature values, and transmits them to the fault judgment module. The fault diagnosis module has a preset unique fault threshold. By comparing the fused feature values ​​with the fault threshold, it is determined whether there is a fault in the line and the type of fault. At the same time, the fault is verified again through S-transform. The trend prediction module has a built-in adaptive ARIMA prediction model. It receives dynamic fusion feature values ​​from multiple sliding windows, forms a fusion feature time series, and performs short-term prediction. When the predicted value reaches the warning threshold, it outputs a warning signal to the alarm and control module. When the alarm and control module confirms the presence of a residual current fault, it immediately issues an audible and visual alarm signal and simultaneously outputs a control signal to the circuit breaker to disconnect the faulty line; when the trend prediction module outputs a warning signal, it issues a warning prompt but does not perform a tripping operation. The self-diagnostic module periodically injects preset self-test signal waveforms and uses closed-loop detection to check the response output of each module to determine whether each module is working properly. When any module responds abnormally or its output deviation exceeds the preset self-test threshold, a module fault code is generated and uploaded to the remote monitoring center. At the same time, the faulty module is automatically bypassed and the backup algorithm channel is activated. The data storage and communication module is used to store the acquired raw residual current signal, preprocessed multi-source digital signal, characteristic parameters, fused characteristic values ​​and fault judgment results. It also supports communication with the remote monitoring center to upload fault information in real time.

7. A residual current fault monitoring system based on multi-feature fusion as described in claim 5, characterized in that: Sampling frequency of the signal acquisition module The frequency is fixed at 1000Hz, the sampling accuracy is 16 bits, the measurement range of the zero-sequence current transformer is 0-500mA, the measurement range of the temperature sensor is -20℃~80℃, the absolute value of the measurement error is no greater than 0.5℃, the measurement range of the current sensor is 0-100A, and the absolute value of the measurement error is no greater than ±1%.

8. A residual current fault monitoring system based on multi-feature fusion as described in claim 5, characterized in that: Fault threshold in the fault diagnosis module The value was obtained by training with 1,000 sets of sample data of different fault types and was set to 0.

72. It was then fine-tuned through the remote monitoring center according to the actual line operating conditions. In the adaptive feature fusion module, the basic weight coefficients are determined through sample training, and are respectively =0.35, , , , Dynamic adjustment coefficient The value is 1.

0.

9. A residual current fault monitoring system based on multi-feature fusion as described in claim 5, characterized in that: The self-diagnostic module's self-test cycle is set to 24 hours, and the self-test signal waveform includes an amplitude of =50Hz fundamental and second harmonic amplitudes =100Hz, third harmonic amplitude The composite waveform has an amplitude of 50% of the full scale; the absolute value of the normal response deviation threshold of each module does not exceed 5%, otherwise it is judged as a module abnormality.

Images