Power supply and distribution safety monitoring method and system for transformer substation

Abstract

The invention relates to the technical field of transformer substation monitoring, and discloses a power supply and distribution safety monitoring method and system for a transformer substation. The method comprises the following steps: acquiring various data in a transformer substation, and performing time synchronization calibration on the three types of data; dividing a power supply and distribution operation mode into a steady-state mode, a transient mode and an overload mode based on the calibrated data, extracting a harmonic distortion rate, a dielectric loss angle tangent value and a vibration spectrum entropy value of a current waveform in each mode, and generating an operation characteristic matrix; detecting an operation mode switching event, and calculating a characteristic offset between modes; decomposing a degradation characteristic component in the operation characteristic matrix according to a matching degree between an insulation aging stage and current dielectric loss data in a historical fault case library; coupling compensation correction is performed on the degradation characteristic component in combination with the cross influence relationship between the electrical parameters and the mechanical parameters of the equipment; and reversely tracing the abnormal signal source equipment node along the topological structure of the power supply and distribution network based on the corrected degradation characteristic component, and outputting an equipment maintenance sequence.

Description

Technical Field

[0001] This invention relates to the field of substation monitoring technology, specifically to a method and system for monitoring the safety of power supply and distribution in substations. Background Technology

[0002] As a critical node in the power system for energy conversion, voltage transformation, and power distribution, the safe and stable operation of the power supply and distribution equipment within substations is a vital foundation for ensuring the reliability of the power grid. With rapid economic and social development, electricity load demand continues to grow, and load characteristics are becoming increasingly complex. Transformers, circuit breakers, disconnectors, switchgear, and other power supply and distribution equipment within substations are operating under high load and frequent operation for extended periods. Simultaneously, the large-scale integration of new energy sources and the increase in nonlinear loads have led to a more complex power grid operating environment, significantly increasing the electrical, thermal, and mechanical stresses faced by substation equipment. Against this backdrop, traditional equipment maintenance models, primarily based on periodic inspections, preventative testing, and reactive maintenance, are no longer sufficient to meet the high reliability requirements of modern power grids.

[0003] Existing substation monitoring systems primarily rely on threshold monitoring of single physical quantities, such as triggering alarms for exceeding limits of current and voltage RMS values, or online monitoring of equipment temperature. These methods have significant limitations: threshold alarms are reactive and cannot predict equipment status trends or provide early warnings of faults; single-parameter monitoring struggles to comprehensively reflect the true health status of equipment; for example, potential defects such as slow aging of insulation or slight loosening of mechanical structures often do not cause significant changes in electrical parameters in their early stages; and substation equipment operating conditions exhibit distinct multimodal characteristics, with significant differences in parameter characteristics under different conditions such as normal steady-state operation, transient operation, and overload operation. Using fixed thresholds for judgment can easily lead to false alarms or missed alarms.

[0004] Failures in power supply and distribution equipment are usually the result of multiple coupled factors. Taking switchgear as an example, its insulation performance deterioration may be simultaneously affected by multiple factors such as electrical aging, mechanical vibration, and environmental temperature and humidity. Current monitoring systems lack the ability to deeply integrate and correlate multi-source heterogeneous data. Electrical, mechanical, and insulation monitoring data often form information silos, failing to effectively reveal the inherent laws governing equipment condition evolution. Equipment degradation is a gradual process. Extracting characteristic indicators related to equipment lifespan loss from massive amounts of monitoring data and establishing correlation models between these indicators and historical failure cases is crucial for achieving predictive maintenance. Therefore, there is an urgent need for an intelligent monitoring method that can integrate multi-source information, identify operating modes, quantify the degree of equipment degradation, and achieve precise fault location to improve the safe operation level of substations. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for monitoring the safety of power supply and distribution in substations, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for power supply and distribution safety monitoring in substations, the method comprising: Collect current waveform data, insulation dielectric loss data, and switchgear mechanical vibration data of power supply and distribution equipment in the substation, and perform time synchronization calibration on the three types of data. Based on the calibrated data, the power supply and distribution operation modes are divided into steady-state mode, transient mode and overload mode. The harmonic distortion rate, dielectric loss tangent and vibration spectrum entropy of the current waveform under each mode are extracted to generate the operation feature matrix. Detect running mode switching events. If the number of mode switching times per unit time exceeds the dynamic threshold, perform dynamic alignment on the feature matrices of adjacent modes and calculate the feature offset between modes. Based on the matching degree between the insulation aging stage in the historical fault case library and the current dielectric loss data, the deterioration feature components in the operation feature matrix are decomposed. By combining the cross-influence relationship between the electrical and mechanical parameters of the equipment, coupled compensation correction is performed on the degradation characteristic components; Based on the corrected degradation feature components, the abnormal signal source device node is traced backward along the power supply and distribution network topology, and the device maintenance sequence is output.

[0007] Preferably, the time synchronization calibration of the three types of data includes: The sampling start point is aligned using the zero-crossing detection method for current waveform data; the insulation dielectric loss data is interpolated in segments according to the power frequency period; and the fundamental frequency phase is extracted from mechanical vibration data through short-time Fourier transform. A sliding time window is used to compensate for the time scale deviation of the three types of data, generating a synchronized and calibrated data stream.

[0008] Preferably, the step of extracting the harmonic distortion rate, dielectric loss tangent, and vibration spectrum entropy of the current waveform under each mode to generate the operating feature matrix includes: The proportion of odd harmonic content is calculated from the current waveform data in steady-state mode, the peak value of polarization loss is extracted from the dielectric loss data in transient mode, and the proportion of high-frequency energy is calculated from the vibration data in overload mode. The three types of features are aggregated into a three-dimensional matrix according to the time window. The matrix dimensions correspond to the harmonic distortion rate, the dielectric loss tangent, and the vibration spectrum entropy.

[0009] Preferably, the dynamic alignment of the feature matrices of adjacent patterns includes: The time axis of the feature matrix of adjacent modes is adjusted by dynamic time warping algorithm, and the gradient of harmonic distortion rate change, the offset of dielectric loss tangent and the difference of vibration spectrum entropy are calculated. The weighted sum of the three types of offsets is output as the inter-mode feature offset.

[0010] Preferably, the degraded feature components in the decomposed running feature matrix include: When the matching degree between dielectric loss data and insulation aging stage is higher than the preset threshold, the low-frequency drift component of dielectric loss tangent is separated from the operating feature matrix. When the matching degree is lower than the preset threshold, the components of the high-frequency vibration energy concentration band are extracted more effectively based on the correlation between the vibration spectrum entropy value and the wear degree of the switching mechanism.

