Electrical fire monitoring method based on inherent residual current automatic compensation

Abstract

The invention discloses an electrical fire monitoring method based on inherent residual current automatic compensation, and relates to the technical field of electrical safety monitoring, and the method comprises the following steps: S100, collecting a residual current signal, executing multi-component time-frequency deconstruction processing, extracting an energy distribution characteristic through wavelet packet transformation, and extracting a mutability index in combination with short-time spectrum entropy analysis, and constructing a preliminary distribution map of higher harmonic interference, and determining boundary features of interference signals in a time domain and a frequency domain. According to the method, high-frequency interference signals are accurately positioned through time-frequency deconstruction, wavelet packet analysis and spectral entropy indexes, non-fault harmonic components are effectively eliminated in combination with amplitude-frequency coupling recognition and a recursive rejection strategy, closed-loop control is constructed by introducing an adaptive compensation and stability backtracking mechanism, accurate recognition and dynamic correction of real electric leakage risks are achieved, and the method is suitable for large-scale popularization and application. The identification precision and the safety reliability of the monitoring system in a complex industrial environment are obviously improved, and the method has good engineering adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical safety monitoring, and in particular to an electrical fire monitoring method based on automatic compensation of inherent residual current. Background Art

[0002] "Electrical fire monitoring based on automatic compensation of inherent residual current" refers to the ability to identify and model inherent residual current in the lines during electrical system operation (i.e., non-fault residual current generated under normal operating conditions by factors such as cable structure, equipment leakage inductance, and capacitive coupling). This allows for dynamic differentiation between normal residual current and abnormal leakage behavior, and automatically eliminates the inherent component through a compensation mechanism, thereby enabling accurate identification and early warning of real leakage risks. This method effectively avoids the false alarms or missed alarms caused by inherent residual current in traditional monitoring systems, improving the sensitivity, accuracy, and reliability of fire hazard identification. It is particularly suitable for intelligent monitoring and early warning responses to early electrical fire signs (such as insulation damage and ground faults) in complex power distribution scenarios.

[0003] The existing technology has the following deficiencies: In industrial load environments, the widespread use of nonlinear loads such as inverters, welding equipment, and power rectifiers can easily introduce high-order harmonic signals with multiple frequency components into the power distribution system. These harmonic signals are superimposed on the normal operating current through channels such as conductors and ground coupling capacitors, and are mixed into the monitoring signal path during the residual current acquisition process, causing the residual current waveform collected by the system to contain non-power frequency components. Because the spectral characteristics of high-order harmonics overlap with the waveform of the actual leakage current, current automatic compensation algorithms based on waveform amplitude and frequency thresholds cannot effectively distinguish between the two sources. As a result, high-order harmonic interference may be mistakenly identified as inherent residual current components for compensation, or mistakenly identified as abnormal leakage signals, triggering alarms. Ultimately, this can lead to misjudgments and missed detections, seriously affecting the accuracy and reliability of electrical fire monitoring systems.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure. Summary of the Invention

[0005] The purpose of the present invention is to provide an electrical fire monitoring method based on automatic compensation of inherent residual current to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: an electrical fire monitoring method based on automatic compensation of inherent residual current, comprising the following steps: S100 collects the residual current signal and performs multi-component time-frequency deconstruction processing. The energy distribution characteristics are extracted through wavelet packet transform, and the mutation index is extracted by combining short-time spectral entropy analysis. The preliminary distribution map of high-order harmonic interference is constructed to determine the boundary characteristics of the interference signal in the time and frequency domains. S200, based on the preliminary distribution map of high-order harmonic interference, performs amplitude-frequency coupling relationship identification, analyzes the phase offset and response delay between the power frequency and interference frequency bands, identifies the dynamic correlation between atypical leakage in the residual current and high-order harmonic interference, and determines the location of the interference frequency band and its change path; S300 performs recursive harmonic interference removal based on the interference frequency band location and change path, uses frequency-differential filtering to remove interference components, and retains low-frequency non-steady-state components through local waveform reconstruction to obtain a structurally purified residual current signal. S400 performs adaptive compensation control based on the residual current signal after structural purification. It combines historical operating conditions with the current rate of change to generate a compensation reference curve for the inherent residual current and dynamically adjusts the compensation amplitude. S500, based on the adaptive compensation control process, builds a stability index backtracking mechanism to track signal stability over multiple sampling periods, analyze response lag, amplitude offset, and error accumulation, and correct adaptive compensation control parameters; S600, based on the revised adaptive compensation control parameter system, performs joint closed-loop control, integrating time-frequency deconstruction, coupling identification, interference rejection, compensation control and backtracking mechanism to accurately identify and alarm the real leakage risk under high-order harmonic interference conditions.

[0007] Preferably, step S100 includes: The zero-sequence current signal in the three-phase distribution line is collected and filtered out by an anti-aliasing low-pass filter to remove high-frequency interference. The signal is then digitized by an analog-to-digital conversion chip at a sampling rate of more than 10,000 times per second. The digitized residual current signal is decomposed using a four-layer wavelet packet decomposition. The db6-order function in the Daubechies wavelet function is used to decompose the residual current signal, dividing the signal into 16 sub-bands and calculating the frequency domain energy in each sub-band. Based on the results of frequency band energy distribution, short-time spectral entropy analysis is introduced to process the signal by windowing it according to the power frequency period and sliding the window to extract the energy concentration and structural change characteristics of each frame signal. A comprehensive time-frequency two-dimensional disturbance map is constructed to identify the frequency boundaries, duration and disturbance intensity of high-frequency interference signals, and to complete the construction of a preliminary distribution map of high-order harmonic interference.

[0008] Preferably, step S200 includes: Extract frequency sub-bands whose energy density is higher than a preset threshold and whose spectral entropy value is lower than a preset threshold, and establish a pairing relationship between the working frequency band and the interference frequency band; Calculate the instantaneous phase offset between corresponding frequency bands based on Hilbert transform and determine their phase coupling characteristics; Taking the time position of the residual current mutation point as the anchor point, the response delay of the signal in each frequency band is calculated to determine its time domain response dependency; The above analysis results are combined to form a frequency band behavior identification result set, marking the coupling type and change path of the interference frequency band.

[0009] Preferably, step S300 includes: Construct a frequency template set of the interference frequency band to be removed, and extract the corresponding frequency band node coefficients to reconstruct the interference component signal; Dynamic time warping is used to perform frequency differential filtering to preliminarily remove interference frequency band signals; Set a sliding time window to perform local waveform reconstruction, retain low-frequency non-stationary components and perform amplitude correction; The elimination iteration threshold is set based on the high-frequency energy proportion and reconstruction error, and multiple rounds of interference elimination are recursively performed to output the residual current signal after structural purification.

