Primary and secondary fusion complete ring main unit fault diagnosis method

Abstract

The invention discloses a primary and secondary fusion complete ring main unit fault diagnosis method, and particularly relates to the technical field of power distribution fault diagnosis, and the method comprises the steps: collecting multi-path original data, carrying out the frequency domain and time domain combined calibration, carrying out the comprehensive evaluation according to two preset discrimination factors, namely, a data stability deviation amplitude and a multi-path waveform time deviation degree, and obtaining a fault diagnosis result. Two types of feature data sets are constructed subsequently, input data of two interference modeling networks are calculated respectively, then the interference modeling networks are input to output interference grade values, sampling precision dynamic adjustment and fault recognition strategy switching operation are executed according to the interference influence grade values, and diagnosis accuracy and stability in a complex interference environment are improved. According to the method, unified normalization processing of multi-path sensing data is realized, and the sensing accuracy of fault features is improved; through double-factor triggering and feature fusion evaluation, the stability of interference identification is enhanced; and sampling adjustment and strategy switching are executed based on the interference level value, so that the robustness and reliability of diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution fault diagnosis, and more particularly to a method for diagnosing faults in a primary-secondary integrated ring network box. Background Art

[0002] As distribution automation systems continue to evolve toward intelligent, integrated, and edge-aware integration, the integrated primary and secondary ring network enclosure, a key carrier for the deep integration of sensing, protection, control, and communication functions, has become the core unit of the next-generation distribution terminal within the intelligent distribution network architecture. Its integrated structure enables the physical integration and functional coupling of power main equipment and information subsystems, providing a richer data foundation and execution capabilities for status awareness, fault prediction, and adaptive control of distribution equipment.

[0003] However, in the context of the full deployment of converged architectures, fault diagnosis technology generally remains limited to static identification of single-path features, failing to fully leverage the cross-information value of sensor data and its ability to identify interconnected characteristics. In multi-disturbance scenarios, complex signal behaviors such as amplitude mutations, phase shifts, harmonic superposition, and localized interference often occur simultaneously across sensor paths. Using traditional single-channel identification and fixed threshold triggering methods can easily lead to false alarms, missed alarms, or misidentified faults, severely restricting the evolution of intelligent diagnosis in ring network box equipment.

[0004] Especially in application scenarios such as secondary-side reset interference during operation, electrical disturbance chain conduction caused by load switching on one side, and synchronous response offset caused by local poor contact at the ring network interface, the perception data often exhibits interference characteristics with high correlation but low consistency, making it difficult for traditional discrimination logic to effectively distinguish between "structural co-interference behavior" and "isolated fault response". The diagnostic model frequently fluctuates on the discrimination boundary, resulting in insufficient algorithm stability, passive execution of the control strategy, and difficulty in achieving breakthroughs in recognition accuracy. Therefore, the present invention proposes a primary and secondary integrated ring network box fault diagnosis method to solve the above problems. Summary of the Invention

[0005] To achieve the above object, the present invention provides the following technical solutions: The method for diagnosing faults of a primary and secondary integrated ring network box includes the following steps: The system collects raw data generated by multiple sensing paths during the operation of distribution equipment. The raw data includes instantaneous current curves, high-order harmonic change sequences, partial discharge pulse signal sequences, and low-frequency ground potential disturbance trajectories. The data is then normalized and standardized using a joint frequency-domain and time-domain calibration algorithm. Comprehensive evaluation is performed based on two preset discrimination factors: data stability deviation amplitude and multipath waveform time offset. When both discrimination factors exceed the set trigger threshold range at the same time, the interference impact level assessment process is automatically initiated. Construct the first and second characteristic data sets. The first characteristic data set includes the transient energy concentration distribution, the waveform phase coupling strength between sensing paths, and the probability of overlapping events between abnormal segments. The second characteristic data set includes the signal residual compression rate, the spectrum phase continuity factor, and the median stability of the fluctuation segment. The multipath interference synergy evaluation value and the signal anti-interference structure robustness factor are calculated based on the first and second characteristic data groups, respectively. The two evaluation results are used as inputs and imported into a pre-built interference degree modeling network. The network outputs a comprehensive grade value representing the degree of interference impact, which is used to quantify the interference interference intensity on the fault diagnosis results. The following two linked control behaviors are performed based on the interference impact level value: First, the signal sampling accuracy of the current perception path is automatically adjusted. When the impact level exceeds the intervention threshold, the signal sensitivity is reduced or the high-noise segment data is frozen. Second, the fault identification strategy is dynamically switched. When the impact level exceeds the judgment threshold, the fault-tolerant judgment mechanism is activated, and the accuracy and reliability of fault judgment are enhanced through multi-channel redundancy verification and response delay correction processes.

[0006] In a preferred embodiment, the process of collecting raw data generated by multiple sensing paths during the operation of the power distribution equipment includes the following operations: By integrating a current acquisition unit for sensing instantaneous load changes, a harmonic identification unit for extracting frequency domain features, a pulse perception unit for capturing dielectric local breakdown signals, and a low-frequency measurement unit for detecting ground potential perturbation characteristics in each sensing path, four types of raw data are collected respectively: instantaneous current curve, high-order harmonic change sequence, partial discharge pulse signal sequence, and low-frequency ground potential disturbance trajectory; the four types of raw data are synchronously input into the joint frequency domain and time domain calibration algorithm for normalization standard processing.

[0007] In a preferred embodiment, the joint frequency domain and time domain calibration algorithm includes the following processing steps: The frequency components of various raw data are calculated through spectrum transformation, and the frequency domain energy concentration position of each sensing path is extracted as the frequency reference basis; Extract the rising edge starting point, peak position and duration of the original data in the time series to build a time baseline and achieve time alignment across perception paths; Based on the frequency reference basis and time baseline, the amplitude, phase and sampling points of the original data are jointly transformed, so that the data output by all perception paths have consistent amplitude scale, time structure and phase direction in structure, forming a unified data base that meets the needs of subsequent discrimination and feature extraction.

[0008] In a preferred embodiment, the process of performing comprehensive evaluation based on two preset discriminant factors, namely, the data stability deviation amplitude and the multipath waveform time offset degree, includes the following operations: For each sensing path output data processed by the joint frequency domain and time domain calibration algorithm, a sliding time window with a fixed sampling length is set. The real-time fluctuation root mean square value of the sensing path within each window is calculated. The difference between the real-time fluctuation root mean square value and the baseline root mean square value within the corresponding window under the historical stable operation state is calculated. The absolute value of the difference is used as the data stability deviation amplitude of the current sensing path. Perform waveform first-order derivative analysis on the data of all sensing paths within the same sliding time window, identify the first position in the waveform where the rate of change of the first-order derivative exceeds the set mutation threshold as the main response inflection point, record the absolute timestamp of the inflection point on the time axis, and perform pairwise difference calculation on the main response inflection point timestamps corresponding to all sensing paths to extract their maximum time difference; The maximum time difference is calculated by comparing it with the synchronization response reference time interval set during the equipment operation phase. The result is used as the multipath waveform time offset degree. This offset degree is used to represent the degree of synchronization response deviation between the sensing paths in the current time window. When the calculated results of the data stability deviation amplitude and the multipath waveform time offset degree simultaneously exceed their respective preset interference trigger thresholds, the interference impact level assessment process is automatically triggered, and the subsequent feature data group construction and impact level judgment operations are entered.