[0011] Preferably, the coupling compensation correction performed on the degraded feature components includes: Compensation is provided for temperature rise deviation caused by harmonic distortion rate based on the current thermal effect coefficient. The amplitude-frequency characteristic of the vibration spectrum entropy value is corrected based on the mechanical vibration transfer function; The temperature rise deviation after compensation and the amplitude-frequency characteristic after correction are combined to generate the coupling compensation correction result.

[0012] Preferably, the device nodes for tracing the abnormal signal source along the power supply and distribution network topology include: Construct an impedance connection diagram of the power supply and distribution network topology, where nodes represent devices and edges represent impedance values ​​and fault propagation weights; Identify abnormal harmonic distortion rate, dielectric loss tangent or vibration spectrum entropy from the coupling compensation correction results, and match the corresponding nodes in the impedance connection diagram. Traverse the energy transfer in the reverse direction along the impedance connection diagram to locate the device node of the abnormal signal source.

[0013] Preferably, the output device maintenance sequence includes: The number of times the harmonic distortion rate of the abnormal signal source equipment node exceeds the standard, the magnitude of the dielectric loss tangent value exceeding the standard, and the frequency of the vibration spectrum entropy value exceeding the standard are statistically analyzed. The equipment maintenance priority sequence is generated by sorting the three categories of exceedance indicators in descending order by weighted sum.

[0014] Preferably, the method further includes: Perform an integrity check on the synchronized data stream; if the check fails, trigger a data re-acquisition command. Mark the timestamp abnormal segments in the validated data stream for exclusion in the subsequent feature extraction stage.

[0015] Preferably, the present invention also includes a power supply and distribution safety monitoring system for substations, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the aforementioned power supply and distribution safety monitoring method for substations.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves refined monitoring and intelligent diagnosis of the status of substation power distribution equipment by constructing a complete technology chain encompassing multi-source data fusion, operation mode identification, degradation feature extraction, and fault location. Time synchronization calibration is performed on three types of heterogeneous data: current waveform, insulation dielectric loss, and mechanical vibration. This solves the problem of inconsistent time scales in multi-sensor data acquisition, providing a unified time benchmark for subsequent correlation analysis and ensuring data comparability and analytical accuracy. Based on the calibrated data, three operation modes—steady-state, transient, and overload—are defined, and characteristic parameters of each mode are extracted to generate an operation feature matrix. This allows for differentiated status assessment for different operating conditions, avoiding misjudgments by a single assessment standard under different conditions and significantly improving the adaptability of status identification.

[0017] A mechanism for detecting operating mode switching events is introduced. When the number of mode switching times per unit time exceeds a dynamic threshold, the feature matrices of adjacent modes are automatically dynamically aligned and the feature offset is calculated. This effectively captures abnormal response characteristics of equipment during frequent operating condition transitions. This dynamic analysis mechanism enhances the monitoring capability of equipment transient processes and transition states, providing an effective means to identify intermittent faults and poor operating condition adaptability. By combining the matching degree between insulation aging stages in the historical fault case library and current dielectric loss data, the degradation feature components in the operating feature matrix are decomposed. This allows the system to learn from historical experience and more accurately identify early fault symptoms related to insulation aging, achieving intelligent diagnosis based on case-based reasoning.

[0018] Considering the cross-influence between electrical and mechanical parameters of equipment, coupled compensation correction is applied to the degradation characteristic components, eliminating the limitations of single-parameter analysis and making the diagnostic results more consistent with the actual operating mechanism of the equipment. This multi-physics coupled analysis method can more comprehensively reflect the overall health status of the equipment and improve the reliability of the diagnostic results. Based on the corrected degradation characteristic components, the abnormal signal source equipment nodes are traced backward along the power supply and distribution network topology, which can accurately locate faulty equipment or potential risk points. The output equipment maintenance sequence provides clear guidance for operation and maintenance decisions, achieving precise positioning from "surface" to "point".

[0019] This method elevates substation monitoring from traditional single-parameter threshold alarms to multi-source data fusion analysis, from static assessment to dynamic pattern recognition, and from reactive maintenance to predictive maintenance. By establishing a complete "data acquisition-feature extraction-state assessment-fault location" technical system, it significantly improves the intelligence level of substation equipment status monitoring, providing strong support for the safe and stable operation of the power grid. Simultaneously, this method effectively reduces false alarm and missed alarm rates, improves operation and maintenance efficiency, and reduces unnecessary power outage maintenance time, demonstrating significant engineering application value. Attached Figure Description

[0020] Figure 1 Flowchart of the core steps of a power supply and distribution safety monitoring method for substations; Figure 2 A comprehensive analysis diagram of the operating status of the substation power supply and distribution safety monitoring system; Figure 3 A flowchart for time synchronization calibration of three types of data; Figure 4 A flowchart for dynamically aligning adjacent pattern feature matrices; Figure 5 This is a diagram showing the characteristic decomposition and coupling compensation analysis of insulating dielectric loss. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1This invention provides a method and system for power supply and distribution safety monitoring in substations. The method includes collecting current waveform data, insulation dielectric loss data, and switchgear mechanical vibration data of power supply and distribution equipment within the substation via a sensor network. A high-precision analog-to-digital converter module is used during the acquisition process to ensure the original data fidelity. After data acquisition, time synchronization calibration is performed on the three types of data to eliminate time scale deviations caused by sensor response delays or sampling frequency differences. The synchronized data stream is input into an operation mode classification module, which classifies the power supply and distribution operation modes into steady-state mode, transient mode, and overload mode based on preset threshold conditions. In each mode, the harmonic distortion rate of the current waveform is calculated using a fast Fourier transform, the dielectric loss tangent of the insulation dielectric loss data is extracted using a bridge method, and the vibration spectrum entropy value of the switchgear mechanical vibration data is derived based on the Shannon entropy formula. The three types of feature values ​​are aggregated into an operation feature matrix according to a time window, with the matrix dimension fixed as a triplet of harmonic distortion rate, dielectric loss tangent, and vibration spectrum entropy value. The operation mode switching event is monitored by the event detection module. If the number of mode switching times exceeds the dynamic threshold within a unit of time, the dynamic alignment module performs dynamic time warping on the feature matrices of adjacent modes and calculates the feature offset between modes. The historical fault case library stores typical data patterns of the insulation aging stage. The matching degree calculation module compares the current dielectric loss data with the case library. When the matching degree is higher than the set threshold, the feature decomposition module separates the deteriorated feature components from the operation feature matrix. The coupling compensation module performs temperature rise deviation compensation and amplitude-frequency characteristic correction on the deteriorated feature components based on the cross-influence relationship between the electrical and mechanical parameters of the equipment. The anomaly tracing module combines the power supply and distribution network topology and traverses backward along the impedance connection diagram to locate the abnormal signal source equipment node. The maintenance sequence generation module counts the node exceeding indicators and outputs the equipment maintenance priority sequence in descending order of weighted sum. The entire method flow is implemented on an embedded processor, and the data flow is scheduled by a real-time operating system to ensure timeliness.