[0010] Preferably, step S400 includes: Build an operating condition feature library containing residual current feature vectors and compensation reference waveforms under historical operating scenarios; Extract the residual current characteristics in the current cycle and calculate the Euclidean distance between it and the historical template, determine the optimal matching template and generate a dynamic compensation target curve; Compare the current signal with the compensation target curve point by point and perform amplitude compensation and delay adjustment operations within the set threshold range; The error indicators are counted over multiple cycles and the compensation reference curve is reconstructed based on the error feedback results to achieve closed-loop optimization.

[0011] Preferably, step S500 includes: Ten consecutive sampling periods are set as the analysis window, and the maximum amplitude, minimum amplitude, root mean square value, average rising slope and energy center frequency of each period are extracted to construct the period feature vector set; Based on the periodic characteristic vector set, the range, standard deviation and mean square deviation of each parameter are calculated, and the current window stability index is constructed according to the set weight combination; When the stability index is lower than the set threshold, the compensation output is compared with the compensation reference curve cycle by cycle, and the response delay point, amplitude offset and error accumulation value are extracted to identify the cause of instability; The compensation control parameters are adjusted according to the recognition results and the correction values ​​are enabled in the next cycle. At the same time, the correction information is recorded for subsequent tracking and analysis.

[0012] Preferably, step S600 includes: A unified sampling time base and data transmission link are set, and the residual current signal is decomposed by wavelet packets to extract the frequency domain energy matrix and disturbance spectrum as input; Based on the frequency domain spectrum, it performs amplitude-frequency coupling identification processing, extracts the phase offset, response delay, and energy index of the interference frequency band, drives three rounds of recursive interference removal processing, and outputs a purified signal; Load the current compensation parameter system based on the purified signal, generate the target compensation curve by combining the change rate and dynamic adjustment coefficient, adjust the signal point by point and perform amplitude control; At the end of the cycle, the compensation effect and stability index are analyzed. If the compensation accuracy decreases, the dynamic parameters and template matching are corrected, the parameter system is updated and loaded into the next cycle to complete the closed-loop control.

[0013] In the above technical solution, the technical effects and advantages provided by the present invention are: This invention achieves accurate positioning of high-frequency interference signals by introducing multi-component time-frequency deconstruction, wavelet packet energy extraction, and spectral entropy mutation analysis. It effectively distinguishes and eliminates non-fault harmonic components by establishing an amplitude-frequency coupling behavior recognition mechanism and a recursive interference elimination strategy. Furthermore, by combining historical operating conditions with dynamic change trends, it implements adaptive compensation and stability retrospective adjustment, constructing closed-loop feedback control logic so that the system can continuously optimize judgment criteria and correct compensation strategies online. Overall, this solution solves the problem of misjudgment and omission caused by the inability to distinguish between high-order harmonics and true leakage signals in traditional systems, achieves detailed modeling and intelligent response to abnormal current behavior in complex power distribution scenarios, and has a high degree of engineering adaptability and safety protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0015] Figure 1 This is a flow chart of the method for electrical fire monitoring based on automatic compensation of inherent residual current of the present invention. DETAILED DESCRIPTION

[0016] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0017] The present invention provides Figure 1 The electrical fire monitoring method based on inherent residual current automatic compensation shown includes the following steps: S100, collecting a residual current signal and performing multi-component time-frequency deconstruction processing on the residual current signal, including extracting multi-band energy distribution characteristics through wavelet packet transform, and extracting signal mutation indexes in combination with short-time spectral entropy analysis, so as to construct a preliminary distribution map of high-order harmonic interference and determine the boundary characteristics of the interference signal in the time domain and frequency domain; Collecting the residual current signal and performing multi-component time-frequency deconstruction processing on the signal specifically includes the following steps: High-precision acquisition of residual current in the electrical system. The specific operation is to install a zero-sequence current sensor in the three-phase distribution line. The sensor is used to monitor the unbalanced current difference between the three-phase currents in real time, and introduce the residual current signal into the signal acquisition circuit through analog signal transmission. The acquisition circuit is equipped with an anti-aliasing low-pass filter. The cutoff frequency of the filter is set to less than half of the maximum sampling frequency of the system to suppress high-frequency interference noise outside the frequency band above 50Hz. The filtered analog signal is introduced into a 16-bit high-precision analog-to-digital conversion chip and digitized at a sampling rate of more than 10,000 times per second to ensure that no less than 200 sampling points are obtained in each 50Hz power frequency cycle, thereby fully recording the detailed change characteristics of the signal. During the signal acquisition process, an insulated and isolated shielded transmission cable is used to prevent external electromagnetic interference from distorting the signal.

[0018] The digitized residual current signal undergoes wavelet packet decomposition to analyze the energy characteristics of its various frequency bands from a frequency perspective. This process uses a Daubechies wavelet function, which boasts excellent symmetry and compact support, as the basis function. Through a four-layer wavelet packet decomposition operation, the original time series signal is divided into 16 subbands. These subbands cover different frequency ranges from 0 Hz to the sampling bandwidth. For example, the first subband might cover 0 to 312.5 Hz, the second subband from 312.5 to 625 Hz, and so on, until the last subband covers the frequency range closest to the sampling bandwidth. Within each subband, the energy of the corresponding signal—the sum of the squared amplitudes of the sample points—is calculated to quantify the signal activity within that frequency range. The resulting band energy spectrum clearly reflects the residual current intensity distribution within each frequency range. In particular, abnormal energy concentration in the 250 Hz to 3000 Hz band suggests potential interference from higher-order harmonic signals.

[0019] The Daubechies wavelet function is a type of wavelet basis function with good time-frequency localization ability and orthogonality, and is often used to process the multi-scale decomposition of non-stationary signals. In this step, it is used as the basis function of wavelet packet decomposition to decompose the collected residual current signal into multiple frequency bands, thereby extracting the local energy characteristics of different frequency components. The Daubechies wavelet can effectively retain the transient changes and high-frequency interference characteristics in the signal, and is particularly suitable for identifying high-order harmonics introduced by equipment such as frequency converters and welding machines in industrial electrical systems. Specific optional functions include db4, db6, db8, etc., among which db6 achieves a good balance between signal fidelity and computational complexity, and is the preferred function for constructing a multi-component time-frequency deconstruction model in this step.

