Big data-based substation equipment fault detection method and system

Abstract

The application relates to the technical field of power transformation monitoring, and discloses a power transformation equipment fault detection method and system based on big data, which comprises the following steps: acquiring a wide-frequency signal of a main loop of power transformation equipment, extracting a power frequency current sequence and a high-frequency transient sequence; extracting an energy envelope curvature extreme point of the high-frequency transient sequence to form a real-time characteristic sequence; determining a transient electromagnetic damping correction coefficient according to a power frequency current load deviation; implementing a nonlinear scale transformation on a health benchmark sequence time axis by using the coefficient to generate a time-based benchmark sequence suitable for an adaptive load working condition; and calculating an alignment distance between the real-time characteristic sequence and the time-based benchmark sequence, and issuing a pre-warning when the distance exceeds a preset threshold value. The application compensates for electromagnetic response lag caused by load through time-based remodeling at a physical level, eliminates a heavy load monitoring blind area, and captures early weak insulation deterioration precursors.

Description

Technical Field

[0001] This invention belongs to the field of substation monitoring technology, and in particular relates to a big data-based method and system for detecting substation equipment faults. Background Technology

[0002] Current power equipment health management typically uses multi-dimensional physical quantity monitoring data such as current, voltage, transformer oil temperature, and casing vibration to identify equipment defects by constructing a multi-modal fusion model. This type of monitoring mode has fault early warning capabilities under stable operating conditions, but it has a fundamental limitation in identifying early and subtle insulation degradation. Early failure of the insulation medium often produces microsecond-level high-frequency transient characteristics, whose response time is significantly earlier than changes in thermodynamic or mechanical states. Conventional monitoring modes synchronize heterogeneous data within a unified time window, causing electrical precursor features with early warning value to be diluted by large inertial physical quantities, reducing the system's perception resolution for early and subtle lesions.

[0003] To avoid false alarms caused by large load fluctuations, the industry generally adopts a statistical error avoidance strategy based on adjusting alarm thresholds according to load variance. In addition to hardware sensing and synchronization constraints, the signal matching algorithm logic ignores the modification of signal morphology by physical operating conditions. For example, Chinese invention patent application CN120971895A discloses a method for locating fault sections in distribution networks based on an improved dynamic time warping algorithm. It introduces derivative sequence optimization DTW algorithm to enhance waveform trend capture accuracy. However, the underlying logic is based on the assumption that the signal morphology only undergoes topological distortion or geometric shift, without addressing the nonlinear evolution mechanism of the strong electromagnetic coupling environment in substations. Under heavy load or core magnetic saturation conditions, transient signals exhibit nonlinear extensions in the time domain driven by electromagnetic damping drift. This load-dependent temporal tailing phenomenon is highly coupled with waveform distortion caused by insulation degradation at the dimensional level. Due to the lack of a dynamic pre-compensation mechanism for underlying operating parameters, the calculated alignment distance includes redundancy in non-fault conditions, causing the system to avoid false alarms by raising the warning threshold. This results in the equipment being in a blind zone of monitoring sensitivity deficiency during the insulation stress operation cycle.

[0004] Therefore, the technical problem to be solved by this invention is how to decouple the interference of complex load fluctuations on high-frequency electrical configurations from the physical mechanism level while maintaining constant detection sensitivity, and to achieve accurate identification of early weak electrical degradation trajectories of power equipment. Summary of the Invention

[0005] This invention proposes a big data-based method for detecting faults in power equipment, comprising the following steps: Step S1: Obtain the broadband electrical signal of the main circuit of the power equipment, and extract the power frequency current sequence and high frequency transient sequence from the broadband electrical signal; Step S2: Extract the energy envelope of the high-frequency transient sequence, identify the local curvature extrema in the energy envelope, and construct a real-time feature sequence that characterizes the transient energy distribution structure. Step S3: Obtain the locally stored health baseline feature sequence. The health baseline feature sequence consists of the transient feature anchor points of the power equipment under no-load conditions and their corresponding initial timestamps. Step S4: Calculate the effective value of the fundamental current in the power frequency current sequence within the characteristic time window, and determine the amplitude of the power frequency current at the moment of triggering the high-frequency transient sequence. Determine the transient electromagnetic damping correction coefficient based on the numerical deviation between the effective value of the fundamental current and the amplitude of the power frequency current. Step S5: Based on the transient electromagnetic damping correction coefficient, perform nonlinear scaling on the time interval between adjacent transient feature anchor points in the health baseline feature sequence to obtain the temporal extended baseline sequence. Step S6: Using the dynamic time warp algorithm, calculate the alignment path distance between the real-time feature sequence and the temporal extension reference sequence, and output the electrical degradation warning signal of the substation when the alignment path distance exceeds the preset constant absolute warning threshold.

[0006] Preferably, step S2, which extracts the energy envelope of the high-frequency transient sequence, includes: performing a Hilbert transform on the high-frequency transient sequence to obtain the instantaneous amplitude envelope; and performing a smoothing filter on the instantaneous amplitude envelope using a sliding window.

