An arc fault protection method based on intelligent algorithm of circuit breaker

Abstract

The application relates to the technical field of electrical safety, and particularly discloses an arc fault protection method based on a circuit breaker intelligent algorithm, which comprises the following steps: S1, collecting a line current signal in real time, eliminating frequency converter harmonic interference of an original signal through an anti-aliasing filter, and converting the original signal into a digital signal; S2, processing the current signal through a three-stage cascaded wavelet analysis module, and outputting a dynamic peak-to-average ratio to S4; S3, detecting a current disturbance based on a double sliding rail window, outputting a disturbance positioning mark and a time-frequency feature to S4, determining a disturbance event, and positioning a starting point; S4, receiving the dynamic peak-to-average ratio of S2 and the disturbance positioning mark and the time-frequency feature of S3, processing according to a characteristic analysis result, and adopting a direct interruption method and an electromagnetic radiation-current frequency domain cross verification method; and S5, constructing an arc energy accumulation prediction model; the application solves the problems of weak arc missing detection, electromagnetic interference misoperation and delay.

Description

Technical Field

[0001] This invention relates to the field of electrical fault detection technology, and specifically to an arc fault protection method based on a circuit breaker intelligent algorithm. Background Technology

[0002] Among the many potential safety hazards in power distribution, electric arc faults, due to their instantaneous nature, high energy, and concealment, are like a sword hanging over one's head. The enormous heat energy released by an electric arc in an instant is enough to ignite surrounding materials, leading to catastrophic electrical fires and causing incalculable casualties and property losses. Faced with such a severe threat, traditional protection mechanisms such as thermomagnetic tripping or electronic overcurrent protection are inadequate for arc detection. Existing technologies mainly rely on current amplitude threshold detection or simple waveform analysis. However, these methods have inherent limitations. When problems such as intermittent poor contact, aging wire insulation damage, or carbonization paths occur in the circuit, low-energy series arcs are often generated, such as arcs <5A caused by line aging or poor contact. The amplitude of these arc currents may not exceed the circuit breaker's set threshold, and the waveform distortion is insufficient to be effectively captured by conventional Fourier analysis. Traditional protection devices may fail, allowing dangerous arcs to remain latent in the system. Common arc fault circuit breakers rely heavily on detecting high-frequency noise components in the current or the degree of harmonic distortion at specific frequencies. Such algorithms may be effectively verified in a clean load environment in the laboratory. However, real-world conditions are far more complex. Nonlinear loads such as frequency converters, switching power supplies, energy-saving lamps, and motors generate a large number of high-frequency harmonics and electromagnetic interference during use and operation. Their spectral characteristics often overlap and interfere with the signal components generated by early or weak fault arcs. This phenomenon directly leads to frequent malfunctions in circuit breaker protection systems, causing unnecessary power outages and disrupting production and daily life. Conversely, an even more dangerous situation is that when there is a real risk of electric arc, the system fails to accurately identify weak but real arc characteristic signals in a complex interference background, resulting in missed detection and failure, thus creating a major safety hazard.