[0020] Based on frequency band energy distribution analysis, short-time spectral entropy analysis is introduced to further reveal the temporal characteristics of residual current signals. This is achieved by windowing the complete signal sequence according to the length of a power frequency cycle, with each analysis window sliding forward by a fixed sample interval to obtain multiple overlapping signal frames. The frequency band energy distribution within each frame is normalized, allowing analysis of energy concentration variations across frequency bands. The changing trends in spectral entropy values ​​reveal the stability and structural complexity of the signal. If a frequency band consistently exhibits high energy concentration and strong structural regularity across multiple sliding windows, the signal component in that frequency band may originate from periodic harmonic interference generated by electrical equipment. If the energy distribution in a frequency band is unstable, fluctuates dramatically, and exhibits strong transient mutation characteristics, it is often closely related to abnormal leakage processes. By integrating the spectral variation results in different time windows, a two-dimensional time-frequency disturbance distribution map is constructed to describe the persistence of the interference signal over time and its distribution pattern in frequency space.

[0021] Finally, by combining the frequency energy distribution characteristics obtained from wavelet packet decomposition with the time-varying disturbance characteristics obtained from short-time spectral entropy analysis, a preliminary distribution map of high-order harmonic interference is constructed. This map is presented in the form of a two-dimensional matrix, where each row corresponds to an analysis period and each column corresponds to a frequency sub-interval. The values ​​in the matrix represent the disturbance intensity of the frequency band during that period. Based on this map, frequency bands with high disturbance persistence and energy concentration can be identified, and parameter information such as the frequency boundary, duration, frequency point of the energy peak, and degree of interference coupling with adjacent frequency bands of the high-order harmonic signal can be further extracted.

[0022] The main purpose of this step is to provide high-resolution, structured time-frequency information support for subsequent high-order harmonic interference identification and compensation. Specifically, it performs a multi-dimensional, detailed analysis of the collected residual current signal to accurately reveal the hidden non-power frequency interference characteristics. Through wavelet packet transform, the original current signal can be decomposed into multiple sub-signals with different frequency ranges, and the local energy distribution of each frequency band can be extracted to identify whether there is abnormal energy accumulation concentrated in the high-frequency range, which lays the foundation for the frequency domain location of harmonic interference. At the same time, combined with short-time spectral entropy analysis, the energy variation trend of each frequency band can be captured over time, and the signal can be judged to be sudden or regular, thereby distinguishing periodic harmonic disturbances from non-periodic leakage behavior. The combination of the two forms a preliminary distribution map of high-order harmonic interference, which not only clearly defines the distribution boundaries of the interference signal on the time and frequency axes, but also extracts its key behavioral characteristics, such as the frequency band center, disturbance duration, and frequency jump area. The establishment of this map provides accurate, detailed and traceable basic data support for subsequent amplitude-frequency coupling analysis, interference behavior identification, misjudgment avoidance and dynamic compensation control, and is a key prerequisite for the entire process of intelligent identification of real leakage risks in the present invention.

[0023] S200, based on the preliminary distribution map of high-order harmonic interference, performs amplitude-frequency coupling relationship identification processing. By analyzing the phase offset trend and response delay characteristics between the power frequency band and the interference frequency band, it identifies the dynamic behavioral correlation between the atypical leakage component and the high-order harmonic interference component in the residual current signal, and determines the frequency band location and change path of the harmonic interference; The process of performing amplitude-frequency coupling relationship identification based on the preliminary distribution spectrum of high-order harmonic interference may specifically include the following steps: Based on the preliminary distribution map of high-order harmonic interference constructed in the previous steps, frequency component identification and structured partitioning are performed. Specifically, frequency sub-bands that exhibit energy density higher than a set threshold (such as the 80th percentile) and spectral entropy lower than a set stability threshold (such as 0.3) in multiple consecutive sampling periods are selected from the map and defined as the interference frequency band to be judged. This region is usually concentrated in the sub-band between 150Hz and 4500Hz, including the distribution frequency bands of common industrial harmonic sources such as the 3rd, 5th, 7th, and 11th harmonics. At the same time, the wavelet packet sub-band corresponding to the 50Hz industrial frequency signal is used as the reference frequency band, and its position and change characteristics in the map are extracted to establish an analytical pairing relationship between the industrial frequency band and each interference frequency band, clarifying the frequency domain position and time domain distribution basis between the analysis objects.

[0024] For the paired frequency sub-bands mentioned above, the instantaneous phase changes between the power frequency band and each interference frequency band are calculated respectively. The specific method is to use Hilbert transform to extract the analytical signal representation of each frequency band signal, and then calculate its corresponding instantaneous phase trajectory. For each pairing of power frequency and interference frequency bands, within four adjacent complete sampling cycles, with a sliding window step of 10 milliseconds, the phase difference change trend is statistically analyzed. If a certain interference frequency band and the power frequency band maintain a stable phase offset of less than 30 degrees in most sampling windows, and show a synchronous change with a consistent trend over time, it is determined to be an interference component with coupled behavior; on the contrary, if the phase difference fluctuates randomly or shows an obvious asynchronous trend in different cycles, it is determined that the interference frequency band has no strong correlation with the power frequency component in the phase dimension, and is suspected to be an independent harmonic source signal.

[0025] Further, a time delay response analysis is performed on frequency bands with phase coupling characteristics. The location of the mutation point is identified in the residual current signal, and the energy surge threshold method is used. A mutation point is defined as a signal energy increase exceeding the mean level by more than 30% in any sampling period. Using the mutation point as the anchor point, the response start times of the power frequency band signal and the signals of each interference frequency band before and after this location are detected, and the time difference is calculated as the interference response delay value. If the delay time is between 5 milliseconds and 20 milliseconds, and the delay value remains consistent across multiple mutation point scenarios, it indicates that there is a time domain response dependency between the interference frequency band and the power frequency band, and it is determined to be a slave coupling frequency band. If the delay time differs by more than 10 milliseconds between one mutation point and another, or the response start time fluctuates significantly, it indicates that the frequency band is not affected by the main frequency band and is considered a harmonic disturbance with behavioral independence.

[0026] The phase offset analysis results and time delay analysis results are combined to construct a frequency band behavior identification result set. This result set includes dimensions such as the interference band number, frequency range, phase coupling score (e.g., a quantitative value between 0 and 1), time delay consistency indicator (e.g., standard deviation), and comprehensive coupling confidence label (e.g., strong coupling, medium coupling, no coupling). Based on the identification results, all interference bands in the spectrum are classified and labeled, and an interference behavior map with behavioral labels is plotted. This map can be used to clearly identify the coupling type, response path, and behavioral changes over time for each interference band, providing an accurate input basis for subsequent recursive interference removal and identification of true leakage behavior.

