Multi-dimensional electrical equipment insulation analysis and evaluation method

Abstract

The invention provides a multi-dimensional electrical equipment insulation analysis and evaluation method, and relates to the technical field of electrical equipment insulation online detection and intelligent diagnosis. According to the method, an electrically isolated direct current detection channel is constructed in a neutral point small resistance grounding system, controllable direct current detection signals are injected, micro direct current leakage response signals are collected, and after zero drift correction, temperature compensation and power frequency ripple suppression are carried out, layered attribution is carried out in combination with primary system topology and an equipment ledger, and a multi-dimensional response matrix is formed. According to the method, environment and working condition normalization is realized through factor rejection and robust pull-back, a mechanism model containing leakage channel equivalent parameters and path constraints is established, multi-working-condition data is fused to calculate equivalent insulation parameters and credibility indication quantity, and a multi-dimensional insulation representation vector is generated. And performing time sequence analysis on the representation vector, extracting degradation and abrupt change characteristics, outputting a health index and a risk level, and realizing quantitative evaluation and intelligent early warning of the insulation state.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of online detection and intelligent diagnosis of electrical equipment insulation, and particularly relates to a multi-dimensional electrical equipment insulation analysis and evaluation method. BACKGROUND

[0002] In power systems and large industrial plants, the insulation performance of electrical equipment is an important foundation for safe operation. At present, the neutral point small resistance grounding mode is widely used to balance the current control and protection sensitivity when single-phase grounding fault occurs. The detection of insulation state mainly relies on traditional test methods after power failure, such as megohmmeter measurement, dielectric loss test, etc., or indirect evaluation using operating period voltage and current data. However, these methods can only reflect the overall insulation level of the bus, and it is difficult to distinguish the aging state of different branches or specific equipment. At the same time, the on-site detection is often affected by environmental humidity, temperature and electromagnetic interference, etc., resulting in unstable measurement results and unable to continuously reflect the dynamic changes of equipment insulation performance.

[0003] With the promotion of smart grid and condition-based maintenance concept, the industry is gradually transforming from periodic maintenance to state-based predictive maintenance. Future insulation detection technology needs to realize online evaluation under the condition that the equipment does not stop operating, which not only guarantees electrical safety, but also has high sensitivity and high resolution capability. The research trend mainly focuses on two directions: one is to realize insulation state recognition under non-interrupted operation condition through safe decoupling structure and controllable signal injection; the other is to combine artificial intelligence with physical model, and use multi-dimensional data (such as environmental conditions, operating conditions, transient events, etc.) to build a unified evaluation system, so as to realize real-time quantification, trend prediction and early warning of insulation state.

[0004] III. Disadvantages of Prior Art The existing online monitoring schemes generally have the following problems: First, the detection signal is coupled with the system grounding and mutual inductance loop, resulting in large measurement error and easy false alarm; Second, microampere level leakage signal is difficult to accurately capture in complex noise environment, and the sensitivity and reliability are insufficient; Third, it is mostly targeted at the system as a whole, lacking resolution capability for branches and individual equipment; Fourth, there is a lack of adaptive processing for environmental and operating condition factors such as temperature and humidity, and load changes, and the evaluation results are easily disturbed by external interference; Fifth, the evaluation model lacks interpretability, and it is difficult to form a unified scale among different plants and stations, resulting in arbitrary threshold setting and difficulty in realizing transferable and traceable risk warning. SUMMARY

[0005] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide a multi-dimensional electrical equipment insulation analysis and evaluation method, by constructing decoupling detection channels, hierarchical topology mapping and mechanism-data fusion modeling, realizing high-precision quantitative evaluation and intelligent early warning of the insulation state of electrical equipment, and significantly improving the stability, positioning and predictability of insulation monitoring.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following scheme: A multi-dimensional electrical equipment insulation analysis and evaluation method, comprising: In a neutral point small resistance grounding system, a direct current detection channel electrically isolated is constructed, electromagnetic decoupling control is implemented on the neutral point and the voltage mutual inductance loop, a controllable direct current detection signal is injected under the condition of meeting insulation coordination and safety constraints, and a decoupled detection loop and a safety injection parameter set are obtained; A micro direct current leakage response signal corresponding to the controllable direct current detection signal is collected, zero drift correction, temperature compensation and power frequency ripple suppression are performed, a bidirectional comparison sequence based on a reference channel is constructed, and a leakage response reference sequence and a real-time sequence are obtained; According to the topology structure and equipment account of the primary system, the monitoring object is divided into a system layer, a branch layer and a device layer, the micro direct current leakage response signal is decomposed and attributed according to the topological reachable relationship, and a response matrix identified by dimension and a corresponding object mapping relationship are formed; Synchronous collection of meteorological parameters and operating condition data, factor weight elimination and robust pullback strategy are used to standardize and robustly process the response matrix, and event labels are generated based on start-stop, load fluctuation and grounding transient record, a multi-dimensional matrix normalized by environment and operating condition and the event label are obtained; Based on the corresponding relationship between the controllable direct current detection signal and the leakage response reference sequence and the real-time sequence, a mechanism model containing leakage channel equivalent parameters and path constraints is established, and a fusion modeling framework is constructed combined with multi-condition data, equivalent insulation parameters and credibility weighted indicators are calculated for the system layer, the branch layer and the device layer respectively, and a multi-dimensional insulation representation vector is obtained; The multi-dimensional insulation representation vector is subjected to time series analysis, slow degradation features and mutation features are extracted, health indexes and risk levels are generated according to a unified evaluation scale, suspicious object positioning and early warning output are realized combined with the topological reachable relationship and the event label, and insulation health evaluation results of the system layer, the branch layer and the device layer are obtained.

[0007] Preferably, in a neutral point small resistance grounding system, a direct current detection channel electrically isolated is constructed, electromagnetic decoupling control is implemented on the neutral point and the voltage mutual inductance loop, a controllable direct current detection signal is injected under the condition of meeting insulation coordination and safety constraints, and a decoupled detection loop and a safety injection parameter set are obtained, comprising: An isolation branch is arranged in parallel with a voltage mutual inductance loop on a neutral point path of the neutral point small resistance grounding system; the isolation branch is composed of a high impedance isolation unit and a controllable switch unit, and is used for maintaining electrical isolation in a power frequency state and forming the DC detection channel in a DC detection state; A detection injection node is established in the DC detection channel, and a controllable DC detection signal is output by a programmable voltage source, which is used for exciting the insulation loop without interrupting the system operation; An impedance matching and reverse isolation unit is arranged on the side of the voltage mutual inductance loop, so that the controllable DC detection signal and the power frequency signal are not coupled with each other, so as to realize electromagnetic decoupling of the DC detection channel and the operation channel, and obtain the decoupled detection loop; According to the insulation coordination condition and the grounding resistance parameter of the neutral point small resistance grounding system, the upper limit of the injection voltage and the current constraint threshold are dynamically limited, and the safe injection parameter set is obtained.

[0008] Preferably, a micro DC leakage response signal corresponding to the controllable DC detection signal is collected, zero drift correction, temperature compensation and power frequency ripple suppression are performed, a bidirectional comparison sequence based on a reference channel is constructed, and a leakage response reference sequence and a real-time sequence are obtained, including: In the DC detection channel of the neutral point small resistance grounding system, a leakage current signal synchronized with the controllable DC detection signal is collected in real time, and a reference channel signal is used as a synchronous reference to obtain an original leakage response data set; Zero drift correction is performed on the original leakage response data set, a baseline model is established according to a static calibration section, and baseline regression compensation is performed on the running period data, so as to eliminate the DC drift error of the detection link, and obtain a first processed data set; According to the temperature characteristic parameter of the neutral point small resistance grounding system, temperature compensation is performed on the first processed data set, and a temperature normalized data sequence is generated; The temperature normalized data sequence is subjected to power frequency ripple suppression and low frequency noise filtering, and a leakage response comparison model is constructed by using a bidirectional difference method of the reference channel and the detection channel; The reference channel result output by the leakage response comparison model is defined as the leakage response reference sequence, and the real-time output of the detection channel is defined as the real-time sequence of the leakage response.