[0003] On the other hand, a deeper examination of the current state of the industry reveals a deep-seated bottleneck in existing arc fault protection technologies: a lack of effective monitoring and trend prediction of the development process of fault energy. Arc fault detection modules in circuit breakers typically analyze signal characteristics in isolation within a single time segment, passively waiting for certain preset characteristic quantities to exceed fixed thresholds before reacting. However, reality shows that truly dangerous arcs with the potential to cause fires do not possess all their destructive energy at the moment of eruption. On the contrary, such faults often originate from extremely weak or even intermittent initial discharges. If the protection system cannot keenly identify the existence and development of these early low-energy discharges, and cannot capture subtle abnormal trends in discharge intensity, frequency, and duration, by the time the accumulated energy triggers an electrical fire, the protection action is already too late. This lag and inability to identify the process of fault energy changing from quantitative to qualitative makes it difficult for existing protection mechanisms to achieve intrinsic safety by moving the threshold forward. It is important to emphasize that complex circuits contain a large number of normal electromagnetic transient phenomena, whose spectral and temporal characteristics are highly similar to the low-energy arc discharges in their initial stages. For example, transient overvoltage sparks generated during switching of inductive loads, commutation sparks from carbon brushes in old motors under specific operating conditions, and even the inherent power switching oscillations in some high-frequency switching power supplies can all produce noise characteristics or distorted waveforms similar to real series-type low-energy arcs. Existing techniques that rely on differences in energy distribution across a single or small number of fixed frequency bands often struggle with such complex, variable, and highly similar aliasing scenarios. They cannot reliably determine the existence of early-stage low-energy arcs by penetrating the noise fog, and they are even less able to scientifically quantify the potential hazards and energy accumulation rate of these minute discharge events in specific circuit environments. When the fault arc is still in its low-energy latency period, current technical solutions have significant limitations. Crucially, although the arc at this stage does not yet generate a significant temperature rise that could ignite, each weak discharge subtly accelerates circuit degradation, such as exacerbating the carbonization of insulation materials or damaging the surface condition of contacts. This cumulative degradation, like water dripping on a stone, will eventually create a low-impedance path, triggering a catastrophic arc explosion. Traditional protection methods are neither aware of nor capable of monitoring this process, remaining inactive in its early stages. This lack of monitoring and technical countermeasures for the accumulation trend of intermittent discharges is a critical gap that urgently needs to be filled in the current field of arc fault protection. In summary, the severe predicament encountered by traditional arc fault protection technology, particularly the deep-seated bottleneck in detecting early low-energy arcs and predicting their development trends under complex load environments, has become a core barrier hindering a substantial leap in electrical safety levels. If this bottleneck cannot be overcome, the goals of improving the reliability of arc fault identification and preventing early electrical fires will remain mere empty talk.

[0004] Therefore, there is an urgent need for an arc fault protection method based on intelligent circuit breaker algorithms in existing technologies. Summary of the Invention

[0005] The purpose of this invention is to provide an arc fault protection method based on a circuit breaker intelligent algorithm, in order to solve the problems of traditional arc protection devices in the prior art that fail to detect small arcs generated by old lines or poor contact when identifying early and weak potential arc faults, fail to identify electromagnetic noise generated by normal equipment similar to early arc signals, and lack predictive ability for small discharges that are accumulating damage.

[0006] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution:

[0007] An arc fault protection method based on a circuit breaker intelligent algorithm includes:

[0008] S1. Real-time acquisition of high-frequency current signals and dynamic noise reduction: Real-time acquisition of line current signals, the original signal is filtered by an anti-aliasing filter to eliminate inverter harmonic interference, and then converted into a digital signal by the MCU's built-in ADC.

[0009] S2, Peak-to-Average Ratio Calculation: Three-level wavelet decomposition: The current signal is processed by a three-level cascaded wavelet analysis module, and the dynamic peak-to-average ratio is output to S4;

[0010] S3. Improved Cumulative Sum CUSUM Algorithm for Disturbance Localization and Time-Frequency Feature Extraction: Based on dual sliding window detection of current disturbance, the disturbance localization marker and time-frequency features are output to S4. The time-frequency features include the current standard deviation σ and kurtosis β.

[0011] S4. Multimodal hierarchical decision-making and protection execution: Receive the dynamic peak-to-average power ratio of S2 and the disturbance location markers and time-frequency characteristics of S3, and process them according to the feature analysis results;

[0012] S5. Arc energy accumulation prediction: Construct an arc energy accumulation prediction model and output the processing results; S2 and S3 are executed in parallel, and the output results are synchronously input into S4, which is implemented through a dual-core MCU or real-time operating system.

[0013] Furthermore, S1 includes: acquiring the line current signal in real time through a high-precision Rogowski coil current sensor, calculating the load current change rate dI / dt in real time, with the unit being A / ms, and activating a dynamic noise reduction algorithm when dI / dt>5 A / ms. The activation of the dynamic noise reduction algorithm includes using adaptive wavelet threshold noise reduction, suppressing inverter harmonic interference by adjusting the filtering parameters in real time, and improving the signal-to-noise ratio to above 35 dB.

[0014] Furthermore, S2 includes: the wavelet analysis module uses the db4 wavelet basis function, with high-pass group coefficients of [-0.2304, 0.7148, -0.6309] and low-pass group coefficients of [0.0273, -0.0322, 0.1884]. The output is set as follows: the 50–100kHz high-frequency component H1 is the first level, the 25–50 kHz frequency band component H2 is the second level, and the 12.5–25 kHz frequency band component H3 is the third level.

[0015] Furthermore, S2 also includes: real-time calculation of dynamic peak-to-average power ratio (PAR), with the formula as follows:

[0016] ;

[0017] in Take the largest amplitude value among H1–H3. The value is the average of the three channels, and the dynamic threshold is set to 15 dB + 0.3×|dI / dt|. When the standard deviation of the load current σ>5 A, the coefficient is adjusted to 0.5.