[0027] This step, based on the preliminary distribution map of high-order harmonic interference constructed in the previous stage, conducts an in-depth analysis of the behavioral correlations between various frequency bands in the residual current signal, thereby accurately distinguishing between high-order harmonic components that are behaviorally coupled with the power frequency signal and uncoupled, independently distributed, atypical leakage components. In complex industrial power environments, high-order harmonics introduced by nonlinear load devices (such as inverters, welding machines, and rectifiers) often embed into the normal current with a certain phase delay or characteristic synchronization. Although their frequency components are higher than the power frequency range, they often exhibit a time-domain response pattern that is synchronized or lagged with power frequency changes. This step quantifies the phase offset trends between the power frequency band and each interference frequency band, combined with the delay calculation of the signal mutation response moment, effectively identifying whether the interference frequency band is excited by the main frequency signal and thus determining its source attributes. If a certain frequency band is highly synchronized with the power frequency band in the phase dimension, and the response delay is stable and predictable over multiple sampling cycles, then the interference frequency band can be determined to be an electromagnetic harmonic that is highly coupled to the power frequency behavior; on the contrary, if it shows strong behavioral independence, further attention should be paid to whether it may come from an abnormal leakage path or a non-periodic discharge signal caused by deterioration of insulation performance. Through this step, the system can not only more accurately locate the specific position of the interference frequency band in the spectrum, but also track its change path based on the law of time evolution, providing the necessary behavioral boundary division and frequency domain dynamic mapping support for the subsequent high-precision harmonic elimination and true leakage signal extraction. This identification mechanism breaks through the limitations of traditional static frequency threshold and amplitude judgment methods, making residual current monitoring more intelligent and adaptable in the context of complex electromagnetic interference.

[0028] S300: Based on the harmonic interference frequency band location and change path, recursive harmonic interference removal processing is performed. The removal processing includes using a different-frequency differential filtering method to remove the identified interference frequency band signal and retaining the low-frequency non-steady-state component through a local waveform reconstruction method, thereby obtaining a residual current signal after structural purification; Based on the frequency band position and change path of the high-order harmonic interference, a recursive harmonic interference elimination process is performed, which specifically includes the following steps: Construct a set of frequency templates for the interference bands to be removed. Based on the frequency band behavior identified during the previous amplitude-frequency coupling relationship identification process, select frequency subbands that exhibit strong coupling behavior over multiple sampling periods. For example, with a sampling frequency of 10 kHz and a four-level wavelet packet decomposition, 16 subbands of equal bandwidth are obtained. By calculating the phase offset, response delay, and energy density ratio between the power frequency reference band (0–312.5 Hz) and high-frequency bands (e.g., 625–937.5 Hz, 937.5–1250 Hz, and 1250–1562.5 Hz), these three frequency ranges are selected as bands of significant high-order harmonic interference. These frequency bands, along with their respective start and end frequencies, center frequencies, energy means, and trend markers, form a set of interference frequency templates, which serve as the target frequency set for subsequent differential filtering.

[0029] The interference components corresponding to the above-mentioned frequency templates are subjected to heterogeneous frequency differential filtering. The specific steps are as follows: within each sampling period, the original residual current signal is decomposed through a 4-layer wavelet packet to extract the node coefficients corresponding to the above-mentioned interference frequency band (such as the 5th, 6th, and 7th nodes); this group of node coefficients is separately reconstructed into an interference component signal sequence; then, a point-to-point subtraction method is used to subtract the reconstructed interference signal from the original residual current signal sample by sample to obtain the initial filtering result. To prevent the phase alignment error from causing the overlapping frequency attenuation error, the dynamic time warping method (Dynamic Time Warping) is used to locally align the starting point and peak point of the original signal and the interference signal to ensure that the differential operation maintains maximum synchronization at the key waveform feature points, thereby improving the accuracy of interference removal.

[0030] Based on the initial filtered signal, local waveform reconstruction is performed to protect low-frequency, non-stationary waveform structures that may contain true leakage characteristics. This step uses a sliding energy detection method to identify locations where sudden changes in the signal occur by setting a local time window (e.g., 500 sampling points). The detection logic is as follows: when the signal energy surge within the local window exceeds 1.5 times the average value of the window in the previous cycle and lasts for more than 10 milliseconds, this segment is considered a non-stationary segment that may contain abnormal leakage behavior. Within the non-stationary segment, a forward predictive interpolation algorithm is used to compensate for energy peaks that may have been mistakenly deleted in the local waveform, and a backward amplitude recalibration mechanism is used to perform amplitude correction based on the historical peak range in the previous 10 power frequency cycles. This method ensures that the sudden behavior characteristics of the true leakage signal are retained to the greatest extent possible, even when the interference frequency band and the abnormal signal frequency band boundary overlap.

[0031] A recursive elimination process is constructed to perform multiple rounds of iterative purification on the eliminated residues. The high-frequency energy retention rate and waveform reconstruction error of two consecutive rounds of elimination are set as the judgment criteria. Specifically, if the energy proportion of the frequency band above 625Hz in the signal after elimination is still greater than 5%, or the local window reconstruction error (i.e., the mean square error of the difference between the reconstructed waveform and the original signal waveform) exceeds 3% of the original signal mean amplitude, a second round of elimination is performed, and the above-mentioned difference and reconstruction steps are repeated. After each round of iteration, the energy statistical parameters of the interference template are updated, and the frequency band coverage is dynamically adjusted to adapt to the actual interference change path. The iterative process is set to a maximum of three rounds. When the purification criteria are met, that is, the high-frequency energy proportion is less than 5%, the reconstruction error is less than 2%, and the mutation point detection accuracy is higher than 90%, the purification is determined to be complete, and the residual current signal after the final structure purification is output.