[0009] Preferably, according to the one-time system topology structure and the equipment account, the monitoring object is divided into a system layer, a branch layer and a device layer, the micro DC leakage response signal is decomposed and attributed according to a topological reachable relationship, a response matrix identified by dimension and a corresponding object mapping relationship are formed, including: Based on the primary system topology and the equipment account, a hierarchical topology graph containing bus segments, feeder segments and typical equipment nodes is established, each node is given a hierarchical identification of system layer, branch layer and equipment layer, and a hierarchical annotated topology model is obtained; Taking the injection node of the direct current detection channel as a source, a topological reachable closure is calculated according to the conduction state and the grounding resistance parameter, a reachable node set and a path constraint are determined, and the topological reachable relationship is obtained; The micro-direct current leakage response signal is path attributed according to the topological reachable relationship, a projection operator from the measurement domain to the hierarchical object domain is constructed, and an initial response vector arranged according to the system layer, the branch layer and the equipment layer is output; Based on the object electrical parameter and the path constraint, the initial response vector is implemented with uniformization and decoupling calculation, and a response matrix identified by dimension is formed; The response matrix is corresponded to the hierarchical annotated topology model, a pairing relationship of object identification and hierarchical index is generated, and is defined as the object mapping relationship.

[0010] Preferably, taking the injection node of the direct current detection channel as a source, a topological reachable closure is calculated according to the conduction state and the grounding resistance parameter, a reachable node set and a path constraint are determined, and the topological reachable relationship is obtained, including: The nodes of the primary system topology are weighted to obtain an admittance matrix , wherein, ; in the formula, is the conduction state indicator between node and node , and the value is one, indicating that the conduction is allowed and the direct current is reached, and the value is zero, indicating that the conduction is not allowed; is the direct current equivalent resistance of the branch between node and node ; is the direct current equivalent admittance between node and node ; is defined as zero to avoid interference of the self-loop on the admittance matrix; The external admittance of each node is normalized to obtain a transition matrix , wherein, ; in the formula, is the topological transition coefficient from node to node , and the value reflects the relative reachable strength of the leakage channel from node to node under direct current excitation; is the sum of the external admittance of node ; Taking the injection node of the direct current detection channel as , with the neutral grounding resistance as , with the resistance of each node to ground as , defining the equivalent outgoing resistance of the injection node , and defining the propagation coefficient ; in the formula, is the injection node; is the grounding resistance of the neutral small resistance grounding system; is the resistance of node i to ground; is the equivalent outgoing resistance of the injection node to the external network under direct current; is the propagation coefficient obtained by combining the grounding constraint and the network external connection admittance, used to characterize the effective diffusion proportion of the direct current detection energy in the network; , a unit vector is used to represent the unit injection at node , and the steady-state reachable intensity vector is solved, and the formula is: ; in the formula, is the reachable intensity vector of each node; is the unit matrix of the same order as ; in the formula, is the transpose of the transition matrix; is a unit injection vector with only the component being one and the rest being zero; , a detection sensitivity threshold is used to determine the reachable node set and the path constraint, and is obtained; wherein, ; in the formula, is the topologically reachable node set; is the reachable edge set that meets the path non-zero constraint; is a judgment threshold determined according to the current detection lower limit and the injection voltage; is the current resolution lower limit of the micro direct current leakage detection; is the injection voltage amplitude of the controllable direct current detection signal.

[0011] Preferably, meteorological parameters and operating condition data are synchronously collected, factor weighting and robust pullback strategies are used to standardize and robustly process the response matrix, and event labels are generated based on start-stop, load fluctuation and grounding transient records, to obtain a multi-dimensional matrix normalized by environment and operating conditions and the event labels, including: Within the operating cycle of the neutral small resistance grounding system, a meteorological parameter set containing temperature, humidity, air pressure and environmental electric field strength, and an operating condition data set containing load current, bus voltage, device start-stop state and grounding transient characteristics are established, and time alignment and sampling synchronization are performed on the meteorological parameter set and the operating condition data set; Taking the response matrix as input, a weight function is constructed according to the correlation between the set of meteorological parameters and the set of operating condition data, weight reduction is implemented on high correlation or periodic disturbance factors, and a factor-weighted response matrix is obtained; Robust pullback processing is implemented on the factor-weighted response matrix, sudden noise and short-time fluctuation are suppressed through time smoothing and median regression, and a robustly pulled-back response matrix is obtained; According to the device start-stop information, load change rate and grounding potential fluctuation characteristics in the set of operating condition data, disturbance events related to insulation state changes are identified, and corresponding event label sequences are generated. The robustly pulled-back response matrix is associated with the event label sequence to form a multi-dimensional response matrix normalized by environment and condition and the event label.

[0012] Preferably, based on the correspondence between the controllable direct current detection signal and the leakage response reference sequence and the real-time sequence, a mechanism model containing equivalent parameters of leakage channels and path constraints is established, and a fusion modeling framework is constructed combined with multi-condition data, equivalent insulation parameters and credibility weighted indicators are calculated for the system layer, the branch layer and the device layer respectively, and a multi-dimensional insulation characterization vector is obtained, including: According to the controllable direct current detection signal and the leakage response reference sequence and the real-time sequence, time alignment and amplitude normalization of excitation and response are completed, an excitation-response pairing set is established, which is used as the input set of mechanism modeling; Under the constraints of the topological reachability relationship and the object mapping relationship, a mechanism model containing equivalent parameters of leakage channels and path constraints is constructed, and the initial equivalent insulation parameter set of the layered object is output by identifying the constrained excitation-response pairing set; The set of meteorological parameters, the set of operating condition data and the event label are introduced into the fusion modeling framework, the scenario-based estimation of equivalent insulation parameters is calculated for each multi-condition segment, and the corresponding credibility measure is generated based on residual consistency and signal-to-noise stability; According to the object mapping relationship, the scenario-based equivalent insulation parameters and the corresponding credibility are weighted and aggregated in the system layer, the branch layer and the device layer, and the hierarchical equivalent insulation parameters and the hierarchical credibility weighted indicators are obtained. The equivalent insulation parameters and the credibility weighted indicators of the system layer, the branch layer and the device layer are vectorized and combined in a predetermined order to obtain the multi-dimensional insulation characterization vector.

[0013] Preferably, the expression of the mechanism model adopts a constrained mapping based on the difference between the reference sequence and the real-time sequence and a closed-form identification, the constrained mapping is: The closed-form identification is: ; wherein, is an observed current difference vector obtained by subtracting the leakage response reference sequence from the real-time sequence; is a topological propagation observation operator from equivalent admittance difference to current difference; is an injected voltage amplitude of the controllable DC detection signal; is an identity matrix; is a propagation coefficient determined by neutral grounding parameters and external admittance; is a topological transition matrix constructed according to conduction states and branch resistances and normalized; is an observation selection operator determined by the object mapping relationship; is a difference vector of equivalent admittances to ground of each reachable node, used to represent insulation changes; is an equivalent admittance difference estimate identified under path constraints and non-negativity constraints; is a non-negative projection operator; is a screening operator set according to the topological reachability relationship, used to impose zeroing constraints on unreachable nodes; is a path smoothing regularization coefficient; is a graph Laplacian matrix composed of a set of reachable edges, used to impose smoothing constraints on equivalent admittances along paths.

[0014] Preferably, the multi-dimensional insulation characterization vector is subjected to time series analysis to extract slow degradation features and sudden change features, a health index and a risk level are generated according to a unified evaluation scale, suspicious object positioning and early warning output are realized in combination with the topological reachability relationship and the event label, and insulation health evaluation results of the system layer, the branch layer and the device layer are obtained, including: A time series model is established for the multi-dimensional insulation characterization vector to extract long-term trend components and transient change components, and the slow degradation features and the sudden change features of insulation performance are identified; In combination with the event label, the sudden change features caused by start-stop operations, load fluctuations and grounding transients are compared and analyzed with the slow degradation features under the reference working condition to determine stable and abnormal intervals of equivalent insulation parameter changes; According to a unified evaluation scale, the change amplitude of the equivalent insulation parameters in the stable and abnormal intervals is mapped to a health index, and a risk level is calculated based on the time change rate of the health index and the duration of the abnormal interval, to obtain an evaluation index set reflecting the quantitative level of insulation state; In combination with the topological reachability relationship, the health index and the risk level of each object are matched and clustered in the spatial distribution of the system layer, the branch layer and the device layer to determine the area with relatively concentrated risk levels and the corresponding objects, and an object positioning result is generated; The object positioning result is outputted, and the insulation health evaluation result containing the hierarchical identification, the health index value, the risk level interval and the suspicious object identification is obtained. Figure One And output, get including hierarchical identification, health index value, risk level interval and suspicious object identification The insulation health evaluation result.