[0009] In a preferred embodiment, the process of calculating the multipath interference cooperation evaluation value based on the first type of feature data group includes the following steps: Within the set time window, the following three indicators are extracted for each perception path: Transient energy concentration distribution: It is defined as the minimum continuous time period length corresponding to the energy accumulation exceeding 80% of the total energy in the instantaneous current curve of the sensing path, recorded as ; Waveform phase coupling strength: defined as the maximum value of the mutual correlation coefficient of the main frequency band signals of the high-order harmonics between the perception path i and the perception path j in the frequency domain, denoted as , whose value range is -1 to 1; The probability of an abnormal segment overlap event is defined as the ratio of the number of times the abnormal wave packets detected in the partial discharge pulse signal sequence in the sensing path i and the sensing path j overlap on the time axis to the total number of abnormal wave packets, denoted as , ranging from 0 to 1; The following interference cooperation function is constructed to characterize the overall cooperative disturbance degree of the sensing path group: ; N is the total number of sensing paths, α, β, and γ are all preset positive real coefficients, which control the degree of effect of energy distribution difference suppression term, phase coupling enhancement term, and abnormal overlap weight term respectively. is the multipath interference cooperation evaluation value.

[0010] In a preferred embodiment, the process of calculating the signal interference rejection structure robustness factor based on the second type of characteristic data group includes the following steps: Within the set time window, three indicators are extracted from the second type of feature data group: The signal residual compression ratio, denoted as R, is defined as the ratio of the average residual value obtained after the original signal of the perception path is compressed by autoencoding to the standard deviation of the original signal amplitude; The spectral phase continuity factor, denoted as P, is defined as the average phase difference variance between adjacent frequency points in the main frequency band of the signal, which is normalized after taking the inverse; The stability of the median of the fluctuation segment, denoted as M, is defined as the inverse of the standard deviation of the median of each segment after the signal is divided into sliding segments of equal length, which is used as a local stability indicator; Introducing the worst disturbance tolerance modeling structure, construct the following combination function: ; is the signal anti-interference structure robustness factor, and λ is a preset adjustment parameter with a value range of 0 to 1, which is used to balance the relationship between “extremely weak indicator suppression” and “cooperative optimization gain”.

[0011] In a preferred embodiment, the multipath interference synergy evaluation value and the signal anti-interference structure robustness factor are input as input to the interference degree modeling network. The interference degree modeling network is a prediction model constructed using a gradient boosting regression tree structure. The prediction model is constructed through offline training based on a historical fault perception data set. During the training process, the diagnostic error rate under interference conditions is used as the supervision target, and the optimization loss function is the square error function between the predicted value and the labeled interference level. During the training phase, the interference degree modeling network uses pairs of measured samples of multipath interference synergy evaluation values and signal anti-interference structure robustness factors as feature vectors, and interference level values based on expert evaluation or historical misjudgment rate normalization results as training labels. The network then fits the regression tree structure through feature splitting and residual iteration. In the inference stage, the interference degree modeling network accepts the multipath interference synergy evaluation value and signal anti-interference structure robustness factor calculated in real time as input, and outputs a comprehensive grade value used to represent the intensity of interference intervention on the fault diagnosis result. The comprehensive grade value is a quantitative output in a real value domain, which serves as the basis for judging subsequent linkage control behavior.

[0012] In a preferred embodiment, performing the following two linked control actions according to the interference impact level value refers to: The first control behavior is to automatically adjust the signal sampling accuracy of the current perception path, including the following operations: when the interference impact level value is lower than the sensitivity threshold, the original sampling accuracy is maintained unchanged; when the interference impact level value is between the sensitivity threshold and the intervention threshold, the signal sampling accuracy is reduced by one level to reduce the impact of high-frequency noise; when the interference impact level value exceeds the intervention threshold, the data segment of the current perception path within the evaluation window is frozen, and the data segment does not participate in the subsequent diagnostic reasoning process during the freezing period. The second control behavior is the dynamic switching of fault identification strategies, which includes the following operations: when the interference impact level value exceeds the strategy switching threshold, the original single-path anomaly judgment method is switched to a fault-tolerant diagnosis mode based on a multi-path redundancy verification mechanism. The fault-tolerant diagnosis mode analyzes the consistency and confidence strength of the diagnosis results between multiple perception paths, uses a weighted voting mechanism to determine the final diagnosis conclusion, and delays the diagnosis output to improve fault tolerance robustness.

[0013] Technical effects and advantages of the present invention: By constructing multiple sensing paths and jointly collecting instantaneous current curves, high-order harmonic change sequences, partial discharge pulse signal sequences, and low-frequency ground potential disturbance trajectories, this invention achieves multi-dimensional information coverage and physical scene fusion of the operating status of the integrated primary and secondary ring network box, effectively improving the perceptibility and timing alignment capabilities of multiple types of fault characteristics. Based on a joint frequency domain and time domain calibration algorithm, the raw data is uniformly processed to ensure the consistency of the data output by different types of sensing paths in terms of amplitude scale, phase structure, and sampling time axis. This establishes a unified data base for subsequent interference discrimination and feature extraction, significantly reducing the risk of diagnostic errors caused by sampling asynchrony or signal structure imbalance, and enhancing the data structure's ability to characterize changes in electrical disturbances.

[0014] The present invention sets up a dual discriminant factor based on the amplitude of data stability deviation and the degree of multi-path waveform time offset. When the perception path fluctuates or the response is inconsistent, it can trigger the interference level assessment process, thereby improving the recognition sensitivity of non-structural interference or common-mode disturbance. By introducing the first and second types of feature data groups, the inter-path synergy and single-path robustness are quantified respectively, which can effectively take into account the local structural stability and the overall coordinated disturbance behavior. Then, through the interference degree modeling network, deep level modeling is carried out to output the comprehensive level value, realizing the transformation of interference identification from rule judgment to multi-feature fusion quantitative output, and enhancing the adaptability and diagnostic credibility of the method in complex operation scenarios.

[0015] The present invention sets multi-level control behavior thresholds based on the interference impact level value, and can dynamically adjust the sampling accuracy of the perception path and switch the diagnostic strategy, effectively improving the robustness of fault identification and the flexibility of the response strategy during the operation of the ring network box. When the interference level is at a mild stage, noise suppression can be performed by reducing the sampling frequency or signal smoothing. When the interference level is severe, the path data is frozen to prevent distortion in the diagnosis and judgment. At the diagnostic level, by switching to a fault-tolerant diagnostic mode with a multi-path redundant verification mechanism, a diagnostic voting and delayed output strategy is introduced to reduce the risk of misjudgment due to individual path inaccuracies, thereby achieving closed-loop control of the ring network box fault identification link under interference background, ensuring stable fault location capabilities in high-interference scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of the fault diagnosis method for the primary and secondary integrated ring network box in the present invention. DETAILED DESCRIPTION

[0017] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] Reference Figure 1 The following examples were obtained: Example 1: The present invention proposes a method for diagnosing faults of a primary-secondary integrated ring network box, comprising the following steps: The raw data generated by multiple sensing paths during the operation of distribution equipment is collected and normalized to standardize the data. To fully perceive the electrical and structural status of distribution equipment during operation, a sensing path system covering different physical quantities, including current, voltage, partial discharge, and ground potential, is required. By collecting instantaneous current curves, high-order harmonic change sequences, partial discharge pulse signal sequences, and low-frequency ground potential disturbance trajectories, a multi-source raw data foundation is formed. Because various signals have different sampling methods, amplitude ranges, and time bases, it is necessary to use a joint frequency domain and time domain calibration algorithm to uniformly process the amplitude, phase, and timing of all path data to form a normalized data base with consistent structure and strong comparability, ensuring the accuracy and coordination of subsequent analysis.