[0018] Furthermore, the improved CUSUM algorithm in S3 uses a dual sliding window. The first window W1 has a window length of 20 ms corresponding to 4000 points, and the second window W2 has a window length of 10 ms corresponding to 2000 points. The mean effective value μ of the current in window W1 is calculated, and the offset threshold K = 1.5 × μ is set. The offset accumulation value within window W2 is iteratively calculated. The formula is ,like If H > 5 × K, it is determined to be a disturbance event, and the starting point is located.

[0019] Furthermore, in S3, the starting point for locating the disturbance point includes extracting 20 ms of data before and after the disturbance point, calculating the current standard deviation σ and kurtosis β, and taking the point with the maximum standard deviation as the center. If its kurtosis is the minimum value of the 5 neighboring points, it is marked as the starting point. The threshold σ>0.01 and the threshold β>3.5.

[0020] Furthermore, the path-based processing in S4 based on the feature analysis results includes: if the real-time calculated dynamic peak-to-average power ratio (PAR) > 15 dB and the current standard deviation (σ) > 0.01, then the direct interruption method is adopted; if the real-time calculated dynamic peak-to-average power ratio (PAR) ∈ [10 dB, 15 dB] and the kurtosis (β) > 3.5, then the electromagnetic radiation-current frequency domain cross-validation method is initiated.

[0021] Furthermore, the direct interruption method includes:

[0022] SA41 uses a half-cycle event counter for accumulation, with a half-cycle of 10 ms;

[0023] SA42: If the cumulative number of events is ≥8 within 100 ms, the magnetic latching relay will be triggered to trip.

[0024] SA43, drives the MOSFET switching circuit, with a conduction time ≤10 ns;

[0025] SA44 records the fault waveform and illuminates the red LED indicator.

[0026] Furthermore, the electromagnetic radiation-current frequency domain cross-verification method includes:

[0027] SB41, synchronously acquires electromagnetic radiation signals in the 200–300 MHz range, using a fourth-order Hilbert antenna with a gain ≥5 dBi and a sensitivity of -110 dBm;

[0028] SB42 extracts the current energy E_c from 10–30 kHz and the electromagnetic energy E_m from 200–300 MHz, and constructs the feature vector [E_c, E_m].

[0029] SB43 calculates membership degrees using a pre-trained FCM clustering model, with fault cluster centers at [0.85, 1.2] and normal cluster centers at [0.15, 0.3].

[0030] SB44: If the membership degree is greater than 0.8 and the condition is met for three consecutive windows, the trip is triggered and the fault location is reported.

[0031] Furthermore, the arc energy accumulation prediction model in S5 uses the following formula: time interval T = 10ms. If it > 200J / 100ms, it is determined to be an energy accumulation type fault. The compressed fault characteristic data is uploaded to the cloud via the wireless communication module with a delay of ≤10ms, and the complete waveform is output to the local monitoring terminal via the wired communication interface. The wireless communication module is a Bluetooth Low Energy (BLE) protocol, and the wired communication interface is an RS485 industrial bus with a transmission delay of <10ms to update the FCM clustering center.

[0032] Furthermore, it also includes deploying a fourth-order Hilbert antenna to capture electromagnetic radiation signals of 200–300 MHz and using an infrared thermal imaging sensor to monitor the temperature rise of the contact points, thus constructing a three-modal sensing system of current, electromagnetic, and temperature.

[0033] Furthermore, S5 also includes joint training using LS-SVM and dynamic threshold mechanism, setting 8,000 sets of samples with uniform distribution of arc type and load scenario as training set, injecting 30% noise into 2,000 sets of samples to simulate electromagnetic interference as test set, and in simulated environment and photovoltaic DC system test, the action time meets the requirements of metallic short circuit ≤30 ms and series arc ≤100 ms, and the positioning error is <10 cm within 200m cable.