[0032] This step, based on the identification of the location of the harmonic interference band and its time-varying path, implements targeted signal purification processing, effectively removing non-leakage interference components introduced by higher-order harmonic coupling from the residual current signal while maximally preserving the non-steady-state, non-periodic characteristics of potential abnormal leakage behavior. In industrial power distribution systems, higher-order harmonics introduced by nonlinear loads such as inverters, welding equipment, and power rectifiers often superimpose on the normal power frequency current to form periodic disturbances. These disturbance signals have specific frequency band concentration and time persistence. If not identified and removed, they can easily be misidentified as leakage phenomena and trigger false alarms. This step first uses the interference frequency bands identified in the previous stage to accurately remove these bands from the original signal through a heterogeneous frequency differential filtering method. Unlike traditional fixed-bandwidth filters, this method dynamically constructs frequency templates based on the actual interference behavior, ensuring removal accuracy and range control. Subsequently, in order to avoid the loss of the true abnormal current signal due to false filtering, a local waveform reconstruction method was introduced to perform time domain reconstruction and amplitude compensation on the abrupt waveform in the low-frequency band, ensuring that the true leakage signal maintains structural integrity during the interference removal process. Through this "removal + reconstruction" recursive processing mechanism, the residual current signal after structural purification is finally output. This signal not only has significantly weakened noise interference, but also has highly fidelity original abnormal characteristics, providing an accurate and reliable signal basis for subsequent compensation control and risk assessment. This processing process achieves fine coordination in the frequency domain, time domain and behavioral characteristics, significantly improving the electrical fire monitoring system's response capability to real leakage risks and the misjudgment suppression effect under complex interference backgrounds.

[0033] S400, based on the residual current signal after the structure is purified, performs adaptive compensation control processing, which includes combining historical operating condition characteristics with the current signal change rate to dynamically generate a compensation reference curve for the inherent residual current, and adjusts the residual current compensation amplitude in real time according to the compensation reference curve; The process of performing adaptive compensation control processing based on the residual current signal after structural purification includes the following steps: Construct a historical operating condition feature library for compensating for the generation of the benchmark curve, and extract typical inherent residual current behavior templates related to the power frequency operating state. In the initial stage of system operation or the regular inspection stage, select multiple time periods in normal operation and without abnormal leakage events. Sample the residual current signal after structural purification point by point within each sampling cycle, and extract five parameters including the peak amplitude of the signal, the cycle mean, the slope of change, the frequency domain energy concentration rate, and the zero crossing position within the cycle to form a feature vector. These feature vectors are sorted and organized by time period, load type, and current level. For example, the daytime high-load operation stage, the midday low-load standby stage, and the nighttime maintenance operation stage are classified separately, and the corresponding extracted residual current average waveform is saved as a reference template. Each template is stored in the feature database in the form of a vector as a reference for subsequent comparison with the current operating condition.

[0034] The residual current signal after structural purification, which is currently collected in real time, is obtained. Within a complete power frequency cycle, its latest waveform features are extracted in units of 200 sampling points, and compared item by item with the historical operating condition templates in the feature library. The Euclidean distance algorithm is used to calculate the distance between the current cycle feature vector and all templates in the database, and the template with the smallest distance is selected as the reference for matching the current operating condition. For example, if the peak value of the current residual current waveform is 38mA, the cycle mean is 12mA, and the average change rate is 5mA / ms, then the template corresponding to the "medium load stable operation scenario" in the library has the highest match. The corresponding compensation reference waveform in this template is used as the preliminary compensation reference curve, and further based on the change trend of the current waveform in 10 consecutive sampling points, the instantaneous adjustment coefficient is calculated to correct the response characteristics of the compensation curve at the local mutation point or slow change section to form the final usable dynamic compensation target curve.

[0035] Based on the generated compensation target curve, the current residual current signal is compensated point-by-point. At each sampling point, the current signal amplitude is compared in real time with the reference value at the same time point on the compensation target curve. The amplitude difference is calculated and a determination is made as to whether it exceeds the dynamic adjustment threshold. For example, if the current amplitude is 42mA and the reference value is 35mA, and the difference exceeds the set error limit of 5mA, amplitude reduction is performed. If the difference is within the ±3mA range and the waveform is stable, standard proportional compensation is performed. The amplitude adjustment limits are set to ±15% to prevent misjudgment and overcompensation caused by short-term fluctuations. Furthermore, a compensation buffer is set at local waveform abrupt changes (e.g., when the slope change rate exceeds 10mA / ms), delaying the gradual adjustment by two sampling points to prevent sudden and drastic changes in the compensation curve from causing unbalanced response. The adjustment process utilizes real-time sliding window logic to ensure that the compensation output remains continuous, smooth, and closely adheres to the dynamic characteristics of the original waveform within each cycle.

[0036] The system performs short-term statistics and self-correction on the compensation execution results over multiple consecutive power frequency cycles, constructing a feedback mechanism based on error accumulation and response deviation to correct the construction parameters of the compensation baseline curve for the next cycle. Within a time window consisting of 10 cycles, the average deviation, maximum error, mean square error, and slope deviation indicators of the current compensated residual current signal are calculated and compared with the data from the previous window. For example, if the mean square error exceeds twice the value of the previous window, or the proportion of high-frequency energy in the residual signal increases by more than 10%, it is determined that the current compensation curve has a structural mismatch and requires reconstruction of the compensation baseline. At this point, the feature matching process is re-executed, using the template curve with the second closest match, and the dynamic adjustment coefficient is recalculated. At the same time, the error feedback value is recorded in the historical record and used to participate in weight allocation in the next parameter tuning strategy, thus forming a closed-loop regulation mechanism with continuous learning and self-adaptation capabilities.

[0037] This step aims to further identify and compensate for any remaining inherent non-fault residual current components within the residual current signal after structural purification. This ensures that the electrical fire monitoring system can more accurately identify abnormal current fluctuations caused by actual leakage, effectively reducing the risk of false alarms and missed alarms. In industrial power distribution systems, inherent residual current is widely present in various load conditions during normal operation. It is typically caused by factors such as cable-to-ground capacitance, electrical equipment leakage inductance, and current transformer zero deviation. While its manifestation is stable, it exhibits some overlap in amplitude and waveform with earlier abnormal leakage signals. This step constructs a historical operating condition feature database, statistically models the residual current under typical load operating scenarios, extracts representative non-fault current behavior curves as compensation benchmarks, and dynamically selects the reference model that best matches the current operating conditions, combining the rate of change and waveform characteristics of the current signal after structural purification, generating a time-varying target compensation curve. The system then adjusts the amplitude of the current residual current signal in real time based on this curve, dynamically removing the inherent background current component. This makes the remaining current variation more directional and valuable for anomaly identification. This step realizes the personalized, adaptive and dynamic optimization of residual current compensation processing in the time domain and behavioral levels, effectively avoiding the response lag or misjudgment problems caused by traditional fixed threshold or static template methods, and providing a purer and more reliable basic signal for subsequent leakage risk warning algorithms. It is a key link in realizing intelligent monitoring and accurate alarm.