[0019] A comprehensive evaluation is performed based on two preset discriminant factors: the data stability deviation amplitude and the multi-path waveform time offset degree, to determine whether to start the interference impact level assessment process. During the operation of the power distribution system, non-fault abnormal data fluctuations may occur due to factors such as the environment, electromagnetic interference, and equipment aging. In order to avoid misjudgment, two discriminant factors are designed as trigger conditions in this step. Among them, the data stability deviation amplitude is used to reflect whether a single path signal has experienced drastic fluctuations, and the waveform time offset degree is used to determine whether there is a trend of time response asynchrony between multiple paths. When these two factors exceed the set threshold at the same time, it indicates that there may be cross-path common-mode interference or complex disturbance phenomena. At this time, starting the interference impact level assessment process will help to include "atypical interference background" in the fault diagnosis considerations and improve the robustness of the system.

[0020] The first and second types of feature data sets are constructed to characterize multi-path collaborative disturbances and single-path signal stability. Interference may manifest in two forms: "overall coordination" or "local sensitivity." Therefore, two types of feature data sets are constructed to comprehensively describe these two phenomena. The first type of feature data set is used to measure whether multiple paths show consistency in energy, phase, and anomaly distribution under the same disturbance event, thereby determining whether it is common-mode interference. The second type of feature data set focuses on the anti-interference stability of a certain perception path signal itself, including structural compressibility, spectral phase smoothness, and local median fluctuation, to assess whether it is susceptible to interference. This dual-channel feature extraction design enables the system to grasp the overall trend and locate local vulnerable links, laying a stable input structure for subsequent modeling.

[0021] Two evaluation results are calculated based on the feature data set and fed into the interference degree modeling network, which outputs a comprehensive grade. A prediction model is then constructed based on the two input feature evaluation values, outputting a quantitative result of the interference impact. This modeling process not only integrates the collaborative behavior of the entire path with the interference resistance of a single path, but also forms a prediction system highly correlated with the actual risk of misjudgment based on historical data training. The output value of the interference degree modeling network represents the risk level of the current perception path within the context of the overall system interference. Through this mechanism, complex multi-source features can be converted into grade values with practical reference value, providing a clear control basis for subsequent execution behavior.

[0022] Based on the interference impact level value, coordinated control behavior is executed to achieve local adjustment and diagnostic strategy switching. After the system recognizes that the interference level has reached the intervention threshold, it no longer relies directly on raw data for diagnosis. Instead, it first automatically adjusts the sampling accuracy of the current path, such as reducing sensitivity or temporarily freezing high-noise segments, to reduce the interference of interference data on system decision-making. When the interference level further rises to the judgment threshold, it indicates that the system is in a serious unstable state. At this time, it switches to fault-tolerant diagnostic mode, and through multi-path redundancy verification and delay judgment mechanisms, the stability and reliability of the overall diagnostic results are improved. This behavioral design highly couples the perception link with the diagnostic strategy to ensure that the system can maintain the continuity and accuracy of fault identification in high-interference environments.

[0023] In the process of collecting raw data generated by multiple sensing paths during the operation of the power distribution equipment, the collection process includes the following specific operation steps: first, four functional structures are set in each sensing path, respectively used to collect different types of physical changes in the power distribution operation state. Specifically: 1. Establish a current acquisition unit to sense current fluctuations in the conductor caused by load changes. This unit can be implemented using a Hall effect current sensor or a micro-transformer. A high-speed sampling circuit records the instantaneous amplitude of the current signal per unit time, forming an instantaneous current curve that reflects the dynamic behavior of the distribution circuit during load changes and switching operations.

[0024] Second, establish a harmonic identification unit, which extracts the frequency domain characteristics of the voltage or current signals in the line. This unit can use a spectrum acquisition module based on fast Fourier transform or a bandpass filter array. Its output is the amplitude and phase information of each harmonic component. Through continuous sampling over time, a high-order harmonic variation sequence is generated, which is used to identify harmonic disturbance characteristics related to loose metal structures and nonlinear effects at connection points.

[0025] Third, establish a pulse sensing unit to detect transient breakdown in insulation structures caused by dielectric aging or damage. This unit can be used with an ultra-high-frequency discharge sensor or a high-sensitivity impulse voltage detector to perform time-domain compression and peak identification on the captured high-frequency discharge pulses, generating a partial discharge pulse signal sequence to characterize potential discharge defects, particularly micro-breakdown caused by insulation damage at the fastening points of stainless steel cable ties.

[0026] 4. Set up a low-frequency measurement unit, which is used to detect micro-disturbances in the grounding loop during operation. This is usually achieved by installing a low-frequency vibration acquisition probe at the grounding point of the equipment. The slow fluctuations in the ground potential with a frequency below 50 Hz are sampled and the baseline is adjusted for a long time, eventually forming a low-frequency ground potential disturbance trajectory. This information can reveal the slow stability imbalance trend caused by changes in external grounding conditions or loose grounding structure.

[0027] After completing the acquisition of the above four types of raw data, the instantaneous current curve, high-order harmonic change sequence, partial discharge pulse signal sequence and low-frequency ground potential disturbance trajectory are synchronously input into the subsequent joint frequency domain and time domain calibration algorithm for unified processing. The calibration algorithm is used to normalize the original signals of different physical quantities in terms of amplitude scale, phase structure and sampling time axis, ensuring that the data obtained by multiple sensing paths under different channels have a unified structural basis, equal reference standards and time alignment capabilities, and provide strict consistency data input conditions for subsequent data stability judgment, feature construction and interference level modeling. For example, in actual deployment, the four units can be respectively set at the cable joint area, busbar terminal board, tie wrapping area and grounding lead-out at the outlet end of the ring network box to realize distributed perception and centralized data collection, forming a complete structural status perception chain.

[0028] The joint frequency domain and time domain calibration algorithm includes the following processing steps: The frequency components of various types of raw data are calculated through spectrum transformation, and the frequency domain energy concentration position of each sensing path is extracted as the frequency reference basis. Among them, "various types of raw data" refers to the four types of signals collected in multiple sensing paths, including instantaneous current curves, high-order harmonic change sequences, local discharge pulse signal sequences, and low-frequency ground potential disturbance trajectories. In order to achieve normalization of the signal in the frequency structure, a spectrum transformation operation is performed on each type of raw data. The "spectrum transformation" refers to a processing method that maps continuous or discrete signals in the time domain to frequency domain signals, and is often calculated using fast Fourier transform or short-time Fourier transform. After each sensing path completes the spectrum transformation, its main energy concentration position, that is, the frequency band with the highest energy distribution on the frequency axis, is extracted according to the spectrum energy density function curve as the "frequency domain energy concentration position" of the path. This position is the "frequency reference basis" in the subsequent normalization process, indicating the main frequency response center of the path in the current time window.