[0034] Compared with existing technologies, this invention has the following advantages: First, addressing the shortcomings of traditional detection methods in responding slowly to weak series arcs, it improves the extraction accuracy of high-frequency arc features by more than 40% based on the synergistic mechanism of three-level wavelet dynamic decomposition and kurtosis analysis. When the load current change rate dI / dt>5 A / ms, the adaptive noise reduction algorithm improves the signal-to-noise ratio under inverter harmonic interference to more than 35dB by adjusting the wavelet threshold in real time, thereby reducing the false detection rate to less than 5% in industrial inverter equipment-dense scenarios. Second, facing the problem of malfunctions caused by complex electromagnetic environments, an innovative electromagnetic-current frequency domain cross-validation path is designed: simultaneously capturing 200-300MHz electromagnetic radiation signals and 10-30kHz current energy, and constructing feature vectors [E_c, E_m] to input into the pre-trained FCM clustering model. This model maintains 95% accuracy even under 30dB noise by quantifying the membership difference between fault cluster centers [0.85, 1.2] and normal centers [0.15, 0.3], reducing the false alarm rate by 60% compared to traditional single current detection. By clearly defining the data flow from S2 / S3 to S4, the model solves the problem of aggregation delay in parallel processing results: S4 can directly call the PAR feature of S2 and the time-domain location result of S3, avoiding redundant calculations; the decision response time is shortened to ≤100ms, an improvement of 19% compared to schemes without clearly defined flow direction; the dual-path data synchronization mechanism eliminates inter-core communication conflicts, reducing the false alarm rate to 2.1%. More importantly, this invention breaks through the technical bottleneck of fault energy accumulation prediction. The improved accumulation and CUSUM algorithm uses a dual-sliding window to accurately locate the disturbance initiation point, combined with an arc energy integral model, enabling the system to issue an early warning 500ms before the fault energy reaches the critical value. Simultaneously, the dynamic threshold mechanism solves the adaptability problem under light / heavy load conditions by using Gaussian process regression to self-update coefficients every 24 hours. Actual measurements show that this solution reduces the series arcing time to ≤100ms and the metallic short-circuit response to ≤30ms in photovoltaic DC systems, and reduces the positioning error of a 200-meter cable to <10cm, providing core technical support for accurate fault diagnosis. Attached Figure Description

[0035] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0036] Figure 1 This is a flowchart illustrating the method described in Embodiment 1 of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] The concepts involved in this application will first be described with reference to the accompanying drawings. It should be noted that the following descriptions of various concepts are only for the purpose of making the content of this application easier to understand and do not constitute a limitation on the scope of protection of this application; furthermore, the embodiments and features in the embodiments of this application can be combined with each other unless otherwise specified. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0039] Example 1

[0040] like Figure 1 As shown, this invention provides an arc fault protection method based on a circuit breaker intelligent algorithm, comprising the following steps:

[0041] S1. Acquisition of high-frequency current signals and dynamic noise reduction processing: Real-time acquisition of line current signals. The original signal is filtered by an anti-aliasing filter to eliminate harmonic interference from the frequency converter, and then converted into a digital signal by the MCU's built-in ADC.

[0042] Specifically, the line current signal is acquired in real time by a high-precision Rogowski coil current sensor with a fixed sampling rate of 200 kHz and a sampling bit width of 16 bits. The original signal is filtered by an anti-aliasing filter (cutoff frequency 150 kHz, attenuation slope -60 dB / dec) to eliminate inverter harmonic interference, and then converted into a digital signal by the built-in ADC of the 32-bit MCU.

[0043] In S1, the sampled data is grouped in 20 ms intervals (corresponding to a complete power frequency cycle of a 50 Hz system), with each group containing 4000 sampling points. The load current change rate dI / dt is calculated in real time, with the unit being A / ms. When dI / dt > 5A / ms, the dynamic noise reduction algorithm is activated.

[0044] S2. Peak-to-Average Ratio (PAR) Calculation: Three-level wavelet decomposition and adaptive peak: A three-level cascaded wavelet analysis module (wavelet basis function db4, high-pass coefficient group: [-0.2304, 0.7148, -0.6309], low-pass coefficient group: [0.0273, -0.0322, 0.1884]) is used to process the current signal; S2 outputs the dynamic PAR to the S4 decision module as a criterion for high-frequency arc characteristics; PAR refers to the logarithmic ratio of the maximum amplitude of the high-frequency component to the root mean square value.