[0015] The present application discloses the following technical effects: The present application constructs an electrically isolated DC detection channel in a neutral point small resistance grounding system and implements electromagnetic decoupling control, injects a controllable DC detection signal under the premise of meeting insulation coordination and safety constraints, and significantly overcomes the defects of traditional power frequency online monitoring which is susceptible to harmonic, electromagnetic coupling and operation transient interference. The previous method often relies on power frequency side measurement and experience threshold, and it is difficult to stably extract insulation degradation signals in a complex electromagnetic environment, and false positives and false negatives are prone to occur. The present application minimizes the influence of the detection circuit on the network side disturbance through the isolation channel and decoupling control, and ensures that the detection signal amplitude, duty cycle and injection timing do not trigger protection devices or introduce additional risks with the "safe injection parameter set", thereby improving the detection signal-to-noise ratio and engineering implementability from the source.

[0016] The present application introduces zero drift correction, temperature compensation and power frequency ripple suppression in the signal processing link, and constructs a bidirectional comparison sequence based on the reference channel, effectively solving the measurement deviation and false image problems caused by sensor temperature drift, environmental temperature change and power frequency ripple superposition in the background technology. Traditional solutions often use single-channel, static filtering and fixed thresholds, which are difficult to adapt to day-night temperature differences and seasonal fluctuations. The present application establishes a "leakage response reference sequence" based on the reference channel, and performs bidirectional comparison with the real-time sequence, which can simultaneously eliminate slow drift and periodic disturbance, so that the micro-DC leakage response can be reconstructed stably, providing high-quality input for subsequent equivalent parameter identification.

[0017] The present application attributes the measurement signal to the system layer, branch layer and device layer according to the one-time system topology and equipment account, and forms a response matrix and object mapping identified by dimension, breaking through the positioning bottleneck of "only seeing abnormalities but not knowing where they come from" in the prior art. Traditional methods often give insulation state scores for the whole station or a single point, lack of reachable relationship coupling with physical topology, and fault attribution often relies on human experience. The present application divides the signal contribution degree by the topological reachable relationship, so that the leakage path has a structured mapping, and the abnormal measurement of the whole station can be decomposed to specific branches and devices, realizing the coarse-grained judgment of "whether abnormal" and improving to the traceable positioning of "where abnormal", providing direct guidance for maintenance decision-making.

[0018] The application fuses factor weighting, robust pullback and event label construction in data robustness and modeling level, and establishes a mechanism model containing equivalent parameters of leakage channel and path constraints, which solves the pain points of background technology, such as sensitivity to environment and working condition, poor consistency across working conditions, and easy overfitting of pure data-driven. Traditional experience method is unstable in threshold value when facing start-stop, load fluctuation and grounding transient, and it is difficult to compare across time periods. The application first standardizes and robustly processes meteorological and working condition factors, then generates event labels for start-stop, load disturbance and grounding transient, and evaluates by scenario; meanwhile, equivalent parameter identification and multi-condition fusion modeling are introduced, and equivalent insulation parameters and credibility weighted indicators are output at system level, branch level and device level, realizing quantization absorption and credibility expression of multi-source uncertainty, improving physical interpretability of parameters and transferability of conclusions.

[0019] The application performs time series analysis on the multi-dimensional insulation characterization vector at the decision output end, jointly identifies slow degradation and mutation, generates health index and risk level according to a unified evaluation scale, and realizes suspicious object positioning and early warning combined with topological accessibility and event label, which makes up for the defects of background technology, such as focusing on static cross-section evaluation, lacking of evolution trend and early warning. The previous scheme often determines the state by single time score, which is difficult to capture the signs of insulation slow degradation and sudden breakdown in time. The application extracts trend items and mutation items in parallel, which can form early warning for chronic aging and quickly trigger early warning for abnormal transition; the unified scale ensures the comparability across time periods, across devices and across voltage levels, so that the operation and maintenance unit can carry out hierarchical management, risk closed loop and maintenance optimization, thereby realizing the measurability, judgment and controllability of system, branch and device at multiple levels without increasing the cost of power outage and test. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 The method flowchart provided by the embodiments of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] The application aims to provide a multi-dimensional electrical equipment insulation analysis and evaluation method, which can accurately characterize the electrical equipment insulation degradation mechanism through multi-dimensional signal fusion and topological correlation analysis, and can build a quantifiable, traceable and early warning insulation health evaluation system, thereby improving the safety and intelligent level of system operation.

[0024] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the application will be further described in detail below in combination with the drawings and specific embodiments.

[0025] Figure 1 The method flowchart provided by the embodiment of the application is shown in Figure 1 The application provides a multi-dimensional electrical equipment insulation analysis and evaluation method, which comprises the following steps: Step 100: In a neutral point small resistance grounding system, a direct current detection channel is constructed, electromagnetic decoupling control is performed on the neutral point and the voltage mutual inductance loop, a controllable direct current detection signal is injected under the condition of meeting the insulation coordination and safety constraint conditions, and a decoupled detection loop and a safety injection parameter set are obtained; Step 200: The micro direct current leakage response signal corresponding to the controllable direct current detection signal is collected, zero drift correction, temperature compensation and power frequency ripple suppression are performed, a bidirectional comparison sequence based on a reference channel is constructed, and a leakage response reference sequence and a real-time sequence are obtained; Step 300: According to the topological structure and the equipment account of the primary system, the monitoring object is divided into a system layer, a branch layer and a device layer, the micro direct current leakage response signal is decomposed and attributed according to the topological reachable relationship, and a response matrix identified by dimension and a corresponding object mapping relationship are formed; Step 400: The meteorological parameters and the operation condition data are synchronously collected, the response matrix is standardized and robustly processed by using factor weight elimination and robust pullback strategy, and event labels are generated based on start-stop, load fluctuation and grounding transient record, so that a multi-dimensional matrix normalized by environment and condition and an event label are obtained; Step 500: Based on the corresponding relationship between the controllable direct current detection signal and the leakage response reference sequence and the real-time sequence, a mechanism model containing equivalent parameters of the leakage channel and path constraints is established, a fusion modeling framework is constructed in combination with the multi-condition data, equivalent insulation parameters and credibility weighted indicators are calculated for the system layer, the branch layer and the device layer respectively, and a multi-dimensional insulation characterization vector is obtained; Step 600: The multi-dimensional insulation characterization vector is subjected to time series analysis, slow degradation features and mutation features are extracted, a health index and a risk level are generated according to a unified evaluation scale, suspicious object positioning and early warning output are realized in combination with the topological reachable relationship and the event label, and insulation health evaluation results of the system layer, the branch layer and the device layer are obtained.

[0026] Specifically, in step 100 of the embodiment, an isolation branch is arranged in parallel on the neutral point path, and the isolation branch is composed of a high-impedance isolation unit and a controllable switch unit. The high-impedance isolation unit presents an equivalent impedance of not less than 10^7 ohms to the system in a power frequency state, so as to avoid forming a bypass current. The structure of the high-impedance isolation unit is composed of a plurality of precise resistors in series and a surge suppressor in parallel, and is matched with a shielding lead (shielding layer coverage rate ≥ 95%) and a grounding terminal to suppress induced interference. The controllable switch unit adopts a structure in which a bipolar solid-state relay and a mechanical relay are connected in series, so as to ensure that the controllable switch unit is turned on in direct current detection and is turned off in a power frequency state, and an insulation gap is ≥ 8 mm. The interlocking and timing control of the two levels of switches are realized through a control logic, so that the isolation branch is turned on in a direct current detection window and remains in an isolated state at other times. In order to ensure operation safety, state indicator lamps and test terminals are arranged at both ends of the isolation branch, the indication response time is ≤ 100 ms, and the on state can be confirmed before and after detection, so that a direct current detection channel with controllable start and stop and electrical isolation performance is obtained.