[0023] Example 1: Refer to the figure Figure 2 and Figure 3The current waveform data employs a zero-crossing detection method to align the sampling start point. This method uses a high-speed comparator circuit to monitor the zero-crossing moment of the current signal in real time. When the current waveform crosses the zero line from a negative value into the positive value region, the comparator outputs a digital pulse signal. This pulse signal serves as the synchronous trigger signal for the data acquisition system, and the sampling start point is strictly aligned to this trigger edge. The digital comparator circuit uses a Schmitt trigger structure to suppress noise interference, and the trigger threshold is set to a small hysteresis window near the zero value to avoid false triggering due to signal jitter. Insulation dielectric loss data is interpolated in segments according to the power frequency cycle, which is fixed at 20 milliseconds corresponding to a 50 Hz power grid frequency. The segmented interpolation process divides one power frequency cycle into 200 equally spaced time points. Within each segment, a linear interpolation algorithm is used to resample the original sampled data. The linear interpolation algorithm inserts new data points by calculating the slope between adjacent sampling points, ensuring that the sampling rate of the dielectric loss data remains consistent with the 10 kHz of the current waveform data. The fundamental frequency phase of the mechanical vibration data is extracted by short-time Fourier transform. The short-time Fourier transform uses a Hanning window with a length of 10 milliseconds to perform frame processing on the vibration signal. The window function overlap is set to 50% to balance the time resolution and frequency resolution. The fundamental frequency phase is identified from the vibration spectrum by a peak detection algorithm to determine the phase angle corresponding to the fundamental frequency component.

[0024] A sliding time window compensates for timescale deviations in the three types of data. The sliding time window is configured to be 1 second long and contains 1000 data samples. The timescale deviation compensation algorithm estimates the relative delay by calculating the correlation coefficient between the timestamps of the three types of data. The delay estimation uses the maximum value method of the cross-correlation function. The time synchronization calibration module integrates a high-precision temperature-controlled crystal oscillator clock with a clock frequency stability of 0.1 ppm. The clock signal is synchronized with the GPS clock source via the IEEE 1588 protocol. During data stream synchronization, current waveform data, insulation dielectric loss data, and mechanical vibration data are stored in three first-in-first-out buffers, with a buffer depth of 10,000 sample points. The data readout controller sequentially extracts the aligned data from the buffers according to a unified time base. The digital comparator circuit of the zero-crossing detection method adopts a differential input structure. The input signal is preprocessed by a bandpass filter with a cutoff frequency set to 45 Hz to 55 Hz to eliminate high-frequency noise and power frequency harmonic interference. The power frequency periodic interpolation of the insulation dielectric loss data is performed by a dedicated interpolation processor. The interpolation processor has a built-in linear interpolation calculation unit, inserting 199 new data points in each power frequency period. The short-time Fourier transform of the mechanical vibration data is implemented in a digital signal processor, which is equipped with a dedicated FFT calculation unit. The fundamental frequency phase extraction is performed by calculating the ratio of the real part to the imaginary part of the spectrum using the arctangent function.

[0025] The generated data stream after synchronization calibration employs a data fusion algorithm, which packages the three types of calibrated data into a unified data frame. Each data frame contains a timestamp, data value, and quality flag. Data stream transmission is achieved via Gigabit Industrial Ethernet, using a precise time protocol to synchronize the clocks of each node. A data synchronization verification unit performs integrity checks before data flow, including data length verification, timestamp continuity verification, and CRC error detection. Zero-crossing detection of current waveform data is implemented in parallel processing within a field-programmable gate array (FPGA), which is configured with multiple comparator channels to simultaneously process multiple current signals. The power frequency period segmented interpolation of insulation dielectric loss data is triggered by a timer interrupt service routine in the microcontroller, with the timer interrupt period strictly matching the power frequency period. The short-time Fourier transform of mechanical vibration data is accelerated using a dedicated spectrum analysis chip, which supports real-time window function processing and FFT transform.

[0026] The hardware platform of the time synchronization calibration system adopts a multi-core processor architecture. One processor core is dedicated to zero-crossing detection and sampling of current waveform data, another processor core handles interpolation calculations for insulation dielectric loss data, and a third processor core performs spectral analysis of mechanical vibration data. The processor cores exchange synchronization information via shared memory, which stores a unified time reference counter. The data acquisition card is equipped with a synchronous sampling analog-to-digital converter (ADC), whose sampling clock is driven by the same clock source, eliminating sampling time deviations between channels. The synchronized data stream is stored in a circular buffer with a configurable size, and the data outflow rate is kept stable through a flow control algorithm. The timestamp generation unit uses a high-precision time counter with nanosecond-level resolution, and each data sample is accompanied by a 64-bit timestamp. The data synchronization calibration algorithm periodically performs timescale deviation estimation and compensation. The compensation value is dynamically adjusted by a PID controller, whose proportional, integral, and derivative parameters are tuned according to the system delay characteristics.

[0027] The zero-crossing detection method for current waveform data includes a digital filtering preprocessing stage. The digital filter employs a finite-length unit impulse response filter structure, and the filter coefficients are stored in read-only memory. The power frequency periodic interpolation algorithm for insulation dielectric loss data includes an outlier removal function, which is based on the standard deviation detection of data within a sliding window. The fundamental frequency phase extraction for mechanical vibration data includes a phase unwrapping step, and the phase unwrapping algorithm corrects phase jumps to ensure phase continuity. The time scale deviation compensation for the sliding time window employs an adaptive filtering algorithm, and the step size parameter of the adaptive filtering algorithm is dynamically adjusted according to the signal-to-noise ratio. The synchronously calibrated data stream output interface supports multiple industrial standard protocols, including Modbus TCP, OPCUA, and IEC61850, facilitating integration with higher-level monitoring systems. The data acquisition front end is equipped with signal conditioning circuitry, which provides isolation, amplification, and filtering functions to ensure input signal quality. The entire time synchronization calibration system is installed in an IP54-rated enclosure with electromagnetic shielding and heat dissipation design, adapting to the harsh environment of substations. The calibration system uses a software configuration tool to set parameters. This tool provides a graphical interface for setting the sampling rate, window length, and synchronization mode. System operating status is displayed via front panel indicator lights, with the light color indicating the data synchronization quality level. The time synchronization calibration module supports hot-swapping, allowing for online replacement in case of module failure without affecting continuous system operation. Data stream time synchronization accuracy testing employs a standard signal source injection method, and test results are recorded in the system log for maintenance and analysis.