[0029] Extract the rising edge starting point, peak position and duration of the original data in the time series to construct a time baseline and achieve time alignment across perception paths. "Time series" refers to the data arrangement structure of the signal sampled in the perception path that changes over time. In order to ensure the temporal consistency of signals of different paths at key event points, it is necessary to extract several time feature points from each time series, including: "rising edge starting point": refers to the starting position where the signal amplitude changes from a stable state to a sudden change state, usually identified by judging that its first-order derivative exceeds the set threshold; "peak position": the time coordinate corresponding to the maximum absolute amplitude point of the signal in the current time window, usually representing the outbreak moment of the abnormal event; "duration": defined as the complete event time span experienced by the signal from the rising edge starting point to the fall back to a stable level, used to evaluate the duration of the event.

[0030] Using the three aforementioned time parameters as the core, a "time baseline" is constructed for each perception path. This baseline reflects the temporal characteristics of key response events along that path. By performing operations such as translation, stretching, and interpolation on the time baselines of multiple perception paths, "time alignment across perception paths" is achieved. This synchronization of key time points across multiple paths onto a unified time axis ensures structural temporal consistency. For example, if multiple perception paths all sense a circuit breaker trip or abnormal cable connection, but the response peak position is offset due to sensor delays or different sampling frequencies, the aforementioned time baseline construction and alignment can achieve precise synchronization and avoid misjudgments.

[0031] Based on the frequency reference basis and time baseline, the amplitude, phase and sampling points of the original data are jointly transformed, so that the data output by all sensing paths has a consistent amplitude scale, time structure and phase direction in structure, forming a unified data base that meets the needs of subsequent discrimination and feature extraction. "Joint transformation" means that in the same processing flow, based on the two unified standards of "frequency reference basis" and "time baseline", the three key structural attributes of the original data are normalized and corrected: "Amplitude" normalization: standardizing the maximum amplitude or root mean square value of the original signal to fall into a unified numerical range, which is often used to eliminate the sensitivity differences between different sensors; "Phase" adjustment: referencing the phase offset of each path signal in the main frequency band to a unified phase starting point to ensure consistent frequency response structure; "Sampling point" reconstruction: unifying the number of samples and sampling time points of different paths in the same time window through interpolation, resampling, etc., to facilitate subsequent window-based synchronous processing.

[0032] After the joint transformation of the above three aspects is completed, a unified data output structure is formed, which is called the "unified data base." This base refers to the standardized data set output by all sensing paths after processing. It has structural consistency, time alignment, and physical reference. It is a prerequisite for subsequent data stability analysis, interference trigger judgment, feature construction, and modeling prediction. For example, if there is an instantaneous voltage fluctuation within a certain time window, but this fluctuation is weakened or delayed by some paths, the unified data base can use a multi-dimensional correction mechanism of time and frequency to synchronously amplify the anomaly to the standard representation of all paths, realizing collaborative discrimination.

[0033] The comprehensive evaluation process based on two preset discriminant factors, data stability deviation amplitude and multipath waveform time offset, includes the following operations: For each sensing path's output data processed by the joint frequency-domain and time-domain calibration algorithm, a sliding time window with a fixed sampling length is set. The real-time RMS fluctuation of the sensing path within each window is calculated and then the difference is calculated with the baseline RMS value within the corresponding window under historical stable operation. The absolute value of this difference is used as the data stability deviation amplitude of the current sensing path. The "output data of each sensing path" refers to the standardized time series formed based on structural consistency after the instantaneous current curve, high-order harmonic change sequence, partial discharge pulse signal sequence, and low-frequency ground potential disturbance trajectory are processed by the joint frequency-domain and time-domain calibration algorithm. For this time series, the "sliding time window with a fixed sampling length" refers to a continuous analysis window of a fixed length (e.g., one second, fifty milliseconds) set on the time axis, used to analyze the signal change characteristics segment by segment.

[0034] Within each time window, the signal's "real-time RMS fluctuation value" is calculated. This value reflects the signal's energy intensity and degree of fluctuation within that time period. Subsequently, the corresponding "baseline RMS value" for the path during a historically stable operation within a time window of the same length is retrieved as a comparison value. The difference between the two values is calculated, and the absolute value is taken as the "data stability deviation amplitude," which measures whether the current path has experienced abnormal short-term disturbances. This method is capable of identifying small but consistent abnormal trends and has a strong response capability at the onset of disturbances.

[0035] First-order derivative analysis of the waveform data for all sensing paths within the same sliding time window is performed. The location where the first first-order derivative rate of change exceeds a set mutation threshold is identified as the primary response inflection point. The absolute timestamp of this inflection point on the time axis is recorded. Pairwise difference calculations are performed on the timestamps of the primary response inflection points corresponding to all sensing paths, extracting the maximum time difference. "First-order derivative analysis" refers to the mathematical modeling of the signal's rate of change over a time series to determine the mutation characteristics of its rising or falling speed. The "set mutation threshold" refers to the preset first-order derivative value boundary. When the signal's rate of change exceeds this threshold, a significant response is considered to have occurred at that point. The "primary response inflection point" is the first characteristic point within an analysis window that meets the mutation condition. It typically reflects the initial sensing behavior of an external disturbance or internal device fault. By recording the "absolute timestamp" of this point on the time axis, the initial response time of each sensing path in the interference event can be obtained. Afterwards, the timestamps recorded in all paths are combined in pairs for "difference calculation", that is, the time difference between the response start moments of any two paths is compared, and finally the largest time difference among all combinations is extracted, which represents the worst performance of the "response synchronization" of the perception path in this event.

[0036] The maximum time difference is calculated by comparing it with the synchronization response reference time interval set during the equipment operation phase. The result is used as the degree of multipath waveform time offset. This offset is used to characterize the degree of deviation in the synchronization response between sensing paths within the current time window. The "synchronization response reference time interval set during the equipment operation phase" refers to the maximum allowable response time difference of multipath signals in response to the same disturbance event when the equipment is in normal operation. For example, in stable power grid operation, the response error of similar sensing paths should not exceed five milliseconds. The "degree of multipath waveform time offset" can be obtained by comparing the actual maximum time difference with this reference value. The larger the ratio, the greater the difference in response between paths, reflecting the deterioration of the time alignment capability of the multipath signal, and the possible existence of severe interference or partial path data distortion.

[0037] When the calculated results of the data stability deviation amplitude and the multipath waveform time offset simultaneously exceed their respective preset interference trigger thresholds, the interference impact level assessment process is automatically triggered, and the subsequent feature data set construction and impact level determination process begins. The "interference trigger thresholds" set the critical boundaries for determining whether the signal stability deviation is severe and whether the response time difference is out of control, respectively. They are generally adjusted dynamically based on empirical data or training samples. When these two indicators "simultaneously" exceed the thresholds, it indicates not only local data fluctuations but also abnormal multipath response offsets, indicating a high risk of interference. At this point, the "automatic triggering of the interference impact level assessment process" means entering the subsequent construction process based on the first and second feature data sets, completing the calculation of the multipath interference coherence evaluation value and the signal interference rejection structure robustness factor, thereby providing reliable judgment for the final interference level output and control strategy. For example, if one sensing path detects continuous high-amplitude current disturbances, while other paths experience significant response delays, and this phenomenon is not caused by actual load changes, this method can identify it as possible external interference, triggering subsequent model intervention.