[0027] In the direct current detection channel, a detection injection node is arranged, and the node is connected with a programmable voltage source and a current limiting module. The voltage source output signal adopts a slow rise and slow fall strategy, and the rise time is controlled to be 1-3 seconds, so as to avoid the influence of step disturbance on the primary circuit; the duty cycle is controlled to be less than 20%, and the current limiting threshold is not higher than 5 mA. A fast fuse and a transient suppression device (clamping voltage ≤ 75 volts) are configured at the front end of the injection node, so as to form a protection mechanism against overcurrent and reverse flow. A high-precision shunt sampling and isolation amplification circuit is configured at the rear end of the node, so as to realize 24-bit precision acquisition of the microampere level leakage current signal, and the time reference error of the acquisition terminal is not higher than 1 ms. Through the cooperation of the injection node and the acquisition link, the embodiment can gently stimulate the insulation circuit and accurately acquire the leakage response signal without interrupting the system operation, so as to form synchronous recorded injection and measurement data, and provide reliable input for subsequent mechanism modeling.

[0028] The embodiment sets impedance matching unit and reverse isolation unit on the voltage mutual inductance loop side, the former keeps balance between injected signal and system impedance through adjustable voltage dividing network, avoiding reflection distortion between loops; the latter prevents running side voltage from backfilling detection side through diode limiting conduction device. When injection channel and running channel run in parallel, electromagnetic decoupling is realized between channels through common mode inductor (common mode rejection ratio ≥ 80 decibels) and low pass filter (cutoff frequency ≤ 2 hertz), ensuring that power frequency signal and DC detection signal are not coupled with each other. Control logic dynamically limits safe injection boundary based on primary system insulation coordination and grounding resistance parameters, sets upper limit of injected voltage to not higher than 48 volts, single injection duration to not higher than 30 seconds, and repetition period to not less than 120 seconds. When detection voltage or current is out of limit, control logic immediately triggers cut-off, and cut-off action time is not higher than 10 milliseconds. Through the above configuration, the embodiment realizes complete electromagnetic decoupling between running channel and detection channel, obtains decoupled detection loop and dynamic safe injection parameter set, thereby ensuring insulation detection safety and data reliability under all working conditions.

[0029] Specifically, in step 200 of the embodiment, in the DC detection channel of the neutral point small resistance grounding system, the controllable DC detection signal is taken as the trigger reference, the corresponding micro DC leakage current signal is collected in real time, and the reference channel signal is collected synchronously as a comparison. The collection process adopts a high input impedance and low noise sampling module, and the sampling rate is set to more than 1000 times per second to ensure that the leakage response signal still has sufficient time resolution when the leakage is below the milliamper level; the sampling accuracy is not less than 24 bits, and the sampling error between the detection channel and the reference channel is ensured to be not more than 1 millisecond through the time stamp synchronization device. The original leakage response data set is formatted and abnormal point marked after collection, and stored according to injection timing and working condition label. The reference channel signal is usually taken from the same non-injection branch as the neutral point insulation equivalent path, which is used to reflect the system common mode interference background. Through synchronous collection and reference establishment, real-time comparison of detection channel signal and environmental noise is realized, providing data basis for subsequent compensation and comparison.

[0030] The original leakage response data set collected is corrected for zero drift to eliminate baseline drift errors of the detection link and acquisition device during long-term operation. First, a static calibration section (duration not less than 30 seconds) in the no injection state is selected as a reference to establish a baseline model. The baseline model fits the steady-state average value of the detection channel output in a time sliding window to form a dynamic baseline sequence; the data during operation will be matched by time for regression compensation, and the data will be offset corrected according to the baseline drift trend, so that the corrected direct current component tends to zero. The first processed data set formed after correction shows a zero point drift of less than 0.1 microamperes in multiple rounds of verification, significantly reducing the error accumulation caused by amplifier zero point drift, temperature change and stress aging. Through this process, the subsequent temperature compensation and ripple suppression process are carried out on a stable reference, improving data repeatability and reliability.

[0031] The first processed data set is temperature compensated and power frequency ripple suppressed according to the temperature characteristic parameters of the neutral point small resistance grounding system. In the temperature compensation stage, the current drift caused by resistivity change is proportionally corrected according to the real-time data of the system temperature sensor, and the compensation coefficient is dynamically adjusted according to the temperature sensitivity constant (value range 0.002 to 0.005 per degree Celsius). The output temperature normalized data sequence. Subsequently, the temperature normalized data sequence is subjected to power frequency ripple suppression and low frequency noise filtering, and a combination of bandpass filter and wavelet smoothing algorithm is used to automatically identify and weaken the power frequency component in the range of 50 Hz ± 1 Hz. To further improve the anti-interference performance, the embodiment constructs a bidirectional differential comparison model of the reference channel and the detection channel, and takes the difference between the two as the net leakage response signal. In the comparison model output, the filtered result of the reference channel is defined as the leakage response reference sequence, and the real-time output of the detection channel is defined as the leakage response real-time sequence. According to the actual measurement verification, the signal-to-noise ratio after bidirectional differential comparison is improved by more than 25 decibels, and the leakage response resolution is better than 0.05 microamperes, which can accurately reflect the subtle changes of the insulation loop and provide high confidence input for subsequent mechanism modeling.

[0032] Further, in step 300 of the embodiment, first, a hierarchical topology graph containing busbar segments, feeder segments, circuit breakers, disconnectors, transformers, arresters, cable terminals and important load nodes is established based on the single-line diagram of the system and the equipment account, and each node is given a hierarchical identification of the system layer, branch layer and equipment layer to form a hierarchical annotated topology model. To ensure consistency between the model and the field, the topology graph is automatically imported from the account with rated voltage, rated current, installation location, commissioning date and maintenance record, and the on-off state is updated in real time by the switch quantity remote signaling; the model refresh cycle is set to not more than 60 seconds, the number of nodes is not less than 10,000, and the hierarchical depth is not less than 10 levels. To avoid incorrect connection, the embodiment performs three verifications in the import stage: busbar phase consistency verification, equipment unique identification verification, and loop closure verification, and each verification failure generates a diagnosis list and a positioning path, which can be entered into the running state after manual confirmation.

[0033] The embodiment takes the injection node of the direct current detection channel as the source, integrates the current switch position and the grounding loop parameters, calculates the topology reachable closure, and obtains the reachable node set and its path constraint which exist in the current on-off state with the injection node. The calculation of the topology reachable closure is divided into two steps: first, search for the on-off path on the hierarchical topology graph in a breadth-first manner; second, remove high resistance paths and suspended paths according to the grounding resistance and connection resistance threshold. The upper limit of the search depth is set to 20 levels, and the time consumption of a single calculation is controlled within 1 second. The path constraint records the on-off state, equivalent impedance level, cross-zone connection information and parallel system characteristics of each segment in the path, and performs loop breaking processing and priority sorting on the area with loop; when the number of reachable nodes exceeds 1,000, the system automatically triggers partition calculation to maintain real-time performance, and marks the unreachable area as a gray area to prevent signal misattribution.

[0034] The embodiment attributes the collected micro direct current leakage response signal to the topology reachable relationship, constructs a projection operator that maps the measurement domain to the hierarchical object domain, and realizes the initial response vector output from the multi-measurement-point data to the system layer, branch layer and equipment layer. To improve the physical consistency of the attribution result, the embodiment implements the consistency and decoupling calculation of the initial response vector according to the rated parameters, grounding mode, shielding structure and operating load of the object: first, normalize according to the equipment category and rated voltage; second, apportion the parallel and series contributions according to the path constraint; finally, decouple the cross-influence of the shared path. The decoupling calculation adopts an iterative method, the maximum number of iterations is set to 50, and the result is considered to be converged if the change between two consecutive results is less than 1%; if it does not converge, the system automatically reverts to the last stable result and marks the list of objects that need to be retested. After processing, a response matrix is formed according to the dimension identification, and the object mapping relationship is generated one by one corresponding to the hierarchical annotated topology model, including object unique identification, hierarchical index, attribution path number and data timestamp.