[0007] Preferably, the identification of local curvature extrema in step S2 includes the following sub-steps: step S21, calculating the curvature value of each sampling point on the energy envelope; step S22, identifying local maxima and local minima in the curvature values ​​as transient feature anchors; step S23, retaining the transient feature anchors and filtering out the sampling points between adjacent transient feature anchors to form a real-time feature sequence.

[0008] Preferably, in step S3, the health benchmark feature sequence is constructed in the following way: multiple sets of discharge pulse signals are collected during the no-load commissioning phase of the power equipment; the envelope curvature extreme points of the discharge pulse signals are extracted; and the time axis distribution of the same extreme points is statistically averaged to generate a standard time mapping template.

[0009] Preferably, the specific logic of step S5 is as follows: when the amplitude of the power frequency current is higher than the effective value of the fundamental current, the timestamp interval in the standard time mapping template is stretched using the transient electromagnetic damping correction coefficient to establish a reference benchmark that includes electromagnetic damping delay characteristics.

[0010] Preferably, step S6, calculating the alignment path distance, includes: constructing an element spacing matrix between the real-time feature sequence and the temporal extended reference sequence; using a dynamic programming algorithm to search for the regular path with the minimum cumulative path cost in the spacing matrix, and determining the minimum cumulative path cost as the alignment path distance.

[0011] Preferably, step S1 involves extracting the high-frequency transient sequence, which includes: using a high-pass filter to perform filtering processing on the broadband electrical signal with a cutoff frequency of 500kHz to separate the power frequency fundamental component.

[0012] Preferably, the constant absolute warning threshold is calibrated based on the rated insulation level of the power equipment. In the context of implementing temporal extension in step S5, the constant absolute warning threshold remains constant under different load rate conditions. It is used to determine the non-operating condition distortion of the real-time feature sequence relative to the temporal extension reference sequence and output an electrical degradation warning signal, including: counting the cumulative frequency of alignment path distance exceeding the constant absolute warning threshold; and pushing a diagnostic instruction containing degradation rating and fault location index to the operation and maintenance terminal when the cumulative frequency exceeds the set frequency threshold within a preset continuous observation period.

[0013] A big data-driven fault detection system for power equipment, comprising: The signal acquisition module is used to acquire the broadband electrical signals of the main circuit of the power equipment and to extract the power frequency current sequence and high frequency transient sequence from the broadband electrical signals. The feature extraction module, whose input is connected to the signal acquisition module, is used to extract the energy envelope of the high-frequency transient sequence and identify the local curvature extrema in the energy envelope to form a real-time feature sequence. The benchmark management module is used to retrieve the health benchmark feature sequence stored locally. The health benchmark feature sequence consists of transient feature anchor points of the power equipment under no-load conditions and the initial timestamps corresponding to each transient feature anchor point. The damping calculation module is used to determine the transient electromagnetic damping correction coefficient for compensating electromagnetic response lag based on the numerical deviation between the effective value of the fundamental current of the power frequency current sequence within the characteristic time window and the amplitude of the power frequency current at the moment of triggering the high-frequency transient sequence. The temporal reconstruction module, which is connected to the benchmark management module and the damping calculation module respectively, is used to perform nonlinear scaling on the timestamps of the health benchmark characteristic sequence using transient electromagnetic damping correction coefficients to generate a temporally extended benchmark sequence. The fault early warning module is connected to the feature extraction module and the temporal reconstruction module respectively. It is used to calculate the alignment path distance between the real-time feature sequence and the temporal extension reference sequence using the dynamic time warp algorithm, and output the electrical degradation early warning signal of the substation when the alignment path distance exceeds the preset constant absolute early warning threshold.

[0014] Compared with existing technologies, the big data-based substation equipment fault detection method of this invention has the following advantages: 1. In the fault detection of power equipment, by combining high-frequency energy envelope extraction of broadband current with skeleton downsampling based on local curvature extrema, the amount of big data is reduced while retaining the non-stationary morphological evolution characteristics reflecting insulation degradation. Combined with the dynamic time warp algorithm, it can achieve sensitive capture of transient electrical defect topological distortion, solving the technical problem in traditional big data detection where weak precursor features are obscured or diluted due to forced alignment of high and low frequency data.

[0015] 2. An electromagnetic damping correction coefficient based on the instantaneous amplitude of the main circuit power frequency is introduced, and the pre-stored health benchmark skeleton sequence is broadened and mapped on the time axis according to the coefficient. This cancels out the non-faulty tailing interference of the waveform caused by the magnetic saturation of the iron core and the change of electromagnetic damping, ensuring that the system maintains the same limit detection sensitivity as the no-load state under the heavy load or impact conditions of the equipment subjected to electric field and thermal stress. This eliminates the blind spot of heavy load monitoring caused by the statistical compromise strategy of raising the warning threshold commonly used in the industry. Attached Figure Description