[0038] S500: Based on the execution process of the adaptive compensation control process, a residual current stability index backtracking mechanism is established to track the stability of the residual current signal after structural purification over multiple sampling periods, analyze the response lag, amplitude offset, and error accumulation in the compensation results, and modify the adaptive compensation control process parameter system based on the analysis results; Based on the execution process of the adaptive compensation control process, a residual current stability index backtracking mechanism is constructed, which specifically includes the following steps: A fixed number of sampling cycles was set as the stability assessment window, and the residual current signal after structural cleanup was collected in segments. Using a 50 Hz power frequency cycle as the basic unit, with 200 sampling points per cycle, 10 consecutive cycles were selected as the analysis window, processing a total of 2000 data points. Within each cycle, five characteristic parameters of the residual current signal after cleanup were extracted: maximum amplitude (in mA), minimum amplitude (in mA), root mean square value (in mA), average rise slope within the cycle (in mA / ms), and frequency energy center distribution value (in Hz). Taking the sample data as an example, the maximum amplitude in cycle 1 is 42.3mA, the minimum amplitude is 18.5mA, the RMS is 30.1mA, the average rising slope within the cycle is 6.2mA / ms, and the frequency energy center distribution value is 140Hz. The maximum amplitude in cycle 2 is 41.7mA, the minimum amplitude is 17.9mA, the RMS is 30.5mA, the average rising slope within the cycle is 5.9mA / ms, and the frequency energy center distribution value is 135Hz. And so on and so forth until cycle 10. These parameters are organized into 10 sets of periodic feature vectors to form a behavioral benchmark dataset within the current time window, providing a quantitative basis for subsequent stability analysis.

[0039] Based on the above characteristic parameters, the fluctuation amplitude and behavioral consistency of the residual current signal within the current window are calculated to construct a comprehensive stability index. Specifically, the range (maximum minus minimum), standard deviation (reflecting cycle-to-cycle fluctuations), and mean square deviation (reflecting the degree of deviation from the average) are calculated for each parameter. For example, for the RMS value sequence {30.1, 30.5, 30.3, …}, a standard deviation of 0.7 mA and a range of 1.6 mA indicate moderate fluctuation. The standardized results of the five parameters are then combined to calculate a comprehensive index using pre-defined weights, such as 0.35 for amplitude fluctuation, 0.25 for slope fluctuation, 0.25 for energy distribution stability, 0.1 for RMS stability, and 0.05 for minimum value stability. This weighted calculation yields the stability score for the current window. The score ranges from 0 to 1, with a score ≥0.8 considered highly stable, between 0.6 and 0.8 considered moderately stable, and <0.6 indicating instability, triggering the parameter backtracking process.

[0040] After determining an unstable state, the current compensation execution process is analyzed for cycle-by-cycle response differences to identify the specific cause of the instability. This is done by aligning the actual compensation output signal of each cycle with the adaptive compensation reference curve of the previous cycle, comparing the response delay point (the time difference between the first rising edge, measured in sampling points), amplitude offset value (the difference between the actual peak value and the target peak value of each cycle, measured in mA), and error accumulation (the sum of the absolute values ​​of the errors over the past three cycles, measured in mA / cycle). For example, if the first rising edge of cycle 8 lags by 4 sampling points, the amplitude deviation reaches 7.5 mA, and the cumulative error for three consecutive cycles exceeds 20 mA / cycle, the current compensation strategy is considered to have insufficient configuration for hysteresis control and amplitude prediction, and the compensation control parameter system needs to be revised immediately.

[0041] Based on the specific deviation type and numerical feedback identified above, the compensation control strategy parameters are corrected and the results are applied to the next sampling cycle. If response lag is identified as the primary issue, the compensation adjustment slope is increased from the current setting of 6mA / ms to 8mA / ms to accelerate response. If amplitude deviation is identified, the dynamic adjustment coefficient weight is adjusted, with the waveform change rate weight for the current cycle increased from 0.3 to 0.45 to enhance the ability to adapt to local sudden changes. If error accumulation is the cause, a suboptimal operating condition template is used in the template matching phase, and the matching threshold is relaxed from 0.9 to 0.85 to improve adaptation flexibility. The corrected parameters are immediately applied in the next cycle. The hysteresis value, deviation value, and correction amount during this correction process are recorded in the stability tracking log for subsequent retrospective analysis and long-term model optimization. The entire retrospective correction logic operates continuously in a sliding window manner during the compensation control process, ensuring that compensation accuracy and output stability are dynamically maintained even under complex operating conditions such as frequent load switching and increased external disturbances.

[0042] This step establishes a dynamic monitoring and feedback optimization mechanism for residual current signals after structural purification. This mechanism continuously evaluates and adjusts the effectiveness of the adaptive compensation control process, ensuring that the electrical fire monitoring system maintains high accuracy and stability even during long-term operation and in complex interference environments. Due to the frequent load switching and dynamic changes in the grid harmonic environment in industrial power plants, the adaptive compensation algorithm, while highly responsive, can exhibit response lag, under-compensation, or over-compensation if its parameters deviate from the current optimal state during actual operation, compromising the identification of abnormal leakage behavior. This step uses an analysis window spanning multiple consecutive sampling cycles to track the stability of the purified signal in real time across multiple metrics (such as RMS value, maximum amplitude, slope, and energy distribution trend). The system further calculates the fluctuation amplitude, standard deviation, response time offset, and error accumulation between cycles to generate a quantifiable stability index. When the index falls below the set stability threshold (for example, a score below 0.6), a retrospective analysis process is triggered to locate key parameter points in the current compensation strategy that cause stability degradation, including insufficient response rate, accumulated errors in dynamic adjustment factors, and mismatched operating conditions. Based on the retrospective results, the compensation parameter system is automatically corrected, such as adjusting the dynamic compensation slope, reselecting the compensation curve template, and optimizing the sampling and comparison strategy within the cycle. Ultimately, this mechanism enables online self-optimization of the compensation strategy without relying on human intervention. It maintains high adaptability and reliability of compensation control through continuous learning and feedback adjustment. It is an important support link for realizing intelligent, closed-loop residual current precision monitoring and has significant technical necessity and practical value.

[0043] S600, based on the revised adaptive compensation control processing parameter system, performs a joint closed-loop control process, integrating multi-component time-frequency deconstruction, amplitude-frequency coupling relationship identification, recursive harmonic interference elimination, adaptive compensation control, and stability backtracking to form a control closed loop, thereby accurately identifying and alarming real leakage risks under high-order harmonic interference conditions; Based on the revised adaptive compensation control processing parameter system, the process of executing joint closed-loop control processing includes the following steps, which aims to integrate the previous processing links into a complete control process with real-time response and self-adjustment capabilities, and realize the recognition and alarm of real leakage signals under high-order harmonic interference conditions.