[0028] Example 2: See Figure 4 In steady-state mode, the current waveform data is used to calculate the proportion of odd harmonics, which refers to the 3rd, 5th, and 7th harmonic components. The proportion of harmonic content is calculated by summing the percentage ratios of each odd harmonic amplitude to the fundamental amplitude after obtaining the spectrum through Fast Fourier Transform (FFT). The FFT point count is set to 1024, and the Hanning window is used to reduce spectral leakage in the spectrum analysis. The odd harmonic detection range covers the 3rd to 25th harmonics. In transient mode, the dielectric loss data is used to extract the polarization loss peak value. The polarization loss peak value is identified by a differential algorithm to pinpoint abrupt changes in the dielectric loss tangent. The differential algorithm calculates the first derivative of adjacent sampling points, with the peak value determined by the derivative crossing zero and the second derivative being negative. In overload mode, the vibration data is used to calculate the proportion of high-frequency energy, defined as the vibration energy in the 1kHz to 5kHz frequency band. The energy calculation uses the power spectral density integration method, and the power spectral density is estimated using the Welch method.

[0029] The three types of features are aggregated into a three-dimensional matrix according to a fixed time window of 10 seconds containing 10,000 data points. Each row vector of the matrix corresponds to a feature triplet at a sampling time. The column dimensions of the matrix are arranged sequentially as harmonic distortion rate, dielectric loss tangent, and vibration spectrum entropy. The running feature matrix is ​​stored in dual-port random access memory, with the memory address mapped to the matrix's row and column indices. The mode switching event detection module monitors the temporal changes of the feature matrix. The detection algorithm is based on a hidden Markov model to identify running state transitions, and the unit time threshold is dynamically updated based on the moving average of historical data. The dynamic alignment process adjusts the time axis of adjacent mode feature matrices through a dynamic time warping algorithm. The dynamic time warping algorithm constructs a cost matrix to store the Euclidean distance between the row vectors of two feature matrices and finds the curved path with the minimum cumulative cost through dynamic programming.

[0030] The harmonic distortion rate gradient is calculated using the central difference method, with the gradient value being the average of the harmonic distortion rate differences between five sampling points before and after the current moment. The dielectric loss tangent offset is measured using a structural similarity index, which compares the brightness, contrast, and structural information of the dielectric loss tangent sequences within two time windows. The vibration spectrum entropy difference is calculated based on the Jensen-Shannon divergence, which measures the statistical distance between two vibration spectrum distributions. The weighted sum of the three types of offsets is calculated using a linear weighting model, with the weighting coefficients preset according to the equipment operating characteristics: 0.4 for the harmonic distortion rate gradient, 0.35 for the dielectric loss tangent offset, and 0.25 for the vibration spectrum entropy difference.

[0031] The dynamic time warping algorithm employs DTW-Band constraints for curved path search, with the constraint bandwidth set to 10% of the time axis length to limit the path search range and improve computational efficiency. The row vectors of the feature matrix are normalized using a minimum-maximum scaling method to map eigenvalues ​​to the [0,1] interval. The inter-mode feature offsets are output to the decision module, with offset thresholds set according to equipment safety operating specifications. The dynamic alignment process of the feature matrix is ​​parallelized on the graphics processing unit (GPU), with multiple GPU cores simultaneously handling matrix alignment tasks across different time periods. The aligned feature matrix is ​​used for mode transition analysis, which visualizes the mode switching trajectory through principal component analysis (PCA) for dimensionality reduction.

[0032] The update cycle of the feature matrix is ​​synchronized with the data acquisition cycle. Each new sampling point triggers a rolling update of the matrix, removing the oldest row and appending the new row to the end of the matrix. Matrix data is managed using a circular buffer, with the buffer pointer automatically overwriting historical data cyclically. Harmonic distortion rate calculation includes a data preprocessing step, which uses digital filters to eliminate DC components and high-frequency noise. The filters are implemented using a cascaded high-pass filter with a cutoff frequency of 0.5Hz and a low-pass filter with a cutoff frequency of 2kHz. Polarization loss peak detection of the dielectric loss tangent includes a smoothing stage, which uses a moving average filter to suppress random fluctuations. Vibration spectrum entropy calculation is based on the probability distribution estimation of the power spectrum, which is obtained as a discrete probability mass function through spectral energy normalization.

[0033] The dimension of the cost matrix in the dynamic time warping algorithm is proportional to the number of rows in the two feature matrices. The cost matrix is ​​stored in static random access memory to improve access speed. Curved path backtracking is implemented through a path pointer matrix, which records the predecessor position of each grid point. The weighted sum calculation module uses fixed-point arithmetic to improve computational efficiency, with the fixed-point numbers in Q15 format to represent the fractional part. The hidden Markov model for mode switching event detection includes three hidden states corresponding to the steady-state mode, transient mode, and overload mode; the state transition probabilities are obtained through training on historical data. Threshold comparison of feature offsets is implemented using a hysteresis comparator circuit, with a hysteresis window preventing frequent switching near the threshold.

[0034] The persistent storage of the feature matrix utilizes non-volatile memory, with a structured binary file format including a file header recording matrix dimensions and timestamp information. Real-time performance of the dynamic alignment process is ensured through a pipelined architecture, with pipeline stages including feature extraction, matrix construction, path calculation, and offset calculation. The graphics processor is configured with 256 threads, each handling the distance calculation for a pair of feature matrix rows. Visualization of the feature matrix is ​​achieved through a human-machine interface (HMI), displaying the pattern clustering distribution in the 3D feature space. The time series of feature offsets between patterns is recorded in the system log for long-term trend analysis of equipment degradation. The parameters of the dynamic alignment algorithm can be adjusted via a configuration interface, including curved path constraint bandwidth, weighting coefficients, and offset thresholds. The entire feature matrix generation and dynamic alignment system is integrated into a field monitoring device, which communicates with the upper-level monitoring center via industrial Ethernet. The device chassis meets IP65 protection standards, adapting to the high-temperature and high-humidity environment of substations. A self-test process is executed upon system startup, verifying the normal access to the feature matrix memory and the correctness of the dynamic time warping algorithm calculations. During operation, the utilization rate of computing resources is monitored in real time. When the utilization rate exceeds the limit, a load balancing strategy is triggered to adjust the allocation of computing tasks.