[0035] Further, in the embodiment, when determining the topology reachable relationship, first, the nodes of the primary system topology are weighted to establish an admittance matrix. The elements of the admittance matrix are based on the conduction state between nodes and the direct current equivalent resistance, wherein the conduction state between node i and node j is represented by a conduction indicator, and the value of 1 indicates that the node interconnection is closed and the direct current signal is reachable, and the value of 0 indicates that the conduction is not conducted or there is insulation isolation. The direct current equivalent resistance between nodes is determined by the bus length, conductor cross section and material, and the typical value range is 0.01-1 ohm; thus the equivalent admittance of the node pair is obtained, and the value is the inverse of the resistance. In order to avoid the influence of self-loop, the admittance of the node itself is defined as zero, thereby forming a symmetric admittance matrix without self-coupling. The matrix reflects the direct current reachability of the system under the current conduction state, and lays a foundation for subsequent calculation of the topology reachable closure.

[0036] In the embodiment, each row of the admittance matrix is normalized to establish a transition matrix between nodes to reflect the relative reachability strength of the leakage channel under the direct current excitation. The element of the transition matrix is defined as the proportion of the admittance of a node to its adjacent node to the total external admittance of the node, which represents the possibility of the leakage current being branched at the node along different branches. The sum of the external admittance of each node is obtained by accumulating the admittance of the adjacent nodes, and when the node is connected to n branches, the normalization coefficient is the sum of the admittances. Through the normalization processing, the sum of all transition coefficients is 1, thereby realizing the probabilistic topology transition description. The transition matrix is equivalent to the diffusion operator of the network under the direct current condition, and is used to describe the directionality and strength relationship of energy or current propagation in the network, thereby providing a basis for subsequent solution of the steady-state reachability strength vector.

[0037] In the embodiment, the injection node of the direct current detection channel is taken as the starting point, which is recorded as v0, and the equivalent outgoing resistance of the injection node is calculated in combination with the neutral point grounding resistance and the ground resistance of each node. The neutral point grounding resistance takes the system design value, and the typical range is 1-10 ohm; the ground resistance of each node takes the field measurement or account value, which is generally between 100 kilo-ohm and 10 mega-ohm. The equivalent outgoing resistance of the external network of the injection node under the direct current is obtained, which reflects the overall impedance of the energy diffusion of the injection point to the outside. According to the result, the propagation coefficient is defined, and the value of the propagation coefficient is between 0 and 1. When the neutral point grounding is good, the typical value is about 0.2-0.4, which represents the effective diffusion proportion of the injection energy in the system. The unit excitation at the injection node is represented by a unit vector, and the steady-state reachability strength vector is solved in combination with the transition matrix. Each component of the vector corresponds to the relative reachability degree of each node in the network. According to the detection sensitivity threshold, the reachable node set and path constraint are determined, wherein the lower limit of the detection current resolution is 0.1-1 microampere, and the injection voltage amplitude is set to 24-48 volts, thereby calculating the determination threshold, and the value is usually between 10 -6 -10 -4The range of the voltage is in the range of 0.1-10 V. The set of reachable nodes includes all nodes with a reachable strength higher than the threshold, and the path constraint is constituted by node pairs with non-zero transition coefficients. Through the above calculation, the effective diffusion range and topological propagation path of the DC leakage signal in the neutral point small resistance grounding system can be comprehensively and quantitatively described.

[0038] Optionally, in step 400 of the embodiment, the meteorological parameters and operating condition data are synchronously collected within the operating cycle: the meteorological parameter set includes temperature, humidity, air pressure and ambient electric field strength, and the sampling frequencies are 1 Hz, 1 Hz, 0.2 Hz and 1 Hz respectively; the operating condition data set includes load current, bus voltage, equipment start-stop state and grounding transient characteristics, and the sampling frequencies are 10 Hz, 10 Hz, event-triggered type and 1 kHz respectively. All data are aligned with a unified time base, the clock source is GPS or a time service network, and the time drift correction is not more than 1 ms. The data with different sampling rates are resampled at a common step of 1 s: linear interpolation is used for slowly varying variables (temperature, humidity, air pressure), anti-aliasing low-pass filtering is used for fast variables (current, voltage), and event variables (start-stop and grounding transient) are dropped into the corresponding 1 s window by timestamp; when the missing measurement ratio exceeds 5%, a missing mask is triggered and the nearest valid section is forward filled for not more than 3 s to ensure the hourly comparability of the response matrix and external factors.

[0039] The aligned response matrix is taken as the input in the embodiment, and the meteorological and operating condition factors are combined to perform "factor weighting" and "robust pullback". Among them, factor weighting refers to reducing the participation weight of factors prone to bias based on correlation and periodicity: within a sliding window of 30 min and a step of 5 min, if the absolute correlation coefficient of any factor and the response variable is ≥0.6, or there is a significant peak (peak energy ≥20% of total energy) in the power spectrum in the day cycle (about 24 h) and the power frequency multiple frequency neighborhood, the weight of the factor is lowered from 1 to 0.3-0.5; for factors that meet both conditions, the weight is lowered to 0.1-0.3. The response matrix after factor weighting is subjected to robust pullback: first, a median filter window of 61 s is used to suppress impulse noise, and then a first-order exponential smoothing (equivalent time constant 10-30 s) is used to regress to a short-term robust baseline, while outliers exceeding the 1% and 99% quantiles of the window distribution are truncated and filled back; if the proportion of outliers within 5 minutes exceeds 10%, it is marked as "high risk of disturbance" for subsequent event interpretation. The robustly pulled back response matrix is obtained after processing, and the short-time standard deviation is reduced by not less than 30% relative to the original sequence.

[0040] The embodiment automatically generates event label sequences according to operating condition sets, and associates them with the response matrix after robust pullback to form a "multi-dimensional response matrix + event label normalized by environment and operating condition". Device start-stop events are identified by rising / falling edges of switch quantity, with 2s debouncing; load fluctuation events are determined by the rate of change per unit time, and when the relative change in 1 min is ≥10%, it is recorded as "moderate fluctuation", and when it is ≥20%, it is recorded as "severe fluctuation"; ground transient events are triggered by the joint of ground potential step and change rate, and when the potential transition is ≥5V and the rise time is ≤200ms, or the cumulative transition in 1s is ≥10V, it is marked as "ground transient". The start time, end time, peak amplitude and involved hierarchical object set of each event are recorded, and are archived together with the multi-dimensional response vector in the corresponding time window. The final output data structure includes: a normalized multi-dimensional response matrix (arranged according to system layer / branch layer / device layer, time resolution 1s), and a parallel event label sequence (including event type, intensity level and object list), which provides reliable input for subsequent mechanism modeling, positioning and early warning.

[0041] Further, in step 500 of the embodiment, strict pairing of excitation and response is first completed: taking the timestamp of the controllable direct current detection signal as the main time base, the leakage response reference sequence and the leakage response real-time sequence are aligned point by point, and the amplitudes of the three are unified in dimension and range (for example, both voltage and current are converted to per second average and peak value to maintain dual-track representation). After alignment, a "excitation-reference-real-time" paired set is formed, which is used as the input of mechanism modeling. To avoid magnification of alignment error, the embodiment sets two levels of review: first, the maximum time deviation is allowed to be not more than 1 millisecond, and the samples exceeding the limit are directly excluded; second, the static stable section of 5 seconds before and after the injection window is used to check the baseline consistency, and if the baseline deviation exceeds 0.05 microamperes, the samples in this section are zero-point corrected before being input into the model. After the above processing, the data entering the modeling has stable time domain registration and amplitude consistency, ensuring the repeatability of the subsequent identification link.

[0042] The embodiment constructs a mechanism model under the dual constraints of topological reachability and object mapping. The observation mapping of the model consists of three parts in sequence: the first part is the injection voltage amplitude parameter, which is used to map the network injection magnitude to the observation domain; the second part is the inverse mapping operator of network diffusion, which is obtained by inverting “unit mapping minus the product of propagation coefficient and topological transition matrix transpose”, and is used to describe the diffusion and reflux of direct current energy in the network; the third part is the observation selection operator, which is used to select the observable nodes corresponding to the sensors from all network nodes (given by the object mapping relationship). Under this observation mapping, the observation current difference vector of “real-time sequence minus reference sequence” is used as the target quantity, and the difference estimate of the equivalent admittance of each reachable node to ground is obtained by constraint identification. In order to ensure physical rationality, the embodiment applies zeroization shielding to unreachable nodes, and applies non-negative constraint to all equivalent admittance differences, so as to avoid the occurrence of physically unreasonable negative admittance increment.