[0038] The process of calculating the multipath interference cooperation evaluation value based on the first type of feature data group includes the following steps: Within the set time window, the following three indicators are extracted for each perception path: Transient energy concentration distribution: It is defined as the minimum continuous time period length corresponding to the energy accumulation exceeding 80% of the total energy in the instantaneous current curve of the sensing path, recorded as ; This indicator is used to measure the temporal concentration of energy changes. The shorter the time period, the more concentrated the energy is, indicating that there may be sudden interference behavior.

[0039] Waveform phase coupling strength: defined as the maximum value of the mutual correlation coefficient of the main frequency band signals of the high-order harmonics between the perception path i and the perception path j in the frequency domain, denoted as , whose value range is -1 to 1; this indicator is used to characterize the phase synchronization of the two paths in the main frequency band. The closer the value is to 1, the more likely the two paths are subject to co-source interference or common-mode response.

[0040] The probability of an abnormal segment overlap event is defined as the ratio of the number of times the abnormal wave packets detected in the partial discharge pulse signal sequence in the sensing path i and the sensing path j overlap on the time axis to the total number of abnormal wave packets, denoted as , ranging from 0 to 1; this indicator is used to determine whether multiple paths perceive local abnormal events in the same time period and is used to identify the probabilistic characteristics of synchronous disturbances.

[0041] The following interference cooperation function is constructed to characterize the overall cooperative disturbance degree of the sensing path group: ; N is the total number of sensing paths, α, β, and γ are all preset positive real coefficients, which control the degree of effect of energy distribution difference suppression term, phase coupling enhancement term, and abnormal overlap weight term respectively. is the multipath interference cooperation evaluation value. This means that when the difference in energy distribution between perception path i and perception path j is greater, the exponential decay is stronger, suppressing the contribution of asynchronous behavior to synergy; It represents the weighted amplification of the phase coupling term. If the two paths have strong frequency domain similarity, the weight of this term is increased, which enhances the recognition of common-mode interference. It represents the weighted amplification mechanism when there is anomaly overlap, which enhances the contribution of simultaneous perception of abnormal events to synergy. This function calculates the multipath interference synergy evaluation value, which is a real number result. Its design logic is to fuse three non-redundant feature items (energy concentration, phase synchronization and abnormal overlap) through nonlinear combination to avoid the information masking problem caused by simple weighting, and introduce cross-path bidirectional combination to ensure that the interaction between all paths is taken into account. The three indicators are all derived from different physical signal levels, but they jointly reflect whether there is a unified disturbance structure between the paths. By exponentially suppressing the difference items, the deviation of a certain path is avoided from amplifying the overall interference judgment. The abnormal overlap weighted items are used to identify group disturbances rather than isolated noise, thereby improving the credibility of the judgment. For example, during the operation of a section of a distribution ring network, if three paths detect high-frequency partial discharge signals almost at the same time, and their frequency main energy distribution is close, the phase response is basically synchronized, and the energy distribution time period difference is extremely small, then the formula output value It will increase significantly, indicating that there is strong interference synergy among multiple paths, thus entering the subsequent modeling and control process.

[0042] The process of calculating the signal anti-interference structure robustness factor based on the second type of characteristic data group includes the following steps: Within the set time window, three indicators are extracted from the second type of feature data group: The signal residual compression rate, denoted as R, is defined as the ratio of the average residual value obtained after auto-encoding compression of the original signal of the perception path to the standard deviation of the original signal amplitude; this indicator is used to measure the integrity of the structural information retained by the signal data after compression and reconstruction. The smaller the residual and the higher the structural compression rate, the stronger the redundancy and consistency of the signal, and the more it can be correctly restored after interference, and has strong structural robustness.

[0043] The spectral phase continuity factor, denoted as P, is defined as the average phase difference variance between adjacent frequency points in the main frequency band of the signal, normalized after taking the inverse; this indicator reflects the smoothness of the signal's frequency domain structure. If the phase change in the main frequency region shows a continuous trend, it means that the signal has not been subjected to strong non-structural disturbances, has good phase robustness, and has a more stable spectral structure.

[0044] The stability of the median of the fluctuating segment, denoted as M, is defined as the inverse of the standard deviation of the median of each segment after the signal is divided into sliding segments of equal length, which is used as a local stability indicator. The degree of median change is evaluated by a sliding window method. If the median fluctuation is small, it means that the signal maintains a stable central trend in different sub-segments and has stronger anti-interference ability, which is suitable as a reference factor for structural robustness under low-frequency interference background.

[0045] Introducing the worst disturbance tolerance modeling structure, construct the following combination function: ; is the signal anti-interference structure robustness factor, and λ is a preset adjustment parameter with a value range of 0 to 1, which is used to balance the relationship between "extremely weak index suppression" and "cooperative optimization gain". Indicates that the weakest of the three indicators is selected as the main evaluation benchmark to reflect the "weakest indicator suppression" logic; is a weighted compensation term used to balance the joint optimization enhancement effect brought by other better indicators. The design of this formula follows the idea of combining the "short board dominant principle" and the "cooperative optimization principle" in structural robustness modeling: first, the minimum function is used to ensure that the overall robustness evaluation will not be masked when a certain structural indicator deteriorates significantly; secondly, when multiple indicators perform well, the square root product term is used to reflect the synergistic enhancement effect between multiple structural stabilities. Specifically, when there is a high compression residual (i.e., R is large), a sharp jump in the spectral phase (i.e., P is small), or an unstable median (i.e., M is small) in the signal structure, the first term directly lowers the overall value of the robustness factor H, promptly reflecting the potential interference risk; and when the signal performs well in both the frequency domain and the time domain, i.e., when P and M are large, the second term improves the final output result through the geometric mean term, thereby identifying stable structural signals with high confidence.

[0046] For example, during the operation of a ring network box, if the original signal output by a sensing path exhibits extremely low residual error after compression (indicating a stable signal structure), minimal phase variation in the main frequency band of the spectrum (indicating the absence of sudden interference), and a controllable median fluctuation, then all three indicators are excellent, and the output of function H is high, indicating that the current signal on this path has good structural robustness. Based on this, the interference level modeling network may determine that this path is less likely to be misidentified as a fault even in an overall interference environment, thereby improving diagnostic accuracy.

[0047] The multipath interference cooperation evaluation value and the signal interference rejection structural robustness factor are input into the interference degree modeling network. The multipath interference cooperation evaluation value is a composite indicator constructed based on the cooperative behavior of multiple perception paths in terms of energy distribution, phase coupling, and abnormal overlap events, which is used to reflect whether the interference is a holistic common-mode trend. The signal interference rejection structural robustness factor is constructed based on the compression residual, spectral phase smoothness, and sliding median volatility of the perception path itself, which is used to reflect the structural stability of a single path under interference. Together, they constitute a joint quantitative expression of the "global structural behavior" and "local signal robustness" of the same perception window.