[0044] At the structural level, a unified sampling time base and data transmission link are established to ensure that multi-component time-frequency deconstruction, amplitude-frequency coupling relationship identification, recursive harmonic interference elimination, adaptive compensation control, and stability backtracking can be efficiently executed sequentially within the same cycle. The sampling frequency is set to 10,000 points per second, corresponding to 200 sampling points per 50Hz power frequency cycle. After signal acquisition is completed, the db6-order Daubechies wavelet function is immediately called to perform a four-layer wavelet packet decomposition, splitting the residual current signal into 16 fixed-band subbands. The energy density and short-time spectral entropy of each frequency band are calculated, and the frequency domain energy distribution matrix and disturbance identification map are output. This map is represented as a two-dimensional matrix, with each row corresponding to a frequency band and each column corresponding to a period window. The values ​​in the matrix represent normalized energy and disturbance intensity indicators, providing accurate and continuous input data for subsequent interference identification.

[0045] Amplitude-frequency coupling relationship identification is performed, and the output is used to drive recursive harmonic interference removal. This process extracts the phase offset, response delay (expressed in sampling points), burst duration, and center frequency identifier for each frequency band as a parameter set. A set of identifiers, such as "band number = 8, center frequency = 1250 Hz, phase offset = 27°, and response delay = 4 points," forms the list of frequency bands to be removed. After entering the recursive interference removal process, three rounds of frequency-differential filtering are performed on that frequency band by matching the wavelet packet coefficient numbers. In the first round, the energy of the frequency band is subtracted to reconstruct the signal. If the energy of the frequency band in the remaining signal still accounts for more than 5% of the total energy of the structured cleansed signal, the signal enters the second round of processing. If interference energy still remains after two rounds of processing, the third round combines the weighted matching of the coefficients of adjacent frequency bands and removes them in a coordinated manner. All parameters are filtered to six digits of accuracy to ensure that no false negatives are removed. The final cleansed signal significantly eliminates periodic interference in the frequency domain while fully preserving low-frequency non-stationary anomalies.

[0046] The adaptive compensation control parameter system corresponding to the current cycle is loaded, and precise compensation is performed based on the purified signal. This parameter system includes five key control factors: the best matching template number (e.g., template T3 - medium load steady state), a dynamic adjustment coefficient (set in the range of 0.8–1.2), upper and lower compensation limits (±10mA, respectively), a response time lag correction value (fixed to the difference between the sampling points in the previous cycle), and a local sudden change point tracking window width (set to 20 sampling points). A sliding window is used to analyze the change rate, peak amplitude position, and slope of the current cycle signal compared to the previous cycle's baseline value to dynamically generate the current cycle's compensation target curve. Compensation adjustment is then performed at each sampling point: if the current amplitude exceeds the target value by more than 5mA, the compensation is reduced to the target value; if it falls below the target value by less than 3mA, the compensation is increased, with the compensation amplitude not exceeding the ±10mA limit. Linear interpolation is used during the adjustment process to ensure consistent response between sudden and gradual changes.

[0047] At the end of the current cycle, stability backtesting is performed to analyze the consistency of compensation results with historical cycles and trigger parameter correction based on the stability score. From the residual signal over 10 consecutive cycles, indicators such as the RMS value sequence, slope trend, spectral center of gravity shift rate, and burst point volatility are extracted. If the calculated comprehensive stability index falls below 0.65, compensation accuracy is considered to have degraded. If the lag response persists beyond four sampling points and the cumulative error exceeds 30% of the average error of the previous three cycles, the dynamic adjustment coefficient for the current cycle is adjusted from 1.0 to 1.1, the burst point tracking window is shortened to 16 sampling points, and the compensation template is switched to the second-best model closest to the current cycle (for example, from T3 to T2). After the correction is completed, the parameters are immediately loaded into the next cycle, forming a complete closed loop of parameter-compensation-feedback-correction. By executing this sequence in each cycle, the system can dynamically adapt to grid noise, load changes, and environmental disturbances during continuous operation, achieving self-learning and self-optimization.

[0048] This step integrates the multiple key processes involved in the electrical fire monitoring process—including multi-component time-frequency deconstruction, amplitude-frequency coupling relationship identification, recursive harmonic interference removal, adaptive compensation control, and stability backtracking—into a closed-loop control system using a unified data structure, time window, and execution sequence. This enables the system to continuously, dynamically, and accurately identify true leakage signals and issue alarms in complex distribution environments where high-order harmonic interference is prevalent. Traditional electrical fire monitoring relies on single criteria, such as amplitude thresholds or frequency domain templates, which struggle to maintain high recognition rates in the presence of overlapping spectra or high noise. The closed-loop mechanism constructed in this step first extracts the time-frequency eigenvalues ​​of high-frequency disturbances from the raw residual current signal, dynamically identifying their behavioral boundaries and change paths. Non-leakage interference components are then accurately removed through recursive processing. Based on the cleaned-up signal, an adaptive compensation algorithm is used to dynamically correct the inherent background current. Finally, a stability backtracking mechanism continuously evaluates compensation accuracy and adjusts compensation strategy parameters in a closed-loop manner to avoid misjudgments and delayed responses. Crucially, this closed-loop structure isn't a static series connection; instead, it utilizes parameter feedback and time-domain response linkage to ensure that each processing step is not only effective for the current signal but also adapts to non-stationary operating conditions such as load changes and interference frequency shifts in the next cycle. This closed-loop control mechanism empowers the system with self-learning, self-adaptation, and self-correction capabilities, enabling it to maintain high sensitivity and accuracy in identifying real electrical fire hazards even in the face of long-term operation, complex operating conditions, and multi-source interference. This is a core support component for ensuring the practicality, reliability, and engineering deployability of the monitoring system.

[0049] The above-mentioned electrical fire monitoring method based on automatic compensation of inherent residual current can significantly improve the monitoring system's recognition accuracy and alarm reliability for real leakage risks in high-order harmonic interference environments. This method accurately locates high-frequency interference signals by introducing multi-component time-frequency deconstruction, wavelet packet energy extraction, and spectral entropy mutation analysis. It effectively distinguishes and eliminates non-fault harmonic components by establishing an amplitude-frequency coupling behavior identification mechanism and a recursive interference elimination strategy. It further combines historical operating conditions with dynamic change trends to implement adaptive compensation and stability retrospective adjustment, and constructs closed-loop feedback control logic, enabling the system to continuously optimize judgment criteria and correct compensation strategies online. Overall, this solution solves the problem of misjudgment and underreporting caused by the "inability to distinguish between high-order harmonics and real leakage signals" in traditional systems, and achieves detailed modeling and intelligent response to abnormal current behavior in complex distribution scenarios, with a high degree of engineering adaptability and safety protection.