[0016] Figure 1 This is a flowchart of the fault detection method integrating adaptive correction of electromagnetic damping according to the present invention; Figure 2 This is a structural diagram of the fault detection system with temporal reshaping function of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0018] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0019] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0020] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0021] A method for fault detection in power equipment using big data includes the following steps: Step S1: Obtain the broadband electrical signal of the main circuit of the power equipment, and extract the power frequency current sequence and high frequency transient sequence from the broadband electrical signal; Step S2: Extract the energy envelope of the high-frequency transient sequence, identify the local curvature extrema in the energy envelope, and construct a real-time feature sequence that characterizes the transient energy distribution structure. Step S3: Obtain the locally stored health baseline feature sequence. The health baseline feature sequence consists of the transient feature anchor points of the power equipment under no-load conditions and their corresponding initial timestamps. Step S4: Calculate the effective value of the fundamental current in the power frequency current sequence within the characteristic time window, and determine the amplitude of the power frequency current at the moment of triggering the high-frequency transient sequence. Determine the transient electromagnetic damping correction coefficient based on the numerical deviation between the effective value of the fundamental current and the amplitude of the power frequency current. Step S5: Based on the transient electromagnetic damping correction coefficient, perform nonlinear scaling on the time interval between adjacent transient feature anchor points in the health baseline feature sequence to obtain the temporal extended baseline sequence. Step S6: Using the dynamic time warp algorithm, calculate the alignment path distance between the real-time feature sequence and the temporal extension reference sequence, and output the electrical degradation warning signal of the substation when the alignment path distance exceeds the preset constant absolute warning threshold.

[0022] Preferably, in step S4, the transient electromagnetic damping correction coefficient is determined by the following logic: obtaining the effective value of the fundamental current of the power frequency current sequence within the characteristic time window. And the amplitude of the power frequency current at the moment of triggering the high-frequency transient sequence. According to the formula Calculate the transient electromagnetic damping correction factor ,in, This is the preset inherent damping response constant.

[0023] Preferably, step S2, which extracts the energy envelope of the high-frequency transient sequence, includes: performing a Hilbert transform on the high-frequency transient sequence to obtain the instantaneous amplitude envelope; and performing a smoothing filter on the instantaneous amplitude envelope using a sliding window.

[0024] Preferably, the identification of local curvature extrema in step S2 includes the following sub-steps: step S21, calculating the curvature value of each sampling point on the energy envelope; step S22, identifying local maxima and local minima in the curvature values ​​as transient feature anchors; step S23, retaining the transient feature anchors and filtering out the sampling points between adjacent transient feature anchors to form a real-time feature sequence.

[0025] Preferably, in step S3, the health benchmark feature sequence is constructed in the following way: multiple sets of discharge pulse signals are collected during the no-load commissioning phase of the power equipment; the envelope curvature extreme points of the discharge pulse signals are extracted; and the time axis distribution of the same extreme points is statistically averaged to generate a standard time mapping template.

[0026] Preferably, the specific logic of step S5 is as follows: when the amplitude of the power frequency current is higher than the effective value of the fundamental current, the timestamp interval in the standard time mapping template is stretched using the transient electromagnetic damping correction coefficient to establish a reference benchmark that includes electromagnetic damping delay characteristics.

[0027] Preferably, step S6, calculating the alignment path distance, includes: constructing an element spacing matrix between the real-time feature sequence and the temporal extended reference sequence; using a dynamic programming algorithm to search for the regular path with the minimum cumulative path cost in the spacing matrix, and determining the minimum cumulative path cost as the alignment path distance.

[0028] Preferably, step S1 involves extracting the high-frequency transient sequence, which includes: using a high-pass filter to perform filtering processing on the broadband electrical signal with a cutoff frequency of 500kHz to separate the power frequency fundamental component.

[0029] Preferably, the constant absolute warning threshold is calibrated based on the rated insulation level of the power equipment. In the context of implementing temporal extension in step S5, the constant absolute warning threshold remains constant under different load rate conditions. It is used to determine the non-operating condition distortion of the real-time feature sequence relative to the temporal extension reference sequence and output an electrical degradation warning signal, including: counting the cumulative frequency of alignment path distance exceeding the constant absolute warning threshold; and pushing a diagnostic instruction containing degradation rating and fault location index to the operation and maintenance terminal when the cumulative frequency exceeds the set frequency threshold within a preset continuous observation period.

[0030] A big data-driven fault detection system for power equipment, comprising: The signal acquisition module is used to acquire the broadband electrical signals of the main circuit of the power equipment and to extract the power frequency current sequence and high frequency transient sequence from the broadband electrical signals. The feature extraction module, whose input is connected to the signal acquisition module, is used to extract the energy envelope of the high-frequency transient sequence and identify the local curvature extrema in the energy envelope to form a real-time feature sequence. The benchmark management module is used to retrieve the health benchmark feature sequence stored locally. The health benchmark feature sequence consists of transient feature anchor points of the power equipment under no-load conditions and the initial timestamps corresponding to each transient feature anchor point. The damping calculation module is used to determine the transient electromagnetic damping correction coefficient for compensating electromagnetic response lag based on the numerical deviation between the effective value of the fundamental current of the power frequency current sequence within the characteristic time window and the amplitude of the power frequency current at the moment of triggering the high-frequency transient sequence. The temporal reconstruction module, which is connected to the benchmark management module and the damping calculation module respectively, is used to perform nonlinear scaling on the timestamps of the health benchmark characteristic sequence using transient electromagnetic damping correction coefficients to generate a temporally extended benchmark sequence. The fault early warning module is connected to the feature extraction module and the temporal reconstruction module respectively. It is used to calculate the alignment path distance between the real-time feature sequence and the temporal extension reference sequence using the dynamic time warp algorithm, and output the electrical degradation early warning signal of the substation when the alignment path distance exceeds the preset constant absolute early warning threshold.