[0045] S3. Improved Cumulative Sum CUSUM Algorithm for Disturbance Localization and Time-Frequency Feature Extraction: Based on dual sliding window detection of current disturbances, the disturbance localization marker and time-frequency features (standard deviation σ and kurtosis β) are output to S4 as the event triggering basis; assuming the length of the front window W1 is 20 ms (4000 points) and the length of the rear window W2 is 10 ms (2000 points), the mean effective current value μ of window W1 is calculated, and the offset threshold K = 1.5 × μ is set. The cumulative offset value S within window W2 is iteratively calculated. n The formula is ,like >H (H=5×K) is determined to be a disturbance event, and the starting point is located; where The cumulative offset of the current window, dimensionless; S n-1 x is the accumulated value of the offset from the previous window, dimensionless; n The current value at the current sampling point is given by μ, where μ is the mean of the effective current values ​​within window W1 (window length 20ms), in amperes; K is the offset threshold, in amperes. max(0, S) n-1 + (x n - μ - K)) is a function that takes the maximum value to ensure that the accumulated value is non-negative and dimensionless, thus preventing negative offset from affecting the detection.

[0046] S2 and S3 are executed in parallel. Specifically, after the main controller completes the acquisition of the S1 signal, it synchronously starts the S2 and S3 processing threads. In the dual-core MCU architecture, S2 and S3 are assigned to Core0 and Core1 respectively. In the single-core system, pseudo-parallelism is achieved through RTOS, and the task switching cycle is ≤100μs.

[0047] Specifically, an STM32H743 dual-core microcontroller is used: Core0 runs task S2: performing db4 wavelet transform using the CMSIS-DSP library; Core1 runs task S3: implementing the improved CUSUM algorithm; shared memory region: 0x24000000-0x24000FFF; synchronization mechanism: ensuring data consistency through HSEM semaphores.

[0048] S4, Multimodal Hierarchical Decision Making and Protection Execution: Receives the dynamic peak-to-average power ratio (PAR) from S2 and the disturbance location markers and time-frequency characteristics from S3. Processes the data according to the feature analysis results. If the real-time calculated dynamic PAR > 15 dB and the current standard deviation σ > 0.01 (normal arc), a direct interruption method is used. If the real-time calculated dynamic PAR ∈ [10 dB, 15 dB] and the kurtosis β > 3.5 (weak arc), an electromagnetic radiation-current frequency domain cross-validation method is used. This design ensures that the parallel calculation results are efficiently converged to the decision layer, reducing instruction cycle delay (measured ≤ 0.8 ms).

[0049] S5. Arc Energy Accumulation Prediction: Construct an arc energy accumulation prediction model, using the following formula: If Where T is the time interval in seconds, fixed at 10ms (0.01s), and the length of each integration window is... >200 J / 100ms indicates an energy accumulation fault. A compressed feature vector (containing...) is extracted using the BLE module. Location markers and cluster membership (with a latency of ≤10ms) are uploaded to the cloud; a 20ms complete current waveform is output at a baud rate of 1Mbps via the RS485 interface to update the FCM cluster center; among which The cumulative value of arc energy is expressed in joules; the summation operation, k from 1 to 8, represents the summation of 8 time intervals, covering an 80ms window (T=10ms); the integration operation, time from (k-1)T to kT, calculates the energy integral within each time interval; It is a square function of current, in amperes², k is the time window index, dimensionless, with a value range of 1-8, corresponding to a total duration of 80ms.

[0050] The dual-channel redundancy design includes:

[0051] ① Key characteristics of BLE transmission (data size ≤ 256 bytes) ensure real-time cloud-based early warning;

[0052] ② RS485 transmits the raw waveform (8KB data) for local in-depth analysis;

[0053] Specifically, the communication module implementation includes an unlimited channel: using a Nordic nRF5340 BLE 5.2 chip, configured in 1M PHY mode, with a data packet interval ≤7.5ms, and transmitting feature vectors [ [, β, cluster membership degree]; Wired channel: using ADI ADM2587E isolated RS485 transceiver, baud rate 1Mbps, transmitting 20ms / 4000 points of raw waveform (time taken 8.2ms).

[0054] The db4 wavelet basis functions use the Daubechies wavelet standard coefficient set, and the high-pass coefficients [-0.2304, 0.7148, -0.6309] conform to the IEEE 1241-2010 standard.

[0055] Specifically, the direct interruption method includes the following steps:

[0056] SA41, half-cycle event counter accumulation, half-cycle is 10 ms;

[0057] SA42: If the cumulative number of events is ≥8 within 100 ms, the magnetic latching relay will be triggered to trip.

[0058] SA43, driving MOSFET switch (on-time ≤ 10 ns) disconnection circuit;

[0059] SA44 records the fault waveform and illuminates the red LED indicator.