[0050] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. An electrical fire monitoring method based on automatic compensation of inherent residual current, characterized in that: The following steps are involved: S100 collects the residual current signal and performs multi-component time-frequency deconstruction processing. The energy distribution characteristics are extracted through wavelet packet transform, and the mutation index is extracted by combining short-time spectral entropy analysis. The preliminary distribution map of high-order harmonic interference is constructed to determine the boundary characteristics of the interference signal in the time and frequency domains. S200, based on the preliminary distribution map of high-order harmonic interference, performs amplitude-frequency coupling relationship identification, analyzes the phase offset and response delay between the power frequency and interference frequency bands, identifies the dynamic correlation between atypical leakage in the residual current and high-order harmonic interference, and determines the location of the interference frequency band and its change path; S300 performs recursive harmonic interference removal based on the interference frequency band location and change path, uses frequency-differential filtering to remove interference components, and retains low-frequency non-steady-state components through local waveform reconstruction to obtain a structurally purified residual current signal. S400 performs adaptive compensation control based on the residual current signal after structural purification. It combines historical operating conditions with the current rate of change to generate a compensation reference curve for the inherent residual current and dynamically adjusts the compensation amplitude. S500, based on the adaptive compensation control process, builds a stability index backtracking mechanism to track signal stability over multiple sampling periods, analyze response lag, amplitude offset, and error accumulation, and correct adaptive compensation control parameters; S600, based on the revised adaptive compensation control parameter system, performs joint closed-loop control, integrating time-frequency deconstruction, coupling identification, interference rejection, compensation control and backtracking mechanism to accurately identify and alarm the real leakage risk under high-order harmonic interference conditions.

2. The electrical fire monitoring method based on inherent residual current automatic compensation according to claim 1, characterized in that: Step S100 includes: The zero-sequence current signal in the three-phase distribution line is collected and filtered out by an anti-aliasing low-pass filter to remove high-frequency interference. The signal is then digitized by an analog-to-digital conversion chip at a sampling rate of more than 10,000 times per second. The digitized residual current signal is decomposed using a four-layer wavelet packet decomposition. The db6-order function in the Daubechies wavelet function is used to decompose the residual current signal, dividing the signal into 16 sub-bands and calculating the frequency domain energy in each sub-band. Based on the results of frequency band energy distribution, short-time spectral entropy analysis is introduced to process the signal by windowing it according to the power frequency period and sliding the window to extract the energy concentration and structural change characteristics of each frame signal. A comprehensive time-frequency two-dimensional disturbance map is constructed to identify the frequency boundaries, duration and disturbance intensity of high-frequency interference signals, and to complete the construction of a preliminary distribution map of high-order harmonic interference.

3. The electrical fire monitoring method based on inherent residual current automatic compensation according to claim 1, characterized in that: Step S200 includes: Extract frequency sub-bands whose energy density is higher than a preset threshold and whose spectral entropy value is lower than a preset threshold, and establish a pairing relationship between the working frequency band and the interference frequency band; Based on the Hilbert transform, the instantaneous phase offset between the corresponding frequency bands is calculated to determine their phase coupling characteristics; Taking the time position of the residual current mutation point as the anchor point, the response delay of the signal in each frequency band is calculated to determine its time domain response dependency; The above analysis results are combined to form a frequency band behavior identification result set, marking the coupling type and change path of the interference frequency band.

4. The electrical fire monitoring method based on inherent residual current automatic compensation according to claim 1, characterized in that: Step S300 includes: Construct a frequency template set of the interference frequency band to be removed, and extract the corresponding frequency band node coefficients to reconstruct the interference component signal; Dynamic time warping is used to perform frequency differential filtering to preliminarily remove interference frequency band signals; Set a sliding time window to perform local waveform reconstruction, retain low-frequency non-stationary components and perform amplitude correction; The elimination iteration threshold is set based on the high-frequency energy proportion and reconstruction error, and multiple rounds of interference elimination are recursively performed to output the residual current signal after structural purification.

5. The electrical fire monitoring method based on inherent residual current automatic compensation according to claim 1, characterized in that: Step S400 includes: Build an operating condition feature library containing residual current feature vectors and compensation reference waveforms under historical operating scenarios; Extract the residual current characteristics in the current cycle and calculate the Euclidean distance between it and the historical template, determine the optimal matching template and generate a dynamic compensation target curve; Compare the current signal with the compensation target curve point by point and perform amplitude compensation and delay adjustment operations within the set threshold range; The error indicators are counted over multiple cycles and the compensation reference curve is reconstructed based on the error feedback results to achieve closed-loop optimization.

6. The electrical fire monitoring method based on inherent residual current automatic compensation according to claim 1, characterized in that: Step S500 includes: Ten consecutive sampling periods are set as the analysis window, and the maximum amplitude, minimum amplitude, root mean square value, average rising slope and energy center frequency of each period are extracted to construct the period feature vector set; Based on the periodic characteristic vector set, the range, standard deviation and mean square deviation of each parameter are calculated, and the current window stability index is constructed according to the set weight combination; When the stability index is lower than the set threshold, the compensation output is compared with the compensation reference curve cycle by cycle, and the response delay point, amplitude offset and error accumulation value are extracted to identify the cause of instability; The compensation control parameters are adjusted according to the recognition results and the correction values ​​are enabled in the next cycle. At the same time, the correction information is recorded for subsequent tracking and analysis.

7. The electrical fire monitoring method based on inherent residual current automatic compensation according to claim 1, characterized in that: Step S600 includes: A unified sampling time base and data transmission link are set, and the residual current signal is decomposed by wavelet packets to extract the frequency domain energy matrix and disturbance spectrum as input; Based on the frequency domain spectrum, it performs amplitude-frequency coupling identification processing, extracts the phase offset, response delay, and energy index of the interference frequency band, drives three rounds of recursive interference removal processing, and outputs a purified signal; Load the current compensation parameter system based on the purified signal, generate the target compensation curve by combining the change rate and dynamic adjustment coefficient, adjust the signal point by point and perform amplitude control; At the end of the cycle, the compensation effect and stability index are analyzed. If the compensation accuracy decreases, the dynamic parameters and template matching are corrected, the parameter system is updated and loaded into the next cycle to complete the closed-loop control.

Images