[0031] Example 1: In a hub substation with multiple 220kV transformers operating under heavy load, the main circuit current surges to over 120% of its rated value due to peak summer electricity demand. This causes the transformer core to enter the transient magnetic saturation region, resulting in a drift in the intrinsic electromagnetic damping coefficients of the distributed inductance and capacitance circuits. Under normal insulation conditions, the high-frequency transient electrical components generated by the transformer exhibit a non-fault-related waveform time-axis broadening phenomenon, i.e., the transient energy relaxation process exhibits a tailing broadening in the time dimension. Specifically, the relative permeability of the transformer core decreases sharply in a deeply saturated state. This nonlinear abrupt change in the overall steady-state magnetic characteristics directly alters the local high-frequency skin depth within the core laminations. This surface skin effect... The changes lead to an increase in the equivalent impedance of eddy current losses on the surface of the iron core due to high-frequency electromagnetic waves. At the same time, it causes a dynamic lengthening of the charging and discharging time constant of the high-frequency stray capacitance circuit between the winding and ground and between turns. As a result, the overall 50Hz steady-state magnetic saturation effect is actually coupled to the megahertz frequency band in the physical transmission chain. Ultimately, this manifests as a nonlinear physical extension of the relaxation period of the microsecond-level discharge signal at the electromagnetic radiation level. When faced with this kind of waveform distortion induced by load fluctuations, conventional monitoring methods usually adopt a method of synchronously raising the fault judgment threshold to reduce the false alarm rate. This method artificially causes a decrease in detection sensitivity during the operation cycle of the equipment under thermal stress and strong electric field, making it impossible to identify the microsecond-level discharge characteristics caused by early weak insulation degradation.

[0032] The power equipment fault detection system acquires broadband electrical signals from the main circuit of the power equipment in real time. A high-pass filter separates components with frequencies greater than 500kHz from the broadband electrical signals to extract high-frequency transient sequences reflecting changes in the state of the insulating medium. To obtain power frequency current data containing the true load phase characteristics without inducing high-frequency aliasing, the system diverts the real-time acquired broadband electrical signals to a parallel low-frequency extraction channel. Within this channel, the front-end hardware is equipped with an anti-aliasing low-pass filter circuit, and the back-end digital signal processing unit uses a Butterworth digital low-pass filter algorithm with a cutoff frequency set to 1kHz to smooth and denoise the downsampled signal, accurately removing transient glitches from switching operations and discharge radio frequency interference superimposed on the fundamental frequency. From this, the power frequency current sequence for load characteristic analysis is extracted without distortion. The system's calculation module calculates the effective value of the fundamental current within a preset characteristic time window. And extract the instantaneous power frequency current amplitude corresponding to the triggering time of the high-frequency transient sequence. The transient electromagnetic damping correction coefficient used to compensate for time-domain waveform deviations is determined according to the following calculation formula. : ,in, This is the transient electromagnetic damping correction factor. The damping response constant is pre-calibrated. This represents the instantaneous power frequency current amplitude at the trigger moment of the high-frequency transient sequence. The effective value of the fundamental current within the characteristic time window is used. In the actual operation logic, since the power frequency current itself oscillates periodically with the phase, directly capturing the absolute amplitude of any phase will cause drastic random jumps in the calculation results. Therefore, when determining the instantaneous power frequency current amplitude, the system actually locks the local peak current within a complete power frequency half-wave interval where the transient sequence is triggered. By taking the peak value of the half-wave envelope for calculation, it is ensured that the current data involved in the correction can truly and uniquely represent the maximum magnetization depth borne by the core in the current cycle. This avoids the collapse of the damping correction logic caused by the occasional high-frequency transient discharge near the voltage zero crossing point, thus enabling the electromagnetic damping compensation mechanism to achieve a logical closed loop at the surface trigger level and the overall load level.