[0060] Specifically, the electromagnetic radiation-current frequency domain cross-verification method includes the following steps:

[0061] SB41, synchronously acquires electromagnetic radiation signals in the 200–300 MHz range, using a fourth-order Hilbert antenna (gain ≥5 dBi, sensitivity -110 dBm).

[0062] SB42 extracts the current energy E_c from 10–30 kHz and the electromagnetic energy E_m from 200–300 MHz, and constructs the feature vector [E_c, E_m].

[0063] SB43 calculates membership degrees using a pre-trained FCM clustering model (fault cluster centers [0.85, 1.2], normal cluster centers [0.15, 0.3]);

[0064] SB44: If the membership degree is greater than 0.8 and the condition is met for three consecutive windows, trigger tripping and report the fault location.

[0065] The principle and effect of the above technical solution are as follows: It solves the problem of missed detection of weak electric arcs by using three-level wavelet dynamic peak-to-average ratio (PAPR) detection, combines improved accumulation and CUSUM algorithms for time-frequency positioning to resist electromagnetic noise interference, and achieves action 100 ms earlier than traditional methods. It utilizes electromagnetic-current feature fusion to achieve early energy accumulation prediction. The outputs of the S2 wavelet decomposition module and the S3 disturbance localization module are both connected to the input of the S4 decision module, forming a direct "feature extraction-decision" link. The PAPR value output by S2 is directly input to the threshold comparator of S4, and the disturbance marker output by S3 triggers the window truncation unit of S4, realizing real-time fusion of dual-channel data.

[0066] In one embodiment, activating the dynamic noise reduction algorithm includes employing adaptive wavelet threshold noise reduction, suppressing inverter harmonic interference by adjusting filter parameters in real time, increasing the signal-to-noise ratio to above 35 dB, deploying a fourth-order Hilbert antenna to capture 200–300 MHz electromagnetic radiation signals, using an infrared thermal imaging sensor to monitor contact temperature rise, and constructing a current-electromagnetic-temperature three-modal sensing system.

[0067] In one embodiment, locating the starting point includes extracting 20 ms of data before and after the disturbance point, calculating the current standard deviation σ (threshold σ>0.01) and kurtosis β (threshold β>3.5), and marking the point with the maximum standard deviation as the center if its kurtosis is the minimum of the 5 neighboring points.

[0068] In one embodiment, processing the current signal includes setting the output of a 50–100 kHz high-frequency component H1 as the first stage, the output of a 25–50 kHz frequency band component H2 as the second stage, and the output of a 12.5–25 kHz frequency band component H3, and calculating the dynamic peak-to-average power ratio (PAR) in real time as the third stage; the formula used for calculating the dynamic PAR of the output of the 12.5–25 kHz frequency band component H3 in real time is:

[0069] ,in Take the largest amplitude value among H1–H3, in amperes; This represents the root mean square average of the amplitudes of the three channels H1-H3, in amperes.

[0070] The dynamic threshold is set to 15 dB + 0.3 × |dI / dt|. When the standard deviation of the load current σ > 5 A, the coefficient is adjusted to 0.5. The threshold coefficient is updated every 24 hours through a Gaussian process regression model, with the formula: new threshold = original threshold × (1 + 0.1 × |actual state - predicted state|). When the standard deviation of the load current σ > 5 A, the dynamic threshold coefficient switches to 0.5, and this switch is fed back to S2 in real time by the S3 module.

[0071] In one embodiment, S5 includes:

[0072] S501 uses LS-SVM and dynamic threshold mechanism for joint training. It sets 8,000 samples with uniform distribution of arc type and load scenario as training set, and injects 30% noise into 2,000 samples to simulate electromagnetic interference as test set.

[0073] S502, in the simulation environment (Matlab / Simulink) and photovoltaic DC system test, the metallic short circuit action time is set to ≤30 ms and the series arc action time is set to ≤100 ms. The photovoltaic DC system test meets the requirement of positioning error <10 cm within a 200m cable.