[0043] The embodiment introduces meteorological parameter sets, operating condition data sets and event labels into the fusion modeling framework, and calculates the scenario-based equivalent insulation parameters and corresponding confidence for each multi-condition segment. The specific process is as follows: first, split according to the scene (such as “normal temperature steady state”, “high temperature and heavy load”, “equipment start-stop transient”, “grounding transient”), and run the mechanism identification independently in each scene to obtain the equivalent insulation parameters of the scene; then, calculate two types of consistency indexes, which are residual consistency and signal-to-noise stability. The residual consistency measures the consistency degree of the model output and the observation difference (measured by the combination of mean square residual and absolute residual quantile), and the signal-to-noise stability measures the parameter fluctuation of repeated tests in the same scene (measured by the combination of coefficient of variation and autocorrelation decay constant). After standardization, the two types of indexes are combined into the confidence measure, with a value range of 0-1; when there is a sharp load change or grounding transient in the scene, the confidence will be automatically adjusted downward to avoid the amplification effect of transient mismatch on the parameters. In engineering, the embodiment suggests that the confidence target in the normal temperature steady state scene should not be less than 0.8, and the confidence target in the transient scene should not be less than 0.6.

[0044] The embodiment performs weighted aggregation within the hierarchy to obtain equivalent insulation parameters and credibility weighted indicators of the system layer, branch layer and device layer. The weighting rules follow the order of "scene credibility first, observation coverage second, time freshness third": the parameters of the same object in multiple scenes are first linearly aggregated as the main weight according to the credibility; if there are multiple repeated observations in the same scene, the weight is then subdivided according to the observation coverage (a joint indicator of the number of sensing points and the number of reachable paths); when the observation time span is large, a time freshness decay factor (for example, the weight decreases by 0.9 every 24 hours) is introduced to ensure that the latest state gets a higher weight. To suppress extreme jumps of adjacent objects in space, the embodiment adds path smoothing regularization before hierarchical aggregation: a graph structure composed of a set of reachable edges is used to apply gentle smoothing to the equivalent parameters of adjacent objects along the path (the regularization strength can be 0.001-0.01 in engineering), which improves the overall stability without losing positioning sensitivity.

[0045] The embodiment finally vectorizes and combines the results of the three layers in a predetermined order to obtain a multi-dimensional insulation characterization vector. The combination order is system layer first, branch layer in the middle, and device layer last; each layer arranges three groups of items in order: "equivalent insulation parameter, credibility weighted indicator, scene proportion abstract", to ensure comparability and traceability across time periods, scenes and objects. To support engineering review, the characterization vector also carries metadata, including observation window start and end time, participating scene list, dominant weight source and maximum residual position; when the credibility of any object is less than 0.5 or the residual is higher than the 95th percentile of the scene distribution, a "retest suggestion" label is added to the characterization vector, and the preferred retest scene and injection amplitude suggestion (for example, 24-36 volts) are given. After field trial operation, the embodiment has stable detection capability for insulation degradation with a relative change of not less than 10% after multi-condition fusion, and the signal-to-noise ratio can be improved by 15-25 decibels in typical scenes. The intra-day repeatability of hierarchical parameters is better than 5%.

[0046] Further, in step 600 of the embodiment, a time series model is established to extract long-term trends and transient changes using the multi-dimensional insulation characterization vector as input: each object is processed according to the "denoising-decomposition-mutation detection" process. First, median filtering (window 61s) and first-order exponential smoothing (time constant 20-40s) are used to suppress pulses and high-frequency jitter; then additive decomposition is used to split the sequence into a long-term trend component and a residual component, and the trend component is fitted with a sliding polynomial regression (window 30-120min); on the residual component, mutation detection (based on cumulative deviation and based on piecewise fitting) is run in parallel, with the trigger condition being a relative change of ≥10% within 60s or a normalized residual of ≥3. The trend component is defined as "slow degradation feature", and the mutation detection output is defined as "mutation feature", both of which retain amplitude, duration and start and end timestamps for subsequent alignment and interpretation with event labels.

[0047] This embodiment aligns event labels with features, and distinguishes stable intervals and abnormal intervals: with three types of event sources, device start-stop, load fluctuation, and grounding transient, the "mutation feature" is attributed to preferential matching (start-stop → grounding transient → load fluctuation), and the matching window is event center ± 30s; If it hits in the matching window, it is marked as "event-driven mutation", otherwise it is marked as "non-event mutation". The stable interval is composed of time slices with "no event, no mutation", and the duration needs to be ≥300s; The abnormal interval is composed of "non-event mutation" or "the slope of slow degradation feature exceeds the upper threshold of steady state", and the slope threshold is set to the 95th percentile of the steady state distribution. The time slice with significant deviation of temperature and humidity (temperature exceeding baseline +10℃ or humidity exceeding baseline +20%) is additionally labeled with "environmental warning" to reduce the weight in the subsequent process. Through this alignment process, observable changes are divided into two categories: explainable and unexplainable, providing a clear boundary for subsequent health index and risk calculation.

[0048] This embodiment generates a health index and calculates a risk level based on a unified evaluation scale. The health index uses a 0-100 scale, with 100 indicating optimal insulation and 0 indicating a serious failure risk; It is obtained by linear combination of two parts: the equivalent insulation parameter amplitude mapping in the stable interval (accounting for 0.7 weight) and the parameter offset penalty in the abnormal interval (accounting for 0.3 weight). The mapping interval can set the upper and lower limits according to the device category and voltage level, and update adaptively by quantile; When there are "non-event mutations" ≥3 times in a single day or a single abnormality lasts ≥300s, the health index is additionally deducted by 5-15 points. The risk level is output in 5 levels (I low, II lower, III medium, IV high, V extremely high), with two indicators for hierarchical judgment: the time change rate threshold of the health index (absolute value ≥5 points per hour) and the abnormal duration threshold (≥600s); If both thresholds are exceeded and the peak offset exceeds the upper limit of the reference by 20%, it is directly upgraded to level IV or above; When no abnormality is triggered for 24 consecutive hours and the health index rebounds by ≥10 points, the risk level is automatically reduced by 1 level and the historical short-term penalty is cleared.

[0049] The embodiment combines the topological reachability in the system layer, branch layer and device layer to realize the suspicious object positioning and early warning output. First, the health index and risk level of each object are mapped to the topological graph, and the "topological adjacency weighted density clustering" is used to identify the risk concentration area: the minimum cluster size is greater than or equal to 3 objects, the average shortest path length in the cluster is less than or equal to 3, and the average risk level in the cluster is greater than or equal to IV or the average health index is less than or equal to 60 to trigger the "high risk cluster"; at the same time, the bridging threshold is set for the chain risk across the branch, and the reachable strength of the bridging edge needs to be greater than or equal to 0.3 to be merged. The positioning result outputs the object list, cluster level, dominant path and recommended treatment order (system layer→branch layer→device layer), and generates a risk level distribution graph and a time axis early warning card; the early warning includes severity, recommended retest scene and recommended injection amplitude (for example, 24-36V), and the default release throttling period is 10 minutes and the repeated alarm suppression is 30 minutes, so as to ensure that the information is executable and not overloaded.

[0050] The beneficial effects of the present application are as follows: The present application constructs an electrically isolated DC detection channel and an electromagnetic decoupling control mechanism in a neutral point small resistance grounding system, realizes safe injection and high sensitivity detection under system operation state, and breaks through the technical bottleneck that traditional power frequency monitoring is seriously disturbed by harmonics and cannot distinguish local insulation degradation from systematic leakage online. Through the dynamic constraint design of the safe injection parameter set, the detection signal obtains a high signal-to-noise ratio response without triggering the relay protection and changing the system operating condition, thereby providing a basis for continuous and disturbance-free online insulation monitoring.

[0051] The present application introduces a three-dimensional robust algorithm of zero drift correction, temperature compensation and power frequency ripple suppression at the signal processing level, and forms a reference sequence and a real-time sequence through bidirectional comparison of the reference channel and the detection channel, thereby fundamentally eliminating the influence of zero point drift and environmental temperature drift of the measurement link. The traditional insulation monitoring relies on static threshold discrimination, which is easy to produce false alarms in temperature and humidity fluctuations or day-night operating condition changes; the dynamic comparison mechanism of the present application can stably detect the insulation degradation trend at a microampere level current resolution, thereby significantly improving the monitoring accuracy and long-term stability.