[0033] The feature extraction module in the system identifies transient feature anchors by calculating the local curvature extrema of the energy envelope of the high-frequency transient sequence, and removes the sampled data between two adjacent transient feature anchors to construct a real-time feature sequence; the temporal reconstruction module retrieves the health baseline feature sequence generated by the power equipment under no-load and healthy conditions from the local memory, and reads the first... Initial timestamp of each transient feature anchor point and the Initial timestamp of each transient feature anchor point Calculate the initial time difference between the initial timestamps. Using transient electromagnetic damping correction coefficients A scaling transformation is applied to the initial time difference, and the iterative formula is used. Determine the target timestamp after reconstruction To generate a temporally extended reference sequence adapted to the current heavy-load operating conditions, wherein, For the reconstructed first The timestamp of the _th transient feature anchor point is the _th _th _th_ ... The timestamp of each transient feature anchor point As the initial time difference, the fault warning module runs a dynamic time warp algorithm to calculate the alignment path distance by searching for the minimum cumulative path cost after aligning the real-time feature sequence with the temporal extension reference sequence on the time axis. Since the system pre-compensates the waveform time axis deviation caused by magnetic saturation conditions through temporal reshaping, the finally calculated alignment path distance can be stripped of load interference and directly characterize the non-stationary evolution trend of the insulating medium. When the alignment path distance exceeds the preset constant absolute judgment threshold based on the rated insulation level, the system outputs an electrical degradation warning signal for the substation equipment. Even under 120% heavy load impact conditions, the system still maintains the same detection sensitivity as the no-load state, identifying the characteristic distortion caused by extremely weak losses in the insulating medium, thereby realizing the monitoring of equipment health status under extreme power load scenarios.

[0034] Example 2: On a 220kV substation equipment simulation platform built based on a physical prototype, the test environment was implemented by simulating the electromagnetic environment of a hub substation. This platform includes an oil-immersed transformer model with adjustable load characteristics, a high-frequency transient pulse injection unit, and a data acquisition unit. The sampling frequency of the data acquisition unit is set to 10MHz to capture weak insulation degradation signals in the frequency range of 500kHz to 2MHz, with a measurement accuracy of no less than 12 bits to maintain signal quantization fidelity. Gaussian white noise with a signal-to-noise ratio of 20dB is injected into the acquisition circuit, and power frequency harmonic disturbances with a frequency of 50Hz and an amplitude fluctuation range within 5% are superimposed to verify the stability of the technical solution under strong background noise. In this test platform, the sampling period... The settings follow the Nyquist sampling theorem and the logic of balancing data processing load, by identifying the highest effective frequency component of the monitored high-frequency transient sequence. To determine the sampling reference and to avoid signal aliasing, the sampling frequency... Meet no less than The criterion, for partial discharge signals with a maximum frequency of 2MHz, is to set the sampling frequency... The sampling frequency was set to 10MHz, which is the sampling period. The value is 0.1μs. This value ensures that the details of the rising edge of the transient waveform are captured, while controlling the amount of data generated by a single sampling to be within the real-time computing capability of the processor.

[0035] The experiment was conducted simultaneously in three dimensions: the experimental group of this invention, the control group, and the partially missing control group. The experimental group of this invention fully employed a detection method incorporating a temporal reconstruction mechanism, the control group used a traditional fixed threshold determination method, and the partially missing control group removed the transient electromagnetic damping correction coefficient. In the compensation stage, only the skeleton downsampling processing is retained; insulation degradation signals of different intensities are injected into the transformer model through a high-frequency transient pulse injection unit, while the main circuit current is adjusted to generate dynamic changes from 50% to 130% of the rated load. Under the heavy load condition of 120%, the waveform broadening phenomenon of the high-frequency transient sequence in the original input data due to the response lag caused by core saturation is observed, and its wave tail relaxation time is extended from 4.5μs under no-load conditions to 6.2μs; the sample group of this invention uses the calculation module to obtain the effective value of the fundamental current. The amplitude of the power frequency current at the moment of triggering is 1000A. The value is 1697A, and the damping response constant is... Under the condition of 0.15, the transient electromagnetic damping correction coefficient is determined according to the calculation formula. The coefficient is 1.104. The temporal reconstruction module applies nonlinear stretching to the time axis of the health baseline feature sequence based on this coefficient, so that the reconstructed temporal extended baseline sequence is aligned with the real-time feature sequence in waveform topology.

[0036] To verify the correlation between technical effectiveness and problem severity, the experiment was designed to cover insulation degradation gradients of low, medium, and high strength, corresponding to simulated discharge quantities of 50pC, 200pC, and 1000pC, respectively. The results showed that under a 120% heavy load environment, the control group experienced increased alignment errors due to waveform distortion, resulting in an accuracy rate of only 62.4% at the low strength gradient of 50pC. In contrast, the present invention's sample maintained a stable accuracy rate of 97.8% under the same conditions, demonstrating its ability to capture weak precursory features. Empirical verification of the parameter boundaries revealed that when the main circuit current was within the operating window of 80% to 120% of the rated load, the alignment path distance of the present invention's sample consistently remained below 1.5, demonstrating adaptability to various operating conditions. However, when the load further increased and... and After the ratio exceeds the performance inflection point of 1.3, the permeability decreases due to the transformer core entering the deep saturation region, and the electromagnetic response exhibits a nonlinear saturation effect. At this point, the transient electromagnetic damping correction coefficient... The compensation effect tends to saturate, and the alignment path distance jumps from 1.6 to 3.8, indicating that the system has exceeded the physical linear compensation boundary. Through the experimental verification of the above multi-dimensional comparison system, it is confirmed that the transient electromagnetic damping correction coefficient is effective. The technical logic of temporal reshaping of the health baseline feature sequence can offset the waveform time axis deviation under heavy load conditions. Compared with the partially missing control group, the detection sensitivity of the sample group of this invention is improved by more than 35.4% under heavy load conditions, and the determinism of the output results is still maintained under a background noise environment of 20dB. The experimental results show that the collaborative alignment mechanism of temporal extended baseline sequence and real-time feature sequence solves the technical problem of mutual coupling between load fluctuation and insulation abnormality characteristics, verifies the practical value of the technical solution in improving the accuracy of fault prediction of substation equipment, and provides engineering data support for achieving precise health management in complex power grid environments.