[0074] The technical principle of the above scheme is as follows: The Rogowski coil sensor is based on the principle of electromagnetic induction and can capture high-frequency transient current signals without distortion. The anti-aliasing filter is designed for the harmonic interference characteristics of nonlinear loads such as frequency converters, blocking noise above the Nyquist frequency and preventing spectral aliasing. The dynamic noise reduction triggering mechanism originates from the current mutation characteristics of the fault arc. The db4 wavelet basis function has tight support and regularity, which is suitable for extracting the high-frequency transient characteristics of the arc. The dynamic peak-to-average power ratio (PAR) quantifies the amplitude fluctuation of high-frequency components. The fault arc manifests as a sudden amplitude change. The Gaussian process regression model dynamically adjusts the threshold coefficient according to historical data to adapt to load fluctuation scenarios. The dual sliding window design is based on the power frequency cycle characteristics. The front window establishes a steady-state baseline μ, and the rear window detects the accumulated offset. The kurtosis β>3.5 identifies the spike characteristics of the current waveform; the standard deviation σ>0.01 reflects the current dispersion and helps distinguish between electric arcs and motors using transient processes; the direct interruption path targets typical parallel arcs; electromagnetic-current cross-validation utilizes the electromagnetic radiation characteristics of the fault arc and forms redundant criteria with the current frequency domain characteristics to reduce the false tripping rate; the FCM clustering model quantifies the fault probability by calculating membership degrees to avoid misjudgment by a single sensor; the energy integral model reflects the Joule heat accumulation during the fault duration; the threshold is set based on the ignition point experiment of insulating materials; the LS-SVM training samples cover 8 load scenarios, such as capacitive / inductive loads, motors, and LED drivers, to ensure the model's generalization ability; and Bluetooth + RS485 dual-channel transmission meets the real-time early warning requirements.

[0075] The technical effects of the above solution are as follows: The signal-to-noise ratio is increased to ≥35dB, significantly reducing the false alarm rate. The adaptive noise reduction algorithm can suppress broadband interference from equipment such as frequency converters and switching power supplies in industrial scenarios, improving signal purity. The detection sensitivity for weak series arcs is improved by 40%. In a 30dB noise environment, the measured false alarm rate using Tektronix MDA800 is 4.8%. The dynamic threshold mechanism avoids the problem of malfunctions under light / heavy load conditions with fixed thresholds. The positioning error is <0.1ms, which is 3 times more accurate than the traditional zero-crossing detection method. The anti-electromagnetic noise interference capability is enhanced, maintaining 95% accuracy in a 30dB noise environment. The typical arc action time is ≤100ms. Electromagnetic radiation-assisted verification improves the accuracy of weak arc identification to 92%. The early warning lead time for energy accumulation faults reaches 500ms, reducing the risk of fire. In photovoltaic DC system testing, the Fluke 1738 recording error is 8.5cm, supporting accurate fault diagnosis.

[0076] In summary, through a closed-loop design involving high-frequency signal processing (S1), multi-scale feature extraction (S2–S3), hierarchical decision-making (S4), and energy prediction (S5), the problems of missed detection, false alarms, and delays in traditional AFDD are solved. The detection rate of weak electric arcs is improved by using three-level wavelet decomposition and kurtosis analysis, the false alarm rate is reduced by electromagnetic radiation cross-validation and FCM clustering, and the action time is compressed to within 100ms by CUSUM positioning and direct interruption path.

[0077] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method of arc fault protection based on intelligent algorithm of circuit breaker, characterized in that, Comprise: S1, real-time acquisition of high-frequency current signal and dynamic noise reduction processing: real-time acquisition of line current signal, the original signal is filtered by an anti-aliasing filter to eliminate the harmonic interference of the frequency converter, and is converted into a digital signal by a built-in ADC of MCU; S2, peak-to-average ratio calculation three-level wavelet decomposition: the current signal is processed by a three-level cascaded wavelet analysis module, and the dynamic peak-to-average ratio is output to S4; S3, improved cumulative sum CUSUM algorithm disturbance positioning and time-frequency domain feature extraction: based on double sliding window detection of current disturbance, the disturbance positioning marker and time-frequency feature are output to S4, and the time-frequency domain feature includes current standard deviation σ and kurtosis β; S4, multi-modal hierarchical decision and protection execution: receiving the dynamic peak-to-average ratio of S2 and the disturbance positioning marker and time-frequency feature of S3, according to the characteristic analysis result, the path is processed; S5, arc energy accumulation prediction: an arc energy accumulation prediction model is constructed, and the processing result is output; wherein S2 and S3 are executed in parallel.