[0035] Example 3: The matching degree calculation between the insulation aging stage in the historical fault case library and the current dielectric loss data adopts the dynamic time warping algorithm. The dynamic time warping algorithm obtains the matching degree value by comparing the morphological similarity between the current dielectric loss angle tangent value sequence and the typical aging mode sequence in the case library. The preset threshold is set to 0.8. When the matching degree is higher than 0.8, the low-frequency drift component separation process of the dielectric loss angle tangent is triggered. The low-frequency drift component separation is implemented using a Butterworth low-pass filter with a cutoff frequency of 0.1 Hz. The high-frequency residual component is obtained by subtracting the filtered signal from the original signal. When the matching degree is lower than 0.8, the high-frequency vibration energy concentration band is extracted more effectively. The high-frequency vibration energy concentration band is limited to 2kHz to 5kHz. The enhanced extraction process adopts a method combining a bandpass filter and Hilbert-Huang transform. First, the target frequency band signal is extracted through the bandpass filter, and then the instantaneous frequency and amplitude are calculated by applying the Hilbert transform.

[0036] After decomposition of the degradation characteristic components, coupling compensation correction is performed. This correction compensates for the temperature rise deviation caused by harmonic distortion rate based on the current thermal effect coefficient. The current thermal effect coefficient is calculated using the temperature coefficient of resistance of the conductor material and heat dissipation conditions. The temperature rise deviation compensation uses the following formula:

[0037] in, This represents the compensated temperature rise deviation, expressed in degrees Celsius. This represents the coefficient of thermal effect of electric current, and its unit is degrees Celsius per ampere square. This indicates the effective value of the current, and the unit is ampere. This indicates the current measured value of harmonic distortion rate; This represents the reference harmonic distortion rate value; This represents the thermal resistance coefficient, with units of degrees Celsius per watt.

[0038] The mechanical vibration transfer function (MJF) is used to correct the amplitude-frequency response of the vibration spectrum entropy. The MJF is obtained through experimental measurements of the equipment structure's vibration response characteristics at different frequencies. The amplitude-frequency response correction employs an inverse filter design to compensate for the influence of the mechanical transmission path. During the generation of the coupled compensation correction results, the temperature rise deviation compensation value is superimposed on the harmonic distortion rate characteristic value. The amplitude-frequency response correction value is adjusted for the amplitude and phase of the vibration spectrum entropy value through complex multiplication. The compensated deteriorated characteristic components are merged into a new eigenvector. The eigenvector dimension maintains three dimensions, corresponding to the corrected harmonic distortion rate, dielectric loss tangent, and vibration spectrum entropy value.

[0039] The historical fault case database stores typical data patterns for various insulation aging stages, including initial aging, intermediate aging, and final aging stages. Each stage includes a typical curve showing the change of the dielectric loss tangent over time. The matching degree calculation uses a dynamic time warping algorithm to determine the minimum bending path cost between two time series; the bending path cost is then normalized and converted into a similarity score. During low-frequency drift component separation, the Butterworth low-pass filter is set to order 4 to provide sufficient stopband attenuation, and a forward-backward filtering method is used to eliminate phase distortion. For high-frequency vibration energy enhancement extraction, the bandpass filter is designed as a finite-length unit impulse response filter with an order of 100 and a ripple coefficient controlled below 0.01.

[0040] The current-thermal effect coefficient, corrected for coupling compensation, was determined through materials testing. The tests measured the conductor's resistance at different temperatures, and the thermal effect coefficient was calculated as the average of the curve slopes. The thermal resistance coefficient was calculated based on the thermal design parameters of the equipment structure, including material thermal conductivity, surface area, and heat dissipation conditions. The mechanical vibration transfer function was measured using the impact hammer method. This method involves exciting the equipment structure with a hammer while simultaneously measuring the input force and response acceleration. The transfer function was calculated as the frequency domain ratio of the output acceleration to the input force. The inverse filter design was based on the Wiener filtering principle. Wiener filtering calculates an estimate of the inverse transfer function in the frequency domain while considering noise suppression to avoid excessive amplification of high-frequency noise.

[0041] The decomposition and compensation of degraded feature components are performed cyclically, with decomposition parameters updated after each operating mode cycle. The decomposition parameter updates are based on feature statistics within the most recent time window, including the mean, variance, and autocorrelation function of the feature values. The coupling compensation correction result is smoothed using a Kalman filter, with the Kalman filter state variables set as the three components of the feature vector and the observation matrix set as the identity matrix. The corrected feature vector is output to the anomaly tracing module, which calculates the deviation of the feature vector from the normal state based on Mahalanobis distance. The entire degraded feature decomposition and coupling compensation correction process is implemented in a digital signal processor (DSP), which is equipped with a dedicated floating-point unit to accelerate the filter and compensation calculations. The decomposition algorithm adopts a modular design, with modules including data preprocessing, matching degree calculation, feature decomposition, coupling compensation, and result verification. The data preprocessing module standardizes the input features using z-score standardization to eliminate the influence of dimensions. The matching degree calculation module employs a parallel processing architecture, simultaneously calculating the matching degree between the current data and multiple aging mode templates in the case library. The feature decomposition module automatically selects a decomposition strategy based on the matching degree results. When the matching degree is high, low-frequency drift decomposition is enabled, and when the matching degree is low, high-frequency vibration enhancement decomposition is enabled.

[0042] The coupling compensation and correction module includes a temperature compensation unit and a vibration compensation unit. The temperature compensation unit calculates the temperature rise effect caused by harmonic distortion rate in real time, while the vibration compensation unit adjusts the vibration spectrum entropy value according to the mechanical characteristics of the equipment. Compensation parameters are stored in non-volatile memory, supporting online updates and calibration. The result verification module judges the correction effect by comparing the statistical characteristics of eigenvectors before and after compensation. Verification indicators include the uniformity of eigenvalue distribution and correlation coefficient. During system operation, the system monitors the usage of computing resources in real time. When the computing load is too high, it automatically adjusts the computational accuracy of the filter and compensation algorithm to balance performance and resource consumption. The degraded feature decomposition and coupling compensation correction system periodically performs a self-calibration process, which verifies the normal functioning of each module by injecting standard test signals. Calibration results are recorded in the system log for fault diagnosis and maintenance analysis. The entire system adopts a redundant design, with backup units for critical modules. In the event of a primary unit failure, it automatically switches to the backup unit to ensure continuous operation.