[0037] Example 3: In the commissioning environment of a 220kV hub substation, the inherent damping response constant of the power equipment is determined by measuring the electrical parameters of the power equipment under no-load and 80% rated load conditions. An initial health baseline feature sequence is constructed, and the calculation module obtains the effective value of the fundamental current under no-load conditions. And the initial relaxation time when the amplitude of the corresponding high-frequency transient sequence decays to 10% of its peak value; read the instantaneous power frequency current amplitude under 80% load conditions. And the disturbed relaxation time of the corresponding high-frequency transient sequence; the transient electromagnetic damping correction coefficient is determined by the ratio of the disturbed relaxation time to the initial relaxation time. The measured value is used, and the damping response constant is derived in reverse according to the following calculation formula. : ,in, This is the transient electromagnetic damping correction factor. The inherent damping response constant is This refers to the instantaneous power frequency current amplitude. This represents the effective value of the fundamental current.

[0038] The feature extraction module extracts the instantaneous energy envelope of the high-frequency transient sequence using Hilbert transform; the discrete three-point moving window algorithm is used to calculate the sampling points. The numerical curvature at the sampling point is calculated by... With sampling points Sampling points The amplitude difference between the two values ​​determines the first-order difference. A second-order difference operation is performed on the first-order difference sequence to obtain the numerical curvature. Due to the physical limitations of the microprocessor's computing power at the substation monitoring site, the system abandons theoretical curvature functions involving complex derivative calculations and directly defines the obtained discrete second-order difference sequence results as an engineering curvature index characterizing the intensity of transient energy envelope topological abrupt changes. The system compares the absolute value of the numerical curvature with a fluctuation threshold calibrated based on the background electromagnetic noise level. If the sampling point... If the absolute value of the numerical curvature at a point reaches a local maximum and exceeds the fluctuation threshold, the sampling point is determined as a transient feature anchor point and its corresponding timestamp is recorded by the reference management module, thereby converting the broadband electrical signal into a real-time feature sequence characterizing the transient energy distribution structure.

[0039] The benchmark management module updates the health benchmark feature sequence using a sliding window method triggered by a safety judgment interval. The system sets a sliding time window of 30 natural days. The fault early warning module calculates the average daily alignment path distance within the window. When the average value is continuously lower than 20% of the early warning threshold, the system triggers an update based on the equipment status: extracting 50 sets of transient feature sequences collected within the window after power frequency load deviation compensation, and generating a reconstructed health benchmark feature sequence by calculating the arithmetic mean of the timestamps of the transient feature anchor points with the same index in each set of sequences. This reconstructed health benchmark feature sequence is stored in the local memory. The health benchmark feature sequence shifts with the change of the intrinsic impedance of the transformer equipment insulation, enabling the system to maintain the detection sensitivity to small increments of insulation degradation without manual intervention, thus achieving long-term monitoring of the operating status of the transformer equipment.

[0040] Example 4: In the field deployment and commissioning scenario for a new type of oil-immersed transformer, the control data acquisition unit of the power equipment fault detection system continuously collects 60 power frequency cycles of the intrinsic electromagnetic noise sequence in the silent state when the power equipment is not closed. The calculation module extracts the energy envelope of the intrinsic electromagnetic noise sequence through Hilbert transform and calculates the standard deviation σ of the curvature distribution of the full-time domain sampling points. The fluctuation threshold H is determined based on three times the standard deviation σ of the curvature distribution when the feature extraction module identifies the transient feature anchor point. In the thermal stability stage after the power equipment is closed and enters the rated voltage no-load operation, the system collects 100 consecutive sets of high-frequency transient sequences and identifies the original anchor point in each set of sequences based on the fluctuation threshold H. The timestamp and curvature amplitude of the original anchor point are uniformized using the arithmetic mean method to generate the initial health benchmark feature sequence.