[0048] The interference degree modeling network is a prediction model constructed using a gradient-boosted regression tree structure. This "gradient-boosted regression tree structure" is a nonlinear regression modeling method based on the principles of ensemble learning. Its basic concept is to iteratively construct multiple weak regression subtrees and gradually fit the residuals of the previous model, thereby improving overall prediction accuracy and generalization. This structure is suitable for modeling problems involving nonlinearities and interactions between features. It has the advantages of strong interpretability, high adjustability, and insensitivity to feature scale. It is suitable for modeling the complex relationship between multipath interference characteristics and single-path robustness in the present invention.

[0049] The prediction model is built using an offline training method based on a historical fault perception dataset. This dataset refers to multi-sensory path collaborative data collected and annotated during the long-term operation of the ring network box. It contains multiple rounds of actual operating conditions and manual verification results, and is used to simulate the robustness of the fault identification mechanism under interference scenarios. The offline training method involves fully training the model using a large amount of known data in a non-deployed state, ensuring sufficient learning capabilities upon deployment without the need for online retraining.

[0050] During training, the diagnostic error rate under interference conditions serves as the supervisory objective, and the optimized loss function is the squared error between the predicted value and the annotated interference level. In other words, the training objective is no longer to classify whether a fault exists, but to accurately fit the continuous output of the interference level, thereby reducing the risk of the model overestimating minor interference or underestimating major interference. The "squared error function" used is a typical regression loss function and helps control the bias-variance trade-off during model training.

[0051] During the training phase, the interference level modeling network uses pairs of measured samples of multipath interference coordination evaluation values and signal anti-interference structure robustness factors as feature vectors to form the input vector space. Interference level values, based on expert assessments or normalized historical error rates, are used as training labels. These training labels are real-valued and range from a continuous range of values within the set interference level domain. The regression tree structure is fitted through a feature splitting and residual iteration mechanism. The "feature splitting mechanism" partitions the sample space based on the feature values and constructs regression subtree nodes based on the principle of maximizing residual reduction. The "residual iteration mechanism" fits the prediction residuals of the previous model to each newly constructed subtree, improving prediction accuracy.

[0052] During the inference phase, the interference degree modeling network accepts the multipath interference synergy evaluation value and signal anti-interference structure robustness factor calculated in real time as input, and outputs a comprehensive level value that represents the intensity of interference interference on the fault diagnosis results. The "comprehensive level value" is a quantitative output in the real-value domain, representing the degree to which the target perception path is affected by the current overall interference structure within a certain analysis time window, and is used to dynamically adjust the perception strategy or diagnostic mode. This level value serves as the basis for subsequent linkage control behavior judgments, and can be used to trigger the first behavior (automatic adjustment of signal sampling accuracy) and the second behavior (diagnostic strategy switching). Its design purpose is to establish a quantitative basis for control triggering through data-driven interference perception results, and to achieve a unified integration of path-level response and strategy-level control.

[0053] For example, during a given sensing cycle, if a path exhibits high consistency with other paths in the first-category feature (a high interference cooperativity evaluation value), while also exhibiting high compression ratio but spectral phase discontinuity in the second-category feature (a low robustness factor), the interference modeling network will output a medium to high overall rating. If this rating exceeds a set threshold, the signal sampling accuracy of the current path will be reduced and a redundant verification mechanism will be activated to prevent false faults due to single-path instability.

[0054] The following two linkage control actions are executed based on the interference impact level value: The first control behavior is to automatically adjust the signal sampling accuracy of the current perception path, including the following operations: when the interference impact level value is lower than the sensitivity threshold, the original sampling accuracy is maintained unchanged; when the interference impact level value is between the sensitivity threshold and the intervention threshold, the signal sampling accuracy is reduced by one level to reduce the impact of high-frequency noise; when the interference impact level value exceeds the intervention threshold, the data segment of the current perception path within the evaluation window is frozen, and the data segment does not participate in the subsequent diagnostic reasoning process during the freezing period. The second control behavior is to dynamically switch the fault identification strategy, which includes the following operations: when the interference impact level value exceeds the strategy switching threshold, the original single-path anomaly judgment method is switched to a fault-tolerant diagnosis mode based on a multi-path redundancy verification mechanism. The fault-tolerant diagnosis mode analyzes the consistency and confidence level of the diagnosis results between multiple perception paths, uses a weighted voting mechanism to determine the final diagnosis conclusion, and delays the diagnosis output to improve fault tolerance and robustness. More specifically: The first control behavior is to automatically adjust the signal sampling accuracy of the current perception path, including the following operations: When the interference impact level falls below the sensitivity threshold, the original sampling accuracy is maintained. The "interference impact level" here refers to the real-valued output by the interference severity modeling network, reflecting the degree of impact of the current sensing path in the overall interference environment. The "sensitivity threshold" is a preset first-level response threshold used to determine whether interference suppression is necessary. If the current sensing path's level does not reach this threshold, it indicates that the interference fluctuation is within an acceptable range, signal quality is acceptable, and no sampling strategy adjustment is required.

[0055] When the interference impact level is between the sensitivity threshold and the intervention threshold, the signal sampling accuracy is reduced by one level to reduce the impact of high-frequency noise. "Reducing the sampling accuracy by one level" means that after the interference impact level exceeds the preset sensitivity threshold, the signal sampling parameters of the current sensing path are graded downward to reduce the impact of high-frequency noise interference on feature extraction and diagnostic judgment. This downward adjustment operation can be implemented based on one or more combinations of the following three methods, as follows: Sampling frequency reduction: The original sampling frequency of the perception path is lowered from the current level to the next set frequency, for example: from the original 10kHz to 5kHz, or from 20kHz to 10kHz. Lowering the sampling frequency by one level means that the signal acquisition accuracy in the high-frequency bandwidth is reduced, but it can effectively suppress the influence of high-frequency interference while retaining low-frequency trend information.

[0056] Averaging within a time window: While keeping the sampling frequency unchanged, use a sliding average or median filter to smooth the original signal: set the sliding window length to a fixed percentage of the current signal period (such as 5%, 10%), and replace the value of each time point with the average or median value within its sliding window. This processing method reduces the sensitivity of signal changes without reducing the number of sampling points, especially when high-frequency interference is strong, it improves the stability of feature extraction.

[0057] Wavelet threshold filtering suppression: Use a fixed wavelet basis function (such as Daubechies-4 or Symlets-6) to decompose the signal in the current time period and set the high-frequency coefficients of a predetermined number of layers to zero: For example, in the five-layer decomposition, the first three levels of coefficients are retained and the high-frequency part is ignored, so as to achieve a downward shift of the passband range and suppress sharp disturbances. This method is suitable for non-stationary signals and can weaken local pulse interference while retaining the main structural information.

[0058] When the interference impact level value exceeds the intervention threshold, the data segment of the current perception path within the evaluation window period is frozen, and the data segment does not participate in the subsequent diagnostic reasoning process during the freezing period. Here, the "intervention threshold" is the set second-level threshold. When the interference level exceeds this value, it indicates that the signal of the perception path has been severely interfered with, and continuing to participate in the diagnosis may introduce misjudgment. The "freeze operation" refers to marking the diagnostic data segment of the path as "temporarily untrustworthy" and removing it from the current reasoning cycle, so that the diagnostic model is not dragged by distorted data and the overall recognition stability is guaranteed. For example, in a cable joint partial discharge interference, a high-frequency channel receives a burst pulse sequence. If the level value continues to rise, the system first reduces its sampling accuracy to filter the high frequency, and then freezes the data window when the intervention value is triggered, thereby avoiding short-term misleading of the fault model.