2. The circuit breaker intelligent algorithm based arc fault protection method of claim 1, wherein, S1 includes: real-time acquisition of line current signal by high-precision Rogowski coil current sensor, real-time calculation of load current rate of change dI / dt, unit A / ms, when dI / dt > 5 A / ms, activate dynamic noise reduction algorithm, the activated dynamic noise reduction algorithm includes using adaptive wavelet threshold denoising, suppressing frequency converter harmonic interference by adjusting filter parameters in real time, and improving signal-to-noise ratio to more than 35 dB.

3. The circuit breaker intelligent algorithm based arc fault protection method of claim 1, wherein, S2 includes: The wavelet analysis module uses db4 wavelet base function, the high-pass group coefficient is [-0.2304, 0.7148, -0.6309], and the low-pass group coefficient is [0.0273, -0.0322, 0.1884], and the output is set to 50-100 kHz high-frequency component H1 as the first level, 25-50 kHz frequency band component H2 as the second level, and 12.5-25 kHz frequency band component H3 as the third level.

4. The circuit breaker intelligent algorithm based arc fault protection method of claim 3, wherein, S2 further includes: real-time calculation of dynamic peak-to-average ratio PAR, formula: ； wherein Taking the maximum amplitude in H1-H3, is the three-channel average, and the dynamic threshold is set to 15 dB + 0.3 x |dl / dt|, with the coefficient adjusted from 0.3 to 0.5 when the load current standard deviation σ > 5 A.

5. The circuit breaker intelligent algorithm based arc fault protection method of claim 1, wherein, The improved cumulative sum algorithm in S3 uses double sliding windows, the front window W1 has a window length of 20 ms corresponding to 4000 points, and the rear window W2 has a window length of 10 ms corresponding to 2000 points, the current effective value mean μ of the W1 window is calculated, the offset threshold K is set to 1.5*μ, and the offset accumulation value in the W2 window is iteratively calculated , the formula is , x n is the current sampling point current value, if > H and H=5*K, the disturbance event is determined, the starting point is located, and H is the offset accumulation threshold.

6. The circuit breaker intelligent algorithm based arc fault protection method of claim 5, wherein, The S3 positioning starting point includes intercepting 20 ms data before and after the disturbance point, calculating the current standard deviation σ and kurtosis β, taking the maximum standard deviation point as the center, and if the kurtosis is the minimum value of the neighborhood 5 points, it is marked as the starting point.

7. The circuit breaker intelligent algorithm based arc fault protection method of claim 1, wherein, The S4 path processing according to the characteristic analysis result includes: if the real-time calculation of dynamic peak-to-average ratio PAR > 15 dB and the current standard deviation σ > 0.01, the direct interruption method is used; if the real-time calculation of dynamic peak-to-average ratio PAR ∈ [10 dB, 15 dB] and the kurtosis β > 3.5, the electromagnetic radiation-current frequency domain cross verification method is started.

8. The circuit breaker intelligent algorithm based arc fault protection method of claim 7, wherein, The direct interruption method includes: SA41, accumulate with half cycle event counter, half cycle is 10 ms; SA42, if the cumulative events within 100 ms are ≥8 times, trigger the magnetic latching relay to trip; SA43, drive MOSFET switch disconnection circuit, conduction time ≤10 ns; SA44, record fault waveform, light up red LED indicator.

9. The circuit breaker intelligent algorithm based arc fault protection method of claim 7, wherein, The electromagnetic radiation-current frequency domain cross verification method includes: SB41, synchronously collect 200-300 MHz electromagnetic radiation signals, use a fourth-order Hilbert antenna, gain > 5 dBi, sensitivity -110 dBm; SB42, extract 10-30 kHz current energy E_c and 200-300 MHz electromagnetic energy E_m, and construct a feature vector [E_c, E_m]; SB43, calculate the membership degree by a pre-trained FCM clustering model, the fault clustering center is [0.85, 1.2], and the normal center is [0.15, 0.3]; SB44, if the membership degree > 0.8 and the condition is met for 3 consecutive windows, trigger the trip and report the fault location.

10. The circuit breaker intelligent algorithm based arc fault protection method of claim 1, wherein, The arc energy accumulation prediction model in S5 uses the formula ; time interval T = 10 ms, if > 200 J / 100 ms, it is determined as an energy accumulation type fault, compressed fault feature data is uploaded to the cloud through the wireless communication module, and complete waveforms are output to the local monitoring terminal through the wired communication interface, the wireless communication module is a Bluetooth protocol with low power consumption, the wired communication interface is an RS485 industrial bus, the transmission delay is <10 ms, and the FCM clustering center is updated.

Images