[0041] When the system encounters an intrinsic impedance shift in the insulating medium due to changes in ambient temperature and humidity, the temperature and humidity baseline adaptive calibration procedure is initiated. The data acquisition unit acquires multiple sets of steady-state power frequency leakage current effective values ​​of the substation within the range of ambient temperature from 5℃ to 40℃ and relative humidity from 30% to 90%. The calculation module uses a physical state of 20℃ and 60% relative humidity as the measurement baseline node, calculates the ratio of the leakage current at each other measurement point to the leakage current value under the baseline state, and constructs a two-dimensional spatial interpolation lookup table containing temperature and humidity as dual variables. The system collects the ambient temperature and humidity parameters of the field equipment in real time, retrieves the corresponding bias multiplier from the lookup table, and sets it as the temperature and humidity compensation coefficient. The curvature amplitude of the transient feature anchor points in the real-time feature sequence is adjusted using the preset temperature and humidity compensation coefficient. The values ​​are normalized to eliminate the influence of environmental parameters on the fault judgment logic. After calculating the alignment path distance, the fault early warning module retrieves the insulation loss correlation mapping table stored in the local memory. Based on the numerical range of the alignment path distance, it determines the degradation index of the substation equipment. When the alignment path distance fluctuates within the range of 0.5 to 1.2, the system determines it as a stable operating state in which the insulation medium is in a weak polarization stage and no intervention measures are required. When the alignment path distance exceeds the inflection point of 2.5, the system predicts the physical time span of insulation breakdown by calculating the rate of change of the first derivative of the alignment path distance over time. This converts the collected broadband electrical signals into substation insulation health monitoring parameters that can be directly parsed by the operation and maintenance terminal.

[0042] Example 5: In an engineering scenario where a monitoring system is deployed in a 220kV substation that has been in operation for more than 15 years, the substation equipment fault detection system performs pre-compensation for high-order harmonics and long-distance signal attenuation. A standard discharge pulse with a pulse width of 1μs and a rise time of 100ns is injected into the data acquisition unit using a calibration signal source in a shielded environment. The system adjusts the sliding analysis window length... To capture local extrema of transient energy distribution, where the sliding analysis window length... The rule for determining the value is the product of the rise time of the standard discharge pulse and the sampling frequency. The sliding analysis window length is set at a sampling frequency of 10MHz. To maintain the accuracy of curvature extremum point location while filtering out fluctuations caused by high-frequency random noise, the system employs a time-domain range of 10 sampling points. When constructing an insulation loss correlation mapping table by conducting stepped pressure tests from 20kV to 100kV on insulation samples of the same specification in a controlled laboratory environment, the calculation module records the alignment path distance between the real-time characteristic sequences generated at each voltage level and the health baseline characteristic sequence. It then uses the least squares method to perform linear fitting on the acquired data sequences, establishing a functional relationship between the alignment path distance and discharge intensity within a numerical range of 0.5 to 5.0, and storing this relationship in local memory. Based on this, the system assesses the degree of grid load fluctuation by calculating the steady-state variance of the main circuit power frequency current sequence. When a non-periodic burst pulse with an amplitude change rate exceeding 15% of the rated value is detected in the load current, the time-reconstruction module lowers the transient electromagnetic damping correction coefficient in the iterative formula. The step weights are used to reduce the data fluctuation interference caused by the phase change caused by the switching of high-power reactive power compensation devices on the time domain reshaping logic. The final output insulation health monitoring parameters reflect the intrinsic loss level of the insulation medium of the power equipment.

[0043] The data acquisition unit of the power equipment fault detection system uses a clock reference generated by a temperature-compensated crystal oscillator to synchronously sample broadband electrical signals, ensuring that the extracted power frequency current sequence and high-frequency transient sequence are aligned in a microsecond-level time axis coordinate system. The calculation module identifies the first sampling point in the high-frequency transient sequence whose amplitude exceeds the fluctuation threshold as the trigger point, and extracts the instantaneous sampled value at the corresponding timestamp position of the power frequency current sequence as the instantaneous power frequency current amplitude. This eliminates phase asynchrony deviation during cross-frequency band data parsing. When executing the dynamic time warping algorithm, the fault warning module constructs a cost matrix by calculating the Euclidean distance between each transient feature anchor point in the real-time feature sequence and the corresponding feature anchor point in the temporal extended reference sequence. It then uses dynamic programming to search for the warped path that minimizes the cumulative cost, thereby converting the surface morphological distortion of insulation degradation into an aligned path distance reflecting the degree of physical damage to the equipment. Specifically, the system internally presets a weight multiplier with an initial value of 1 as the step size. When the grid load is stable, the weight multiplier remains at 1 and implicitly participates in the coefficient calculation. When the aforementioned non-periodic sudden pulse exceeds the limit, the system control unit abruptly reduces the step weight to 0.2 and directly multiplies this dynamically changed step weight as an attenuation factor onto the aforementioned inherent damping response constant, forcing the transient electromagnetic damping correction coefficient of the short-term surge to converge. After the load current amplitude change rate falls back to within 15% for 5 consecutive power frequency cycles, the system gradually restores the step weight to the constant 1 with a fixed slope of 0.1 per cycle.