[0043] See Figure 5 This paper demonstrates the feature decomposition and coupled compensation analysis process of insulation dielectric loss data, corresponding to the insulation aging diagnosis and compensation correction algorithms. The figure shows the fluctuation characteristics of the original dielectric loss data. The low-frequency drift component separated by the Butterworth low-pass filter reflects the long-term aging trend of the insulation material, while the high-frequency residual component contains instantaneous anomalies during equipment operation. The compensated dielectric loss curve shows the optimization results after temperature rise deviation compensation and amplitude-frequency characteristic correction. The matching degree curve shows the similarity between the current dielectric loss data and typical aging patterns in the historical fault case library; high matching degree areas identify insulation aging development stages that require special attention. The feature decomposition module automatically selects an appropriate decomposition strategy based on the matching degree results, realizing personalized state assessment. The coupled compensation correction process comprehensively considers the impact of current thermal effects and mechanical vibration transmission characteristics on the monitoring data, eliminating inherent biases in the measurement system through inverse filter design and Wiener filtering principles. The analysis information area provides key statistical indicators of feature decomposition for evaluating the decomposition effect and the effectiveness of the compensation algorithm.

[0044] Example 4: Impedance Connection Diagram Construction of Power Supply and Distribution Network Topology. Based on digital modeling of substation primary system wiring diagrams, nodes in the impedance connection diagram represent specific equipment such as transformers, circuit breakers, disconnectors, current transformers, and voltage transformers. Edges represent the connection impedance values ​​and fault propagation weights between equipment. Impedance values ​​are calculated using a power system short-circuit calculation program, taking into account equipment parameters, connection line specifications, and system operating modes. Fault propagation weights are set based on historical equipment fault statistics and reliability analysis. The impedance connection diagram is stored as a graph data structure, using an adjacency list to record node connection relationships. Each node attribute includes equipment number, equipment type, impedance value, and fault propagation weight.

[0045] Abnormal signal source device node location is achieved by identifying abnormal harmonic distortion rate, dielectric loss tangent, or vibration spectrum entropy value from the coupling compensation correction results. Anomaly identification employs a statistical process control-based method, with control limits set to a range of plus or minus three standard deviations of the historical mean of the characteristic values. Corresponding nodes in the matched impedance connection diagram are identified using a device identifier mapping table, which establishes the correspondence between characteristic measurement points and device nodes in the topology diagram. A modified Dijkstra algorithm is used to traverse the energy transfer direction along the impedance connection diagram, starting at the device node with the identified abnormal characteristic value and proceeding in the opposite direction of power flow from the load side to the power supply side.

[0046] The equipment maintenance sequence generates statistical data on the number of times the harmonic distortion rate, dielectric loss tangent, and vibration spectral entropy of the equipment nodes exceed the standard. The number of exceedances is based on the cumulative number of times the characteristic value exceeds the safety threshold within a sliding time window. The exceedance amplitude is calculated as the absolute difference between the characteristic value and the safety threshold. The exceedance frequency is the frequency of exceedance events per unit time. The weighted sum of the three exceedance indicators is calculated using a linear weighted model, with weight coefficients set according to the importance of the equipment in the power supply and distribution network and the severity of the fault consequences. The equipment maintenance priority sequence is sorted in descending order of the weighted sum value, generating a structured data sequence containing the equipment number, maintenance priority score, and exceedance details.

[0047] The impedance connection diagram is constructed using a hierarchical modeling approach. This modeling divides the substation power distribution network into different layers according to voltage levels, with these layers connected via transformer nodes. Fault propagation weights are assigned based on equipment failure rate statistics and operation and maintenance records, and these weights are updated periodically to reflect changes in equipment status. The statistical process control chart for anomaly identification uses an exponentially weighted moving average control chart, which exhibits high sensitivity to small drifts in eigenvalues. An equipment identifier mapping table maintains the correspondence between equipment and measurement points, and this table supports dynamic updates to adapt to changes in system operation.

[0048] The reverse traversal algorithm performs a breadth-first search on the impedance connection graph, expanding layer by layer towards the power supply side from the anomalous node. During the traversal, the impedance values ​​and fault propagation weights along the paths are accumulated. The maintenance sequence generation module monitors the feature value data stream in real time. The data stream is processed using a sliding window mechanism, with configurable window lengths to adapt to different monitoring needs. The weighting coefficients in the weighted summation calculation are determined using the analytic hierarchy process (AHP), which constructs a judgment matrix to compare the relative importance of each exceeding indicator. The output format of the maintenance priority sequence supports multiple industry standard protocols, facilitating integration with asset management systems.

[0049] The node attributes of the impedance connection diagram include equipment static parameters and dynamic operating data. Static parameters include equipment model, commissioning date, rated parameters, etc., while dynamic data includes real-time measured values, historical fault records, and maintenance history. The anomaly identification module adopts a multivariate statistical process control method, which simultaneously monitors the correlation of multiple characteristic variables to improve the accuracy of anomaly detection. The reverse traversal process considers equipment connection methods and operating states. Connection methods include single busbar segmentation, double busbar with bypass, and other wiring configurations. Operating states include the impact of equipment switching, load transfer, and other operations.

[0050] Referring to Table 1, the maintenance sequence generation cycle can be configured according to operational needs, typically adjustable from 1 hour to 24 hours. Weighting coefficients are assigned based on the severity of equipment failure consequences, including power loss range, repair time, and economic impact. The maintenance sequence visualization interface displays a bar chart of equipment priority ranking and a radar chart of out-of-target indicators to assist maintenance personnel in decision-making. The system supports version management of maintenance sequences, recording sequence change history and reasons for modifications.

[0051] Table 1: Equipment Maintenance Priority Calculation Table

[0052] The impedance connection diagram update mechanism is synchronized with the substation monitoring system. The monitoring system automatically triggers an update when it detects equipment commissioning / decommissioning or changes in operating mode. Anomaly identification thresholds are set differently based on equipment type, taking into account the operating characteristics and tolerance capabilities of different devices. A reverse traversal algorithm optimizes the search strategy, prioritizing paths with high fault propagation weights to improve location efficiency. The maintenance sequence generation module integrates an early warning function, which proactively issues an alarm signal when the weighted total score of an equipment exceeds a set threshold.