[0059] The second control behavior is to dynamically switch the fault identification strategy, which includes the following operations: When the interference impact level exceeds the policy switching threshold, the original single-path anomaly determination method is switched to a fault-tolerant diagnostic mode based on a multi-path redundancy verification mechanism. The "policy switching threshold" here refers to the third type of preset threshold, used to determine whether the diagnostic policy framework needs to be upgraded. The "original single-path anomaly determination method" refers to the traditional diagnostic method, which makes single-point judgments based on threshold violations, waveform anomalies, spectral distortion, and other characteristics of an independent signal on a specific sensing path. This method responds quickly but is susceptible to isolated interference.

[0060] The fault-tolerant diagnostic mode based on a multi-path redundant verification mechanism introduces a cross-validation diagnostic process using data from multiple paths when a single path is no longer reliable. Its operating logic involves simultaneously invoking the current identification results of multiple sensing paths, performing consistency analysis and confidence estimation, and forming a multi-source determination matrix. The final diagnosis is determined through a weighted voting mechanism, with weights set based on path stability scores, historical accuracy, or real-time robustness factors. If the majority of paths agree, the system uses that result as the final decision. If the majority of paths disagree, diagnosis is deferred.

[0061] "Delayed diagnostic output" means postponing the release of the final diagnostic conclusion until the next round of evaluation cycle or event reconfirmation, to avoid premature release of low-credibility conclusions in a strong interference background. "Improved fault tolerance robustness" means that this mode can significantly reduce the false alarm rate and missed alarm rate, and enhance the diagnostic stability in non-ideal environments. For example, if only one path among multiple paths experiences high-frequency fluctuations, if it is a normal load change rather than a fault, the path may be misjudged as abnormal. After the fault tolerance mechanism is enabled, the system will find that other paths are diagnosed as normal, and after comprehensive voting, eliminate the abnormality, thereby improving the accuracy of diagnosis.

[0062] It should be noted that in this method, the multipath interference coordination evaluation value, a global indicator characterizing whether all sensing paths exhibit common-mode interference trends within the current time window, is calculated based on the first type of feature data set. The signal interference rejection structural robustness factor, a path-local indicator assessing whether a specific sensing path maintains structural stability under interference, is calculated based on the second type of feature data set. The interference degree modeling network uses these two heterogeneous metrics as combined inputs. During the training phase, it establishes an interactive relationship between the local robustness of the current path and the coordination of all paths, outputting a global interference impact level. This level is used to determine whether the system is currently in a disturbed environment and, combined with the path activity status, is mapped to a specific control target. Therefore, in the coordinated control behavior, although the interference judgment logic is based on the global multipath feature modeling, the sampling adjustment and discrimination switching behaviors can be triggered locally based on whether the current path participates in feature generation within the analysis window, realizing a coordinated control mechanism with global drive and local response.

[0063] The input-output granularity relationship of the interference degree modeling network is as follows: Input 1 (global feature): multi-path interference synergy evaluation value, which is a collaborative disturbance index from all perception paths and is a global scale feature; Input 2 (local feature): signal anti-interference structure robustness factor, each evaluation focuses on a current perception path and is a local path scale feature; Output result: interference impact level value, which indicates the intensity of interference impact on the "current perception path" under the overall interference background; Control behaviors (such as adjusting sampling and freezing fragments) all occur on a specific perception path, and the output interference level value is also the evaluation result of the path, so the logic of the first control behavior is completely closed: evaluate the current path → output its interference level → determine its sampling adjustment. At the same time, why the second behavior switches from the "single-path anomaly judgment method" to the "multi-path redundant verification mechanism" is a diagnostic logic strategy switching behavior, which does not conflict with the model input and output granularity.

[0064] Traditionally, single-path anomaly determination relies solely on the path's own fluctuations, deviations, and discharge indicators (e.g., excessive discharge or sudden current changes) to determine if a particular data path is faulty. This method identifies a path as faulty even if other paths are operating normally. While this method offers a fast response, it also suffers from susceptibility to false positives due to interference. A multi-path redundant verification mechanism eliminates the need to rely solely on abnormal behavior on a single path if the system determines the current interference level is excessive. Instead, it simultaneously analyzes all paths. If the diagnosis is consistent across multiple paths, the fault is confirmed. If only one path is abnormal and its noise immunity is low, it is suspected to be a false positive. The second action is essentially a switch in diagnostic logic, not a localized control of a specific path. It determines the "trustworthiness" of the current path behavior based on multipath historical characteristics. The first control action (automatic adjustment of signal sampling accuracy) and the second control action (diagnostic strategy switching) are not logically mutually exclusive. In practical applications, they are "used in conjunction," meaning they can be enabled simultaneously and work synergistically within the same evaluation cycle. The key difference lies in their different triggering thresholds, control targets, and mechanisms. For example, signal sampling accuracy adjustment, as a proactive interference suppression measure at the perception path level, is preferentially applied in moderate interference zones to reduce the immediate impact of local noise on the diagnostic system. Diagnostic strategy switching, on the other hand, is used in severe interference scenarios and restructures the diagnostic confidence logic by introducing a multipath redundancy mechanism to enhance the robustness of the results. In actual operation, the two control actions can be triggered simultaneously when conditions permit linkage, improving the accuracy and fault tolerance of fault diagnosis in integrated primary and secondary ring network boxes from the perspectives of data input quality and decision output logic.

[0065] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0066] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0067] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0068] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0069] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for diagnosing faults in a ring network box with primary and secondary integration, characterized in that: The following steps are involved: The system collects raw data generated by multiple sensing paths during the operation of distribution equipment. The raw data includes instantaneous current curves, high-order harmonic change sequences, partial discharge pulse signal sequences, and low-frequency ground potential disturbance trajectories. The data is then normalized and standardized using a joint frequency-domain and time-domain calibration algorithm. Comprehensive evaluation is performed based on two preset discrimination factors: data stability deviation amplitude and multipath waveform time offset. When both discrimination factors exceed the set trigger threshold range at the same time, the interference impact level assessment process is automatically initiated. Construct the first and second characteristic data sets. The first characteristic data set includes the transient energy concentration distribution, the waveform phase coupling strength between sensing paths, and the probability of overlapping events between abnormal segments. The second characteristic data set includes the signal residual compression rate, the spectrum phase continuity factor, and the median stability of the fluctuation segment. The multipath interference synergy evaluation value and the signal anti-interference structure robustness factor are calculated based on the first and second characteristic data groups, respectively. The two evaluation results are used as inputs and imported into a pre-built interference degree modeling network. The network outputs a comprehensive grade value representing the degree of interference impact, which is used to quantify the interference interference intensity on the fault diagnosis results. The following two linkage control behaviors are performed based on the interference impact level value: first, the signal sampling accuracy of the current sensing path is automatically adjusted; second, the fault identification strategy is dynamically switched.