[0052] The present application constructs a multi-level topological reachability and object mapping mechanism according to the primary system topology and device account, realizes quantitative attribution of the leakage response signal between the system layer, branch layer and device layer. The layered mapping changes the insulation anomaly from "detected" to "positioned", solving the problem that the traditional single-point monitoring cannot distinguish the contribution of different branches or devices. Through the reachability path constraint and electrical parameter weighted allocation, the present application realizes the visualization decomposition of the leakage path in the complex system, and provides a physical basis for quickly positioning the hidden trouble device.

[0053] The application adopts factor weighting and robust pullback strategy at the data fusion level, comprehensively standardizes and robustly processes the response matrix by combining meteorological and working condition factors, and generates event labels through start-stop, load fluctuation and grounding transient, thereby realizing adaptive suppression of complex external disturbances. The strategy makes the model robust to temperature, humidity, electric field change and operational transient, ensures that the insulation state evaluation reflects the real material aging and dielectric leakage characteristics, and not the false image caused by external interference.

[0054] The insulation modeling framework proposed by the application fuses mechanism constraints and multi-working condition learning, realizes cross-scene quantitative representation of insulation characteristics by introducing leakage channel equivalent parameters, path constraints and regular smoothing. The output multi-dimensional insulation representation vector has physical interpretability and statistical stability, and can realize credibility weighted aggregation at the system layer, branch layer and device layer. The framework shows high consistency and portability in field application, and provides a directly accessible insulation health quantitative index for intelligent operation and maintenance system.

[0055] The application finally realizes insulation health evaluation and early warning output through time series analysis, trend extraction and topology clustering, can identify slow degradation and sudden abnormality at the same time, and establishes a full closed-loop system from detection, attribution, modeling to evaluation. The unified scale of health index and risk level supports standardized evaluation across devices and voltage levels; combined with the early warning results of topology clustering positioning, the insulation risk management can be realized in a multi-level, traceable and quantifiable manner. The application overall improves the intelligent, fine and forward-looking level of power system insulation state monitoring, and provides reliable technical support for preventing insulation breakdown, reducing power outage risk and prolonging equipment life.

[0056] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between the various embodiments can be referred to each other.

[0057] The principles and implementation modes of the application are described by applying specific examples in this paper, and the above embodiment description is only used to help understand the method and core idea of the application; at the same time, for those skilled in the art, according to the idea of the application, the specific implementation mode and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the application.

Claims

1. A multi-dimensional method for insulation analysis and evaluation of electrical equipment, characterized in that, include: In a neutral point low-resistance grounding system, an electrically isolated DC detection channel is constructed, and electromagnetic decoupling control is implemented on the neutral point and voltage inductance circuit. Under the conditions of insulation coordination and safety constraints, a controllable DC detection signal is injected to obtain the decoupled detection circuit and the set of safety injection parameters. The micro-DC leakage response signal corresponding to the controllable DC detection signal is acquired, and zero drift correction, temperature compensation and power frequency ripple suppression are performed. A bidirectional comparison sequence based on the reference channel is constructed to obtain the leakage response benchmark sequence and the real-time sequence. Based on the primary system topology and equipment register, the monitoring objects are divided into system layer, branch layer and equipment layer. The micro DC leakage response signals are decomposed and assigned according to the topology reachability relationship, forming a response matrix identified by dimension and the corresponding object mapping relationship. Meteorological parameters and operating condition data are collected synchronously. The response matrix is ​​standardized and robustned using a factor weighting and robust pull-back strategy. Event labels are generated based on start-up, shutdown, load fluctuation and grounding transient records to obtain a multi-dimensional matrix normalized by environment and operating conditions and the event labels. Based on the correspondence between the controllable DC detection signal, the leakage response benchmark sequence, and the real-time sequence, a mechanism model containing equivalent parameters of leakage channels and path constraints is established. A fusion modeling framework is constructed by combining multi-condition data. Equivalent insulation parameters and confidence weighted indicators are calculated for the system layer, the branch layer, and the equipment layer, respectively, to obtain a multi-dimensional insulation characterization vector. A time-series analysis is performed on the multidimensional insulation characterization vector to extract slow degradation features and abrupt change features. Based on a unified evaluation scale, a health index and risk level are generated. By combining the topological reachability relationship and event tags, suspicious object location and early warning output are achieved, and the insulation health evaluation results of the system layer, the branch layer and the equipment layer are obtained.

2. The method for multi-dimensional electrical equipment insulation analysis and evaluation according to claim 1, characterized in that, In a neutral-point low-resistance grounding system, an electrically isolated DC detection channel is constructed. Electromagnetic decoupling control is implemented on the neutral point and voltage inductance circuit. Under the conditions of insulation coordination and safety constraints, a controllable DC detection signal is injected to obtain the decoupled detection circuit and the set of safety injection parameters, including: On the neutral point path of the neutral point low resistance grounding system, an isolation branch parallel to the voltage inductance circuit is set up; the isolation branch is composed of a high impedance isolation unit and a controllable switch unit, which is used to maintain electrical isolation under power frequency state and to conduct to form the DC detection channel under DC detection state; A detection injection node is established within the DC detection channel, and a controllable DC detection signal is output by a programmable voltage source to excite the insulation circuit without interrupting system operation. An impedance matching and reverse isolation unit is set on the voltage inductance circuit side to prevent the controllable DC detection signal from coupling with the power frequency signal, thereby achieving electromagnetic decoupling between the DC detection channel and the operating channel, and obtaining the decoupled detection circuit. Based on the insulation coordination conditions and grounding resistance parameters of the neutral point low-resistance grounding system, the upper limit of the injection voltage and the current constraint threshold are dynamically limited to obtain the set of safe injection parameters.

3. The method for multi-dimensional electrical equipment insulation analysis and evaluation according to claim 1, characterized in that, The micro-DC leakage response signal corresponding to the controllable DC detection signal is acquired, and zero-drift correction, temperature compensation, and power frequency ripple suppression are performed. A bidirectional comparison sequence based on the reference channel is constructed to obtain the leakage response benchmark sequence and the real-time sequence, including: In the DC detection channel of the neutral point low resistance grounding system, the leakage current signal synchronized with the controllable DC detection signal is collected in real time, and the original leakage response dataset is obtained by using the reference channel signal as the synchronization reference. Zero drift correction is performed on the original leak response dataset, a baseline model is established based on the static calibration section, and baseline regression compensation is performed on the operational data to eliminate DC drift error of the detection link, thus obtaining the first processed dataset. Based on the temperature characteristic parameters of the neutral point low resistance grounding system, temperature compensation is applied to the first processed dataset to generate a temperature normalized data sequence. Power frequency ripple suppression and low frequency noise filtering are performed on the temperature normalized data sequence, and a leakage response comparison model is constructed using a bidirectional differential method between the reference channel and the detection channel. The reference channel result output by the leakage response comparison model is defined as the leakage response baseline sequence, and the real-time output of the detection channel is defined as the real-time sequence of the leakage response.

4. The method for multi-dimensional electrical equipment insulation analysis and evaluation according to claim 1, characterized in that, Based on the primary system topology and equipment register, the monitored objects are divided into system layer, branch layer, and equipment layer. The micro-DC leakage response signals are decomposed and assigned according to topological reachability, forming a dimension-identified response matrix and corresponding object mapping relationships, including: Based on the primary system topology and equipment ledger, a hierarchical topology diagram including bus segments, feeder segments and typical equipment nodes is established. Each node is assigned a hierarchical identifier of system layer, branch layer and equipment layer to obtain a hierarchical labeled topology model. Using the injection node of the DC detection channel as the source, the topology reachability closure is calculated based on the conduction state and grounding resistance parameters to determine the set of reachable nodes and path constraints, thereby obtaining the topology reachability relationship; The micro DC leakage response signal is path-assigned according to the topological reachability relationship, and a projection operator from the measurement domain to the hierarchical object domain is constructed to output the initial response vector arranged in the system layer, branch layer and device layer. Based on the object's electrical parameters and the path constraints, the initial response vector is uniformized and decoupled to form a response matrix identified by dimension. The response matrix is ​​mapped to the hierarchical annotation topology model to generate a pairing relationship between object identifiers and hierarchical indexes, which is defined as the object mapping relationship.