[0053] The equipment maintenance priority calculation process employs standardized scoring rules, which normalize out-of-limit indicators of different dimensions to the same scale for comparison. Before releasing a maintenance sequence, a consistency check is performed, verifying the integrity of equipment data and the rationality of the calculation process. System maintenance operations are recorded in the audit log, which records the generation time of the maintenance sequence, calculation parameters, and operator information. The entire anomaly localization and maintenance sequence generation system adopts a distributed architecture, allocating impedance calculation, anomaly identification, and sequence generation tasks to different computing nodes for parallel processing. System performance monitoring tracks computing resource usage in real time, including memory usage, CPU utilization, and network throughput. System operation reports are generated periodically, and these reports statistically analyze anomaly localization accuracy and maintenance sequence effectiveness indicators for continuous optimization of algorithm parameters.

[0054] Example 5: The data integrity verification system is built on a layered verification architecture, which includes three layers: physical layer verification, data link layer verification, and application layer verification. Physical layer verification monitors signal quality through hardware circuits. These hardware circuits use high-speed comparators to monitor the voltage amplitude and signal-to-noise ratio of the input signal in real time. When the signal voltage falls below a threshold or the signal-to-noise ratio deteriorates, analog front-end recalibration is triggered. Data link layer verification focuses on verifying the integrity of the data transmission process. The verification method employs a dual verification mechanism combining cyclic redundancy check (CRC) and Hamming code. CRC uses a CRC-32 polynomial to calculate a 32-bit checksum, while Hamming code provides unit error correction and two-bit error detection capabilities. Application layer verification checks the logical consistency of business data, including verification of data timestamp continuity, numerical range rationality, and data correlation.

[0055] The synchronized data stream enters the verification pipeline, which employs a three-stage pipeline structure to improve processing throughput. The first stage pipeline performs frame structure verification, confirming the correctness of the data frame header, frame tail flags, and frame length field. The frame header flag is a fixed four-byte pattern of 0xAA55AA55, and the frame length field must match the actual data payload length. The second stage pipeline performs timestamp continuity verification. The timestamp continuity verification algorithm maintains a sliding time window, configurable from 8 to 64 data points. The algorithm calculates the variance of the timestamp intervals of consecutive data points; when the variance exceeds a threshold, timestamp anomalies are marked. The third stage pipeline performs data validity verification. Data validity verification sets the valid range of data values ​​based on the equipment type: the valid range for current data is 0-5000 amperes, the valid range for dielectric loss tangent is 0.001-0.1, and the valid range for vibration data is 0-10g.

[0056] The data integrity verification system's hardware platform adopts a heterogeneous architecture of FPGA + CPU. The FPGA is responsible for high real-time verification tasks, while the CPU handles complex verification logic. The FPGA chip is a Xilinx Kintex-7. The XC7K325T features 32 parallel verification engines, each processing an independent data stream. It uses an Intel Xeon E5-2680 processor to run verification management software and exception handling. Data storage utilizes dual-port DDR3 memory with a capacity of 8GB, achieving a storage bandwidth of 12.8GB / s.

[0057] When the verification system detects data anomalies, the anomaly handling process executes according to a predetermined strategy. Minor anomalies trigger a data repair procedure, which uses linear interpolation or spline interpolation methods to repair the abnormal data points. Moderate anomalies initiate a data re-acquisition process; the data re-acquisition command is sent to the acquisition terminal via industrial Ethernet, and the acquisition terminal responds to the re-acquisition request within 100 milliseconds. Severe anomalies trigger a system alarm and activate a backup acquisition channel, which uses a completely independent hardware path to ensure data continuity.

[0058] The data integrity verification system includes a self-calibration function, which periodically injects standard test signals to verify the accuracy of the verification. The standard test signals are generated by a high-precision signal source with an accuracy of 0.01%, and include various boundary conditions and anomaly modes. The verification system records the results of each self-calibration, establishes a verification accuracy trend chart, and triggers automatic adjustment of system parameters when the accuracy drift exceeds a threshold.

[0059] The verification results are stored in a structured format, using a time-series database. Each verification result includes fields such as timestamp, data source identifier, anomaly type, and handling measures. The database employs a sharded storage strategy, distributing data across different storage nodes according to time ranges. The verification system provides rich query interfaces, supporting queries by time range, anomaly level, device number, and other dimensions.

[0060] The data integrity verification system interfaces with other modules of the monitoring system using standard communication protocols. Real-time data interfaces use the IEEE 1588 precise time protocol for synchronization, configuration interfaces employ a RESTful API design, and alarm interfaces support the SNMP protocol. The system supports remote configuration and management; administrators can adjust verification parameters and thresholds in real time via a web interface. The abnormal data handling process includes a data isolation mechanism; data marked as abnormal is redirected to an isolation zone before entering subsequent processing. Data in the isolation zone, after manual review or advanced algorithm verification, is confirmed as valid and can be reinjected into the normal processing flow. This mechanism prevents abnormal data from contaminating feature extraction and status assessment results. The data integrity verification system's performance monitoring tracks key indicators in real time, including verification latency, throughput, and error detection rate. Monitoring data is visualized through a dashboard and periodic performance reports are generated. The system supports an alarm escalation mechanism; when a significant anomaly persists, the alarm level is automatically escalated, and higher-level management personnel are notified. The verification system's reliability design incorporates multiple protection measures. The power system uses dual power supplies with UPS backup, the storage system is configured with a RAID-6 disk array, and the network system uses a dual-ring network topology. All critical system modules have hot backups, with a master-slave switchover time of less than 50 milliseconds. All verification operations are logged in detail, meeting auditing and traceability requirements.

[0061] In practice, the data integrity verification system needs to work closely with the data acquisition system. The acquisition system provides data quality metadata, including information such as signal strength, ambient temperature, and device operating status. The verification system integrates this metadata to perform a more accurate data quality assessment. This cross-system collaboration significantly improves the accuracy of data integrity judgments.