[0044] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A method for fault detection in power equipment based on big data, characterized in that, Includes the following steps: Step S1: Obtain the broadband electrical signal of the main circuit of the power equipment, and extract the power frequency current sequence and high frequency transient sequence from the broadband electrical signal; Step S2: Extract the energy envelope of the high-frequency transient sequence, identify the local curvature extrema in the energy envelope, and construct a real-time feature sequence that characterizes the transient energy distribution structure. Step S3: Obtain the locally stored health baseline feature sequence. The health baseline feature sequence consists of the transient feature anchor points of the power equipment under no-load conditions and their corresponding initial timestamps. Step S4: Calculate the effective value of the fundamental current in the power frequency current sequence within the characteristic time window, and determine the amplitude of the power frequency current at the moment of triggering the high-frequency transient sequence. Determine the transient electromagnetic damping correction coefficient based on the numerical deviation between the effective value of the fundamental current and the amplitude of the power frequency current. Step S5: Based on the transient electromagnetic damping correction coefficient, perform nonlinear scaling on the time interval between adjacent transient feature anchor points in the health baseline feature sequence to obtain the temporal extended baseline sequence. Step S6: Using the dynamic time warp algorithm, calculate the alignment path distance between the real-time feature sequence and the temporal extension reference sequence, and output the electrical degradation warning signal of the substation when the alignment path distance exceeds the preset constant absolute warning threshold.

2. The method for detecting faults in power equipment using big data according to claim 1, characterized in that, Step S2 involves extracting the energy envelope of the high-frequency transient sequence, including: performing a Hilbert transform on the high-frequency transient sequence to obtain the instantaneous amplitude envelope; and applying a smoothing filter to the instantaneous amplitude envelope using a sliding window.

3. The method for detecting faults in power equipment using big data according to claim 1, characterized in that, Step S2 identifies local curvature extrema, including the following sub-steps: Step S21, calculate the curvature value of each sampling point on the energy envelope; Step S22, identify local maxima and local minima in the curvature values ​​as transient feature anchors. Step S23: Retain transient feature anchor points and filter out sampling points between adjacent transient feature anchor points to form a real-time feature sequence.

4. The method for detecting faults in power equipment using big data according to claim 1, characterized in that, In step S3, the health benchmark feature sequence is constructed in the following way: multiple sets of discharge pulse signals are collected during the no-load commissioning phase of the substation; the envelope curvature extreme points of the discharge pulse signals are extracted; and the time axis distribution of similar extreme points is statistically averaged to generate a standard time mapping template.

5. The method for detecting faults in power equipment using big data according to claim 1, characterized in that, The specific logic of step S5 is as follows: when the amplitude of the power frequency current is higher than the effective value of the fundamental current, the timestamp interval in the standard time mapping template is stretched using the transient electromagnetic damping correction coefficient to establish a reference benchmark that includes electromagnetic damping delay characteristics.

6. The method for detecting faults in power equipment using big data according to claim 1, characterized in that, Step S6 calculates the alignment path distance, including: constructing an element spacing matrix between the real-time feature sequence and the temporal extended reference sequence; using a dynamic programming algorithm to search for the regular path with the minimum cumulative path cost in the spacing matrix, and determining the minimum cumulative path cost as the alignment path distance.

7. The method for detecting faults in power equipment using big data according to claim 1, characterized in that, Step S1 involves extracting the high-frequency transient sequence, including: using a high-pass filter to filter the broadband electrical signal with a cutoff frequency of 500kHz to separate the power frequency fundamental component.

8. The method for detecting faults in power equipment using big data according to claim 1, characterized in that, The constant absolute warning threshold is calibrated based on the rated insulation level of the substation equipment. In the context of implementing temporal extension in step S5, the constant absolute warning threshold remains constant under different load rate conditions. It is used to determine the non-operating condition distortion of the real-time characteristic sequence relative to the temporal extension reference sequence and output an electrical degradation warning signal, including: counting the cumulative frequency of alignment path distance exceeding the constant absolute warning threshold; and pushing a diagnostic instruction containing degradation rating and fault location index to the operation and maintenance terminal when the cumulative frequency exceeds the set frequency threshold within the preset continuous observation period.

9. A big data-based substation equipment fault detection system, used to implement the big data-based substation equipment fault detection method of claim 1, characterized in that, The system includes: The signal acquisition module is used to acquire the broadband electrical signals of the main circuit of the power equipment and to extract the power frequency current sequence and high frequency transient sequence from the broadband electrical signals. The feature extraction module, whose input is connected to the signal acquisition module, is used to extract the energy envelope of the high-frequency transient sequence and identify the local curvature extrema in the energy envelope to form a real-time feature sequence. The benchmark management module is used to retrieve the health benchmark feature sequence stored locally. The health benchmark feature sequence consists of transient feature anchor points of the power equipment under no-load conditions and the initial timestamps corresponding to each transient feature anchor point. The damping calculation module is used to determine the transient electromagnetic damping correction coefficient for compensating electromagnetic response lag based on the numerical deviation between the effective value of the fundamental current of the power frequency current sequence within the characteristic time window and the amplitude of the power frequency current at the moment of triggering the high-frequency transient sequence. The temporal reconstruction module, which is connected to the benchmark management module and the damping calculation module respectively, is used to perform nonlinear scaling on the timestamps of the health benchmark characteristic sequence using transient electromagnetic damping correction coefficients to generate a temporally extended benchmark sequence. The fault early warning module is connected to the feature extraction module and the temporal reconstruction module respectively. It is used to calculate the alignment path distance between the real-time feature sequence and the temporal extension reference sequence using the dynamic time warp algorithm, and output the electrical degradation early warning signal of the substation when the alignment path distance exceeds the preset constant absolute early warning threshold.

Images