5. The multi-dimensional electrical equipment insulation analysis and evaluation method according to claim 4, characterized in that, Using the injection node of the DC detection channel as the source, the topology reachability closure is calculated based on the conduction state and grounding resistance parameters to determine the set of reachable nodes and path constraints, thus obtaining the topology reachability relationship, including: We obtain the admittance matrix by weighting the node pairs of the system topology. ,in, In the formula, For nodes With nodes The conduction status indicator between them has a value of 1 indicating conduction and allowing DC to pass, and a value of zero indicating non-conduction; For nodes With nodes The DC equivalent resistance of the branch between them; For nodes With nodes DC equivalent admittance between; Defined as zero to avoid interference of self-loops on the admittance matrix; Normalize the external admittance of each node to obtain the transition matrix. ,in, In the formula, For the node To the node The topological transfer coefficient, numerically reflecting the leakage path under DC excitation from the node Flow to Node The relative achievable strength; For nodes The sum of external admittance; The injection node of the DC detection channel is denoted as With the neutral point grounding resistance as Taking the ground resistance of each node as Define the equivalent outgoing resistance of the injected node. And define the propagation coefficient. In the formula, For injection nodes; The grounding resistance of a neutral point low-resistance grounding system; Let be the resistance to ground of node i; The equivalent outgoing resistance of the injection node's external network under DC conditions; The propagation coefficient, obtained by combining grounding constraints and network external admittance, is used to characterize the effective diffusion ratio of DC detection energy within the network; unit vector Indicates at node For unit injection at a given location, the steady-state reachable intensity vector is calculated using the following formula: In the formula, Let be the reachability strength vector of each node; To and Identity matrices of the same order; This is the transpose of the transition matrix; For only in the A unit injection vector with one component and zero others; To detect sensitivity threshold Determine the set of reachable nodes and path constraints to obtain ;in, In the formula, It is the set of topologically reachable nodes; The set of reachable edges that satisfy the non-zero path constraint; The judgment threshold is determined based on the lower limit of current detection and the injection voltage; The lower limit of current resolution for micro-DC leakage detection; The amplitude of the injected voltage for the controllable DC detection signal.

6. The method for multi-dimensional electrical equipment insulation analysis and evaluation according to claim 1, characterized in that, Meteorological parameters and operational condition data are collected synchronously. The response matrix is ​​standardized and robustned using a factor weighting and robust pull-back strategy. Event labels are generated based on start-up / shutdown, load fluctuation, and grounding transient records, resulting in a multi-dimensional matrix normalized to environmental and operational conditions, along with the event labels, including: During the operation cycle of the neutral point low resistance grounding system, a set of meteorological parameters including temperature, humidity, air pressure and ambient electric field strength is established, as well as a set of operating condition data including load current, bus voltage, equipment start-up and shutdown status and grounding transient characteristics. Time alignment and sampling synchronization are performed on the meteorological parameter set and the operating condition data set. Using the response matrix as input, a weighting function is constructed based on the correlation between the meteorological parameter set and the operating condition data set. The weights of highly correlated or periodic disturbance factors are reduced to obtain the response matrix after factor weighting. The factor-reduced response matrix is ​​subjected to robust pullback processing. Sudden noise and short-term fluctuations are suppressed by time smoothing and median regression to obtain the robust pullback response matrix. Based on the equipment start-up and shutdown information, load change rate and grounding potential fluctuation characteristics in the operating condition data set, disturbance events related to insulation state changes are identified, and corresponding event tag sequences are generated; The robust pullback response matrix is ​​associated with the event label sequence to form a multidimensional response matrix normalized to the environment and operating conditions, and the event labels.

7. The method for multi-dimensional electrical equipment insulation analysis and evaluation according to claim 6, characterized in that, Based on the correspondence between the controllable DC detection signal, the leakage response benchmark sequence, and the real-time sequence, a mechanism model containing equivalent parameters of the leakage channel and path constraints is established. A fusion modeling framework is constructed by combining multi-condition data. Equivalent insulation parameters and confidence-weighted indicators are calculated for the system layer, the branch layer, and the equipment layer, respectively, resulting in a multi-dimensional insulation characterization vector, including: Based on the controllable DC detection signal, the leakage response reference sequence, and the real-time sequence, the time alignment and amplitude normalization of the excitation and response are completed, and an excitation-response pairing set is established as the input set for mechanism modeling. Under the constraints of the topological reachability relationship and the object mapping relationship, a mechanism model containing the equivalent parameters of leakage channels and path constraints is constructed, the excitation response pairing set is restricted and the initial equivalent insulation parameter set of the hierarchical object is output. The meteorological parameter set, the operating condition data set, and the event tags are introduced into the fusion modeling framework. The equivalent insulation parameters are estimated in a scenario-based manner for each of the multiple operating condition segments. Based on residual consistency and signal-to-noise stability, the corresponding credibility metrics are generated. Based on the object mapping relationship, the scenario-based equivalent insulation parameters and corresponding credibility are weighted and aggregated in the system layer, the branch layer and the device layer to obtain the weighted indicator of the equivalent insulation parameters and credibility at the level. The equivalent insulation parameters of the system layer, the branch layer, and the device layer are vectorized and combined with the confidence weighted indicator in a predetermined order to obtain the multidimensional insulation characterization vector.

8. The method for multi-dimensional electrical equipment insulation analysis and evaluation according to claim 7, characterized in that, The expression of the mechanism model is identified using a restricted mapping and closed-form recognition based on the difference between the baseline sequence and the real-time sequence. The restricted mapping is as follows: The closed form is identified as: ;in, The observed current difference vector is obtained by subtracting the real-time sequence from the leakage response reference sequence; For topological propagation observation operators from equivalent admittance difference to current difference; The amplitude of the injected voltage of the controllable DC detection signal; It is the identity matrix; The propagation coefficient is determined by the neutral point grounding parameters and the external admittance. This is the topology transition matrix constructed and normalized based on the conduction state and branch resistance; Select the observation operator as determined by the object mapping relationship; This is the difference vector of the equivalent admittance to ground for each reachable node, used to characterize insulation changes; This is the equivalent admittance difference estimate obtained under path constraints and nonnegativity constraints; It is a non-negative projection operator; A masking operator set according to the topological reachability relationship, used to apply nullification constraints to unreachable nodes; These are the path smoothing regularization coefficients; Let be the graph Laplacian matrix consisting of the set of reachable edges, used to impose a smooth constraint on the equivalent admittance along the path.

9. The method for multi-dimensional electrical equipment insulation analysis and evaluation according to claim 1, characterized in that, Temporal analysis is performed on the multidimensional insulation characterization vector to extract slow degradation and abrupt change features. A health index and risk level are generated based on a unified evaluation scale. Suspicious object location and early warning output are achieved by combining the topological reachability relationship and event tags, resulting in insulation health evaluation results for the system layer, branch layer, and equipment layer, including: A time series model is established for the multidimensional insulation characterization vector, and the long-term trend component and transient change component are extracted to identify the slow degradation characteristics and abrupt change characteristics of insulation performance. By combining the event tags, the abrupt change characteristics caused by start-stop operations, load fluctuations and grounding transients are compared and analyzed with the slow degradation characteristics under the baseline operating conditions to determine the stable and abnormal ranges of the equivalent insulation parameter changes. Based on a unified evaluation scale, the variation range of the equivalent insulation parameters in the stable and abnormal ranges is mapped to a health index. The risk level is calculated based on the time variation rate of the health index and the duration of the abnormal range, thus obtaining a set of evaluation indicators that reflect the quantitative level of insulation status. Based on the topological reachability, the health index and risk level of each object are matched and clustered in the spatial distribution of the system layer, the branch layer and the device layer to determine the areas with relatively concentrated risk levels and the corresponding objects, and to generate object location results; The object location result is output together with the risk level distribution map formed by the health index and risk level of each layer to obtain the insulation health evaluation result containing the layer identifier, health index value, risk level range and suspicious object identifier.

Images