[0062] Optimizing verification parameters is an ongoing process. The system analyzes historical verification data using machine learning algorithms to automatically adjust verification thresholds and strategies. The machine learning model is periodically retrained to adapt to the effects of equipment aging and environmental changes. This adaptive verification strategy enables the system to maintain optimal verification performance over the long term. The deployment of the data integrity verification system considers on-site environmental factors. In areas with strong electromagnetic interference, the verification system is installed in a shielded cabinet with a shielding effectiveness of 60dB. In high-temperature environments, the system is equipped with forced air cooling to ensure that the chip junction temperature does not exceed the rated value. In humid environments, the circuit boards are coated with conformal coating for protection, and the connectors are waterproof. Maintenance of the verification system supports both remote and local methods. Remote maintenance is performed via a secure VPN connection, while local maintenance provides a touchscreen interface. The system has self-diagnostic capabilities, enabling it to locate faults at the board level. Maintenance personnel can quickly replace faulty modules based on system prompts, reducing system downtime.

[0063] Performance testing of the data integrity verification system is conducted comprehensively before deployment. Testing includes extreme load testing, anomaly injection testing, and long-term stability testing. Detailed reports are generated from the test results and serve as the basis for system acceptance. Only systems that pass all tests can be put into formal operation. Data quality reports generated by the verification system are regularly sent to relevant management personnel. These reports include data integrity rate, anomaly distribution, and trend analysis. These reports are not only used to monitor system operation status but also provide data support for equipment maintenance decisions. High-quality data is the foundation for subsequent intelligent analysis.

[0064] During system upgrades and expansions, the data integrity verification module employs a modular design for easy upgrades. New verification algorithms can be deployed via software updates, while hardware expansion is achieved through a standard bus interface. This design ensures the system can adapt to the needs of future technological developments.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for power supply and distribution safety monitoring in substations, characterized in that, include: Collect current waveform data, insulation dielectric loss data, and switchgear mechanical vibration data of power supply and distribution equipment in the substation, and perform time synchronization calibration on the three types of data. Based on the calibrated data, the power supply and distribution operation modes are divided into steady-state mode, transient mode and overload mode. The harmonic distortion rate, dielectric loss tangent and vibration spectrum entropy of the current waveform under each mode are extracted to generate the operation feature matrix. Detect running mode switching events. If the number of mode switching times per unit time exceeds the dynamic threshold, perform dynamic alignment on the running feature matrices of adjacent modes and calculate the feature offset between modes. Based on the matching degree between the insulation aging stage in the historical fault case library and the current dielectric loss data, the deterioration feature components in the operation feature matrix are decomposed. By combining the cross-influence relationship between the electrical and mechanical parameters of the equipment, coupled compensation correction is performed on the degradation characteristic components; Based on the corrected degradation feature components, the abnormal signal source device node is traced backward along the power supply and distribution network topology, and the device maintenance sequence is output.

2. The substation power supply and distribution safety monitoring method according to claim 1, characterized in that, The time synchronization calibration of the three types of data includes: The sampling start point is aligned using the zero-crossing detection method for current waveform data; the insulation dielectric loss data is interpolated in segments according to the power frequency period; and the fundamental frequency phase is extracted from mechanical vibration data through short-time Fourier transform. A sliding time window is used to compensate for the time scale deviation of the three types of data, generating a synchronized and calibrated data stream.

3. The substation power supply and distribution safety monitoring method according to claim 1, characterized in that, The extraction of harmonic distortion rate, dielectric loss tangent, and vibration spectrum entropy of the current waveform under each mode to generate the operating feature matrix includes: The proportion of odd harmonic content is calculated from the current waveform data in steady-state mode, the peak value of polarization loss is extracted from the dielectric loss data in transient mode, and the proportion of high-frequency energy is calculated from the vibration data in overload mode. The three types of features are aggregated into a three-dimensional matrix according to the time window. The matrix dimensions correspond to the harmonic distortion rate, the dielectric loss tangent, and the vibration spectrum entropy.

4. The substation power supply and distribution safety monitoring method according to claim 3, characterized in that, The dynamic alignment of the feature matrices of adjacent patterns includes: The time axis of the feature matrix of adjacent modes is adjusted by dynamic time warping algorithm, and the gradient of harmonic distortion rate change, the offset of dielectric loss tangent and the difference of vibration spectrum entropy are calculated. The weighted sum of the three types of offsets is output as the inter-mode feature offset.

5. The substation power supply and distribution safety monitoring method according to claim 1, characterized in that, The degraded feature components in the decomposed feature matrix include: When the matching degree between dielectric loss data and insulation aging stage is higher than the preset threshold, the low-frequency drift component of dielectric loss tangent is separated from the operating feature matrix. When the matching degree is lower than the preset threshold, the components of the high-frequency vibration energy concentration band are extracted more effectively based on the correlation between the vibration spectrum entropy value and the wear degree of the switching mechanism.

6. The substation power supply and distribution safety monitoring method according to claim 1, characterized in that, The coupling compensation correction performed on the degraded feature components includes: Compensation is provided for temperature rise deviation caused by harmonic distortion rate based on the current thermal effect coefficient. The amplitude-frequency characteristic of the vibration spectrum entropy value is corrected based on the mechanical vibration transfer function; The temperature rise deviation after compensation and the amplitude-frequency characteristic after correction are combined to generate the coupling compensation correction result.

7. The substation power supply and distribution safety monitoring method according to claim 1, characterized in that, The equipment nodes for tracing the abnormal signal source along the power supply and distribution network topology include: Construct an impedance connection diagram of the power supply and distribution network topology, where nodes represent devices and edges represent impedance values ​​and fault propagation weights; Identify abnormal harmonic distortion rate, dielectric loss tangent or vibration spectrum entropy from the coupling compensation correction results, and match the corresponding nodes in the impedance connection diagram. Traverse the energy transfer in the reverse direction along the impedance connection diagram to locate the device node of the abnormal signal source.

8. The substation power supply and distribution safety monitoring method according to claim 7, characterized in that, The output device maintenance sequence includes: The number of times the harmonic distortion rate of the abnormal signal source equipment node exceeds the standard, the magnitude of the dielectric loss tangent value exceeding the standard, and the frequency of the vibration spectrum entropy value exceeding the standard are statistically analyzed. The equipment maintenance priority sequence is generated by sorting the three categories of exceedance indicators in descending order by weighted sum.

9. The substation power supply and distribution safety monitoring method according to claim 1, characterized in that, Also includes: Perform an integrity check on the synchronized data stream; if the check fails, trigger a data re-acquisition command. Mark the timestamp abnormal segments in the validated data stream for exclusion in the subsequent feature extraction stage.

10. A power supply and distribution safety monitoring system for a substation, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the substation power supply and distribution safety monitoring method according to any one of claims 1 to 9.

Images