2. The method for diagnosing faults of a primary and secondary fusion ring network box according to claim 1 is characterized in that: The process of collecting raw data generated by multiple sensing paths during the operation of power distribution equipment includes the following operations: By integrating a current acquisition unit for sensing instantaneous load changes, a harmonic identification unit for extracting frequency domain features, a pulse perception unit for capturing dielectric local breakdown signals, and a low-frequency measurement unit for detecting ground potential perturbation characteristics in each sensing path, four types of raw data are collected respectively: instantaneous current curve, high-order harmonic change sequence, partial discharge pulse signal sequence, and low-frequency ground potential disturbance trajectory; the four types of raw data are synchronously input into the joint frequency domain and time domain calibration algorithm for normalization standard processing.

3. The method for diagnosing faults of a primary and secondary fusion ring network box according to claim 2 is characterized in that: The joint frequency domain and time domain calibration algorithm includes the following processing steps: The frequency components of various raw data are calculated through spectrum transformation, and the frequency domain energy concentration position of each sensing path is extracted as the frequency reference basis; Extract the rising edge starting point, peak position and duration of the original data in the time series to build a time baseline and achieve time alignment across perception paths; Based on the frequency reference basis and time baseline, the amplitude, phase and sampling points of the original data are jointly transformed, so that the data output by all perception paths have consistent amplitude scale, time structure and phase direction in structure, forming a unified data base that meets the needs of subsequent discrimination and feature extraction.

4. The method for diagnosing faults of a primary and secondary fusion ring network box according to claim 3 is characterized in that: The comprehensive evaluation process based on two preset discriminant factors, data stability deviation amplitude and multipath waveform time offset, includes the following operations: For each sensing path output data processed by the joint frequency domain and time domain calibration algorithm, a sliding time window with a fixed sampling length is set. The real-time fluctuation root mean square value of the sensing path within each window is calculated. The difference between the real-time fluctuation root mean square value and the baseline root mean square value within the corresponding window under the historical stable operation state is calculated. The absolute value of the difference is used as the data stability deviation amplitude of the current sensing path. Perform waveform first-order derivative analysis on the data of all sensing paths within the same sliding time window, identify the first position in the waveform where the rate of change of the first-order derivative exceeds the set mutation threshold as the main response inflection point, record the absolute timestamp of the inflection point on the time axis, and perform pairwise difference calculation on the main response inflection point timestamps corresponding to all sensing paths to extract their maximum time difference; The maximum time difference is calculated by comparing it with the synchronization response reference time interval set during the equipment operation phase. The result is used as the multipath waveform time offset degree. This offset degree is used to represent the degree of synchronization response deviation between the sensing paths in the current time window. When the calculated results of the data stability deviation amplitude and the multipath waveform time offset degree simultaneously exceed their respective preset interference trigger thresholds, the interference impact level assessment process is automatically triggered, and the subsequent feature data group construction and impact level judgment operations are entered.

5. The method for diagnosing faults of a primary and secondary fused ring network box according to claim 4 is characterized in that: The process of calculating the multipath interference cooperation evaluation value based on the first type of feature data group includes the following steps: Within the set time window, the following three indicators are extracted for each perception path: Transient energy concentration distribution: It is defined as the minimum continuous time period length corresponding to the energy accumulation exceeding 80% of the total energy in the instantaneous current curve of the sensing path, recorded as ; Waveform phase coupling strength: defined as the maximum value of the mutual correlation coefficient of the main frequency band signals of the high-order harmonics between the perception path i and the perception path j in the frequency domain, denoted as , whose value range is -1 to 1; The probability of an abnormal segment overlap event is defined as the ratio of the number of times the abnormal wave packets detected in the partial discharge pulse signal sequence in the sensing path i and the sensing path j overlap on the time axis to the total number of abnormal wave packets, denoted as , ranging from 0 to 1; The following interference cooperation function is constructed to characterize the overall cooperative disturbance degree of the sensing path group: ； N is the total number of sensing paths, α, β, and γ are all preset positive real coefficients, which control the degree of effect of energy distribution difference suppression term, phase coupling enhancement term, and abnormal overlap weight term respectively. is the multipath interference cooperation evaluation value.

6. The method for diagnosing faults of a primary and secondary fused ring network box according to claim 5 is characterized in that: The process of calculating the signal anti-interference structure robustness factor based on the second type of characteristic data group includes the following steps: Within the set time window, three indicators are extracted from the second type of feature data group: The signal residual compression ratio, denoted as R, is defined as the ratio of the average residual value obtained after the original signal of the perception path is compressed by autoencoding to the standard deviation of the original signal amplitude; The spectral phase continuity factor, denoted as P, is defined as the average phase difference variance between adjacent frequency points in the main frequency band of the signal, which is normalized after taking the inverse; The stability of the median of the fluctuation segment, denoted as M, is defined as the inverse of the standard deviation of the median of each segment after the signal is divided into sliding segments of equal length, which is used as a local stability indicator; Introducing the worst disturbance tolerance modeling structure, construct the following combination function: ； is the signal anti-interference structure robustness factor, and λ is a preset adjustment parameter with a value range of 0 to 1, which is used to balance the relationship between "extremely weak indicator suppression" and "cooperative optimization gain".

7. The method for diagnosing faults of a primary and secondary fused ring network box according to claim 6 is characterized in that: The multipath interference synergy evaluation value and the signal anti-interference structure robustness factor are input into the interference degree modeling network. The interference degree modeling network is a prediction model constructed using a gradient boosting regression tree structure. This prediction model is constructed using an offline training method based on a historical fault perception dataset. During the training process, the diagnostic error rate under interference conditions is used as the supervision objective, and the optimized loss function is the squared error function between the predicted value and the labeled interference level. During the training phase, the interference degree modeling network uses pairs of measured samples of multipath interference synergy evaluation values and signal anti-interference structure robustness factors as feature vectors, and interference level values based on expert evaluation or historical misjudgment rate normalization results as training labels. The network then fits the regression tree structure through feature splitting and residual iteration. In the inference stage, the interference degree modeling network accepts the multipath interference synergy evaluation value and signal anti-interference structure robustness factor calculated in real time as input, and outputs a comprehensive grade value used to represent the intensity of interference intervention on the fault diagnosis result. The comprehensive grade value is a quantitative output in a real value domain, which serves as the basis for judging subsequent linkage control behavior.

8. The method for diagnosing faults of a primary and secondary fused ring network box according to claim 7 is characterized in that: The following two linkage control actions are executed based on the interference impact level value: The first control behavior is to automatically adjust the signal sampling accuracy of the current sensing path, including the following operations: when the interference impact level value is lower than the sensitivity threshold, the original sampling accuracy is maintained unchanged; When the interference impact level is between the sensitivity threshold and the intervention threshold, the signal sampling accuracy is reduced by one level to reduce the impact of high-frequency noise. When the interference impact level exceeds the intervention threshold, the data segment of the current sensing path within the evaluation window is frozen, and the data segment does not participate in the subsequent diagnostic reasoning process during the freezing period. The second control behavior is the dynamic switching of fault identification strategies, which includes the following operations: when the interference impact level value exceeds the strategy switching threshold, the original single-path anomaly judgment method is switched to a fault-tolerant diagnosis mode based on a multi-path redundancy verification mechanism. The fault-tolerant diagnosis mode analyzes the consistency and confidence strength of the diagnosis results between multiple perception paths, uses a weighted voting mechanism to determine the final diagnosis conclusion, and delays the diagnosis output to improve fault tolerance robustness.

Images